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 *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" ,
160 static const char *af_family_slock_key_strings[AF_MAX+1] = {
161 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
162 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
163 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
164 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
165 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
166 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
167 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
168 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
169 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
170 "slock-27" , "slock-28" , "slock-AF_CAN" ,
171 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
172 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
175 static const char *af_family_clock_key_strings[AF_MAX+1] = {
176 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
177 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
178 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
179 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
180 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
181 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
182 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
183 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
184 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
185 "clock-27" , "clock-28" , "clock-AF_CAN" ,
186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
187 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
192 * sk_callback_lock locking rules are per-address-family,
193 * so split the lock classes by using a per-AF key:
195 static struct lock_class_key af_callback_keys[AF_MAX];
197 /* Take into consideration the size of the struct sk_buff overhead in the
198 * determination of these values, since that is non-constant across
199 * platforms. This makes socket queueing behavior and performance
200 * not depend upon such differences.
202 #define _SK_MEM_PACKETS 256
203 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
204 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
207 /* Run time adjustable parameters. */
208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
213 /* Maximal space eaten by iovec or ancilliary data plus some space */
214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
215 EXPORT_SYMBOL(sysctl_optmem_max);
217 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
221 if (optlen < sizeof(tv))
223 if (copy_from_user(&tv, optval, sizeof(tv)))
225 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
229 static int warned __read_mostly;
232 if (warned < 10 && net_ratelimit()) {
234 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
235 "tries to set negative timeout\n",
236 current->comm, task_pid_nr(current));
240 *timeo_p = MAX_SCHEDULE_TIMEOUT;
241 if (tv.tv_sec == 0 && tv.tv_usec == 0)
243 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
244 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
248 static void sock_warn_obsolete_bsdism(const char *name)
251 static char warncomm[TASK_COMM_LEN];
252 if (strcmp(warncomm, current->comm) && warned < 5) {
253 strcpy(warncomm, current->comm);
254 printk(KERN_WARNING "process `%s' is using obsolete "
255 "%s SO_BSDCOMPAT\n", warncomm, name);
260 static void sock_disable_timestamp(struct sock *sk, int flag)
262 if (sock_flag(sk, flag)) {
263 sock_reset_flag(sk, flag);
264 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
265 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
266 net_disable_timestamp();
272 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
277 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
278 number of warnings when compiling with -W --ANK
280 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
281 (unsigned)sk->sk_rcvbuf) {
286 err = sk_filter(sk, skb);
290 if (!sk_rmem_schedule(sk, skb->truesize)) {
296 skb_set_owner_r(skb, sk);
298 /* Cache the SKB length before we tack it onto the receive
299 * queue. Once it is added it no longer belongs to us and
300 * may be freed by other threads of control pulling packets
305 skb_queue_tail(&sk->sk_receive_queue, skb);
307 if (!sock_flag(sk, SOCK_DEAD))
308 sk->sk_data_ready(sk, skb_len);
312 EXPORT_SYMBOL(sock_queue_rcv_skb);
314 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
316 int rc = NET_RX_SUCCESS;
318 if (sk_filter(sk, skb))
319 goto discard_and_relse;
324 bh_lock_sock_nested(sk);
327 if (!sock_owned_by_user(sk)) {
329 * trylock + unlock semantics:
331 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
333 rc = sk_backlog_rcv(sk, skb);
335 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
337 sk_add_backlog(sk, skb);
346 EXPORT_SYMBOL(sk_receive_skb);
348 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
350 struct dst_entry *dst = sk->sk_dst_cache;
352 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
353 sk->sk_dst_cache = NULL;
360 EXPORT_SYMBOL(__sk_dst_check);
362 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
364 struct dst_entry *dst = sk_dst_get(sk);
366 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
374 EXPORT_SYMBOL(sk_dst_check);
376 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
378 int ret = -ENOPROTOOPT;
379 #ifdef CONFIG_NETDEVICES
380 struct net *net = sock_net(sk);
381 char devname[IFNAMSIZ];
386 if (!capable(CAP_NET_RAW))
393 /* Bind this socket to a particular device like "eth0",
394 * as specified in the passed interface name. If the
395 * name is "" or the option length is zero the socket
398 if (optlen > IFNAMSIZ - 1)
399 optlen = IFNAMSIZ - 1;
400 memset(devname, 0, sizeof(devname));
403 if (copy_from_user(devname, optval, optlen))
406 if (devname[0] == '\0') {
409 struct net_device *dev = dev_get_by_name(net, devname);
415 index = dev->ifindex;
420 sk->sk_bound_dev_if = index;
432 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
435 sock_set_flag(sk, bit);
437 sock_reset_flag(sk, bit);
441 * This is meant for all protocols to use and covers goings on
442 * at the socket level. Everything here is generic.
445 int sock_setsockopt(struct socket *sock, int level, int optname,
446 char __user *optval, int optlen)
448 struct sock *sk = sock->sk;
455 * Options without arguments
458 if (optname == SO_BINDTODEVICE)
459 return sock_bindtodevice(sk, optval, optlen);
461 if (optlen < sizeof(int))
464 if (get_user(val, (int __user *)optval))
467 valbool = val ? 1 : 0;
473 if (val && !capable(CAP_NET_ADMIN))
476 sock_valbool_flag(sk, SOCK_DBG, valbool);
479 sk->sk_reuse = valbool;
486 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
489 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
492 /* Don't error on this BSD doesn't and if you think
493 about it this is right. Otherwise apps have to
494 play 'guess the biggest size' games. RCVBUF/SNDBUF
495 are treated in BSD as hints */
497 if (val > sysctl_wmem_max)
498 val = sysctl_wmem_max;
500 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
501 if ((val * 2) < SOCK_MIN_SNDBUF)
502 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
504 sk->sk_sndbuf = val * 2;
507 * Wake up sending tasks if we
510 sk->sk_write_space(sk);
514 if (!capable(CAP_NET_ADMIN)) {
521 /* Don't error on this BSD doesn't and if you think
522 about it this is right. Otherwise apps have to
523 play 'guess the biggest size' games. RCVBUF/SNDBUF
524 are treated in BSD as hints */
526 if (val > sysctl_rmem_max)
527 val = sysctl_rmem_max;
529 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
531 * We double it on the way in to account for
532 * "struct sk_buff" etc. overhead. Applications
533 * assume that the SO_RCVBUF setting they make will
534 * allow that much actual data to be received on that
537 * Applications are unaware that "struct sk_buff" and
538 * other overheads allocate from the receive buffer
539 * during socket buffer allocation.
541 * And after considering the possible alternatives,
542 * returning the value we actually used in getsockopt
543 * is the most desirable behavior.
545 if ((val * 2) < SOCK_MIN_RCVBUF)
546 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
548 sk->sk_rcvbuf = val * 2;
552 if (!capable(CAP_NET_ADMIN)) {
560 if (sk->sk_protocol == IPPROTO_TCP)
561 tcp_set_keepalive(sk, valbool);
563 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
567 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
571 sk->sk_no_check = valbool;
575 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
576 sk->sk_priority = val;
582 if (optlen < sizeof(ling)) {
583 ret = -EINVAL; /* 1003.1g */
586 if (copy_from_user(&ling, optval, sizeof(ling))) {
591 sock_reset_flag(sk, SOCK_LINGER);
593 #if (BITS_PER_LONG == 32)
594 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
595 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
598 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
599 sock_set_flag(sk, SOCK_LINGER);
604 sock_warn_obsolete_bsdism("setsockopt");
609 set_bit(SOCK_PASSCRED, &sock->flags);
611 clear_bit(SOCK_PASSCRED, &sock->flags);
617 if (optname == SO_TIMESTAMP)
618 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
620 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
621 sock_set_flag(sk, SOCK_RCVTSTAMP);
622 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
624 sock_reset_flag(sk, SOCK_RCVTSTAMP);
625 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
629 case SO_TIMESTAMPING:
630 if (val & ~SOF_TIMESTAMPING_MASK) {
634 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
635 val & SOF_TIMESTAMPING_TX_HARDWARE);
636 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
637 val & SOF_TIMESTAMPING_TX_SOFTWARE);
638 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
639 val & SOF_TIMESTAMPING_RX_HARDWARE);
640 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
641 sock_enable_timestamp(sk,
642 SOCK_TIMESTAMPING_RX_SOFTWARE);
644 sock_disable_timestamp(sk,
645 SOCK_TIMESTAMPING_RX_SOFTWARE);
646 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
647 val & SOF_TIMESTAMPING_SOFTWARE);
648 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
649 val & SOF_TIMESTAMPING_SYS_HARDWARE);
650 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
651 val & SOF_TIMESTAMPING_RAW_HARDWARE);
657 sk->sk_rcvlowat = val ? : 1;
661 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
665 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
668 case SO_ATTACH_FILTER:
670 if (optlen == sizeof(struct sock_fprog)) {
671 struct sock_fprog fprog;
674 if (copy_from_user(&fprog, optval, sizeof(fprog)))
677 ret = sk_attach_filter(&fprog, sk);
681 case SO_DETACH_FILTER:
682 ret = sk_detach_filter(sk);
687 set_bit(SOCK_PASSSEC, &sock->flags);
689 clear_bit(SOCK_PASSSEC, &sock->flags);
692 if (!capable(CAP_NET_ADMIN))
698 /* We implement the SO_SNDLOWAT etc to
699 not be settable (1003.1g 5.3) */
707 EXPORT_SYMBOL(sock_setsockopt);
710 int sock_getsockopt(struct socket *sock, int level, int optname,
711 char __user *optval, int __user *optlen)
713 struct sock *sk = sock->sk;
721 unsigned int lv = sizeof(int);
724 if (get_user(len, optlen))
729 memset(&v, 0, sizeof(v));
733 v.val = sock_flag(sk, SOCK_DBG);
737 v.val = sock_flag(sk, SOCK_LOCALROUTE);
741 v.val = !!sock_flag(sk, SOCK_BROADCAST);
745 v.val = sk->sk_sndbuf;
749 v.val = sk->sk_rcvbuf;
753 v.val = sk->sk_reuse;
757 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
765 v.val = -sock_error(sk);
767 v.val = xchg(&sk->sk_err_soft, 0);
771 v.val = !!sock_flag(sk, SOCK_URGINLINE);
775 v.val = sk->sk_no_check;
779 v.val = sk->sk_priority;
784 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
785 v.ling.l_linger = sk->sk_lingertime / HZ;
789 sock_warn_obsolete_bsdism("getsockopt");
793 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
794 !sock_flag(sk, SOCK_RCVTSTAMPNS);
798 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
801 case SO_TIMESTAMPING:
803 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
804 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
805 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
806 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
807 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
808 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
809 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
810 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
811 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
812 v.val |= SOF_TIMESTAMPING_SOFTWARE;
813 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
814 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
815 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
816 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
820 lv = sizeof(struct timeval);
821 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
825 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
826 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
831 lv = sizeof(struct timeval);
832 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
836 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
837 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
842 v.val = sk->sk_rcvlowat;
850 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
854 if (len > sizeof(sk->sk_peercred))
855 len = sizeof(sk->sk_peercred);
856 if (copy_to_user(optval, &sk->sk_peercred, len))
864 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
868 if (copy_to_user(optval, address, len))
873 /* Dubious BSD thing... Probably nobody even uses it, but
874 * the UNIX standard wants it for whatever reason... -DaveM
877 v.val = sk->sk_state == TCP_LISTEN;
881 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
885 return security_socket_getpeersec_stream(sock, optval, optlen, len);
897 if (copy_to_user(optval, &v, len))
900 if (put_user(len, optlen))
906 * Initialize an sk_lock.
908 * (We also register the sk_lock with the lock validator.)
910 static inline void sock_lock_init(struct sock *sk)
912 sock_lock_init_class_and_name(sk,
913 af_family_slock_key_strings[sk->sk_family],
914 af_family_slock_keys + sk->sk_family,
915 af_family_key_strings[sk->sk_family],
916 af_family_keys + sk->sk_family);
919 static void sock_copy(struct sock *nsk, const struct sock *osk)
921 #ifdef CONFIG_SECURITY_NETWORK
922 void *sptr = nsk->sk_security;
925 memcpy(nsk, osk, osk->sk_prot->obj_size);
926 #ifdef CONFIG_SECURITY_NETWORK
927 nsk->sk_security = sptr;
928 security_sk_clone(osk, nsk);
932 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
936 struct kmem_cache *slab;
940 sk = kmem_cache_alloc(slab, priority);
942 sk = kmalloc(prot->obj_size, priority);
945 if (security_sk_alloc(sk, family, priority))
948 if (!try_module_get(prot->owner))
955 security_sk_free(sk);
958 kmem_cache_free(slab, sk);
964 static void sk_prot_free(struct proto *prot, struct sock *sk)
966 struct kmem_cache *slab;
967 struct module *owner;
972 security_sk_free(sk);
974 kmem_cache_free(slab, sk);
981 * sk_alloc - All socket objects are allocated here
982 * @net: the applicable net namespace
983 * @family: protocol family
984 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
985 * @prot: struct proto associated with this new sock instance
987 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
992 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
994 sk->sk_family = family;
996 * See comment in struct sock definition to understand
997 * why we need sk_prot_creator -acme
999 sk->sk_prot = sk->sk_prot_creator = prot;
1001 sock_net_set(sk, get_net(net));
1006 EXPORT_SYMBOL(sk_alloc);
1008 void sk_free(struct sock *sk)
1010 struct sk_filter *filter;
1012 if (sk->sk_destruct)
1013 sk->sk_destruct(sk);
1015 filter = rcu_dereference(sk->sk_filter);
1017 sk_filter_uncharge(sk, filter);
1018 rcu_assign_pointer(sk->sk_filter, NULL);
1021 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1022 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1024 if (atomic_read(&sk->sk_omem_alloc))
1025 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1026 __func__, atomic_read(&sk->sk_omem_alloc));
1028 put_net(sock_net(sk));
1029 sk_prot_free(sk->sk_prot_creator, sk);
1031 EXPORT_SYMBOL(sk_free);
1034 * Last sock_put should drop referrence to sk->sk_net. It has already
1035 * been dropped in sk_change_net. Taking referrence to stopping namespace
1037 * Take referrence to a socket to remove it from hash _alive_ and after that
1038 * destroy it in the context of init_net.
1040 void sk_release_kernel(struct sock *sk)
1042 if (sk == NULL || sk->sk_socket == NULL)
1046 sock_release(sk->sk_socket);
1047 release_net(sock_net(sk));
1048 sock_net_set(sk, get_net(&init_net));
1051 EXPORT_SYMBOL(sk_release_kernel);
1053 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1057 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1058 if (newsk != NULL) {
1059 struct sk_filter *filter;
1061 sock_copy(newsk, sk);
1064 get_net(sock_net(newsk));
1065 sk_node_init(&newsk->sk_node);
1066 sock_lock_init(newsk);
1067 bh_lock_sock(newsk);
1068 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1070 atomic_set(&newsk->sk_rmem_alloc, 0);
1071 atomic_set(&newsk->sk_wmem_alloc, 0);
1072 atomic_set(&newsk->sk_omem_alloc, 0);
1073 skb_queue_head_init(&newsk->sk_receive_queue);
1074 skb_queue_head_init(&newsk->sk_write_queue);
1075 #ifdef CONFIG_NET_DMA
1076 skb_queue_head_init(&newsk->sk_async_wait_queue);
1079 rwlock_init(&newsk->sk_dst_lock);
1080 rwlock_init(&newsk->sk_callback_lock);
1081 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1082 af_callback_keys + newsk->sk_family,
1083 af_family_clock_key_strings[newsk->sk_family]);
1085 newsk->sk_dst_cache = NULL;
1086 newsk->sk_wmem_queued = 0;
1087 newsk->sk_forward_alloc = 0;
1088 newsk->sk_send_head = NULL;
1089 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1091 sock_reset_flag(newsk, SOCK_DONE);
1092 skb_queue_head_init(&newsk->sk_error_queue);
1094 filter = newsk->sk_filter;
1096 sk_filter_charge(newsk, filter);
1098 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1099 /* It is still raw copy of parent, so invalidate
1100 * destructor and make plain sk_free() */
1101 newsk->sk_destruct = NULL;
1108 newsk->sk_priority = 0;
1109 atomic_set(&newsk->sk_refcnt, 2);
1112 * Increment the counter in the same struct proto as the master
1113 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1114 * is the same as sk->sk_prot->socks, as this field was copied
1117 * This _changes_ the previous behaviour, where
1118 * tcp_create_openreq_child always was incrementing the
1119 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1120 * to be taken into account in all callers. -acme
1122 sk_refcnt_debug_inc(newsk);
1123 sk_set_socket(newsk, NULL);
1124 newsk->sk_sleep = NULL;
1126 if (newsk->sk_prot->sockets_allocated)
1127 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1132 EXPORT_SYMBOL_GPL(sk_clone);
1134 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1136 __sk_dst_set(sk, dst);
1137 sk->sk_route_caps = dst->dev->features;
1138 if (sk->sk_route_caps & NETIF_F_GSO)
1139 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1140 if (sk_can_gso(sk)) {
1141 if (dst->header_len) {
1142 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1144 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1145 sk->sk_gso_max_size = dst->dev->gso_max_size;
1149 EXPORT_SYMBOL_GPL(sk_setup_caps);
1151 void __init sk_init(void)
1153 if (num_physpages <= 4096) {
1154 sysctl_wmem_max = 32767;
1155 sysctl_rmem_max = 32767;
1156 sysctl_wmem_default = 32767;
1157 sysctl_rmem_default = 32767;
1158 } else if (num_physpages >= 131072) {
1159 sysctl_wmem_max = 131071;
1160 sysctl_rmem_max = 131071;
1165 * Simple resource managers for sockets.
1170 * Write buffer destructor automatically called from kfree_skb.
1172 void sock_wfree(struct sk_buff *skb)
1174 struct sock *sk = skb->sk;
1176 /* In case it might be waiting for more memory. */
1177 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1178 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1179 sk->sk_write_space(sk);
1182 EXPORT_SYMBOL(sock_wfree);
1185 * Read buffer destructor automatically called from kfree_skb.
1187 void sock_rfree(struct sk_buff *skb)
1189 struct sock *sk = skb->sk;
1191 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1192 sk_mem_uncharge(skb->sk, skb->truesize);
1194 EXPORT_SYMBOL(sock_rfree);
1197 int sock_i_uid(struct sock *sk)
1201 read_lock(&sk->sk_callback_lock);
1202 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1203 read_unlock(&sk->sk_callback_lock);
1206 EXPORT_SYMBOL(sock_i_uid);
1208 unsigned long sock_i_ino(struct sock *sk)
1212 read_lock(&sk->sk_callback_lock);
1213 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1214 read_unlock(&sk->sk_callback_lock);
1217 EXPORT_SYMBOL(sock_i_ino);
1220 * Allocate a skb from the socket's send buffer.
1222 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1225 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1226 struct sk_buff *skb = alloc_skb(size, priority);
1228 skb_set_owner_w(skb, sk);
1234 EXPORT_SYMBOL(sock_wmalloc);
1237 * Allocate a skb from the socket's receive buffer.
1239 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1242 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1243 struct sk_buff *skb = alloc_skb(size, priority);
1245 skb_set_owner_r(skb, sk);
1253 * Allocate a memory block from the socket's option memory buffer.
1255 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1257 if ((unsigned)size <= sysctl_optmem_max &&
1258 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1260 /* First do the add, to avoid the race if kmalloc
1263 atomic_add(size, &sk->sk_omem_alloc);
1264 mem = kmalloc(size, priority);
1267 atomic_sub(size, &sk->sk_omem_alloc);
1271 EXPORT_SYMBOL(sock_kmalloc);
1274 * Free an option memory block.
1276 void sock_kfree_s(struct sock *sk, void *mem, int size)
1279 atomic_sub(size, &sk->sk_omem_alloc);
1281 EXPORT_SYMBOL(sock_kfree_s);
1283 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1284 I think, these locks should be removed for datagram sockets.
1286 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1290 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1294 if (signal_pending(current))
1296 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1297 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1298 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1300 if (sk->sk_shutdown & SEND_SHUTDOWN)
1304 timeo = schedule_timeout(timeo);
1306 finish_wait(sk->sk_sleep, &wait);
1312 * Generic send/receive buffer handlers
1315 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1316 unsigned long data_len, int noblock,
1319 struct sk_buff *skb;
1324 gfp_mask = sk->sk_allocation;
1325 if (gfp_mask & __GFP_WAIT)
1326 gfp_mask |= __GFP_REPEAT;
1328 timeo = sock_sndtimeo(sk, noblock);
1330 err = sock_error(sk);
1335 if (sk->sk_shutdown & SEND_SHUTDOWN)
1338 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1339 skb = alloc_skb(header_len, gfp_mask);
1344 /* No pages, we're done... */
1348 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1349 skb->truesize += data_len;
1350 skb_shinfo(skb)->nr_frags = npages;
1351 for (i = 0; i < npages; i++) {
1355 page = alloc_pages(sk->sk_allocation, 0);
1358 skb_shinfo(skb)->nr_frags = i;
1363 frag = &skb_shinfo(skb)->frags[i];
1365 frag->page_offset = 0;
1366 frag->size = (data_len >= PAGE_SIZE ?
1369 data_len -= PAGE_SIZE;
1372 /* Full success... */
1378 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1379 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1383 if (signal_pending(current))
1385 timeo = sock_wait_for_wmem(sk, timeo);
1388 skb_set_owner_w(skb, sk);
1392 err = sock_intr_errno(timeo);
1397 EXPORT_SYMBOL(sock_alloc_send_pskb);
1399 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1400 int noblock, int *errcode)
1402 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1404 EXPORT_SYMBOL(sock_alloc_send_skb);
1406 static void __lock_sock(struct sock *sk)
1411 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1412 TASK_UNINTERRUPTIBLE);
1413 spin_unlock_bh(&sk->sk_lock.slock);
1415 spin_lock_bh(&sk->sk_lock.slock);
1416 if (!sock_owned_by_user(sk))
1419 finish_wait(&sk->sk_lock.wq, &wait);
1422 static void __release_sock(struct sock *sk)
1424 struct sk_buff *skb = sk->sk_backlog.head;
1427 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1431 struct sk_buff *next = skb->next;
1434 sk_backlog_rcv(sk, skb);
1437 * We are in process context here with softirqs
1438 * disabled, use cond_resched_softirq() to preempt.
1439 * This is safe to do because we've taken the backlog
1442 cond_resched_softirq();
1445 } while (skb != NULL);
1448 } while ((skb = sk->sk_backlog.head) != NULL);
1452 * sk_wait_data - wait for data to arrive at sk_receive_queue
1453 * @sk: sock to wait on
1454 * @timeo: for how long
1456 * Now socket state including sk->sk_err is changed only under lock,
1457 * hence we may omit checks after joining wait queue.
1458 * We check receive queue before schedule() only as optimization;
1459 * it is very likely that release_sock() added new data.
1461 int sk_wait_data(struct sock *sk, long *timeo)
1466 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1467 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1468 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1469 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1470 finish_wait(sk->sk_sleep, &wait);
1473 EXPORT_SYMBOL(sk_wait_data);
1476 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1478 * @size: memory size to allocate
1479 * @kind: allocation type
1481 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1482 * rmem allocation. This function assumes that protocols which have
1483 * memory_pressure use sk_wmem_queued as write buffer accounting.
1485 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1487 struct proto *prot = sk->sk_prot;
1488 int amt = sk_mem_pages(size);
1491 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1492 allocated = atomic_add_return(amt, prot->memory_allocated);
1495 if (allocated <= prot->sysctl_mem[0]) {
1496 if (prot->memory_pressure && *prot->memory_pressure)
1497 *prot->memory_pressure = 0;
1501 /* Under pressure. */
1502 if (allocated > prot->sysctl_mem[1])
1503 if (prot->enter_memory_pressure)
1504 prot->enter_memory_pressure(sk);
1506 /* Over hard limit. */
1507 if (allocated > prot->sysctl_mem[2])
1508 goto suppress_allocation;
1510 /* guarantee minimum buffer size under pressure */
1511 if (kind == SK_MEM_RECV) {
1512 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1514 } else { /* SK_MEM_SEND */
1515 if (sk->sk_type == SOCK_STREAM) {
1516 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1518 } else if (atomic_read(&sk->sk_wmem_alloc) <
1519 prot->sysctl_wmem[0])
1523 if (prot->memory_pressure) {
1526 if (!*prot->memory_pressure)
1528 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1529 if (prot->sysctl_mem[2] > alloc *
1530 sk_mem_pages(sk->sk_wmem_queued +
1531 atomic_read(&sk->sk_rmem_alloc) +
1532 sk->sk_forward_alloc))
1536 suppress_allocation:
1538 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1539 sk_stream_moderate_sndbuf(sk);
1541 /* Fail only if socket is _under_ its sndbuf.
1542 * In this case we cannot block, so that we have to fail.
1544 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1548 /* Alas. Undo changes. */
1549 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1550 atomic_sub(amt, prot->memory_allocated);
1553 EXPORT_SYMBOL(__sk_mem_schedule);
1556 * __sk_reclaim - reclaim memory_allocated
1559 void __sk_mem_reclaim(struct sock *sk)
1561 struct proto *prot = sk->sk_prot;
1563 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1564 prot->memory_allocated);
1565 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1567 if (prot->memory_pressure && *prot->memory_pressure &&
1568 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1569 *prot->memory_pressure = 0;
1571 EXPORT_SYMBOL(__sk_mem_reclaim);
1575 * Set of default routines for initialising struct proto_ops when
1576 * the protocol does not support a particular function. In certain
1577 * cases where it makes no sense for a protocol to have a "do nothing"
1578 * function, some default processing is provided.
1581 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1585 EXPORT_SYMBOL(sock_no_bind);
1587 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1592 EXPORT_SYMBOL(sock_no_connect);
1594 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1598 EXPORT_SYMBOL(sock_no_socketpair);
1600 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1604 EXPORT_SYMBOL(sock_no_accept);
1606 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1611 EXPORT_SYMBOL(sock_no_getname);
1613 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1617 EXPORT_SYMBOL(sock_no_poll);
1619 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1623 EXPORT_SYMBOL(sock_no_ioctl);
1625 int sock_no_listen(struct socket *sock, int backlog)
1629 EXPORT_SYMBOL(sock_no_listen);
1631 int sock_no_shutdown(struct socket *sock, int how)
1635 EXPORT_SYMBOL(sock_no_shutdown);
1637 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1638 char __user *optval, int optlen)
1642 EXPORT_SYMBOL(sock_no_setsockopt);
1644 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1645 char __user *optval, int __user *optlen)
1649 EXPORT_SYMBOL(sock_no_getsockopt);
1651 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1656 EXPORT_SYMBOL(sock_no_sendmsg);
1658 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1659 size_t len, int flags)
1663 EXPORT_SYMBOL(sock_no_recvmsg);
1665 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1667 /* Mirror missing mmap method error code */
1670 EXPORT_SYMBOL(sock_no_mmap);
1672 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1675 struct msghdr msg = {.msg_flags = flags};
1677 char *kaddr = kmap(page);
1678 iov.iov_base = kaddr + offset;
1680 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1684 EXPORT_SYMBOL(sock_no_sendpage);
1687 * Default Socket Callbacks
1690 static void sock_def_wakeup(struct sock *sk)
1692 read_lock(&sk->sk_callback_lock);
1693 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1694 wake_up_interruptible_all(sk->sk_sleep);
1695 read_unlock(&sk->sk_callback_lock);
1698 static void sock_def_error_report(struct sock *sk)
1700 read_lock(&sk->sk_callback_lock);
1701 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1702 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1703 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1704 read_unlock(&sk->sk_callback_lock);
1707 static void sock_def_readable(struct sock *sk, int len)
1709 read_lock(&sk->sk_callback_lock);
1710 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1711 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1712 POLLRDNORM | POLLRDBAND);
1713 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1714 read_unlock(&sk->sk_callback_lock);
1717 static void sock_def_write_space(struct sock *sk)
1719 read_lock(&sk->sk_callback_lock);
1721 /* Do not wake up a writer until he can make "significant"
1724 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1725 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1726 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1727 POLLWRNORM | POLLWRBAND);
1729 /* Should agree with poll, otherwise some programs break */
1730 if (sock_writeable(sk))
1731 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1734 read_unlock(&sk->sk_callback_lock);
1737 static void sock_def_destruct(struct sock *sk)
1739 kfree(sk->sk_protinfo);
1742 void sk_send_sigurg(struct sock *sk)
1744 if (sk->sk_socket && sk->sk_socket->file)
1745 if (send_sigurg(&sk->sk_socket->file->f_owner))
1746 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1748 EXPORT_SYMBOL(sk_send_sigurg);
1750 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1751 unsigned long expires)
1753 if (!mod_timer(timer, expires))
1756 EXPORT_SYMBOL(sk_reset_timer);
1758 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1760 if (timer_pending(timer) && del_timer(timer))
1763 EXPORT_SYMBOL(sk_stop_timer);
1765 void sock_init_data(struct socket *sock, struct sock *sk)
1767 skb_queue_head_init(&sk->sk_receive_queue);
1768 skb_queue_head_init(&sk->sk_write_queue);
1769 skb_queue_head_init(&sk->sk_error_queue);
1770 #ifdef CONFIG_NET_DMA
1771 skb_queue_head_init(&sk->sk_async_wait_queue);
1774 sk->sk_send_head = NULL;
1776 init_timer(&sk->sk_timer);
1778 sk->sk_allocation = GFP_KERNEL;
1779 sk->sk_rcvbuf = sysctl_rmem_default;
1780 sk->sk_sndbuf = sysctl_wmem_default;
1781 sk->sk_state = TCP_CLOSE;
1782 sk_set_socket(sk, sock);
1784 sock_set_flag(sk, SOCK_ZAPPED);
1787 sk->sk_type = sock->type;
1788 sk->sk_sleep = &sock->wait;
1791 sk->sk_sleep = NULL;
1793 rwlock_init(&sk->sk_dst_lock);
1794 rwlock_init(&sk->sk_callback_lock);
1795 lockdep_set_class_and_name(&sk->sk_callback_lock,
1796 af_callback_keys + sk->sk_family,
1797 af_family_clock_key_strings[sk->sk_family]);
1799 sk->sk_state_change = sock_def_wakeup;
1800 sk->sk_data_ready = sock_def_readable;
1801 sk->sk_write_space = sock_def_write_space;
1802 sk->sk_error_report = sock_def_error_report;
1803 sk->sk_destruct = sock_def_destruct;
1805 sk->sk_sndmsg_page = NULL;
1806 sk->sk_sndmsg_off = 0;
1808 sk->sk_peercred.pid = 0;
1809 sk->sk_peercred.uid = -1;
1810 sk->sk_peercred.gid = -1;
1811 sk->sk_write_pending = 0;
1812 sk->sk_rcvlowat = 1;
1813 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1814 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1816 sk->sk_stamp = ktime_set(-1L, 0);
1818 atomic_set(&sk->sk_refcnt, 1);
1819 atomic_set(&sk->sk_drops, 0);
1821 EXPORT_SYMBOL(sock_init_data);
1823 void lock_sock_nested(struct sock *sk, int subclass)
1826 spin_lock_bh(&sk->sk_lock.slock);
1827 if (sk->sk_lock.owned)
1829 sk->sk_lock.owned = 1;
1830 spin_unlock(&sk->sk_lock.slock);
1832 * The sk_lock has mutex_lock() semantics here:
1834 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1837 EXPORT_SYMBOL(lock_sock_nested);
1839 void release_sock(struct sock *sk)
1842 * The sk_lock has mutex_unlock() semantics:
1844 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1846 spin_lock_bh(&sk->sk_lock.slock);
1847 if (sk->sk_backlog.tail)
1849 sk->sk_lock.owned = 0;
1850 if (waitqueue_active(&sk->sk_lock.wq))
1851 wake_up(&sk->sk_lock.wq);
1852 spin_unlock_bh(&sk->sk_lock.slock);
1854 EXPORT_SYMBOL(release_sock);
1856 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1859 if (!sock_flag(sk, SOCK_TIMESTAMP))
1860 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1861 tv = ktime_to_timeval(sk->sk_stamp);
1862 if (tv.tv_sec == -1)
1864 if (tv.tv_sec == 0) {
1865 sk->sk_stamp = ktime_get_real();
1866 tv = ktime_to_timeval(sk->sk_stamp);
1868 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1870 EXPORT_SYMBOL(sock_get_timestamp);
1872 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1875 if (!sock_flag(sk, SOCK_TIMESTAMP))
1876 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1877 ts = ktime_to_timespec(sk->sk_stamp);
1878 if (ts.tv_sec == -1)
1880 if (ts.tv_sec == 0) {
1881 sk->sk_stamp = ktime_get_real();
1882 ts = ktime_to_timespec(sk->sk_stamp);
1884 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1886 EXPORT_SYMBOL(sock_get_timestampns);
1888 void sock_enable_timestamp(struct sock *sk, int flag)
1890 if (!sock_flag(sk, flag)) {
1891 sock_set_flag(sk, flag);
1893 * we just set one of the two flags which require net
1894 * time stamping, but time stamping might have been on
1895 * already because of the other one
1898 flag == SOCK_TIMESTAMP ?
1899 SOCK_TIMESTAMPING_RX_SOFTWARE :
1901 net_enable_timestamp();
1906 * Get a socket option on an socket.
1908 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1909 * asynchronous errors should be reported by getsockopt. We assume
1910 * this means if you specify SO_ERROR (otherwise whats the point of it).
1912 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1913 char __user *optval, int __user *optlen)
1915 struct sock *sk = sock->sk;
1917 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1919 EXPORT_SYMBOL(sock_common_getsockopt);
1921 #ifdef CONFIG_COMPAT
1922 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1923 char __user *optval, int __user *optlen)
1925 struct sock *sk = sock->sk;
1927 if (sk->sk_prot->compat_getsockopt != NULL)
1928 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1930 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1932 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1935 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1936 struct msghdr *msg, size_t size, int flags)
1938 struct sock *sk = sock->sk;
1942 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1943 flags & ~MSG_DONTWAIT, &addr_len);
1945 msg->msg_namelen = addr_len;
1948 EXPORT_SYMBOL(sock_common_recvmsg);
1951 * Set socket options on an inet socket.
1953 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1954 char __user *optval, int optlen)
1956 struct sock *sk = sock->sk;
1958 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1960 EXPORT_SYMBOL(sock_common_setsockopt);
1962 #ifdef CONFIG_COMPAT
1963 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1964 char __user *optval, int optlen)
1966 struct sock *sk = sock->sk;
1968 if (sk->sk_prot->compat_setsockopt != NULL)
1969 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1971 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1973 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1976 void sk_common_release(struct sock *sk)
1978 if (sk->sk_prot->destroy)
1979 sk->sk_prot->destroy(sk);
1982 * Observation: when sock_common_release is called, processes have
1983 * no access to socket. But net still has.
1984 * Step one, detach it from networking:
1986 * A. Remove from hash tables.
1989 sk->sk_prot->unhash(sk);
1992 * In this point socket cannot receive new packets, but it is possible
1993 * that some packets are in flight because some CPU runs receiver and
1994 * did hash table lookup before we unhashed socket. They will achieve
1995 * receive queue and will be purged by socket destructor.
1997 * Also we still have packets pending on receive queue and probably,
1998 * our own packets waiting in device queues. sock_destroy will drain
1999 * receive queue, but transmitted packets will delay socket destruction
2000 * until the last reference will be released.
2005 xfrm_sk_free_policy(sk);
2007 sk_refcnt_debug_release(sk);
2010 EXPORT_SYMBOL(sk_common_release);
2012 static DEFINE_RWLOCK(proto_list_lock);
2013 static LIST_HEAD(proto_list);
2015 #ifdef CONFIG_PROC_FS
2016 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2018 int val[PROTO_INUSE_NR];
2021 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2023 #ifdef CONFIG_NET_NS
2024 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2026 int cpu = smp_processor_id();
2027 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2029 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2031 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2033 int cpu, idx = prot->inuse_idx;
2036 for_each_possible_cpu(cpu)
2037 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2039 return res >= 0 ? res : 0;
2041 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2043 static int sock_inuse_init_net(struct net *net)
2045 net->core.inuse = alloc_percpu(struct prot_inuse);
2046 return net->core.inuse ? 0 : -ENOMEM;
2049 static void sock_inuse_exit_net(struct net *net)
2051 free_percpu(net->core.inuse);
2054 static struct pernet_operations net_inuse_ops = {
2055 .init = sock_inuse_init_net,
2056 .exit = sock_inuse_exit_net,
2059 static __init int net_inuse_init(void)
2061 if (register_pernet_subsys(&net_inuse_ops))
2062 panic("Cannot initialize net inuse counters");
2067 core_initcall(net_inuse_init);
2069 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2071 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2073 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2075 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2077 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2079 int cpu, idx = prot->inuse_idx;
2082 for_each_possible_cpu(cpu)
2083 res += per_cpu(prot_inuse, cpu).val[idx];
2085 return res >= 0 ? res : 0;
2087 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2090 static void assign_proto_idx(struct proto *prot)
2092 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2094 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2095 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2099 set_bit(prot->inuse_idx, proto_inuse_idx);
2102 static void release_proto_idx(struct proto *prot)
2104 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2105 clear_bit(prot->inuse_idx, proto_inuse_idx);
2108 static inline void assign_proto_idx(struct proto *prot)
2112 static inline void release_proto_idx(struct proto *prot)
2117 int proto_register(struct proto *prot, int alloc_slab)
2120 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2121 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2124 if (prot->slab == NULL) {
2125 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2130 if (prot->rsk_prot != NULL) {
2131 static const char mask[] = "request_sock_%s";
2133 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2134 if (prot->rsk_prot->slab_name == NULL)
2135 goto out_free_sock_slab;
2137 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2138 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2139 prot->rsk_prot->obj_size, 0,
2140 SLAB_HWCACHE_ALIGN, NULL);
2142 if (prot->rsk_prot->slab == NULL) {
2143 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2145 goto out_free_request_sock_slab_name;
2149 if (prot->twsk_prot != NULL) {
2150 static const char mask[] = "tw_sock_%s";
2152 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2154 if (prot->twsk_prot->twsk_slab_name == NULL)
2155 goto out_free_request_sock_slab;
2157 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2158 prot->twsk_prot->twsk_slab =
2159 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2160 prot->twsk_prot->twsk_obj_size,
2162 SLAB_HWCACHE_ALIGN |
2165 if (prot->twsk_prot->twsk_slab == NULL)
2166 goto out_free_timewait_sock_slab_name;
2170 write_lock(&proto_list_lock);
2171 list_add(&prot->node, &proto_list);
2172 assign_proto_idx(prot);
2173 write_unlock(&proto_list_lock);
2176 out_free_timewait_sock_slab_name:
2177 kfree(prot->twsk_prot->twsk_slab_name);
2178 out_free_request_sock_slab:
2179 if (prot->rsk_prot && prot->rsk_prot->slab) {
2180 kmem_cache_destroy(prot->rsk_prot->slab);
2181 prot->rsk_prot->slab = NULL;
2183 out_free_request_sock_slab_name:
2184 kfree(prot->rsk_prot->slab_name);
2186 kmem_cache_destroy(prot->slab);
2191 EXPORT_SYMBOL(proto_register);
2193 void proto_unregister(struct proto *prot)
2195 write_lock(&proto_list_lock);
2196 release_proto_idx(prot);
2197 list_del(&prot->node);
2198 write_unlock(&proto_list_lock);
2200 if (prot->slab != NULL) {
2201 kmem_cache_destroy(prot->slab);
2205 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2206 kmem_cache_destroy(prot->rsk_prot->slab);
2207 kfree(prot->rsk_prot->slab_name);
2208 prot->rsk_prot->slab = NULL;
2211 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2212 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2213 kfree(prot->twsk_prot->twsk_slab_name);
2214 prot->twsk_prot->twsk_slab = NULL;
2217 EXPORT_SYMBOL(proto_unregister);
2219 #ifdef CONFIG_PROC_FS
2220 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2221 __acquires(proto_list_lock)
2223 read_lock(&proto_list_lock);
2224 return seq_list_start_head(&proto_list, *pos);
2227 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2229 return seq_list_next(v, &proto_list, pos);
2232 static void proto_seq_stop(struct seq_file *seq, void *v)
2233 __releases(proto_list_lock)
2235 read_unlock(&proto_list_lock);
2238 static char proto_method_implemented(const void *method)
2240 return method == NULL ? 'n' : 'y';
2243 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2245 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2246 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2249 sock_prot_inuse_get(seq_file_net(seq), proto),
2250 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2251 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2253 proto->slab == NULL ? "no" : "yes",
2254 module_name(proto->owner),
2255 proto_method_implemented(proto->close),
2256 proto_method_implemented(proto->connect),
2257 proto_method_implemented(proto->disconnect),
2258 proto_method_implemented(proto->accept),
2259 proto_method_implemented(proto->ioctl),
2260 proto_method_implemented(proto->init),
2261 proto_method_implemented(proto->destroy),
2262 proto_method_implemented(proto->shutdown),
2263 proto_method_implemented(proto->setsockopt),
2264 proto_method_implemented(proto->getsockopt),
2265 proto_method_implemented(proto->sendmsg),
2266 proto_method_implemented(proto->recvmsg),
2267 proto_method_implemented(proto->sendpage),
2268 proto_method_implemented(proto->bind),
2269 proto_method_implemented(proto->backlog_rcv),
2270 proto_method_implemented(proto->hash),
2271 proto_method_implemented(proto->unhash),
2272 proto_method_implemented(proto->get_port),
2273 proto_method_implemented(proto->enter_memory_pressure));
2276 static int proto_seq_show(struct seq_file *seq, void *v)
2278 if (v == &proto_list)
2279 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2288 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2290 proto_seq_printf(seq, list_entry(v, struct proto, node));
2294 static const struct seq_operations proto_seq_ops = {
2295 .start = proto_seq_start,
2296 .next = proto_seq_next,
2297 .stop = proto_seq_stop,
2298 .show = proto_seq_show,
2301 static int proto_seq_open(struct inode *inode, struct file *file)
2303 return seq_open_net(inode, file, &proto_seq_ops,
2304 sizeof(struct seq_net_private));
2307 static const struct file_operations proto_seq_fops = {
2308 .owner = THIS_MODULE,
2309 .open = proto_seq_open,
2311 .llseek = seq_lseek,
2312 .release = seq_release_net,
2315 static __net_init int proto_init_net(struct net *net)
2317 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2323 static __net_exit void proto_exit_net(struct net *net)
2325 proc_net_remove(net, "protocols");
2329 static __net_initdata struct pernet_operations proto_net_ops = {
2330 .init = proto_init_net,
2331 .exit = proto_exit_net,
2334 static int __init proto_init(void)
2336 return register_pernet_subsys(&proto_net_ops);
2339 subsys_initcall(proto_init);
2341 #endif /* PROC_FS */