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;
331 bh_lock_sock_nested(sk);
334 if (!sock_owned_by_user(sk)) {
336 * trylock + unlock semantics:
338 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
340 rc = sk_backlog_rcv(sk, skb);
342 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
344 sk_add_backlog(sk, skb);
353 EXPORT_SYMBOL(sk_receive_skb);
355 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
357 struct dst_entry *dst = sk->sk_dst_cache;
359 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
360 sk->sk_dst_cache = NULL;
367 EXPORT_SYMBOL(__sk_dst_check);
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) {
381 EXPORT_SYMBOL(sk_dst_check);
383 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
385 int ret = -ENOPROTOOPT;
386 #ifdef CONFIG_NETDEVICES
387 struct net *net = sock_net(sk);
388 char devname[IFNAMSIZ];
393 if (!capable(CAP_NET_RAW))
400 /* Bind this socket to a particular device like "eth0",
401 * as specified in the passed interface name. If the
402 * name is "" or the option length is zero the socket
405 if (optlen > IFNAMSIZ - 1)
406 optlen = IFNAMSIZ - 1;
407 memset(devname, 0, sizeof(devname));
410 if (copy_from_user(devname, optval, optlen))
413 if (devname[0] == '\0') {
416 struct net_device *dev = dev_get_by_name(net, devname);
422 index = dev->ifindex;
427 sk->sk_bound_dev_if = index;
439 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
442 sock_set_flag(sk, bit);
444 sock_reset_flag(sk, bit);
448 * This is meant for all protocols to use and covers goings on
449 * at the socket level. Everything here is generic.
452 int sock_setsockopt(struct socket *sock, int level, int optname,
453 char __user *optval, unsigned int optlen)
455 struct sock *sk = sock->sk;
462 * Options without arguments
465 if (optname == SO_BINDTODEVICE)
466 return sock_bindtodevice(sk, optval, optlen);
468 if (optlen < sizeof(int))
471 if (get_user(val, (int __user *)optval))
474 valbool = val ? 1 : 0;
480 if (val && !capable(CAP_NET_ADMIN))
483 sock_valbool_flag(sk, SOCK_DBG, valbool);
486 sk->sk_reuse = valbool;
495 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
498 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
501 /* Don't error on this BSD doesn't and if you think
502 about it this is right. Otherwise apps have to
503 play 'guess the biggest size' games. RCVBUF/SNDBUF
504 are treated in BSD as hints */
506 if (val > sysctl_wmem_max)
507 val = sysctl_wmem_max;
509 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
510 if ((val * 2) < SOCK_MIN_SNDBUF)
511 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
513 sk->sk_sndbuf = val * 2;
516 * Wake up sending tasks if we
519 sk->sk_write_space(sk);
523 if (!capable(CAP_NET_ADMIN)) {
530 /* Don't error on this BSD doesn't and if you think
531 about it this is right. Otherwise apps have to
532 play 'guess the biggest size' games. RCVBUF/SNDBUF
533 are treated in BSD as hints */
535 if (val > sysctl_rmem_max)
536 val = sysctl_rmem_max;
538 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
540 * We double it on the way in to account for
541 * "struct sk_buff" etc. overhead. Applications
542 * assume that the SO_RCVBUF setting they make will
543 * allow that much actual data to be received on that
546 * Applications are unaware that "struct sk_buff" and
547 * other overheads allocate from the receive buffer
548 * during socket buffer allocation.
550 * And after considering the possible alternatives,
551 * returning the value we actually used in getsockopt
552 * is the most desirable behavior.
554 if ((val * 2) < SOCK_MIN_RCVBUF)
555 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
557 sk->sk_rcvbuf = val * 2;
561 if (!capable(CAP_NET_ADMIN)) {
569 if (sk->sk_protocol == IPPROTO_TCP)
570 tcp_set_keepalive(sk, valbool);
572 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
576 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
580 sk->sk_no_check = valbool;
584 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
585 sk->sk_priority = val;
591 if (optlen < sizeof(ling)) {
592 ret = -EINVAL; /* 1003.1g */
595 if (copy_from_user(&ling, optval, sizeof(ling))) {
600 sock_reset_flag(sk, SOCK_LINGER);
602 #if (BITS_PER_LONG == 32)
603 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
604 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
607 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
608 sock_set_flag(sk, SOCK_LINGER);
613 sock_warn_obsolete_bsdism("setsockopt");
618 set_bit(SOCK_PASSCRED, &sock->flags);
620 clear_bit(SOCK_PASSCRED, &sock->flags);
626 if (optname == SO_TIMESTAMP)
627 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
629 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
630 sock_set_flag(sk, SOCK_RCVTSTAMP);
631 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
633 sock_reset_flag(sk, SOCK_RCVTSTAMP);
634 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
638 case SO_TIMESTAMPING:
639 if (val & ~SOF_TIMESTAMPING_MASK) {
643 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
644 val & SOF_TIMESTAMPING_TX_HARDWARE);
645 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
646 val & SOF_TIMESTAMPING_TX_SOFTWARE);
647 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
648 val & SOF_TIMESTAMPING_RX_HARDWARE);
649 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
650 sock_enable_timestamp(sk,
651 SOCK_TIMESTAMPING_RX_SOFTWARE);
653 sock_disable_timestamp(sk,
654 SOCK_TIMESTAMPING_RX_SOFTWARE);
655 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
656 val & SOF_TIMESTAMPING_SOFTWARE);
657 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
658 val & SOF_TIMESTAMPING_SYS_HARDWARE);
659 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
660 val & SOF_TIMESTAMPING_RAW_HARDWARE);
666 sk->sk_rcvlowat = val ? : 1;
670 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
674 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
677 case SO_ATTACH_FILTER:
679 if (optlen == sizeof(struct sock_fprog)) {
680 struct sock_fprog fprog;
683 if (copy_from_user(&fprog, optval, sizeof(fprog)))
686 ret = sk_attach_filter(&fprog, sk);
690 case SO_DETACH_FILTER:
691 ret = sk_detach_filter(sk);
696 set_bit(SOCK_PASSSEC, &sock->flags);
698 clear_bit(SOCK_PASSSEC, &sock->flags);
701 if (!capable(CAP_NET_ADMIN))
707 /* We implement the SO_SNDLOWAT etc to
708 not be settable (1003.1g 5.3) */
711 sock_set_flag(sk, SOCK_RXQ_OVFL);
713 sock_reset_flag(sk, SOCK_RXQ_OVFL);
722 EXPORT_SYMBOL(sock_setsockopt);
725 int sock_getsockopt(struct socket *sock, int level, int optname,
726 char __user *optval, int __user *optlen)
728 struct sock *sk = sock->sk;
736 unsigned int lv = sizeof(int);
739 if (get_user(len, optlen))
744 memset(&v, 0, sizeof(v));
748 v.val = sock_flag(sk, SOCK_DBG);
752 v.val = sock_flag(sk, SOCK_LOCALROUTE);
756 v.val = !!sock_flag(sk, SOCK_BROADCAST);
760 v.val = sk->sk_sndbuf;
764 v.val = sk->sk_rcvbuf;
768 v.val = sk->sk_reuse;
772 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
780 v.val = sk->sk_protocol;
784 v.val = sk->sk_family;
788 v.val = -sock_error(sk);
790 v.val = xchg(&sk->sk_err_soft, 0);
794 v.val = !!sock_flag(sk, SOCK_URGINLINE);
798 v.val = sk->sk_no_check;
802 v.val = sk->sk_priority;
807 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
808 v.ling.l_linger = sk->sk_lingertime / HZ;
812 sock_warn_obsolete_bsdism("getsockopt");
816 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
817 !sock_flag(sk, SOCK_RCVTSTAMPNS);
821 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
824 case SO_TIMESTAMPING:
826 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
827 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
828 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
829 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
830 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
831 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
832 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
833 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
834 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
835 v.val |= SOF_TIMESTAMPING_SOFTWARE;
836 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
837 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
838 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
839 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
843 lv = sizeof(struct timeval);
844 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
848 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
849 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
854 lv = sizeof(struct timeval);
855 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
859 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
860 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
865 v.val = sk->sk_rcvlowat;
873 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
877 if (len > sizeof(sk->sk_peercred))
878 len = sizeof(sk->sk_peercred);
879 if (copy_to_user(optval, &sk->sk_peercred, len))
887 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
891 if (copy_to_user(optval, address, len))
896 /* Dubious BSD thing... Probably nobody even uses it, but
897 * the UNIX standard wants it for whatever reason... -DaveM
900 v.val = sk->sk_state == TCP_LISTEN;
904 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
908 return security_socket_getpeersec_stream(sock, optval, optlen, len);
915 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
924 if (copy_to_user(optval, &v, len))
927 if (put_user(len, optlen))
933 * Initialize an sk_lock.
935 * (We also register the sk_lock with the lock validator.)
937 static inline void sock_lock_init(struct sock *sk)
939 sock_lock_init_class_and_name(sk,
940 af_family_slock_key_strings[sk->sk_family],
941 af_family_slock_keys + sk->sk_family,
942 af_family_key_strings[sk->sk_family],
943 af_family_keys + sk->sk_family);
947 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
948 * even temporarly, because of RCU lookups. sk_node should also be left as is.
950 static void sock_copy(struct sock *nsk, const struct sock *osk)
952 #ifdef CONFIG_SECURITY_NETWORK
953 void *sptr = nsk->sk_security;
955 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
956 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt));
957 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
958 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
959 #ifdef CONFIG_SECURITY_NETWORK
960 nsk->sk_security = sptr;
961 security_sk_clone(osk, nsk);
965 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
969 struct kmem_cache *slab;
973 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
976 if (priority & __GFP_ZERO) {
978 * caches using SLAB_DESTROY_BY_RCU should let
979 * sk_node.next un-modified. Special care is taken
980 * when initializing object to zero.
982 if (offsetof(struct sock, sk_node.next) != 0)
983 memset(sk, 0, offsetof(struct sock, sk_node.next));
984 memset(&sk->sk_node.pprev, 0,
985 prot->obj_size - offsetof(struct sock,
990 sk = kmalloc(prot->obj_size, priority);
993 kmemcheck_annotate_bitfield(sk, flags);
995 if (security_sk_alloc(sk, family, priority))
998 if (!try_module_get(prot->owner))
1005 security_sk_free(sk);
1008 kmem_cache_free(slab, sk);
1014 static void sk_prot_free(struct proto *prot, struct sock *sk)
1016 struct kmem_cache *slab;
1017 struct module *owner;
1019 owner = prot->owner;
1022 security_sk_free(sk);
1024 kmem_cache_free(slab, sk);
1031 * sk_alloc - All socket objects are allocated here
1032 * @net: the applicable net namespace
1033 * @family: protocol family
1034 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1035 * @prot: struct proto associated with this new sock instance
1037 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1042 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1044 sk->sk_family = family;
1046 * See comment in struct sock definition to understand
1047 * why we need sk_prot_creator -acme
1049 sk->sk_prot = sk->sk_prot_creator = prot;
1051 sock_net_set(sk, get_net(net));
1052 atomic_set(&sk->sk_wmem_alloc, 1);
1057 EXPORT_SYMBOL(sk_alloc);
1059 static void __sk_free(struct sock *sk)
1061 struct sk_filter *filter;
1063 if (sk->sk_destruct)
1064 sk->sk_destruct(sk);
1066 filter = rcu_dereference(sk->sk_filter);
1068 sk_filter_uncharge(sk, filter);
1069 rcu_assign_pointer(sk->sk_filter, NULL);
1072 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1073 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1075 if (atomic_read(&sk->sk_omem_alloc))
1076 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1077 __func__, atomic_read(&sk->sk_omem_alloc));
1079 put_net(sock_net(sk));
1080 sk_prot_free(sk->sk_prot_creator, sk);
1083 void sk_free(struct sock *sk)
1086 * We substract one from sk_wmem_alloc and can know if
1087 * some packets are still in some tx queue.
1088 * If not null, sock_wfree() will call __sk_free(sk) later
1090 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1093 EXPORT_SYMBOL(sk_free);
1096 * Last sock_put should drop referrence to sk->sk_net. It has already
1097 * been dropped in sk_change_net. Taking referrence to stopping namespace
1099 * Take referrence to a socket to remove it from hash _alive_ and after that
1100 * destroy it in the context of init_net.
1102 void sk_release_kernel(struct sock *sk)
1104 if (sk == NULL || sk->sk_socket == NULL)
1108 sock_release(sk->sk_socket);
1109 release_net(sock_net(sk));
1110 sock_net_set(sk, get_net(&init_net));
1113 EXPORT_SYMBOL(sk_release_kernel);
1115 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1119 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1120 if (newsk != NULL) {
1121 struct sk_filter *filter;
1123 sock_copy(newsk, sk);
1126 get_net(sock_net(newsk));
1127 sk_node_init(&newsk->sk_node);
1128 sock_lock_init(newsk);
1129 bh_lock_sock(newsk);
1130 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1132 atomic_set(&newsk->sk_rmem_alloc, 0);
1134 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1136 atomic_set(&newsk->sk_wmem_alloc, 1);
1137 atomic_set(&newsk->sk_omem_alloc, 0);
1138 skb_queue_head_init(&newsk->sk_receive_queue);
1139 skb_queue_head_init(&newsk->sk_write_queue);
1140 #ifdef CONFIG_NET_DMA
1141 skb_queue_head_init(&newsk->sk_async_wait_queue);
1144 rwlock_init(&newsk->sk_dst_lock);
1145 rwlock_init(&newsk->sk_callback_lock);
1146 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1147 af_callback_keys + newsk->sk_family,
1148 af_family_clock_key_strings[newsk->sk_family]);
1150 newsk->sk_dst_cache = NULL;
1151 newsk->sk_wmem_queued = 0;
1152 newsk->sk_forward_alloc = 0;
1153 newsk->sk_send_head = NULL;
1154 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1156 sock_reset_flag(newsk, SOCK_DONE);
1157 skb_queue_head_init(&newsk->sk_error_queue);
1159 filter = newsk->sk_filter;
1161 sk_filter_charge(newsk, filter);
1163 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1164 /* It is still raw copy of parent, so invalidate
1165 * destructor and make plain sk_free() */
1166 newsk->sk_destruct = NULL;
1173 newsk->sk_priority = 0;
1175 * Before updating sk_refcnt, we must commit prior changes to memory
1176 * (Documentation/RCU/rculist_nulls.txt for details)
1179 atomic_set(&newsk->sk_refcnt, 2);
1182 * Increment the counter in the same struct proto as the master
1183 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1184 * is the same as sk->sk_prot->socks, as this field was copied
1187 * This _changes_ the previous behaviour, where
1188 * tcp_create_openreq_child always was incrementing the
1189 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1190 * to be taken into account in all callers. -acme
1192 sk_refcnt_debug_inc(newsk);
1193 sk_set_socket(newsk, NULL);
1194 newsk->sk_sleep = NULL;
1196 if (newsk->sk_prot->sockets_allocated)
1197 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1202 EXPORT_SYMBOL_GPL(sk_clone);
1204 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1206 __sk_dst_set(sk, dst);
1207 sk->sk_route_caps = dst->dev->features;
1208 if (sk->sk_route_caps & NETIF_F_GSO)
1209 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1210 if (sk_can_gso(sk)) {
1211 if (dst->header_len) {
1212 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1214 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1215 sk->sk_gso_max_size = dst->dev->gso_max_size;
1219 EXPORT_SYMBOL_GPL(sk_setup_caps);
1221 void __init sk_init(void)
1223 if (totalram_pages <= 4096) {
1224 sysctl_wmem_max = 32767;
1225 sysctl_rmem_max = 32767;
1226 sysctl_wmem_default = 32767;
1227 sysctl_rmem_default = 32767;
1228 } else if (totalram_pages >= 131072) {
1229 sysctl_wmem_max = 131071;
1230 sysctl_rmem_max = 131071;
1235 * Simple resource managers for sockets.
1240 * Write buffer destructor automatically called from kfree_skb.
1242 void sock_wfree(struct sk_buff *skb)
1244 struct sock *sk = skb->sk;
1245 unsigned int len = skb->truesize;
1247 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1249 * Keep a reference on sk_wmem_alloc, this will be released
1250 * after sk_write_space() call
1252 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1253 sk->sk_write_space(sk);
1257 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1258 * could not do because of in-flight packets
1260 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1263 EXPORT_SYMBOL(sock_wfree);
1266 * Read buffer destructor automatically called from kfree_skb.
1268 void sock_rfree(struct sk_buff *skb)
1270 struct sock *sk = skb->sk;
1272 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1273 sk_mem_uncharge(skb->sk, skb->truesize);
1275 EXPORT_SYMBOL(sock_rfree);
1278 int sock_i_uid(struct sock *sk)
1282 read_lock(&sk->sk_callback_lock);
1283 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1284 read_unlock(&sk->sk_callback_lock);
1287 EXPORT_SYMBOL(sock_i_uid);
1289 unsigned long sock_i_ino(struct sock *sk)
1293 read_lock(&sk->sk_callback_lock);
1294 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1295 read_unlock(&sk->sk_callback_lock);
1298 EXPORT_SYMBOL(sock_i_ino);
1301 * Allocate a skb from the socket's send buffer.
1303 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1306 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1307 struct sk_buff *skb = alloc_skb(size, priority);
1309 skb_set_owner_w(skb, sk);
1315 EXPORT_SYMBOL(sock_wmalloc);
1318 * Allocate a skb from the socket's receive buffer.
1320 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1323 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1324 struct sk_buff *skb = alloc_skb(size, priority);
1326 skb_set_owner_r(skb, sk);
1334 * Allocate a memory block from the socket's option memory buffer.
1336 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1338 if ((unsigned)size <= sysctl_optmem_max &&
1339 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1341 /* First do the add, to avoid the race if kmalloc
1344 atomic_add(size, &sk->sk_omem_alloc);
1345 mem = kmalloc(size, priority);
1348 atomic_sub(size, &sk->sk_omem_alloc);
1352 EXPORT_SYMBOL(sock_kmalloc);
1355 * Free an option memory block.
1357 void sock_kfree_s(struct sock *sk, void *mem, int size)
1360 atomic_sub(size, &sk->sk_omem_alloc);
1362 EXPORT_SYMBOL(sock_kfree_s);
1364 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1365 I think, these locks should be removed for datagram sockets.
1367 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1371 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1375 if (signal_pending(current))
1377 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1378 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1379 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1381 if (sk->sk_shutdown & SEND_SHUTDOWN)
1385 timeo = schedule_timeout(timeo);
1387 finish_wait(sk->sk_sleep, &wait);
1393 * Generic send/receive buffer handlers
1396 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1397 unsigned long data_len, int noblock,
1400 struct sk_buff *skb;
1405 gfp_mask = sk->sk_allocation;
1406 if (gfp_mask & __GFP_WAIT)
1407 gfp_mask |= __GFP_REPEAT;
1409 timeo = sock_sndtimeo(sk, noblock);
1411 err = sock_error(sk);
1416 if (sk->sk_shutdown & SEND_SHUTDOWN)
1419 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1420 skb = alloc_skb(header_len, gfp_mask);
1425 /* No pages, we're done... */
1429 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1430 skb->truesize += data_len;
1431 skb_shinfo(skb)->nr_frags = npages;
1432 for (i = 0; i < npages; i++) {
1436 page = alloc_pages(sk->sk_allocation, 0);
1439 skb_shinfo(skb)->nr_frags = i;
1444 frag = &skb_shinfo(skb)->frags[i];
1446 frag->page_offset = 0;
1447 frag->size = (data_len >= PAGE_SIZE ?
1450 data_len -= PAGE_SIZE;
1453 /* Full success... */
1459 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1460 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1464 if (signal_pending(current))
1466 timeo = sock_wait_for_wmem(sk, timeo);
1469 skb_set_owner_w(skb, sk);
1473 err = sock_intr_errno(timeo);
1478 EXPORT_SYMBOL(sock_alloc_send_pskb);
1480 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1481 int noblock, int *errcode)
1483 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1485 EXPORT_SYMBOL(sock_alloc_send_skb);
1487 static void __lock_sock(struct sock *sk)
1492 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1493 TASK_UNINTERRUPTIBLE);
1494 spin_unlock_bh(&sk->sk_lock.slock);
1496 spin_lock_bh(&sk->sk_lock.slock);
1497 if (!sock_owned_by_user(sk))
1500 finish_wait(&sk->sk_lock.wq, &wait);
1503 static void __release_sock(struct sock *sk)
1505 struct sk_buff *skb = sk->sk_backlog.head;
1508 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1512 struct sk_buff *next = skb->next;
1515 sk_backlog_rcv(sk, skb);
1518 * We are in process context here with softirqs
1519 * disabled, use cond_resched_softirq() to preempt.
1520 * This is safe to do because we've taken the backlog
1523 cond_resched_softirq();
1526 } while (skb != NULL);
1529 } while ((skb = sk->sk_backlog.head) != NULL);
1533 * sk_wait_data - wait for data to arrive at sk_receive_queue
1534 * @sk: sock to wait on
1535 * @timeo: for how long
1537 * Now socket state including sk->sk_err is changed only under lock,
1538 * hence we may omit checks after joining wait queue.
1539 * We check receive queue before schedule() only as optimization;
1540 * it is very likely that release_sock() added new data.
1542 int sk_wait_data(struct sock *sk, long *timeo)
1547 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1548 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1549 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1550 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1551 finish_wait(sk->sk_sleep, &wait);
1554 EXPORT_SYMBOL(sk_wait_data);
1557 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1559 * @size: memory size to allocate
1560 * @kind: allocation type
1562 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1563 * rmem allocation. This function assumes that protocols which have
1564 * memory_pressure use sk_wmem_queued as write buffer accounting.
1566 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1568 struct proto *prot = sk->sk_prot;
1569 int amt = sk_mem_pages(size);
1572 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1573 allocated = atomic_add_return(amt, prot->memory_allocated);
1576 if (allocated <= prot->sysctl_mem[0]) {
1577 if (prot->memory_pressure && *prot->memory_pressure)
1578 *prot->memory_pressure = 0;
1582 /* Under pressure. */
1583 if (allocated > prot->sysctl_mem[1])
1584 if (prot->enter_memory_pressure)
1585 prot->enter_memory_pressure(sk);
1587 /* Over hard limit. */
1588 if (allocated > prot->sysctl_mem[2])
1589 goto suppress_allocation;
1591 /* guarantee minimum buffer size under pressure */
1592 if (kind == SK_MEM_RECV) {
1593 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1595 } else { /* SK_MEM_SEND */
1596 if (sk->sk_type == SOCK_STREAM) {
1597 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1599 } else if (atomic_read(&sk->sk_wmem_alloc) <
1600 prot->sysctl_wmem[0])
1604 if (prot->memory_pressure) {
1607 if (!*prot->memory_pressure)
1609 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1610 if (prot->sysctl_mem[2] > alloc *
1611 sk_mem_pages(sk->sk_wmem_queued +
1612 atomic_read(&sk->sk_rmem_alloc) +
1613 sk->sk_forward_alloc))
1617 suppress_allocation:
1619 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1620 sk_stream_moderate_sndbuf(sk);
1622 /* Fail only if socket is _under_ its sndbuf.
1623 * In this case we cannot block, so that we have to fail.
1625 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1629 /* Alas. Undo changes. */
1630 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1631 atomic_sub(amt, prot->memory_allocated);
1634 EXPORT_SYMBOL(__sk_mem_schedule);
1637 * __sk_reclaim - reclaim memory_allocated
1640 void __sk_mem_reclaim(struct sock *sk)
1642 struct proto *prot = sk->sk_prot;
1644 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1645 prot->memory_allocated);
1646 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1648 if (prot->memory_pressure && *prot->memory_pressure &&
1649 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1650 *prot->memory_pressure = 0;
1652 EXPORT_SYMBOL(__sk_mem_reclaim);
1656 * Set of default routines for initialising struct proto_ops when
1657 * the protocol does not support a particular function. In certain
1658 * cases where it makes no sense for a protocol to have a "do nothing"
1659 * function, some default processing is provided.
1662 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1666 EXPORT_SYMBOL(sock_no_bind);
1668 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1673 EXPORT_SYMBOL(sock_no_connect);
1675 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1679 EXPORT_SYMBOL(sock_no_socketpair);
1681 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1685 EXPORT_SYMBOL(sock_no_accept);
1687 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1692 EXPORT_SYMBOL(sock_no_getname);
1694 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1698 EXPORT_SYMBOL(sock_no_poll);
1700 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1704 EXPORT_SYMBOL(sock_no_ioctl);
1706 int sock_no_listen(struct socket *sock, int backlog)
1710 EXPORT_SYMBOL(sock_no_listen);
1712 int sock_no_shutdown(struct socket *sock, int how)
1716 EXPORT_SYMBOL(sock_no_shutdown);
1718 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1719 char __user *optval, unsigned int optlen)
1723 EXPORT_SYMBOL(sock_no_setsockopt);
1725 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1726 char __user *optval, int __user *optlen)
1730 EXPORT_SYMBOL(sock_no_getsockopt);
1732 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1737 EXPORT_SYMBOL(sock_no_sendmsg);
1739 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1740 size_t len, int flags)
1744 EXPORT_SYMBOL(sock_no_recvmsg);
1746 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1748 /* Mirror missing mmap method error code */
1751 EXPORT_SYMBOL(sock_no_mmap);
1753 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1756 struct msghdr msg = {.msg_flags = flags};
1758 char *kaddr = kmap(page);
1759 iov.iov_base = kaddr + offset;
1761 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1765 EXPORT_SYMBOL(sock_no_sendpage);
1768 * Default Socket Callbacks
1771 static void sock_def_wakeup(struct sock *sk)
1773 read_lock(&sk->sk_callback_lock);
1774 if (sk_has_sleeper(sk))
1775 wake_up_interruptible_all(sk->sk_sleep);
1776 read_unlock(&sk->sk_callback_lock);
1779 static void sock_def_error_report(struct sock *sk)
1781 read_lock(&sk->sk_callback_lock);
1782 if (sk_has_sleeper(sk))
1783 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1784 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1785 read_unlock(&sk->sk_callback_lock);
1788 static void sock_def_readable(struct sock *sk, int len)
1790 read_lock(&sk->sk_callback_lock);
1791 if (sk_has_sleeper(sk))
1792 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1793 POLLRDNORM | POLLRDBAND);
1794 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1795 read_unlock(&sk->sk_callback_lock);
1798 static void sock_def_write_space(struct sock *sk)
1800 read_lock(&sk->sk_callback_lock);
1802 /* Do not wake up a writer until he can make "significant"
1805 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1806 if (sk_has_sleeper(sk))
1807 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1808 POLLWRNORM | POLLWRBAND);
1810 /* Should agree with poll, otherwise some programs break */
1811 if (sock_writeable(sk))
1812 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1815 read_unlock(&sk->sk_callback_lock);
1818 static void sock_def_destruct(struct sock *sk)
1820 kfree(sk->sk_protinfo);
1823 void sk_send_sigurg(struct sock *sk)
1825 if (sk->sk_socket && sk->sk_socket->file)
1826 if (send_sigurg(&sk->sk_socket->file->f_owner))
1827 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1829 EXPORT_SYMBOL(sk_send_sigurg);
1831 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1832 unsigned long expires)
1834 if (!mod_timer(timer, expires))
1837 EXPORT_SYMBOL(sk_reset_timer);
1839 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1841 if (timer_pending(timer) && del_timer(timer))
1844 EXPORT_SYMBOL(sk_stop_timer);
1846 void sock_init_data(struct socket *sock, struct sock *sk)
1848 skb_queue_head_init(&sk->sk_receive_queue);
1849 skb_queue_head_init(&sk->sk_write_queue);
1850 skb_queue_head_init(&sk->sk_error_queue);
1851 #ifdef CONFIG_NET_DMA
1852 skb_queue_head_init(&sk->sk_async_wait_queue);
1855 sk->sk_send_head = NULL;
1857 init_timer(&sk->sk_timer);
1859 sk->sk_allocation = GFP_KERNEL;
1860 sk->sk_rcvbuf = sysctl_rmem_default;
1861 sk->sk_sndbuf = sysctl_wmem_default;
1862 sk->sk_state = TCP_CLOSE;
1863 sk_set_socket(sk, sock);
1865 sock_set_flag(sk, SOCK_ZAPPED);
1868 sk->sk_type = sock->type;
1869 sk->sk_sleep = &sock->wait;
1872 sk->sk_sleep = NULL;
1874 rwlock_init(&sk->sk_dst_lock);
1875 rwlock_init(&sk->sk_callback_lock);
1876 lockdep_set_class_and_name(&sk->sk_callback_lock,
1877 af_callback_keys + sk->sk_family,
1878 af_family_clock_key_strings[sk->sk_family]);
1880 sk->sk_state_change = sock_def_wakeup;
1881 sk->sk_data_ready = sock_def_readable;
1882 sk->sk_write_space = sock_def_write_space;
1883 sk->sk_error_report = sock_def_error_report;
1884 sk->sk_destruct = sock_def_destruct;
1886 sk->sk_sndmsg_page = NULL;
1887 sk->sk_sndmsg_off = 0;
1889 sk->sk_peercred.pid = 0;
1890 sk->sk_peercred.uid = -1;
1891 sk->sk_peercred.gid = -1;
1892 sk->sk_write_pending = 0;
1893 sk->sk_rcvlowat = 1;
1894 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1895 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1897 sk->sk_stamp = ktime_set(-1L, 0);
1900 * Before updating sk_refcnt, we must commit prior changes to memory
1901 * (Documentation/RCU/rculist_nulls.txt for details)
1904 atomic_set(&sk->sk_refcnt, 1);
1905 atomic_set(&sk->sk_drops, 0);
1907 EXPORT_SYMBOL(sock_init_data);
1909 void lock_sock_nested(struct sock *sk, int subclass)
1912 spin_lock_bh(&sk->sk_lock.slock);
1913 if (sk->sk_lock.owned)
1915 sk->sk_lock.owned = 1;
1916 spin_unlock(&sk->sk_lock.slock);
1918 * The sk_lock has mutex_lock() semantics here:
1920 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1923 EXPORT_SYMBOL(lock_sock_nested);
1925 void release_sock(struct sock *sk)
1928 * The sk_lock has mutex_unlock() semantics:
1930 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1932 spin_lock_bh(&sk->sk_lock.slock);
1933 if (sk->sk_backlog.tail)
1935 sk->sk_lock.owned = 0;
1936 if (waitqueue_active(&sk->sk_lock.wq))
1937 wake_up(&sk->sk_lock.wq);
1938 spin_unlock_bh(&sk->sk_lock.slock);
1940 EXPORT_SYMBOL(release_sock);
1942 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1945 if (!sock_flag(sk, SOCK_TIMESTAMP))
1946 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1947 tv = ktime_to_timeval(sk->sk_stamp);
1948 if (tv.tv_sec == -1)
1950 if (tv.tv_sec == 0) {
1951 sk->sk_stamp = ktime_get_real();
1952 tv = ktime_to_timeval(sk->sk_stamp);
1954 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1956 EXPORT_SYMBOL(sock_get_timestamp);
1958 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1961 if (!sock_flag(sk, SOCK_TIMESTAMP))
1962 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1963 ts = ktime_to_timespec(sk->sk_stamp);
1964 if (ts.tv_sec == -1)
1966 if (ts.tv_sec == 0) {
1967 sk->sk_stamp = ktime_get_real();
1968 ts = ktime_to_timespec(sk->sk_stamp);
1970 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1972 EXPORT_SYMBOL(sock_get_timestampns);
1974 void sock_enable_timestamp(struct sock *sk, int flag)
1976 if (!sock_flag(sk, flag)) {
1977 sock_set_flag(sk, flag);
1979 * we just set one of the two flags which require net
1980 * time stamping, but time stamping might have been on
1981 * already because of the other one
1984 flag == SOCK_TIMESTAMP ?
1985 SOCK_TIMESTAMPING_RX_SOFTWARE :
1987 net_enable_timestamp();
1992 * Get a socket option on an socket.
1994 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1995 * asynchronous errors should be reported by getsockopt. We assume
1996 * this means if you specify SO_ERROR (otherwise whats the point of it).
1998 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1999 char __user *optval, int __user *optlen)
2001 struct sock *sk = sock->sk;
2003 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2005 EXPORT_SYMBOL(sock_common_getsockopt);
2007 #ifdef CONFIG_COMPAT
2008 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2009 char __user *optval, int __user *optlen)
2011 struct sock *sk = sock->sk;
2013 if (sk->sk_prot->compat_getsockopt != NULL)
2014 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2016 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2018 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2021 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2022 struct msghdr *msg, size_t size, int flags)
2024 struct sock *sk = sock->sk;
2028 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2029 flags & ~MSG_DONTWAIT, &addr_len);
2031 msg->msg_namelen = addr_len;
2034 EXPORT_SYMBOL(sock_common_recvmsg);
2037 * Set socket options on an inet socket.
2039 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2040 char __user *optval, unsigned int optlen)
2042 struct sock *sk = sock->sk;
2044 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2046 EXPORT_SYMBOL(sock_common_setsockopt);
2048 #ifdef CONFIG_COMPAT
2049 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2050 char __user *optval, unsigned int optlen)
2052 struct sock *sk = sock->sk;
2054 if (sk->sk_prot->compat_setsockopt != NULL)
2055 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2057 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2059 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2062 void sk_common_release(struct sock *sk)
2064 if (sk->sk_prot->destroy)
2065 sk->sk_prot->destroy(sk);
2068 * Observation: when sock_common_release is called, processes have
2069 * no access to socket. But net still has.
2070 * Step one, detach it from networking:
2072 * A. Remove from hash tables.
2075 sk->sk_prot->unhash(sk);
2078 * In this point socket cannot receive new packets, but it is possible
2079 * that some packets are in flight because some CPU runs receiver and
2080 * did hash table lookup before we unhashed socket. They will achieve
2081 * receive queue and will be purged by socket destructor.
2083 * Also we still have packets pending on receive queue and probably,
2084 * our own packets waiting in device queues. sock_destroy will drain
2085 * receive queue, but transmitted packets will delay socket destruction
2086 * until the last reference will be released.
2091 xfrm_sk_free_policy(sk);
2093 sk_refcnt_debug_release(sk);
2096 EXPORT_SYMBOL(sk_common_release);
2098 static DEFINE_RWLOCK(proto_list_lock);
2099 static LIST_HEAD(proto_list);
2101 #ifdef CONFIG_PROC_FS
2102 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2104 int val[PROTO_INUSE_NR];
2107 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2109 #ifdef CONFIG_NET_NS
2110 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2112 int cpu = smp_processor_id();
2113 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2115 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2117 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2119 int cpu, idx = prot->inuse_idx;
2122 for_each_possible_cpu(cpu)
2123 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2125 return res >= 0 ? res : 0;
2127 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2129 static int sock_inuse_init_net(struct net *net)
2131 net->core.inuse = alloc_percpu(struct prot_inuse);
2132 return net->core.inuse ? 0 : -ENOMEM;
2135 static void sock_inuse_exit_net(struct net *net)
2137 free_percpu(net->core.inuse);
2140 static struct pernet_operations net_inuse_ops = {
2141 .init = sock_inuse_init_net,
2142 .exit = sock_inuse_exit_net,
2145 static __init int net_inuse_init(void)
2147 if (register_pernet_subsys(&net_inuse_ops))
2148 panic("Cannot initialize net inuse counters");
2153 core_initcall(net_inuse_init);
2155 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2157 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2159 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2161 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2163 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2165 int cpu, idx = prot->inuse_idx;
2168 for_each_possible_cpu(cpu)
2169 res += per_cpu(prot_inuse, cpu).val[idx];
2171 return res >= 0 ? res : 0;
2173 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2176 static void assign_proto_idx(struct proto *prot)
2178 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2180 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2181 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2185 set_bit(prot->inuse_idx, proto_inuse_idx);
2188 static void release_proto_idx(struct proto *prot)
2190 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2191 clear_bit(prot->inuse_idx, proto_inuse_idx);
2194 static inline void assign_proto_idx(struct proto *prot)
2198 static inline void release_proto_idx(struct proto *prot)
2203 int proto_register(struct proto *prot, int alloc_slab)
2206 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2207 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2210 if (prot->slab == NULL) {
2211 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2216 if (prot->rsk_prot != NULL) {
2217 static const char mask[] = "request_sock_%s";
2219 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2220 if (prot->rsk_prot->slab_name == NULL)
2221 goto out_free_sock_slab;
2223 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2224 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2225 prot->rsk_prot->obj_size, 0,
2226 SLAB_HWCACHE_ALIGN, NULL);
2228 if (prot->rsk_prot->slab == NULL) {
2229 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2231 goto out_free_request_sock_slab_name;
2235 if (prot->twsk_prot != NULL) {
2236 static const char mask[] = "tw_sock_%s";
2238 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2240 if (prot->twsk_prot->twsk_slab_name == NULL)
2241 goto out_free_request_sock_slab;
2243 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2244 prot->twsk_prot->twsk_slab =
2245 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2246 prot->twsk_prot->twsk_obj_size,
2248 SLAB_HWCACHE_ALIGN |
2251 if (prot->twsk_prot->twsk_slab == NULL)
2252 goto out_free_timewait_sock_slab_name;
2256 write_lock(&proto_list_lock);
2257 list_add(&prot->node, &proto_list);
2258 assign_proto_idx(prot);
2259 write_unlock(&proto_list_lock);
2262 out_free_timewait_sock_slab_name:
2263 kfree(prot->twsk_prot->twsk_slab_name);
2264 out_free_request_sock_slab:
2265 if (prot->rsk_prot && prot->rsk_prot->slab) {
2266 kmem_cache_destroy(prot->rsk_prot->slab);
2267 prot->rsk_prot->slab = NULL;
2269 out_free_request_sock_slab_name:
2270 kfree(prot->rsk_prot->slab_name);
2272 kmem_cache_destroy(prot->slab);
2277 EXPORT_SYMBOL(proto_register);
2279 void proto_unregister(struct proto *prot)
2281 write_lock(&proto_list_lock);
2282 release_proto_idx(prot);
2283 list_del(&prot->node);
2284 write_unlock(&proto_list_lock);
2286 if (prot->slab != NULL) {
2287 kmem_cache_destroy(prot->slab);
2291 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2292 kmem_cache_destroy(prot->rsk_prot->slab);
2293 kfree(prot->rsk_prot->slab_name);
2294 prot->rsk_prot->slab = NULL;
2297 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2298 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2299 kfree(prot->twsk_prot->twsk_slab_name);
2300 prot->twsk_prot->twsk_slab = NULL;
2303 EXPORT_SYMBOL(proto_unregister);
2305 #ifdef CONFIG_PROC_FS
2306 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2307 __acquires(proto_list_lock)
2309 read_lock(&proto_list_lock);
2310 return seq_list_start_head(&proto_list, *pos);
2313 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2315 return seq_list_next(v, &proto_list, pos);
2318 static void proto_seq_stop(struct seq_file *seq, void *v)
2319 __releases(proto_list_lock)
2321 read_unlock(&proto_list_lock);
2324 static char proto_method_implemented(const void *method)
2326 return method == NULL ? 'n' : 'y';
2329 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2331 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2332 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2335 sock_prot_inuse_get(seq_file_net(seq), proto),
2336 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2337 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2339 proto->slab == NULL ? "no" : "yes",
2340 module_name(proto->owner),
2341 proto_method_implemented(proto->close),
2342 proto_method_implemented(proto->connect),
2343 proto_method_implemented(proto->disconnect),
2344 proto_method_implemented(proto->accept),
2345 proto_method_implemented(proto->ioctl),
2346 proto_method_implemented(proto->init),
2347 proto_method_implemented(proto->destroy),
2348 proto_method_implemented(proto->shutdown),
2349 proto_method_implemented(proto->setsockopt),
2350 proto_method_implemented(proto->getsockopt),
2351 proto_method_implemented(proto->sendmsg),
2352 proto_method_implemented(proto->recvmsg),
2353 proto_method_implemented(proto->sendpage),
2354 proto_method_implemented(proto->bind),
2355 proto_method_implemented(proto->backlog_rcv),
2356 proto_method_implemented(proto->hash),
2357 proto_method_implemented(proto->unhash),
2358 proto_method_implemented(proto->get_port),
2359 proto_method_implemented(proto->enter_memory_pressure));
2362 static int proto_seq_show(struct seq_file *seq, void *v)
2364 if (v == &proto_list)
2365 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2374 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2376 proto_seq_printf(seq, list_entry(v, struct proto, node));
2380 static const struct seq_operations proto_seq_ops = {
2381 .start = proto_seq_start,
2382 .next = proto_seq_next,
2383 .stop = proto_seq_stop,
2384 .show = proto_seq_show,
2387 static int proto_seq_open(struct inode *inode, struct file *file)
2389 return seq_open_net(inode, file, &proto_seq_ops,
2390 sizeof(struct seq_net_private));
2393 static const struct file_operations proto_seq_fops = {
2394 .owner = THIS_MODULE,
2395 .open = proto_seq_open,
2397 .llseek = seq_lseek,
2398 .release = seq_release_net,
2401 static __net_init int proto_init_net(struct net *net)
2403 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2409 static __net_exit void proto_exit_net(struct net *net)
2411 proc_net_remove(net, "protocols");
2415 static __net_initdata struct pernet_operations proto_net_ops = {
2416 .init = proto_init_net,
2417 .exit = proto_exit_net,
2420 static int __init proto_init(void)
2422 return register_pernet_subsys(&proto_net_ops);
2425 subsys_initcall(proto_init);
2427 #endif /* PROC_FS */