2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Alan Cox, <A.Cox@swansea.ac.uk>
18 * Alan Cox : Numerous verify_area() problems
19 * Alan Cox : Connecting on a connecting socket
20 * now returns an error for tcp.
21 * Alan Cox : sock->protocol is set correctly.
22 * and is not sometimes left as 0.
23 * Alan Cox : connect handles icmp errors on a
24 * connect properly. Unfortunately there
25 * is a restart syscall nasty there. I
26 * can't match BSD without hacking the C
27 * library. Ideas urgently sought!
28 * Alan Cox : Disallow bind() to addresses that are
29 * not ours - especially broadcast ones!!
30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
32 * instead they leave that for the DESTROY timer.
33 * Alan Cox : Clean up error flag in accept
34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
35 * was buggy. Put a remove_sock() in the handler
36 * for memory when we hit 0. Also altered the timer
37 * code. The ACK stuff can wait and needs major
39 * Alan Cox : Fixed TCP ack bug, removed remove sock
40 * and fixed timer/inet_bh race.
41 * Alan Cox : Added zapped flag for TCP
42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
47 * Rick Sladkey : Relaxed UDP rules for matching packets.
48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
49 * Pauline Middelink : identd support
50 * Alan Cox : Fixed connect() taking signals I think.
51 * Alan Cox : SO_LINGER supported
52 * Alan Cox : Error reporting fixes
53 * Anonymous : inet_create tidied up (sk->reuse setting)
54 * Alan Cox : inet sockets don't set sk->type!
55 * Alan Cox : Split socket option code
56 * Alan Cox : Callbacks
57 * Alan Cox : Nagle flag for Charles & Johannes stuff
58 * Alex : Removed restriction on inet fioctl
59 * Alan Cox : Splitting INET from NET core
60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
62 * Alan Cox : Split IP from generic code
63 * Alan Cox : New kfree_skbmem()
64 * Alan Cox : Make SO_DEBUG superuser only.
65 * Alan Cox : Allow anyone to clear SO_DEBUG
67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
68 * Alan Cox : Allocator for a socket is settable.
69 * Alan Cox : SO_ERROR includes soft errors.
70 * Alan Cox : Allow NULL arguments on some SO_ opts
71 * Alan Cox : Generic socket allocation to make hooks
72 * easier (suggested by Craig Metz).
73 * Michael Pall : SO_ERROR returns positive errno again
74 * Steve Whitehouse: Added default destructor to free
75 * protocol private data.
76 * Steve Whitehouse: Added various other default routines
77 * common to several socket families.
78 * Chris Evans : Call suser() check last on F_SETOWN
79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
81 * Andi Kleen : Fix write_space callback
82 * Chris Evans : Security fixes - signedness again
83 * Arnaldo C. Melo : cleanups, use skb_queue_purge
88 * This program is free software; you can redistribute it and/or
89 * modify it under the terms of the GNU General Public License
90 * as published by the Free Software Foundation; either version
91 * 2 of the License, or (at your option) any later version.
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
116 #include <asm/uaccess.h>
117 #include <asm/system.h>
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/net_namespace.h>
123 #include <net/request_sock.h>
124 #include <net/sock.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
128 #include <linux/filter.h>
135 * Each address family might have different locking rules, so we have
136 * one slock key per address family:
138 static struct lock_class_key af_family_keys[AF_MAX];
139 static struct lock_class_key af_family_slock_keys[AF_MAX];
141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
143 * Make lock validator output more readable. (we pre-construct these
144 * strings build-time, so that runtime initialization of socket
147 static const char *af_family_key_strings[AF_MAX+1] = {
148 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
149 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
150 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
151 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
152 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
153 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
154 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
155 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
157 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" ,
158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
161 static const char *af_family_slock_key_strings[AF_MAX+1] = {
162 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
163 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
164 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
165 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
166 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
167 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
168 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
169 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
170 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
171 "slock-27" , "slock-28" , "slock-29" ,
172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
173 "slock-AF_RXRPC" , "slock-AF_MAX"
175 static const char *af_family_clock_key_strings[AF_MAX+1] = {
176 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
177 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
178 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
179 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
180 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
181 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
182 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
183 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
184 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
185 "clock-27" , "clock-28" , "clock-29" ,
186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
187 "clock-AF_RXRPC" , "clock-AF_MAX"
192 * sk_callback_lock locking rules are per-address-family,
193 * so split the lock classes by using a per-AF key:
195 static struct lock_class_key af_callback_keys[AF_MAX];
197 /* Take into consideration the size of the struct sk_buff overhead in the
198 * determination of these values, since that is non-constant across
199 * platforms. This makes socket queueing behavior and performance
200 * not depend upon such differences.
202 #define _SK_MEM_PACKETS 256
203 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
204 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
207 /* Run time adjustable parameters. */
208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
213 /* Maximal space eaten by iovec or ancilliary data plus some space */
214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
220 if (optlen < sizeof(tv))
222 if (copy_from_user(&tv, optval, sizeof(tv)))
224 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
228 static int warned __read_mostly;
231 if (warned < 10 && net_ratelimit())
233 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
234 "tries to set negative timeout\n",
235 current->comm, current->pid);
238 *timeo_p = MAX_SCHEDULE_TIMEOUT;
239 if (tv.tv_sec == 0 && tv.tv_usec == 0)
241 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
242 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
246 static void sock_warn_obsolete_bsdism(const char *name)
249 static char warncomm[TASK_COMM_LEN];
250 if (strcmp(warncomm, current->comm) && warned < 5) {
251 strcpy(warncomm, current->comm);
252 printk(KERN_WARNING "process `%s' is using obsolete "
253 "%s SO_BSDCOMPAT\n", warncomm, name);
258 static void sock_disable_timestamp(struct sock *sk)
260 if (sock_flag(sk, SOCK_TIMESTAMP)) {
261 sock_reset_flag(sk, SOCK_TIMESTAMP);
262 net_disable_timestamp();
267 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
272 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
273 number of warnings when compiling with -W --ANK
275 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
276 (unsigned)sk->sk_rcvbuf) {
281 err = sk_filter(sk, skb);
286 skb_set_owner_r(skb, sk);
288 /* Cache the SKB length before we tack it onto the receive
289 * queue. Once it is added it no longer belongs to us and
290 * may be freed by other threads of control pulling packets
295 skb_queue_tail(&sk->sk_receive_queue, skb);
297 if (!sock_flag(sk, SOCK_DEAD))
298 sk->sk_data_ready(sk, skb_len);
302 EXPORT_SYMBOL(sock_queue_rcv_skb);
304 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
306 int rc = NET_RX_SUCCESS;
308 if (sk_filter(sk, skb))
309 goto discard_and_relse;
314 bh_lock_sock_nested(sk);
317 if (!sock_owned_by_user(sk)) {
319 * trylock + unlock semantics:
321 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
323 rc = sk->sk_backlog_rcv(sk, skb);
325 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
327 sk_add_backlog(sk, skb);
336 EXPORT_SYMBOL(sk_receive_skb);
338 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
340 struct dst_entry *dst = sk->sk_dst_cache;
342 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
343 sk->sk_dst_cache = NULL;
350 EXPORT_SYMBOL(__sk_dst_check);
352 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
354 struct dst_entry *dst = sk_dst_get(sk);
356 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
364 EXPORT_SYMBOL(sk_dst_check);
366 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
368 int ret = -ENOPROTOOPT;
369 #ifdef CONFIG_NETDEVICES
370 char devname[IFNAMSIZ];
375 if (!capable(CAP_NET_RAW))
382 /* Bind this socket to a particular device like "eth0",
383 * as specified in the passed interface name. If the
384 * name is "" or the option length is zero the socket
387 if (optlen > IFNAMSIZ - 1)
388 optlen = IFNAMSIZ - 1;
389 memset(devname, 0, sizeof(devname));
392 if (copy_from_user(devname, optval, optlen))
395 if (devname[0] == '\0') {
398 struct net_device *dev = dev_get_by_name(devname);
404 index = dev->ifindex;
409 sk->sk_bound_dev_if = index;
422 * This is meant for all protocols to use and covers goings on
423 * at the socket level. Everything here is generic.
426 int sock_setsockopt(struct socket *sock, int level, int optname,
427 char __user *optval, int optlen)
429 struct sock *sk=sock->sk;
430 struct sk_filter *filter;
437 * Options without arguments
440 #ifdef SO_DONTLINGER /* Compatibility item... */
441 if (optname == SO_DONTLINGER) {
443 sock_reset_flag(sk, SOCK_LINGER);
449 if (optname == SO_BINDTODEVICE)
450 return sock_bindtodevice(sk, optval, optlen);
452 if (optlen < sizeof(int))
455 if (get_user(val, (int __user *)optval))
464 if (val && !capable(CAP_NET_ADMIN)) {
468 sock_set_flag(sk, SOCK_DBG);
470 sock_reset_flag(sk, SOCK_DBG);
473 sk->sk_reuse = valbool;
481 sock_set_flag(sk, SOCK_LOCALROUTE);
483 sock_reset_flag(sk, SOCK_LOCALROUTE);
486 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
489 /* Don't error on this BSD doesn't and if you think
490 about it this is right. Otherwise apps have to
491 play 'guess the biggest size' games. RCVBUF/SNDBUF
492 are treated in BSD as hints */
494 if (val > sysctl_wmem_max)
495 val = sysctl_wmem_max;
497 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
498 if ((val * 2) < SOCK_MIN_SNDBUF)
499 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
501 sk->sk_sndbuf = val * 2;
504 * Wake up sending tasks if we
507 sk->sk_write_space(sk);
511 if (!capable(CAP_NET_ADMIN)) {
518 /* Don't error on this BSD doesn't and if you think
519 about it this is right. Otherwise apps have to
520 play 'guess the biggest size' games. RCVBUF/SNDBUF
521 are treated in BSD as hints */
523 if (val > sysctl_rmem_max)
524 val = sysctl_rmem_max;
526 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
528 * We double it on the way in to account for
529 * "struct sk_buff" etc. overhead. Applications
530 * assume that the SO_RCVBUF setting they make will
531 * allow that much actual data to be received on that
534 * Applications are unaware that "struct sk_buff" and
535 * other overheads allocate from the receive buffer
536 * during socket buffer allocation.
538 * And after considering the possible alternatives,
539 * returning the value we actually used in getsockopt
540 * is the most desirable behavior.
542 if ((val * 2) < SOCK_MIN_RCVBUF)
543 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
545 sk->sk_rcvbuf = val * 2;
549 if (!capable(CAP_NET_ADMIN)) {
557 if (sk->sk_protocol == IPPROTO_TCP)
558 tcp_set_keepalive(sk, valbool);
560 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
564 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
568 sk->sk_no_check = valbool;
572 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
573 sk->sk_priority = val;
579 if (optlen < sizeof(ling)) {
580 ret = -EINVAL; /* 1003.1g */
583 if (copy_from_user(&ling,optval,sizeof(ling))) {
588 sock_reset_flag(sk, SOCK_LINGER);
590 #if (BITS_PER_LONG == 32)
591 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
592 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
595 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
596 sock_set_flag(sk, SOCK_LINGER);
601 sock_warn_obsolete_bsdism("setsockopt");
606 set_bit(SOCK_PASSCRED, &sock->flags);
608 clear_bit(SOCK_PASSCRED, &sock->flags);
614 if (optname == SO_TIMESTAMP)
615 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
617 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
618 sock_set_flag(sk, SOCK_RCVTSTAMP);
619 sock_enable_timestamp(sk);
621 sock_reset_flag(sk, SOCK_RCVTSTAMP);
622 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
629 sk->sk_rcvlowat = val ? : 1;
633 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
637 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
640 case SO_ATTACH_FILTER:
642 if (optlen == sizeof(struct sock_fprog)) {
643 struct sock_fprog fprog;
646 if (copy_from_user(&fprog, optval, sizeof(fprog)))
649 ret = sk_attach_filter(&fprog, sk);
653 case SO_DETACH_FILTER:
655 filter = rcu_dereference(sk->sk_filter);
657 rcu_assign_pointer(sk->sk_filter, NULL);
658 sk_filter_release(sk, filter);
659 rcu_read_unlock_bh();
662 rcu_read_unlock_bh();
668 set_bit(SOCK_PASSSEC, &sock->flags);
670 clear_bit(SOCK_PASSSEC, &sock->flags);
673 /* We implement the SO_SNDLOWAT etc to
674 not be settable (1003.1g 5.3) */
684 int sock_getsockopt(struct socket *sock, int level, int optname,
685 char __user *optval, int __user *optlen)
687 struct sock *sk = sock->sk;
695 unsigned int lv = sizeof(int);
698 if (get_user(len, optlen))
705 v.val = sock_flag(sk, SOCK_DBG);
709 v.val = sock_flag(sk, SOCK_LOCALROUTE);
713 v.val = !!sock_flag(sk, SOCK_BROADCAST);
717 v.val = sk->sk_sndbuf;
721 v.val = sk->sk_rcvbuf;
725 v.val = sk->sk_reuse;
729 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
737 v.val = -sock_error(sk);
739 v.val = xchg(&sk->sk_err_soft, 0);
743 v.val = !!sock_flag(sk, SOCK_URGINLINE);
747 v.val = sk->sk_no_check;
751 v.val = sk->sk_priority;
756 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
757 v.ling.l_linger = sk->sk_lingertime / HZ;
761 sock_warn_obsolete_bsdism("getsockopt");
765 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
766 !sock_flag(sk, SOCK_RCVTSTAMPNS);
770 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
774 lv=sizeof(struct timeval);
775 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
779 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
780 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
785 lv=sizeof(struct timeval);
786 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
790 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
791 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
796 v.val = sk->sk_rcvlowat;
804 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
808 if (len > sizeof(sk->sk_peercred))
809 len = sizeof(sk->sk_peercred);
810 if (copy_to_user(optval, &sk->sk_peercred, len))
818 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
822 if (copy_to_user(optval, address, len))
827 /* Dubious BSD thing... Probably nobody even uses it, but
828 * the UNIX standard wants it for whatever reason... -DaveM
831 v.val = sk->sk_state == TCP_LISTEN;
835 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
839 return security_socket_getpeersec_stream(sock, optval, optlen, len);
847 if (copy_to_user(optval, &v, len))
850 if (put_user(len, optlen))
856 * Initialize an sk_lock.
858 * (We also register the sk_lock with the lock validator.)
860 static inline void sock_lock_init(struct sock *sk)
862 sock_lock_init_class_and_name(sk,
863 af_family_slock_key_strings[sk->sk_family],
864 af_family_slock_keys + sk->sk_family,
865 af_family_key_strings[sk->sk_family],
866 af_family_keys + sk->sk_family);
870 * sk_alloc - All socket objects are allocated here
871 * @family: protocol family
872 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
873 * @prot: struct proto associated with this new sock instance
874 * @zero_it: if we should zero the newly allocated sock
876 struct sock *sk_alloc(int family, gfp_t priority,
877 struct proto *prot, int zero_it)
879 struct sock *sk = NULL;
880 struct kmem_cache *slab = prot->slab;
883 sk = kmem_cache_alloc(slab, priority);
885 sk = kmalloc(prot->obj_size, priority);
889 memset(sk, 0, prot->obj_size);
890 sk->sk_family = family;
892 * See comment in struct sock definition to understand
893 * why we need sk_prot_creator -acme
895 sk->sk_prot = sk->sk_prot_creator = prot;
899 if (security_sk_alloc(sk, family, priority))
902 if (!try_module_get(prot->owner))
909 kmem_cache_free(slab, sk);
915 void sk_free(struct sock *sk)
917 struct sk_filter *filter;
918 struct module *owner = sk->sk_prot_creator->owner;
923 filter = rcu_dereference(sk->sk_filter);
925 sk_filter_release(sk, filter);
926 rcu_assign_pointer(sk->sk_filter, NULL);
929 sock_disable_timestamp(sk);
931 if (atomic_read(&sk->sk_omem_alloc))
932 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
933 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
935 security_sk_free(sk);
936 if (sk->sk_prot_creator->slab != NULL)
937 kmem_cache_free(sk->sk_prot_creator->slab, sk);
943 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
945 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
948 struct sk_filter *filter;
950 sock_copy(newsk, sk);
953 sk_node_init(&newsk->sk_node);
954 sock_lock_init(newsk);
956 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
958 atomic_set(&newsk->sk_rmem_alloc, 0);
959 atomic_set(&newsk->sk_wmem_alloc, 0);
960 atomic_set(&newsk->sk_omem_alloc, 0);
961 skb_queue_head_init(&newsk->sk_receive_queue);
962 skb_queue_head_init(&newsk->sk_write_queue);
963 #ifdef CONFIG_NET_DMA
964 skb_queue_head_init(&newsk->sk_async_wait_queue);
967 rwlock_init(&newsk->sk_dst_lock);
968 rwlock_init(&newsk->sk_callback_lock);
969 lockdep_set_class_and_name(&newsk->sk_callback_lock,
970 af_callback_keys + newsk->sk_family,
971 af_family_clock_key_strings[newsk->sk_family]);
973 newsk->sk_dst_cache = NULL;
974 newsk->sk_wmem_queued = 0;
975 newsk->sk_forward_alloc = 0;
976 newsk->sk_send_head = NULL;
977 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
979 sock_reset_flag(newsk, SOCK_DONE);
980 skb_queue_head_init(&newsk->sk_error_queue);
982 filter = newsk->sk_filter;
984 sk_filter_charge(newsk, filter);
986 if (unlikely(xfrm_sk_clone_policy(newsk))) {
987 /* It is still raw copy of parent, so invalidate
988 * destructor and make plain sk_free() */
989 newsk->sk_destruct = NULL;
996 newsk->sk_priority = 0;
997 atomic_set(&newsk->sk_refcnt, 2);
1000 * Increment the counter in the same struct proto as the master
1001 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1002 * is the same as sk->sk_prot->socks, as this field was copied
1005 * This _changes_ the previous behaviour, where
1006 * tcp_create_openreq_child always was incrementing the
1007 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1008 * to be taken into account in all callers. -acme
1010 sk_refcnt_debug_inc(newsk);
1011 newsk->sk_socket = NULL;
1012 newsk->sk_sleep = NULL;
1014 if (newsk->sk_prot->sockets_allocated)
1015 atomic_inc(newsk->sk_prot->sockets_allocated);
1021 EXPORT_SYMBOL_GPL(sk_clone);
1023 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1025 __sk_dst_set(sk, dst);
1026 sk->sk_route_caps = dst->dev->features;
1027 if (sk->sk_route_caps & NETIF_F_GSO)
1028 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1029 if (sk_can_gso(sk)) {
1030 if (dst->header_len)
1031 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1033 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1036 EXPORT_SYMBOL_GPL(sk_setup_caps);
1038 void __init sk_init(void)
1040 if (num_physpages <= 4096) {
1041 sysctl_wmem_max = 32767;
1042 sysctl_rmem_max = 32767;
1043 sysctl_wmem_default = 32767;
1044 sysctl_rmem_default = 32767;
1045 } else if (num_physpages >= 131072) {
1046 sysctl_wmem_max = 131071;
1047 sysctl_rmem_max = 131071;
1052 * Simple resource managers for sockets.
1057 * Write buffer destructor automatically called from kfree_skb.
1059 void sock_wfree(struct sk_buff *skb)
1061 struct sock *sk = skb->sk;
1063 /* In case it might be waiting for more memory. */
1064 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1065 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1066 sk->sk_write_space(sk);
1071 * Read buffer destructor automatically called from kfree_skb.
1073 void sock_rfree(struct sk_buff *skb)
1075 struct sock *sk = skb->sk;
1077 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1081 int sock_i_uid(struct sock *sk)
1085 read_lock(&sk->sk_callback_lock);
1086 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1087 read_unlock(&sk->sk_callback_lock);
1091 unsigned long sock_i_ino(struct sock *sk)
1095 read_lock(&sk->sk_callback_lock);
1096 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1097 read_unlock(&sk->sk_callback_lock);
1102 * Allocate a skb from the socket's send buffer.
1104 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1107 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1108 struct sk_buff * skb = alloc_skb(size, priority);
1110 skb_set_owner_w(skb, sk);
1118 * Allocate a skb from the socket's receive buffer.
1120 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1123 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1124 struct sk_buff *skb = alloc_skb(size, priority);
1126 skb_set_owner_r(skb, sk);
1134 * Allocate a memory block from the socket's option memory buffer.
1136 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1138 if ((unsigned)size <= sysctl_optmem_max &&
1139 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1141 /* First do the add, to avoid the race if kmalloc
1144 atomic_add(size, &sk->sk_omem_alloc);
1145 mem = kmalloc(size, priority);
1148 atomic_sub(size, &sk->sk_omem_alloc);
1154 * Free an option memory block.
1156 void sock_kfree_s(struct sock *sk, void *mem, int size)
1159 atomic_sub(size, &sk->sk_omem_alloc);
1162 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1163 I think, these locks should be removed for datagram sockets.
1165 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1169 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1173 if (signal_pending(current))
1175 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1176 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1177 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1179 if (sk->sk_shutdown & SEND_SHUTDOWN)
1183 timeo = schedule_timeout(timeo);
1185 finish_wait(sk->sk_sleep, &wait);
1191 * Generic send/receive buffer handlers
1194 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1195 unsigned long header_len,
1196 unsigned long data_len,
1197 int noblock, int *errcode)
1199 struct sk_buff *skb;
1204 gfp_mask = sk->sk_allocation;
1205 if (gfp_mask & __GFP_WAIT)
1206 gfp_mask |= __GFP_REPEAT;
1208 timeo = sock_sndtimeo(sk, noblock);
1210 err = sock_error(sk);
1215 if (sk->sk_shutdown & SEND_SHUTDOWN)
1218 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1219 skb = alloc_skb(header_len, gfp_mask);
1224 /* No pages, we're done... */
1228 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1229 skb->truesize += data_len;
1230 skb_shinfo(skb)->nr_frags = npages;
1231 for (i = 0; i < npages; i++) {
1235 page = alloc_pages(sk->sk_allocation, 0);
1238 skb_shinfo(skb)->nr_frags = i;
1243 frag = &skb_shinfo(skb)->frags[i];
1245 frag->page_offset = 0;
1246 frag->size = (data_len >= PAGE_SIZE ?
1249 data_len -= PAGE_SIZE;
1252 /* Full success... */
1258 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1259 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1263 if (signal_pending(current))
1265 timeo = sock_wait_for_wmem(sk, timeo);
1268 skb_set_owner_w(skb, sk);
1272 err = sock_intr_errno(timeo);
1278 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1279 int noblock, int *errcode)
1281 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1284 static void __lock_sock(struct sock *sk)
1289 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1290 TASK_UNINTERRUPTIBLE);
1291 spin_unlock_bh(&sk->sk_lock.slock);
1293 spin_lock_bh(&sk->sk_lock.slock);
1294 if (!sock_owned_by_user(sk))
1297 finish_wait(&sk->sk_lock.wq, &wait);
1300 static void __release_sock(struct sock *sk)
1302 struct sk_buff *skb = sk->sk_backlog.head;
1305 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1309 struct sk_buff *next = skb->next;
1312 sk->sk_backlog_rcv(sk, skb);
1315 * We are in process context here with softirqs
1316 * disabled, use cond_resched_softirq() to preempt.
1317 * This is safe to do because we've taken the backlog
1320 cond_resched_softirq();
1323 } while (skb != NULL);
1326 } while ((skb = sk->sk_backlog.head) != NULL);
1330 * sk_wait_data - wait for data to arrive at sk_receive_queue
1331 * @sk: sock to wait on
1332 * @timeo: for how long
1334 * Now socket state including sk->sk_err is changed only under lock,
1335 * hence we may omit checks after joining wait queue.
1336 * We check receive queue before schedule() only as optimization;
1337 * it is very likely that release_sock() added new data.
1339 int sk_wait_data(struct sock *sk, long *timeo)
1344 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1345 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1346 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1347 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1348 finish_wait(sk->sk_sleep, &wait);
1352 EXPORT_SYMBOL(sk_wait_data);
1355 * Set of default routines for initialising struct proto_ops when
1356 * the protocol does not support a particular function. In certain
1357 * cases where it makes no sense for a protocol to have a "do nothing"
1358 * function, some default processing is provided.
1361 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1366 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1372 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1377 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1382 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1388 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1393 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1398 int sock_no_listen(struct socket *sock, int backlog)
1403 int sock_no_shutdown(struct socket *sock, int how)
1408 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1409 char __user *optval, int optlen)
1414 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1415 char __user *optval, int __user *optlen)
1420 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1426 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1427 size_t len, int flags)
1432 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1434 /* Mirror missing mmap method error code */
1438 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1441 struct msghdr msg = {.msg_flags = flags};
1443 char *kaddr = kmap(page);
1444 iov.iov_base = kaddr + offset;
1446 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1452 * Default Socket Callbacks
1455 static void sock_def_wakeup(struct sock *sk)
1457 read_lock(&sk->sk_callback_lock);
1458 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1459 wake_up_interruptible_all(sk->sk_sleep);
1460 read_unlock(&sk->sk_callback_lock);
1463 static void sock_def_error_report(struct sock *sk)
1465 read_lock(&sk->sk_callback_lock);
1466 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1467 wake_up_interruptible(sk->sk_sleep);
1468 sk_wake_async(sk,0,POLL_ERR);
1469 read_unlock(&sk->sk_callback_lock);
1472 static void sock_def_readable(struct sock *sk, int len)
1474 read_lock(&sk->sk_callback_lock);
1475 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1476 wake_up_interruptible(sk->sk_sleep);
1477 sk_wake_async(sk,1,POLL_IN);
1478 read_unlock(&sk->sk_callback_lock);
1481 static void sock_def_write_space(struct sock *sk)
1483 read_lock(&sk->sk_callback_lock);
1485 /* Do not wake up a writer until he can make "significant"
1488 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1489 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1490 wake_up_interruptible(sk->sk_sleep);
1492 /* Should agree with poll, otherwise some programs break */
1493 if (sock_writeable(sk))
1494 sk_wake_async(sk, 2, POLL_OUT);
1497 read_unlock(&sk->sk_callback_lock);
1500 static void sock_def_destruct(struct sock *sk)
1502 kfree(sk->sk_protinfo);
1505 void sk_send_sigurg(struct sock *sk)
1507 if (sk->sk_socket && sk->sk_socket->file)
1508 if (send_sigurg(&sk->sk_socket->file->f_owner))
1509 sk_wake_async(sk, 3, POLL_PRI);
1512 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1513 unsigned long expires)
1515 if (!mod_timer(timer, expires))
1519 EXPORT_SYMBOL(sk_reset_timer);
1521 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1523 if (timer_pending(timer) && del_timer(timer))
1527 EXPORT_SYMBOL(sk_stop_timer);
1529 void sock_init_data(struct socket *sock, struct sock *sk)
1531 skb_queue_head_init(&sk->sk_receive_queue);
1532 skb_queue_head_init(&sk->sk_write_queue);
1533 skb_queue_head_init(&sk->sk_error_queue);
1534 #ifdef CONFIG_NET_DMA
1535 skb_queue_head_init(&sk->sk_async_wait_queue);
1538 sk->sk_send_head = NULL;
1540 init_timer(&sk->sk_timer);
1542 sk->sk_allocation = GFP_KERNEL;
1543 sk->sk_rcvbuf = sysctl_rmem_default;
1544 sk->sk_sndbuf = sysctl_wmem_default;
1545 sk->sk_state = TCP_CLOSE;
1546 sk->sk_socket = sock;
1548 sock_set_flag(sk, SOCK_ZAPPED);
1551 sk->sk_type = sock->type;
1552 sk->sk_sleep = &sock->wait;
1555 sk->sk_sleep = NULL;
1557 rwlock_init(&sk->sk_dst_lock);
1558 rwlock_init(&sk->sk_callback_lock);
1559 lockdep_set_class_and_name(&sk->sk_callback_lock,
1560 af_callback_keys + sk->sk_family,
1561 af_family_clock_key_strings[sk->sk_family]);
1563 sk->sk_state_change = sock_def_wakeup;
1564 sk->sk_data_ready = sock_def_readable;
1565 sk->sk_write_space = sock_def_write_space;
1566 sk->sk_error_report = sock_def_error_report;
1567 sk->sk_destruct = sock_def_destruct;
1569 sk->sk_sndmsg_page = NULL;
1570 sk->sk_sndmsg_off = 0;
1572 sk->sk_peercred.pid = 0;
1573 sk->sk_peercred.uid = -1;
1574 sk->sk_peercred.gid = -1;
1575 sk->sk_write_pending = 0;
1576 sk->sk_rcvlowat = 1;
1577 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1578 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1580 sk->sk_stamp = ktime_set(-1L, -1L);
1582 atomic_set(&sk->sk_refcnt, 1);
1585 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1588 spin_lock_bh(&sk->sk_lock.slock);
1589 if (sk->sk_lock.owned)
1591 sk->sk_lock.owned = 1;
1592 spin_unlock(&sk->sk_lock.slock);
1594 * The sk_lock has mutex_lock() semantics here:
1596 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1600 EXPORT_SYMBOL(lock_sock_nested);
1602 void fastcall release_sock(struct sock *sk)
1605 * The sk_lock has mutex_unlock() semantics:
1607 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1609 spin_lock_bh(&sk->sk_lock.slock);
1610 if (sk->sk_backlog.tail)
1612 sk->sk_lock.owned = 0;
1613 if (waitqueue_active(&sk->sk_lock.wq))
1614 wake_up(&sk->sk_lock.wq);
1615 spin_unlock_bh(&sk->sk_lock.slock);
1617 EXPORT_SYMBOL(release_sock);
1619 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1622 if (!sock_flag(sk, SOCK_TIMESTAMP))
1623 sock_enable_timestamp(sk);
1624 tv = ktime_to_timeval(sk->sk_stamp);
1625 if (tv.tv_sec == -1)
1627 if (tv.tv_sec == 0) {
1628 sk->sk_stamp = ktime_get_real();
1629 tv = ktime_to_timeval(sk->sk_stamp);
1631 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1633 EXPORT_SYMBOL(sock_get_timestamp);
1635 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1638 if (!sock_flag(sk, SOCK_TIMESTAMP))
1639 sock_enable_timestamp(sk);
1640 ts = ktime_to_timespec(sk->sk_stamp);
1641 if (ts.tv_sec == -1)
1643 if (ts.tv_sec == 0) {
1644 sk->sk_stamp = ktime_get_real();
1645 ts = ktime_to_timespec(sk->sk_stamp);
1647 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1649 EXPORT_SYMBOL(sock_get_timestampns);
1651 void sock_enable_timestamp(struct sock *sk)
1653 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1654 sock_set_flag(sk, SOCK_TIMESTAMP);
1655 net_enable_timestamp();
1658 EXPORT_SYMBOL(sock_enable_timestamp);
1661 * Get a socket option on an socket.
1663 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1664 * asynchronous errors should be reported by getsockopt. We assume
1665 * this means if you specify SO_ERROR (otherwise whats the point of it).
1667 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1668 char __user *optval, int __user *optlen)
1670 struct sock *sk = sock->sk;
1672 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1675 EXPORT_SYMBOL(sock_common_getsockopt);
1677 #ifdef CONFIG_COMPAT
1678 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1679 char __user *optval, int __user *optlen)
1681 struct sock *sk = sock->sk;
1683 if (sk->sk_prot->compat_getsockopt != NULL)
1684 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1686 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1688 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1691 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1692 struct msghdr *msg, size_t size, int flags)
1694 struct sock *sk = sock->sk;
1698 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1699 flags & ~MSG_DONTWAIT, &addr_len);
1701 msg->msg_namelen = addr_len;
1705 EXPORT_SYMBOL(sock_common_recvmsg);
1708 * Set socket options on an inet socket.
1710 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1711 char __user *optval, int optlen)
1713 struct sock *sk = sock->sk;
1715 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1718 EXPORT_SYMBOL(sock_common_setsockopt);
1720 #ifdef CONFIG_COMPAT
1721 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1722 char __user *optval, int optlen)
1724 struct sock *sk = sock->sk;
1726 if (sk->sk_prot->compat_setsockopt != NULL)
1727 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1729 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1731 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1734 void sk_common_release(struct sock *sk)
1736 if (sk->sk_prot->destroy)
1737 sk->sk_prot->destroy(sk);
1740 * Observation: when sock_common_release is called, processes have
1741 * no access to socket. But net still has.
1742 * Step one, detach it from networking:
1744 * A. Remove from hash tables.
1747 sk->sk_prot->unhash(sk);
1750 * In this point socket cannot receive new packets, but it is possible
1751 * that some packets are in flight because some CPU runs receiver and
1752 * did hash table lookup before we unhashed socket. They will achieve
1753 * receive queue and will be purged by socket destructor.
1755 * Also we still have packets pending on receive queue and probably,
1756 * our own packets waiting in device queues. sock_destroy will drain
1757 * receive queue, but transmitted packets will delay socket destruction
1758 * until the last reference will be released.
1763 xfrm_sk_free_policy(sk);
1765 sk_refcnt_debug_release(sk);
1769 EXPORT_SYMBOL(sk_common_release);
1771 static DEFINE_RWLOCK(proto_list_lock);
1772 static LIST_HEAD(proto_list);
1774 int proto_register(struct proto *prot, int alloc_slab)
1776 char *request_sock_slab_name = NULL;
1777 char *timewait_sock_slab_name;
1781 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1782 SLAB_HWCACHE_ALIGN, NULL);
1784 if (prot->slab == NULL) {
1785 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1790 if (prot->rsk_prot != NULL) {
1791 static const char mask[] = "request_sock_%s";
1793 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1794 if (request_sock_slab_name == NULL)
1795 goto out_free_sock_slab;
1797 sprintf(request_sock_slab_name, mask, prot->name);
1798 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1799 prot->rsk_prot->obj_size, 0,
1800 SLAB_HWCACHE_ALIGN, NULL);
1802 if (prot->rsk_prot->slab == NULL) {
1803 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1805 goto out_free_request_sock_slab_name;
1809 if (prot->twsk_prot != NULL) {
1810 static const char mask[] = "tw_sock_%s";
1812 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1814 if (timewait_sock_slab_name == NULL)
1815 goto out_free_request_sock_slab;
1817 sprintf(timewait_sock_slab_name, mask, prot->name);
1818 prot->twsk_prot->twsk_slab =
1819 kmem_cache_create(timewait_sock_slab_name,
1820 prot->twsk_prot->twsk_obj_size,
1821 0, SLAB_HWCACHE_ALIGN,
1823 if (prot->twsk_prot->twsk_slab == NULL)
1824 goto out_free_timewait_sock_slab_name;
1828 write_lock(&proto_list_lock);
1829 list_add(&prot->node, &proto_list);
1830 write_unlock(&proto_list_lock);
1834 out_free_timewait_sock_slab_name:
1835 kfree(timewait_sock_slab_name);
1836 out_free_request_sock_slab:
1837 if (prot->rsk_prot && prot->rsk_prot->slab) {
1838 kmem_cache_destroy(prot->rsk_prot->slab);
1839 prot->rsk_prot->slab = NULL;
1841 out_free_request_sock_slab_name:
1842 kfree(request_sock_slab_name);
1844 kmem_cache_destroy(prot->slab);
1849 EXPORT_SYMBOL(proto_register);
1851 void proto_unregister(struct proto *prot)
1853 write_lock(&proto_list_lock);
1854 list_del(&prot->node);
1855 write_unlock(&proto_list_lock);
1857 if (prot->slab != NULL) {
1858 kmem_cache_destroy(prot->slab);
1862 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1863 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1865 kmem_cache_destroy(prot->rsk_prot->slab);
1867 prot->rsk_prot->slab = NULL;
1870 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1871 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1873 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1875 prot->twsk_prot->twsk_slab = NULL;
1879 EXPORT_SYMBOL(proto_unregister);
1881 #ifdef CONFIG_PROC_FS
1882 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1884 read_lock(&proto_list_lock);
1885 return seq_list_start_head(&proto_list, *pos);
1888 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1890 return seq_list_next(v, &proto_list, pos);
1893 static void proto_seq_stop(struct seq_file *seq, void *v)
1895 read_unlock(&proto_list_lock);
1898 static char proto_method_implemented(const void *method)
1900 return method == NULL ? 'n' : 'y';
1903 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1905 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1906 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1909 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1910 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1911 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1913 proto->slab == NULL ? "no" : "yes",
1914 module_name(proto->owner),
1915 proto_method_implemented(proto->close),
1916 proto_method_implemented(proto->connect),
1917 proto_method_implemented(proto->disconnect),
1918 proto_method_implemented(proto->accept),
1919 proto_method_implemented(proto->ioctl),
1920 proto_method_implemented(proto->init),
1921 proto_method_implemented(proto->destroy),
1922 proto_method_implemented(proto->shutdown),
1923 proto_method_implemented(proto->setsockopt),
1924 proto_method_implemented(proto->getsockopt),
1925 proto_method_implemented(proto->sendmsg),
1926 proto_method_implemented(proto->recvmsg),
1927 proto_method_implemented(proto->sendpage),
1928 proto_method_implemented(proto->bind),
1929 proto_method_implemented(proto->backlog_rcv),
1930 proto_method_implemented(proto->hash),
1931 proto_method_implemented(proto->unhash),
1932 proto_method_implemented(proto->get_port),
1933 proto_method_implemented(proto->enter_memory_pressure));
1936 static int proto_seq_show(struct seq_file *seq, void *v)
1938 if (v == &proto_list)
1939 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1948 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1950 proto_seq_printf(seq, list_entry(v, struct proto, node));
1954 static const struct seq_operations proto_seq_ops = {
1955 .start = proto_seq_start,
1956 .next = proto_seq_next,
1957 .stop = proto_seq_stop,
1958 .show = proto_seq_show,
1961 static int proto_seq_open(struct inode *inode, struct file *file)
1963 return seq_open(file, &proto_seq_ops);
1966 static const struct file_operations proto_seq_fops = {
1967 .owner = THIS_MODULE,
1968 .open = proto_seq_open,
1970 .llseek = seq_lseek,
1971 .release = seq_release,
1974 static int __init proto_init(void)
1976 /* register /proc/net/protocols */
1977 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1980 subsys_initcall(proto_init);
1982 #endif /* PROC_FS */
1984 EXPORT_SYMBOL(sk_alloc);
1985 EXPORT_SYMBOL(sk_free);
1986 EXPORT_SYMBOL(sk_send_sigurg);
1987 EXPORT_SYMBOL(sock_alloc_send_skb);
1988 EXPORT_SYMBOL(sock_init_data);
1989 EXPORT_SYMBOL(sock_kfree_s);
1990 EXPORT_SYMBOL(sock_kmalloc);
1991 EXPORT_SYMBOL(sock_no_accept);
1992 EXPORT_SYMBOL(sock_no_bind);
1993 EXPORT_SYMBOL(sock_no_connect);
1994 EXPORT_SYMBOL(sock_no_getname);
1995 EXPORT_SYMBOL(sock_no_getsockopt);
1996 EXPORT_SYMBOL(sock_no_ioctl);
1997 EXPORT_SYMBOL(sock_no_listen);
1998 EXPORT_SYMBOL(sock_no_mmap);
1999 EXPORT_SYMBOL(sock_no_poll);
2000 EXPORT_SYMBOL(sock_no_recvmsg);
2001 EXPORT_SYMBOL(sock_no_sendmsg);
2002 EXPORT_SYMBOL(sock_no_sendpage);
2003 EXPORT_SYMBOL(sock_no_setsockopt);
2004 EXPORT_SYMBOL(sock_no_shutdown);
2005 EXPORT_SYMBOL(sock_no_socketpair);
2006 EXPORT_SYMBOL(sock_rfree);
2007 EXPORT_SYMBOL(sock_setsockopt);
2008 EXPORT_SYMBOL(sock_wfree);
2009 EXPORT_SYMBOL(sock_wmalloc);
2010 EXPORT_SYMBOL(sock_i_uid);
2011 EXPORT_SYMBOL(sock_i_ino);
2012 EXPORT_SYMBOL(sysctl_optmem_max);
2013 #ifdef CONFIG_SYSCTL
2014 EXPORT_SYMBOL(sysctl_rmem_max);
2015 EXPORT_SYMBOL(sysctl_wmem_max);