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_tx_queue_clear(sk);
361 sk->sk_dst_cache = NULL;
368 EXPORT_SYMBOL(__sk_dst_check);
370 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
372 struct dst_entry *dst = sk_dst_get(sk);
374 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
382 EXPORT_SYMBOL(sk_dst_check);
384 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
386 int ret = -ENOPROTOOPT;
387 #ifdef CONFIG_NETDEVICES
388 struct net *net = sock_net(sk);
389 char devname[IFNAMSIZ];
394 if (!capable(CAP_NET_RAW))
401 /* Bind this socket to a particular device like "eth0",
402 * as specified in the passed interface name. If the
403 * name is "" or the option length is zero the socket
406 if (optlen > IFNAMSIZ - 1)
407 optlen = IFNAMSIZ - 1;
408 memset(devname, 0, sizeof(devname));
411 if (copy_from_user(devname, optval, optlen))
414 if (devname[0] == '\0') {
417 struct net_device *dev = dev_get_by_name(net, devname);
423 index = dev->ifindex;
428 sk->sk_bound_dev_if = index;
440 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
443 sock_set_flag(sk, bit);
445 sock_reset_flag(sk, bit);
449 * This is meant for all protocols to use and covers goings on
450 * at the socket level. Everything here is generic.
453 int sock_setsockopt(struct socket *sock, int level, int optname,
454 char __user *optval, unsigned int optlen)
456 struct sock *sk = sock->sk;
463 * Options without arguments
466 if (optname == SO_BINDTODEVICE)
467 return sock_bindtodevice(sk, optval, optlen);
469 if (optlen < sizeof(int))
472 if (get_user(val, (int __user *)optval))
475 valbool = val ? 1 : 0;
481 if (val && !capable(CAP_NET_ADMIN))
484 sock_valbool_flag(sk, SOCK_DBG, valbool);
487 sk->sk_reuse = valbool;
496 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
499 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
502 /* Don't error on this BSD doesn't and if you think
503 about it this is right. Otherwise apps have to
504 play 'guess the biggest size' games. RCVBUF/SNDBUF
505 are treated in BSD as hints */
507 if (val > sysctl_wmem_max)
508 val = sysctl_wmem_max;
510 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
511 if ((val * 2) < SOCK_MIN_SNDBUF)
512 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
514 sk->sk_sndbuf = val * 2;
517 * Wake up sending tasks if we
520 sk->sk_write_space(sk);
524 if (!capable(CAP_NET_ADMIN)) {
531 /* Don't error on this BSD doesn't and if you think
532 about it this is right. Otherwise apps have to
533 play 'guess the biggest size' games. RCVBUF/SNDBUF
534 are treated in BSD as hints */
536 if (val > sysctl_rmem_max)
537 val = sysctl_rmem_max;
539 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
541 * We double it on the way in to account for
542 * "struct sk_buff" etc. overhead. Applications
543 * assume that the SO_RCVBUF setting they make will
544 * allow that much actual data to be received on that
547 * Applications are unaware that "struct sk_buff" and
548 * other overheads allocate from the receive buffer
549 * during socket buffer allocation.
551 * And after considering the possible alternatives,
552 * returning the value we actually used in getsockopt
553 * is the most desirable behavior.
555 if ((val * 2) < SOCK_MIN_RCVBUF)
556 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
558 sk->sk_rcvbuf = val * 2;
562 if (!capable(CAP_NET_ADMIN)) {
570 if (sk->sk_protocol == IPPROTO_TCP)
571 tcp_set_keepalive(sk, valbool);
573 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
577 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
581 sk->sk_no_check = valbool;
585 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
586 sk->sk_priority = val;
592 if (optlen < sizeof(ling)) {
593 ret = -EINVAL; /* 1003.1g */
596 if (copy_from_user(&ling, optval, sizeof(ling))) {
601 sock_reset_flag(sk, SOCK_LINGER);
603 #if (BITS_PER_LONG == 32)
604 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
605 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
608 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
609 sock_set_flag(sk, SOCK_LINGER);
614 sock_warn_obsolete_bsdism("setsockopt");
619 set_bit(SOCK_PASSCRED, &sock->flags);
621 clear_bit(SOCK_PASSCRED, &sock->flags);
627 if (optname == SO_TIMESTAMP)
628 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
630 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
631 sock_set_flag(sk, SOCK_RCVTSTAMP);
632 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
634 sock_reset_flag(sk, SOCK_RCVTSTAMP);
635 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
639 case SO_TIMESTAMPING:
640 if (val & ~SOF_TIMESTAMPING_MASK) {
644 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
645 val & SOF_TIMESTAMPING_TX_HARDWARE);
646 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
647 val & SOF_TIMESTAMPING_TX_SOFTWARE);
648 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
649 val & SOF_TIMESTAMPING_RX_HARDWARE);
650 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
651 sock_enable_timestamp(sk,
652 SOCK_TIMESTAMPING_RX_SOFTWARE);
654 sock_disable_timestamp(sk,
655 SOCK_TIMESTAMPING_RX_SOFTWARE);
656 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
657 val & SOF_TIMESTAMPING_SOFTWARE);
658 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
659 val & SOF_TIMESTAMPING_SYS_HARDWARE);
660 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
661 val & SOF_TIMESTAMPING_RAW_HARDWARE);
667 sk->sk_rcvlowat = val ? : 1;
671 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
675 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
678 case SO_ATTACH_FILTER:
680 if (optlen == sizeof(struct sock_fprog)) {
681 struct sock_fprog fprog;
684 if (copy_from_user(&fprog, optval, sizeof(fprog)))
687 ret = sk_attach_filter(&fprog, sk);
691 case SO_DETACH_FILTER:
692 ret = sk_detach_filter(sk);
697 set_bit(SOCK_PASSSEC, &sock->flags);
699 clear_bit(SOCK_PASSSEC, &sock->flags);
702 if (!capable(CAP_NET_ADMIN))
708 /* We implement the SO_SNDLOWAT etc to
709 not be settable (1003.1g 5.3) */
712 sock_set_flag(sk, SOCK_RXQ_OVFL);
714 sock_reset_flag(sk, SOCK_RXQ_OVFL);
723 EXPORT_SYMBOL(sock_setsockopt);
726 int sock_getsockopt(struct socket *sock, int level, int optname,
727 char __user *optval, int __user *optlen)
729 struct sock *sk = sock->sk;
737 unsigned int lv = sizeof(int);
740 if (get_user(len, optlen))
745 memset(&v, 0, sizeof(v));
749 v.val = sock_flag(sk, SOCK_DBG);
753 v.val = sock_flag(sk, SOCK_LOCALROUTE);
757 v.val = !!sock_flag(sk, SOCK_BROADCAST);
761 v.val = sk->sk_sndbuf;
765 v.val = sk->sk_rcvbuf;
769 v.val = sk->sk_reuse;
773 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
781 v.val = sk->sk_protocol;
785 v.val = sk->sk_family;
789 v.val = -sock_error(sk);
791 v.val = xchg(&sk->sk_err_soft, 0);
795 v.val = !!sock_flag(sk, SOCK_URGINLINE);
799 v.val = sk->sk_no_check;
803 v.val = sk->sk_priority;
808 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
809 v.ling.l_linger = sk->sk_lingertime / HZ;
813 sock_warn_obsolete_bsdism("getsockopt");
817 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
818 !sock_flag(sk, SOCK_RCVTSTAMPNS);
822 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
825 case SO_TIMESTAMPING:
827 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
828 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
829 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
830 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
831 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
832 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
833 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
834 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
835 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
836 v.val |= SOF_TIMESTAMPING_SOFTWARE;
837 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
838 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
839 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
840 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
844 lv = sizeof(struct timeval);
845 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
849 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
850 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
855 lv = sizeof(struct timeval);
856 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
860 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
861 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
866 v.val = sk->sk_rcvlowat;
874 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
878 if (len > sizeof(sk->sk_peercred))
879 len = sizeof(sk->sk_peercred);
880 if (copy_to_user(optval, &sk->sk_peercred, len))
888 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
892 if (copy_to_user(optval, address, len))
897 /* Dubious BSD thing... Probably nobody even uses it, but
898 * the UNIX standard wants it for whatever reason... -DaveM
901 v.val = sk->sk_state == TCP_LISTEN;
905 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
909 return security_socket_getpeersec_stream(sock, optval, optlen, len);
916 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
925 if (copy_to_user(optval, &v, len))
928 if (put_user(len, optlen))
934 * Initialize an sk_lock.
936 * (We also register the sk_lock with the lock validator.)
938 static inline void sock_lock_init(struct sock *sk)
940 sock_lock_init_class_and_name(sk,
941 af_family_slock_key_strings[sk->sk_family],
942 af_family_slock_keys + sk->sk_family,
943 af_family_key_strings[sk->sk_family],
944 af_family_keys + sk->sk_family);
948 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
949 * even temporarly, because of RCU lookups. sk_node should also be left as is.
951 static void sock_copy(struct sock *nsk, const struct sock *osk)
953 #ifdef CONFIG_SECURITY_NETWORK
954 void *sptr = nsk->sk_security;
956 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
957 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
958 sizeof(osk->sk_tx_queue_mapping));
959 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
960 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
961 #ifdef CONFIG_SECURITY_NETWORK
962 nsk->sk_security = sptr;
963 security_sk_clone(osk, nsk);
967 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
971 struct kmem_cache *slab;
975 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
978 if (priority & __GFP_ZERO) {
980 * caches using SLAB_DESTROY_BY_RCU should let
981 * sk_node.next un-modified. Special care is taken
982 * when initializing object to zero.
984 if (offsetof(struct sock, sk_node.next) != 0)
985 memset(sk, 0, offsetof(struct sock, sk_node.next));
986 memset(&sk->sk_node.pprev, 0,
987 prot->obj_size - offsetof(struct sock,
992 sk = kmalloc(prot->obj_size, priority);
995 kmemcheck_annotate_bitfield(sk, flags);
997 if (security_sk_alloc(sk, family, priority))
1000 if (!try_module_get(prot->owner))
1002 sk_tx_queue_clear(sk);
1008 security_sk_free(sk);
1011 kmem_cache_free(slab, sk);
1017 static void sk_prot_free(struct proto *prot, struct sock *sk)
1019 struct kmem_cache *slab;
1020 struct module *owner;
1022 owner = prot->owner;
1025 security_sk_free(sk);
1027 kmem_cache_free(slab, sk);
1034 * sk_alloc - All socket objects are allocated here
1035 * @net: the applicable net namespace
1036 * @family: protocol family
1037 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1038 * @prot: struct proto associated with this new sock instance
1040 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1045 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1047 sk->sk_family = family;
1049 * See comment in struct sock definition to understand
1050 * why we need sk_prot_creator -acme
1052 sk->sk_prot = sk->sk_prot_creator = prot;
1054 sock_net_set(sk, get_net(net));
1055 atomic_set(&sk->sk_wmem_alloc, 1);
1060 EXPORT_SYMBOL(sk_alloc);
1062 static void __sk_free(struct sock *sk)
1064 struct sk_filter *filter;
1066 if (sk->sk_destruct)
1067 sk->sk_destruct(sk);
1069 filter = rcu_dereference(sk->sk_filter);
1071 sk_filter_uncharge(sk, filter);
1072 rcu_assign_pointer(sk->sk_filter, NULL);
1075 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1076 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1078 if (atomic_read(&sk->sk_omem_alloc))
1079 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1080 __func__, atomic_read(&sk->sk_omem_alloc));
1082 put_net(sock_net(sk));
1083 sk_prot_free(sk->sk_prot_creator, sk);
1086 void sk_free(struct sock *sk)
1089 * We substract one from sk_wmem_alloc and can know if
1090 * some packets are still in some tx queue.
1091 * If not null, sock_wfree() will call __sk_free(sk) later
1093 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1096 EXPORT_SYMBOL(sk_free);
1099 * Last sock_put should drop referrence to sk->sk_net. It has already
1100 * been dropped in sk_change_net. Taking referrence to stopping namespace
1102 * Take referrence to a socket to remove it from hash _alive_ and after that
1103 * destroy it in the context of init_net.
1105 void sk_release_kernel(struct sock *sk)
1107 if (sk == NULL || sk->sk_socket == NULL)
1111 sock_release(sk->sk_socket);
1112 release_net(sock_net(sk));
1113 sock_net_set(sk, get_net(&init_net));
1116 EXPORT_SYMBOL(sk_release_kernel);
1118 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1122 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1123 if (newsk != NULL) {
1124 struct sk_filter *filter;
1126 sock_copy(newsk, sk);
1129 get_net(sock_net(newsk));
1130 sk_node_init(&newsk->sk_node);
1131 sock_lock_init(newsk);
1132 bh_lock_sock(newsk);
1133 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1135 atomic_set(&newsk->sk_rmem_alloc, 0);
1137 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1139 atomic_set(&newsk->sk_wmem_alloc, 1);
1140 atomic_set(&newsk->sk_omem_alloc, 0);
1141 skb_queue_head_init(&newsk->sk_receive_queue);
1142 skb_queue_head_init(&newsk->sk_write_queue);
1143 #ifdef CONFIG_NET_DMA
1144 skb_queue_head_init(&newsk->sk_async_wait_queue);
1147 rwlock_init(&newsk->sk_dst_lock);
1148 rwlock_init(&newsk->sk_callback_lock);
1149 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1150 af_callback_keys + newsk->sk_family,
1151 af_family_clock_key_strings[newsk->sk_family]);
1153 newsk->sk_dst_cache = NULL;
1154 newsk->sk_wmem_queued = 0;
1155 newsk->sk_forward_alloc = 0;
1156 newsk->sk_send_head = NULL;
1157 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1159 sock_reset_flag(newsk, SOCK_DONE);
1160 skb_queue_head_init(&newsk->sk_error_queue);
1162 filter = newsk->sk_filter;
1164 sk_filter_charge(newsk, filter);
1166 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1167 /* It is still raw copy of parent, so invalidate
1168 * destructor and make plain sk_free() */
1169 newsk->sk_destruct = NULL;
1176 newsk->sk_priority = 0;
1178 * Before updating sk_refcnt, we must commit prior changes to memory
1179 * (Documentation/RCU/rculist_nulls.txt for details)
1182 atomic_set(&newsk->sk_refcnt, 2);
1185 * Increment the counter in the same struct proto as the master
1186 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1187 * is the same as sk->sk_prot->socks, as this field was copied
1190 * This _changes_ the previous behaviour, where
1191 * tcp_create_openreq_child always was incrementing the
1192 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1193 * to be taken into account in all callers. -acme
1195 sk_refcnt_debug_inc(newsk);
1196 sk_set_socket(newsk, NULL);
1197 newsk->sk_sleep = NULL;
1199 if (newsk->sk_prot->sockets_allocated)
1200 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1205 EXPORT_SYMBOL_GPL(sk_clone);
1207 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1209 __sk_dst_set(sk, dst);
1210 sk->sk_route_caps = dst->dev->features;
1211 if (sk->sk_route_caps & NETIF_F_GSO)
1212 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1213 if (sk_can_gso(sk)) {
1214 if (dst->header_len) {
1215 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1217 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1218 sk->sk_gso_max_size = dst->dev->gso_max_size;
1222 EXPORT_SYMBOL_GPL(sk_setup_caps);
1224 void __init sk_init(void)
1226 if (totalram_pages <= 4096) {
1227 sysctl_wmem_max = 32767;
1228 sysctl_rmem_max = 32767;
1229 sysctl_wmem_default = 32767;
1230 sysctl_rmem_default = 32767;
1231 } else if (totalram_pages >= 131072) {
1232 sysctl_wmem_max = 131071;
1233 sysctl_rmem_max = 131071;
1238 * Simple resource managers for sockets.
1243 * Write buffer destructor automatically called from kfree_skb.
1245 void sock_wfree(struct sk_buff *skb)
1247 struct sock *sk = skb->sk;
1248 unsigned int len = skb->truesize;
1250 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1252 * Keep a reference on sk_wmem_alloc, this will be released
1253 * after sk_write_space() call
1255 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1256 sk->sk_write_space(sk);
1260 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1261 * could not do because of in-flight packets
1263 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1266 EXPORT_SYMBOL(sock_wfree);
1269 * Read buffer destructor automatically called from kfree_skb.
1271 void sock_rfree(struct sk_buff *skb)
1273 struct sock *sk = skb->sk;
1275 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1276 sk_mem_uncharge(skb->sk, skb->truesize);
1278 EXPORT_SYMBOL(sock_rfree);
1281 int sock_i_uid(struct sock *sk)
1285 read_lock(&sk->sk_callback_lock);
1286 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1287 read_unlock(&sk->sk_callback_lock);
1290 EXPORT_SYMBOL(sock_i_uid);
1292 unsigned long sock_i_ino(struct sock *sk)
1296 read_lock(&sk->sk_callback_lock);
1297 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1298 read_unlock(&sk->sk_callback_lock);
1301 EXPORT_SYMBOL(sock_i_ino);
1304 * Allocate a skb from the socket's send buffer.
1306 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1309 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1310 struct sk_buff *skb = alloc_skb(size, priority);
1312 skb_set_owner_w(skb, sk);
1318 EXPORT_SYMBOL(sock_wmalloc);
1321 * Allocate a skb from the socket's receive buffer.
1323 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1326 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1327 struct sk_buff *skb = alloc_skb(size, priority);
1329 skb_set_owner_r(skb, sk);
1337 * Allocate a memory block from the socket's option memory buffer.
1339 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1341 if ((unsigned)size <= sysctl_optmem_max &&
1342 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1344 /* First do the add, to avoid the race if kmalloc
1347 atomic_add(size, &sk->sk_omem_alloc);
1348 mem = kmalloc(size, priority);
1351 atomic_sub(size, &sk->sk_omem_alloc);
1355 EXPORT_SYMBOL(sock_kmalloc);
1358 * Free an option memory block.
1360 void sock_kfree_s(struct sock *sk, void *mem, int size)
1363 atomic_sub(size, &sk->sk_omem_alloc);
1365 EXPORT_SYMBOL(sock_kfree_s);
1367 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1368 I think, these locks should be removed for datagram sockets.
1370 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1374 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1378 if (signal_pending(current))
1380 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1381 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1382 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1384 if (sk->sk_shutdown & SEND_SHUTDOWN)
1388 timeo = schedule_timeout(timeo);
1390 finish_wait(sk->sk_sleep, &wait);
1396 * Generic send/receive buffer handlers
1399 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1400 unsigned long data_len, int noblock,
1403 struct sk_buff *skb;
1408 gfp_mask = sk->sk_allocation;
1409 if (gfp_mask & __GFP_WAIT)
1410 gfp_mask |= __GFP_REPEAT;
1412 timeo = sock_sndtimeo(sk, noblock);
1414 err = sock_error(sk);
1419 if (sk->sk_shutdown & SEND_SHUTDOWN)
1422 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1423 skb = alloc_skb(header_len, gfp_mask);
1428 /* No pages, we're done... */
1432 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1433 skb->truesize += data_len;
1434 skb_shinfo(skb)->nr_frags = npages;
1435 for (i = 0; i < npages; i++) {
1439 page = alloc_pages(sk->sk_allocation, 0);
1442 skb_shinfo(skb)->nr_frags = i;
1447 frag = &skb_shinfo(skb)->frags[i];
1449 frag->page_offset = 0;
1450 frag->size = (data_len >= PAGE_SIZE ?
1453 data_len -= PAGE_SIZE;
1456 /* Full success... */
1462 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1463 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1467 if (signal_pending(current))
1469 timeo = sock_wait_for_wmem(sk, timeo);
1472 skb_set_owner_w(skb, sk);
1476 err = sock_intr_errno(timeo);
1481 EXPORT_SYMBOL(sock_alloc_send_pskb);
1483 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1484 int noblock, int *errcode)
1486 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1488 EXPORT_SYMBOL(sock_alloc_send_skb);
1490 static void __lock_sock(struct sock *sk)
1495 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1496 TASK_UNINTERRUPTIBLE);
1497 spin_unlock_bh(&sk->sk_lock.slock);
1499 spin_lock_bh(&sk->sk_lock.slock);
1500 if (!sock_owned_by_user(sk))
1503 finish_wait(&sk->sk_lock.wq, &wait);
1506 static void __release_sock(struct sock *sk)
1508 struct sk_buff *skb = sk->sk_backlog.head;
1511 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1515 struct sk_buff *next = skb->next;
1518 sk_backlog_rcv(sk, skb);
1521 * We are in process context here with softirqs
1522 * disabled, use cond_resched_softirq() to preempt.
1523 * This is safe to do because we've taken the backlog
1526 cond_resched_softirq();
1529 } while (skb != NULL);
1532 } while ((skb = sk->sk_backlog.head) != NULL);
1536 * sk_wait_data - wait for data to arrive at sk_receive_queue
1537 * @sk: sock to wait on
1538 * @timeo: for how long
1540 * Now socket state including sk->sk_err is changed only under lock,
1541 * hence we may omit checks after joining wait queue.
1542 * We check receive queue before schedule() only as optimization;
1543 * it is very likely that release_sock() added new data.
1545 int sk_wait_data(struct sock *sk, long *timeo)
1550 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1551 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1552 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1553 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1554 finish_wait(sk->sk_sleep, &wait);
1557 EXPORT_SYMBOL(sk_wait_data);
1560 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1562 * @size: memory size to allocate
1563 * @kind: allocation type
1565 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1566 * rmem allocation. This function assumes that protocols which have
1567 * memory_pressure use sk_wmem_queued as write buffer accounting.
1569 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1571 struct proto *prot = sk->sk_prot;
1572 int amt = sk_mem_pages(size);
1575 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1576 allocated = atomic_add_return(amt, prot->memory_allocated);
1579 if (allocated <= prot->sysctl_mem[0]) {
1580 if (prot->memory_pressure && *prot->memory_pressure)
1581 *prot->memory_pressure = 0;
1585 /* Under pressure. */
1586 if (allocated > prot->sysctl_mem[1])
1587 if (prot->enter_memory_pressure)
1588 prot->enter_memory_pressure(sk);
1590 /* Over hard limit. */
1591 if (allocated > prot->sysctl_mem[2])
1592 goto suppress_allocation;
1594 /* guarantee minimum buffer size under pressure */
1595 if (kind == SK_MEM_RECV) {
1596 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1598 } else { /* SK_MEM_SEND */
1599 if (sk->sk_type == SOCK_STREAM) {
1600 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1602 } else if (atomic_read(&sk->sk_wmem_alloc) <
1603 prot->sysctl_wmem[0])
1607 if (prot->memory_pressure) {
1610 if (!*prot->memory_pressure)
1612 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1613 if (prot->sysctl_mem[2] > alloc *
1614 sk_mem_pages(sk->sk_wmem_queued +
1615 atomic_read(&sk->sk_rmem_alloc) +
1616 sk->sk_forward_alloc))
1620 suppress_allocation:
1622 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1623 sk_stream_moderate_sndbuf(sk);
1625 /* Fail only if socket is _under_ its sndbuf.
1626 * In this case we cannot block, so that we have to fail.
1628 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1632 /* Alas. Undo changes. */
1633 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1634 atomic_sub(amt, prot->memory_allocated);
1637 EXPORT_SYMBOL(__sk_mem_schedule);
1640 * __sk_reclaim - reclaim memory_allocated
1643 void __sk_mem_reclaim(struct sock *sk)
1645 struct proto *prot = sk->sk_prot;
1647 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1648 prot->memory_allocated);
1649 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1651 if (prot->memory_pressure && *prot->memory_pressure &&
1652 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1653 *prot->memory_pressure = 0;
1655 EXPORT_SYMBOL(__sk_mem_reclaim);
1659 * Set of default routines for initialising struct proto_ops when
1660 * the protocol does not support a particular function. In certain
1661 * cases where it makes no sense for a protocol to have a "do nothing"
1662 * function, some default processing is provided.
1665 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1669 EXPORT_SYMBOL(sock_no_bind);
1671 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1676 EXPORT_SYMBOL(sock_no_connect);
1678 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1682 EXPORT_SYMBOL(sock_no_socketpair);
1684 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1688 EXPORT_SYMBOL(sock_no_accept);
1690 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1695 EXPORT_SYMBOL(sock_no_getname);
1697 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1701 EXPORT_SYMBOL(sock_no_poll);
1703 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1707 EXPORT_SYMBOL(sock_no_ioctl);
1709 int sock_no_listen(struct socket *sock, int backlog)
1713 EXPORT_SYMBOL(sock_no_listen);
1715 int sock_no_shutdown(struct socket *sock, int how)
1719 EXPORT_SYMBOL(sock_no_shutdown);
1721 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1722 char __user *optval, unsigned int optlen)
1726 EXPORT_SYMBOL(sock_no_setsockopt);
1728 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1729 char __user *optval, int __user *optlen)
1733 EXPORT_SYMBOL(sock_no_getsockopt);
1735 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1740 EXPORT_SYMBOL(sock_no_sendmsg);
1742 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1743 size_t len, int flags)
1747 EXPORT_SYMBOL(sock_no_recvmsg);
1749 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1751 /* Mirror missing mmap method error code */
1754 EXPORT_SYMBOL(sock_no_mmap);
1756 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1759 struct msghdr msg = {.msg_flags = flags};
1761 char *kaddr = kmap(page);
1762 iov.iov_base = kaddr + offset;
1764 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1768 EXPORT_SYMBOL(sock_no_sendpage);
1771 * Default Socket Callbacks
1774 static void sock_def_wakeup(struct sock *sk)
1776 read_lock(&sk->sk_callback_lock);
1777 if (sk_has_sleeper(sk))
1778 wake_up_interruptible_all(sk->sk_sleep);
1779 read_unlock(&sk->sk_callback_lock);
1782 static void sock_def_error_report(struct sock *sk)
1784 read_lock(&sk->sk_callback_lock);
1785 if (sk_has_sleeper(sk))
1786 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1787 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1788 read_unlock(&sk->sk_callback_lock);
1791 static void sock_def_readable(struct sock *sk, int len)
1793 read_lock(&sk->sk_callback_lock);
1794 if (sk_has_sleeper(sk))
1795 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1796 POLLRDNORM | POLLRDBAND);
1797 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1798 read_unlock(&sk->sk_callback_lock);
1801 static void sock_def_write_space(struct sock *sk)
1803 read_lock(&sk->sk_callback_lock);
1805 /* Do not wake up a writer until he can make "significant"
1808 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1809 if (sk_has_sleeper(sk))
1810 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1811 POLLWRNORM | POLLWRBAND);
1813 /* Should agree with poll, otherwise some programs break */
1814 if (sock_writeable(sk))
1815 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1818 read_unlock(&sk->sk_callback_lock);
1821 static void sock_def_destruct(struct sock *sk)
1823 kfree(sk->sk_protinfo);
1826 void sk_send_sigurg(struct sock *sk)
1828 if (sk->sk_socket && sk->sk_socket->file)
1829 if (send_sigurg(&sk->sk_socket->file->f_owner))
1830 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1832 EXPORT_SYMBOL(sk_send_sigurg);
1834 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1835 unsigned long expires)
1837 if (!mod_timer(timer, expires))
1840 EXPORT_SYMBOL(sk_reset_timer);
1842 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1844 if (timer_pending(timer) && del_timer(timer))
1847 EXPORT_SYMBOL(sk_stop_timer);
1849 void sock_init_data(struct socket *sock, struct sock *sk)
1851 skb_queue_head_init(&sk->sk_receive_queue);
1852 skb_queue_head_init(&sk->sk_write_queue);
1853 skb_queue_head_init(&sk->sk_error_queue);
1854 #ifdef CONFIG_NET_DMA
1855 skb_queue_head_init(&sk->sk_async_wait_queue);
1858 sk->sk_send_head = NULL;
1860 init_timer(&sk->sk_timer);
1862 sk->sk_allocation = GFP_KERNEL;
1863 sk->sk_rcvbuf = sysctl_rmem_default;
1864 sk->sk_sndbuf = sysctl_wmem_default;
1865 sk->sk_state = TCP_CLOSE;
1866 sk_set_socket(sk, sock);
1868 sock_set_flag(sk, SOCK_ZAPPED);
1871 sk->sk_type = sock->type;
1872 sk->sk_sleep = &sock->wait;
1875 sk->sk_sleep = NULL;
1877 rwlock_init(&sk->sk_dst_lock);
1878 rwlock_init(&sk->sk_callback_lock);
1879 lockdep_set_class_and_name(&sk->sk_callback_lock,
1880 af_callback_keys + sk->sk_family,
1881 af_family_clock_key_strings[sk->sk_family]);
1883 sk->sk_state_change = sock_def_wakeup;
1884 sk->sk_data_ready = sock_def_readable;
1885 sk->sk_write_space = sock_def_write_space;
1886 sk->sk_error_report = sock_def_error_report;
1887 sk->sk_destruct = sock_def_destruct;
1889 sk->sk_sndmsg_page = NULL;
1890 sk->sk_sndmsg_off = 0;
1892 sk->sk_peercred.pid = 0;
1893 sk->sk_peercred.uid = -1;
1894 sk->sk_peercred.gid = -1;
1895 sk->sk_write_pending = 0;
1896 sk->sk_rcvlowat = 1;
1897 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1898 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1900 sk->sk_stamp = ktime_set(-1L, 0);
1903 * Before updating sk_refcnt, we must commit prior changes to memory
1904 * (Documentation/RCU/rculist_nulls.txt for details)
1907 atomic_set(&sk->sk_refcnt, 1);
1908 atomic_set(&sk->sk_drops, 0);
1910 EXPORT_SYMBOL(sock_init_data);
1912 void lock_sock_nested(struct sock *sk, int subclass)
1915 spin_lock_bh(&sk->sk_lock.slock);
1916 if (sk->sk_lock.owned)
1918 sk->sk_lock.owned = 1;
1919 spin_unlock(&sk->sk_lock.slock);
1921 * The sk_lock has mutex_lock() semantics here:
1923 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1926 EXPORT_SYMBOL(lock_sock_nested);
1928 void release_sock(struct sock *sk)
1931 * The sk_lock has mutex_unlock() semantics:
1933 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1935 spin_lock_bh(&sk->sk_lock.slock);
1936 if (sk->sk_backlog.tail)
1938 sk->sk_lock.owned = 0;
1939 if (waitqueue_active(&sk->sk_lock.wq))
1940 wake_up(&sk->sk_lock.wq);
1941 spin_unlock_bh(&sk->sk_lock.slock);
1943 EXPORT_SYMBOL(release_sock);
1945 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1948 if (!sock_flag(sk, SOCK_TIMESTAMP))
1949 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1950 tv = ktime_to_timeval(sk->sk_stamp);
1951 if (tv.tv_sec == -1)
1953 if (tv.tv_sec == 0) {
1954 sk->sk_stamp = ktime_get_real();
1955 tv = ktime_to_timeval(sk->sk_stamp);
1957 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1959 EXPORT_SYMBOL(sock_get_timestamp);
1961 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1964 if (!sock_flag(sk, SOCK_TIMESTAMP))
1965 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1966 ts = ktime_to_timespec(sk->sk_stamp);
1967 if (ts.tv_sec == -1)
1969 if (ts.tv_sec == 0) {
1970 sk->sk_stamp = ktime_get_real();
1971 ts = ktime_to_timespec(sk->sk_stamp);
1973 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1975 EXPORT_SYMBOL(sock_get_timestampns);
1977 void sock_enable_timestamp(struct sock *sk, int flag)
1979 if (!sock_flag(sk, flag)) {
1980 sock_set_flag(sk, flag);
1982 * we just set one of the two flags which require net
1983 * time stamping, but time stamping might have been on
1984 * already because of the other one
1987 flag == SOCK_TIMESTAMP ?
1988 SOCK_TIMESTAMPING_RX_SOFTWARE :
1990 net_enable_timestamp();
1995 * Get a socket option on an socket.
1997 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1998 * asynchronous errors should be reported by getsockopt. We assume
1999 * this means if you specify SO_ERROR (otherwise whats the point of it).
2001 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2002 char __user *optval, int __user *optlen)
2004 struct sock *sk = sock->sk;
2006 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2008 EXPORT_SYMBOL(sock_common_getsockopt);
2010 #ifdef CONFIG_COMPAT
2011 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2012 char __user *optval, int __user *optlen)
2014 struct sock *sk = sock->sk;
2016 if (sk->sk_prot->compat_getsockopt != NULL)
2017 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2019 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2021 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2024 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2025 struct msghdr *msg, size_t size, int flags)
2027 struct sock *sk = sock->sk;
2031 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2032 flags & ~MSG_DONTWAIT, &addr_len);
2034 msg->msg_namelen = addr_len;
2037 EXPORT_SYMBOL(sock_common_recvmsg);
2040 * Set socket options on an inet socket.
2042 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2043 char __user *optval, unsigned int optlen)
2045 struct sock *sk = sock->sk;
2047 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2049 EXPORT_SYMBOL(sock_common_setsockopt);
2051 #ifdef CONFIG_COMPAT
2052 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2053 char __user *optval, unsigned int optlen)
2055 struct sock *sk = sock->sk;
2057 if (sk->sk_prot->compat_setsockopt != NULL)
2058 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2060 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2062 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2065 void sk_common_release(struct sock *sk)
2067 if (sk->sk_prot->destroy)
2068 sk->sk_prot->destroy(sk);
2071 * Observation: when sock_common_release is called, processes have
2072 * no access to socket. But net still has.
2073 * Step one, detach it from networking:
2075 * A. Remove from hash tables.
2078 sk->sk_prot->unhash(sk);
2081 * In this point socket cannot receive new packets, but it is possible
2082 * that some packets are in flight because some CPU runs receiver and
2083 * did hash table lookup before we unhashed socket. They will achieve
2084 * receive queue and will be purged by socket destructor.
2086 * Also we still have packets pending on receive queue and probably,
2087 * our own packets waiting in device queues. sock_destroy will drain
2088 * receive queue, but transmitted packets will delay socket destruction
2089 * until the last reference will be released.
2094 xfrm_sk_free_policy(sk);
2096 sk_refcnt_debug_release(sk);
2099 EXPORT_SYMBOL(sk_common_release);
2101 static DEFINE_RWLOCK(proto_list_lock);
2102 static LIST_HEAD(proto_list);
2104 #ifdef CONFIG_PROC_FS
2105 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2107 int val[PROTO_INUSE_NR];
2110 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2112 #ifdef CONFIG_NET_NS
2113 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2115 int cpu = smp_processor_id();
2116 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2118 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2120 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2122 int cpu, idx = prot->inuse_idx;
2125 for_each_possible_cpu(cpu)
2126 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2128 return res >= 0 ? res : 0;
2130 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2132 static int sock_inuse_init_net(struct net *net)
2134 net->core.inuse = alloc_percpu(struct prot_inuse);
2135 return net->core.inuse ? 0 : -ENOMEM;
2138 static void sock_inuse_exit_net(struct net *net)
2140 free_percpu(net->core.inuse);
2143 static struct pernet_operations net_inuse_ops = {
2144 .init = sock_inuse_init_net,
2145 .exit = sock_inuse_exit_net,
2148 static __init int net_inuse_init(void)
2150 if (register_pernet_subsys(&net_inuse_ops))
2151 panic("Cannot initialize net inuse counters");
2156 core_initcall(net_inuse_init);
2158 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2160 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2162 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2164 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2166 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2168 int cpu, idx = prot->inuse_idx;
2171 for_each_possible_cpu(cpu)
2172 res += per_cpu(prot_inuse, cpu).val[idx];
2174 return res >= 0 ? res : 0;
2176 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2179 static void assign_proto_idx(struct proto *prot)
2181 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2183 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2184 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2188 set_bit(prot->inuse_idx, proto_inuse_idx);
2191 static void release_proto_idx(struct proto *prot)
2193 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2194 clear_bit(prot->inuse_idx, proto_inuse_idx);
2197 static inline void assign_proto_idx(struct proto *prot)
2201 static inline void release_proto_idx(struct proto *prot)
2206 int proto_register(struct proto *prot, int alloc_slab)
2209 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2210 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2213 if (prot->slab == NULL) {
2214 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2219 if (prot->rsk_prot != NULL) {
2220 static const char mask[] = "request_sock_%s";
2222 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2223 if (prot->rsk_prot->slab_name == NULL)
2224 goto out_free_sock_slab;
2226 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2227 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2228 prot->rsk_prot->obj_size, 0,
2229 SLAB_HWCACHE_ALIGN, NULL);
2231 if (prot->rsk_prot->slab == NULL) {
2232 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2234 goto out_free_request_sock_slab_name;
2238 if (prot->twsk_prot != NULL) {
2239 static const char mask[] = "tw_sock_%s";
2241 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2243 if (prot->twsk_prot->twsk_slab_name == NULL)
2244 goto out_free_request_sock_slab;
2246 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2247 prot->twsk_prot->twsk_slab =
2248 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2249 prot->twsk_prot->twsk_obj_size,
2251 SLAB_HWCACHE_ALIGN |
2254 if (prot->twsk_prot->twsk_slab == NULL)
2255 goto out_free_timewait_sock_slab_name;
2259 write_lock(&proto_list_lock);
2260 list_add(&prot->node, &proto_list);
2261 assign_proto_idx(prot);
2262 write_unlock(&proto_list_lock);
2265 out_free_timewait_sock_slab_name:
2266 kfree(prot->twsk_prot->twsk_slab_name);
2267 out_free_request_sock_slab:
2268 if (prot->rsk_prot && prot->rsk_prot->slab) {
2269 kmem_cache_destroy(prot->rsk_prot->slab);
2270 prot->rsk_prot->slab = NULL;
2272 out_free_request_sock_slab_name:
2273 kfree(prot->rsk_prot->slab_name);
2275 kmem_cache_destroy(prot->slab);
2280 EXPORT_SYMBOL(proto_register);
2282 void proto_unregister(struct proto *prot)
2284 write_lock(&proto_list_lock);
2285 release_proto_idx(prot);
2286 list_del(&prot->node);
2287 write_unlock(&proto_list_lock);
2289 if (prot->slab != NULL) {
2290 kmem_cache_destroy(prot->slab);
2294 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2295 kmem_cache_destroy(prot->rsk_prot->slab);
2296 kfree(prot->rsk_prot->slab_name);
2297 prot->rsk_prot->slab = NULL;
2300 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2301 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2302 kfree(prot->twsk_prot->twsk_slab_name);
2303 prot->twsk_prot->twsk_slab = NULL;
2306 EXPORT_SYMBOL(proto_unregister);
2308 #ifdef CONFIG_PROC_FS
2309 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2310 __acquires(proto_list_lock)
2312 read_lock(&proto_list_lock);
2313 return seq_list_start_head(&proto_list, *pos);
2316 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2318 return seq_list_next(v, &proto_list, pos);
2321 static void proto_seq_stop(struct seq_file *seq, void *v)
2322 __releases(proto_list_lock)
2324 read_unlock(&proto_list_lock);
2327 static char proto_method_implemented(const void *method)
2329 return method == NULL ? 'n' : 'y';
2332 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2334 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2335 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2338 sock_prot_inuse_get(seq_file_net(seq), proto),
2339 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2340 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2342 proto->slab == NULL ? "no" : "yes",
2343 module_name(proto->owner),
2344 proto_method_implemented(proto->close),
2345 proto_method_implemented(proto->connect),
2346 proto_method_implemented(proto->disconnect),
2347 proto_method_implemented(proto->accept),
2348 proto_method_implemented(proto->ioctl),
2349 proto_method_implemented(proto->init),
2350 proto_method_implemented(proto->destroy),
2351 proto_method_implemented(proto->shutdown),
2352 proto_method_implemented(proto->setsockopt),
2353 proto_method_implemented(proto->getsockopt),
2354 proto_method_implemented(proto->sendmsg),
2355 proto_method_implemented(proto->recvmsg),
2356 proto_method_implemented(proto->sendpage),
2357 proto_method_implemented(proto->bind),
2358 proto_method_implemented(proto->backlog_rcv),
2359 proto_method_implemented(proto->hash),
2360 proto_method_implemented(proto->unhash),
2361 proto_method_implemented(proto->get_port),
2362 proto_method_implemented(proto->enter_memory_pressure));
2365 static int proto_seq_show(struct seq_file *seq, void *v)
2367 if (v == &proto_list)
2368 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2377 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2379 proto_seq_printf(seq, list_entry(v, struct proto, node));
2383 static const struct seq_operations proto_seq_ops = {
2384 .start = proto_seq_start,
2385 .next = proto_seq_next,
2386 .stop = proto_seq_stop,
2387 .show = proto_seq_show,
2390 static int proto_seq_open(struct inode *inode, struct file *file)
2392 return seq_open_net(inode, file, &proto_seq_ops,
2393 sizeof(struct seq_net_private));
2396 static const struct file_operations proto_seq_fops = {
2397 .owner = THIS_MODULE,
2398 .open = proto_seq_open,
2400 .llseek = seq_lseek,
2401 .release = seq_release_net,
2404 static __net_init int proto_init_net(struct net *net)
2406 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2412 static __net_exit void proto_exit_net(struct net *net)
2414 proc_net_remove(net, "protocols");
2418 static __net_initdata struct pernet_operations proto_net_ops = {
2419 .init = proto_init_net,
2420 .exit = proto_exit_net,
2423 static int __init proto_init(void)
2425 return register_pernet_subsys(&proto_net_ops);
2428 subsys_initcall(proto_init);
2430 #endif /* PROC_FS */