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" ,
158 "sk_lock-AF_IEEE802154",
161 static const char *af_family_slock_key_strings[AF_MAX+1] = {
162 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
163 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
164 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
165 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
166 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
167 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
168 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
169 "slock-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 *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 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
281 number of warnings when compiling with -W --ANK
283 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
284 (unsigned)sk->sk_rcvbuf) {
289 err = sk_filter(sk, skb);
293 if (!sk_rmem_schedule(sk, skb->truesize)) {
299 skb_set_owner_r(skb, sk);
301 /* Cache the SKB length before we tack it onto the receive
302 * queue. Once it is added it no longer belongs to us and
303 * may be freed by other threads of control pulling packets
308 skb_queue_tail(&sk->sk_receive_queue, skb);
310 if (!sock_flag(sk, SOCK_DEAD))
311 sk->sk_data_ready(sk, skb_len);
315 EXPORT_SYMBOL(sock_queue_rcv_skb);
317 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
319 int rc = NET_RX_SUCCESS;
321 if (sk_filter(sk, skb))
322 goto discard_and_relse;
327 bh_lock_sock_nested(sk);
330 if (!sock_owned_by_user(sk)) {
332 * trylock + unlock semantics:
334 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
336 rc = sk_backlog_rcv(sk, skb);
338 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
340 sk_add_backlog(sk, skb);
349 EXPORT_SYMBOL(sk_receive_skb);
351 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
353 struct dst_entry *dst = sk->sk_dst_cache;
355 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
356 sk->sk_dst_cache = NULL;
363 EXPORT_SYMBOL(__sk_dst_check);
365 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
367 struct dst_entry *dst = sk_dst_get(sk);
369 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
377 EXPORT_SYMBOL(sk_dst_check);
379 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
381 int ret = -ENOPROTOOPT;
382 #ifdef CONFIG_NETDEVICES
383 struct net *net = sock_net(sk);
384 char devname[IFNAMSIZ];
389 if (!capable(CAP_NET_RAW))
396 /* Bind this socket to a particular device like "eth0",
397 * as specified in the passed interface name. If the
398 * name is "" or the option length is zero the socket
401 if (optlen > IFNAMSIZ - 1)
402 optlen = IFNAMSIZ - 1;
403 memset(devname, 0, sizeof(devname));
406 if (copy_from_user(devname, optval, optlen))
409 if (devname[0] == '\0') {
412 struct net_device *dev = dev_get_by_name(net, devname);
418 index = dev->ifindex;
423 sk->sk_bound_dev_if = index;
435 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
438 sock_set_flag(sk, bit);
440 sock_reset_flag(sk, bit);
444 * This is meant for all protocols to use and covers goings on
445 * at the socket level. Everything here is generic.
448 int sock_setsockopt(struct socket *sock, int level, int optname,
449 char __user *optval, int optlen)
451 struct sock *sk = sock->sk;
458 * Options without arguments
461 if (optname == SO_BINDTODEVICE)
462 return sock_bindtodevice(sk, optval, optlen);
464 if (optlen < sizeof(int))
467 if (get_user(val, (int __user *)optval))
470 valbool = val ? 1 : 0;
476 if (val && !capable(CAP_NET_ADMIN))
479 sock_valbool_flag(sk, SOCK_DBG, valbool);
482 sk->sk_reuse = valbool;
489 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
492 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
495 /* Don't error on this BSD doesn't and if you think
496 about it this is right. Otherwise apps have to
497 play 'guess the biggest size' games. RCVBUF/SNDBUF
498 are treated in BSD as hints */
500 if (val > sysctl_wmem_max)
501 val = sysctl_wmem_max;
503 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
504 if ((val * 2) < SOCK_MIN_SNDBUF)
505 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
507 sk->sk_sndbuf = val * 2;
510 * Wake up sending tasks if we
513 sk->sk_write_space(sk);
517 if (!capable(CAP_NET_ADMIN)) {
524 /* Don't error on this BSD doesn't and if you think
525 about it this is right. Otherwise apps have to
526 play 'guess the biggest size' games. RCVBUF/SNDBUF
527 are treated in BSD as hints */
529 if (val > sysctl_rmem_max)
530 val = sysctl_rmem_max;
532 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
534 * We double it on the way in to account for
535 * "struct sk_buff" etc. overhead. Applications
536 * assume that the SO_RCVBUF setting they make will
537 * allow that much actual data to be received on that
540 * Applications are unaware that "struct sk_buff" and
541 * other overheads allocate from the receive buffer
542 * during socket buffer allocation.
544 * And after considering the possible alternatives,
545 * returning the value we actually used in getsockopt
546 * is the most desirable behavior.
548 if ((val * 2) < SOCK_MIN_RCVBUF)
549 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
551 sk->sk_rcvbuf = val * 2;
555 if (!capable(CAP_NET_ADMIN)) {
563 if (sk->sk_protocol == IPPROTO_TCP)
564 tcp_set_keepalive(sk, valbool);
566 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
570 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
574 sk->sk_no_check = valbool;
578 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
579 sk->sk_priority = val;
585 if (optlen < sizeof(ling)) {
586 ret = -EINVAL; /* 1003.1g */
589 if (copy_from_user(&ling, optval, sizeof(ling))) {
594 sock_reset_flag(sk, SOCK_LINGER);
596 #if (BITS_PER_LONG == 32)
597 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
598 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
601 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
602 sock_set_flag(sk, SOCK_LINGER);
607 sock_warn_obsolete_bsdism("setsockopt");
612 set_bit(SOCK_PASSCRED, &sock->flags);
614 clear_bit(SOCK_PASSCRED, &sock->flags);
620 if (optname == SO_TIMESTAMP)
621 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
623 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
624 sock_set_flag(sk, SOCK_RCVTSTAMP);
625 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
627 sock_reset_flag(sk, SOCK_RCVTSTAMP);
628 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
632 case SO_TIMESTAMPING:
633 if (val & ~SOF_TIMESTAMPING_MASK) {
637 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
638 val & SOF_TIMESTAMPING_TX_HARDWARE);
639 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
640 val & SOF_TIMESTAMPING_TX_SOFTWARE);
641 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
642 val & SOF_TIMESTAMPING_RX_HARDWARE);
643 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
644 sock_enable_timestamp(sk,
645 SOCK_TIMESTAMPING_RX_SOFTWARE);
647 sock_disable_timestamp(sk,
648 SOCK_TIMESTAMPING_RX_SOFTWARE);
649 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
650 val & SOF_TIMESTAMPING_SOFTWARE);
651 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
652 val & SOF_TIMESTAMPING_SYS_HARDWARE);
653 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
654 val & SOF_TIMESTAMPING_RAW_HARDWARE);
660 sk->sk_rcvlowat = val ? : 1;
664 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
668 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
671 case SO_ATTACH_FILTER:
673 if (optlen == sizeof(struct sock_fprog)) {
674 struct sock_fprog fprog;
677 if (copy_from_user(&fprog, optval, sizeof(fprog)))
680 ret = sk_attach_filter(&fprog, sk);
684 case SO_DETACH_FILTER:
685 ret = sk_detach_filter(sk);
690 set_bit(SOCK_PASSSEC, &sock->flags);
692 clear_bit(SOCK_PASSSEC, &sock->flags);
695 if (!capable(CAP_NET_ADMIN))
701 /* We implement the SO_SNDLOWAT etc to
702 not be settable (1003.1g 5.3) */
710 EXPORT_SYMBOL(sock_setsockopt);
713 int sock_getsockopt(struct socket *sock, int level, int optname,
714 char __user *optval, int __user *optlen)
716 struct sock *sk = sock->sk;
724 unsigned int lv = sizeof(int);
727 if (get_user(len, optlen))
732 memset(&v, 0, sizeof(v));
736 v.val = sock_flag(sk, SOCK_DBG);
740 v.val = sock_flag(sk, SOCK_LOCALROUTE);
744 v.val = !!sock_flag(sk, SOCK_BROADCAST);
748 v.val = sk->sk_sndbuf;
752 v.val = sk->sk_rcvbuf;
756 v.val = sk->sk_reuse;
760 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
768 v.val = -sock_error(sk);
770 v.val = xchg(&sk->sk_err_soft, 0);
774 v.val = !!sock_flag(sk, SOCK_URGINLINE);
778 v.val = sk->sk_no_check;
782 v.val = sk->sk_priority;
787 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
788 v.ling.l_linger = sk->sk_lingertime / HZ;
792 sock_warn_obsolete_bsdism("getsockopt");
796 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
797 !sock_flag(sk, SOCK_RCVTSTAMPNS);
801 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
804 case SO_TIMESTAMPING:
806 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
807 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
808 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
809 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
810 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
811 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
812 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
813 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
814 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
815 v.val |= SOF_TIMESTAMPING_SOFTWARE;
816 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
817 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
818 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
819 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
823 lv = sizeof(struct timeval);
824 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
828 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
829 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
834 lv = sizeof(struct timeval);
835 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
839 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
840 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
845 v.val = sk->sk_rcvlowat;
853 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
857 if (len > sizeof(sk->sk_peercred))
858 len = sizeof(sk->sk_peercred);
859 if (copy_to_user(optval, &sk->sk_peercred, len))
867 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
871 if (copy_to_user(optval, address, len))
876 /* Dubious BSD thing... Probably nobody even uses it, but
877 * the UNIX standard wants it for whatever reason... -DaveM
880 v.val = sk->sk_state == TCP_LISTEN;
884 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
888 return security_socket_getpeersec_stream(sock, optval, optlen, len);
900 if (copy_to_user(optval, &v, len))
903 if (put_user(len, optlen))
909 * Initialize an sk_lock.
911 * (We also register the sk_lock with the lock validator.)
913 static inline void sock_lock_init(struct sock *sk)
915 sock_lock_init_class_and_name(sk,
916 af_family_slock_key_strings[sk->sk_family],
917 af_family_slock_keys + sk->sk_family,
918 af_family_key_strings[sk->sk_family],
919 af_family_keys + sk->sk_family);
923 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
924 * even temporarly, because of RCU lookups. sk_node should also be left as is.
926 static void sock_copy(struct sock *nsk, const struct sock *osk)
928 #ifdef CONFIG_SECURITY_NETWORK
929 void *sptr = nsk->sk_security;
931 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
932 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt));
933 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
934 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
935 #ifdef CONFIG_SECURITY_NETWORK
936 nsk->sk_security = sptr;
937 security_sk_clone(osk, nsk);
941 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
945 struct kmem_cache *slab;
949 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
952 if (priority & __GFP_ZERO) {
954 * caches using SLAB_DESTROY_BY_RCU should let
955 * sk_node.next un-modified. Special care is taken
956 * when initializing object to zero.
958 if (offsetof(struct sock, sk_node.next) != 0)
959 memset(sk, 0, offsetof(struct sock, sk_node.next));
960 memset(&sk->sk_node.pprev, 0,
961 prot->obj_size - offsetof(struct sock,
966 sk = kmalloc(prot->obj_size, priority);
969 kmemcheck_annotate_bitfield(sk, flags);
971 if (security_sk_alloc(sk, family, priority))
974 if (!try_module_get(prot->owner))
981 security_sk_free(sk);
984 kmem_cache_free(slab, sk);
990 static void sk_prot_free(struct proto *prot, struct sock *sk)
992 struct kmem_cache *slab;
993 struct module *owner;
998 security_sk_free(sk);
1000 kmem_cache_free(slab, sk);
1007 * sk_alloc - All socket objects are allocated here
1008 * @net: the applicable net namespace
1009 * @family: protocol family
1010 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1011 * @prot: struct proto associated with this new sock instance
1013 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1018 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1020 sk->sk_family = family;
1022 * See comment in struct sock definition to understand
1023 * why we need sk_prot_creator -acme
1025 sk->sk_prot = sk->sk_prot_creator = prot;
1027 sock_net_set(sk, get_net(net));
1028 atomic_set(&sk->sk_wmem_alloc, 1);
1033 EXPORT_SYMBOL(sk_alloc);
1035 static void __sk_free(struct sock *sk)
1037 struct sk_filter *filter;
1039 if (sk->sk_destruct)
1040 sk->sk_destruct(sk);
1042 filter = rcu_dereference(sk->sk_filter);
1044 sk_filter_uncharge(sk, filter);
1045 rcu_assign_pointer(sk->sk_filter, NULL);
1048 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1049 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1051 if (atomic_read(&sk->sk_omem_alloc))
1052 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1053 __func__, atomic_read(&sk->sk_omem_alloc));
1055 put_net(sock_net(sk));
1056 sk_prot_free(sk->sk_prot_creator, sk);
1059 void sk_free(struct sock *sk)
1062 * We substract one from sk_wmem_alloc and can know if
1063 * some packets are still in some tx queue.
1064 * If not null, sock_wfree() will call __sk_free(sk) later
1066 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1069 EXPORT_SYMBOL(sk_free);
1072 * Last sock_put should drop referrence to sk->sk_net. It has already
1073 * been dropped in sk_change_net. Taking referrence to stopping namespace
1075 * Take referrence to a socket to remove it from hash _alive_ and after that
1076 * destroy it in the context of init_net.
1078 void sk_release_kernel(struct sock *sk)
1080 if (sk == NULL || sk->sk_socket == NULL)
1084 sock_release(sk->sk_socket);
1085 release_net(sock_net(sk));
1086 sock_net_set(sk, get_net(&init_net));
1089 EXPORT_SYMBOL(sk_release_kernel);
1091 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1095 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1096 if (newsk != NULL) {
1097 struct sk_filter *filter;
1099 sock_copy(newsk, sk);
1102 get_net(sock_net(newsk));
1103 sk_node_init(&newsk->sk_node);
1104 sock_lock_init(newsk);
1105 bh_lock_sock(newsk);
1106 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1108 atomic_set(&newsk->sk_rmem_alloc, 0);
1110 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1112 atomic_set(&newsk->sk_wmem_alloc, 1);
1113 atomic_set(&newsk->sk_omem_alloc, 0);
1114 skb_queue_head_init(&newsk->sk_receive_queue);
1115 skb_queue_head_init(&newsk->sk_write_queue);
1116 #ifdef CONFIG_NET_DMA
1117 skb_queue_head_init(&newsk->sk_async_wait_queue);
1120 rwlock_init(&newsk->sk_dst_lock);
1121 rwlock_init(&newsk->sk_callback_lock);
1122 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1123 af_callback_keys + newsk->sk_family,
1124 af_family_clock_key_strings[newsk->sk_family]);
1126 newsk->sk_dst_cache = NULL;
1127 newsk->sk_wmem_queued = 0;
1128 newsk->sk_forward_alloc = 0;
1129 newsk->sk_send_head = NULL;
1130 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1132 sock_reset_flag(newsk, SOCK_DONE);
1133 skb_queue_head_init(&newsk->sk_error_queue);
1135 filter = newsk->sk_filter;
1137 sk_filter_charge(newsk, filter);
1139 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1140 /* It is still raw copy of parent, so invalidate
1141 * destructor and make plain sk_free() */
1142 newsk->sk_destruct = NULL;
1149 newsk->sk_priority = 0;
1151 * Before updating sk_refcnt, we must commit prior changes to memory
1152 * (Documentation/RCU/rculist_nulls.txt for details)
1155 atomic_set(&newsk->sk_refcnt, 2);
1158 * Increment the counter in the same struct proto as the master
1159 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1160 * is the same as sk->sk_prot->socks, as this field was copied
1163 * This _changes_ the previous behaviour, where
1164 * tcp_create_openreq_child always was incrementing the
1165 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1166 * to be taken into account in all callers. -acme
1168 sk_refcnt_debug_inc(newsk);
1169 sk_set_socket(newsk, NULL);
1170 newsk->sk_sleep = NULL;
1172 if (newsk->sk_prot->sockets_allocated)
1173 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1178 EXPORT_SYMBOL_GPL(sk_clone);
1180 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1182 __sk_dst_set(sk, dst);
1183 sk->sk_route_caps = dst->dev->features;
1184 if (sk->sk_route_caps & NETIF_F_GSO)
1185 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1186 if (sk_can_gso(sk)) {
1187 if (dst->header_len) {
1188 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1190 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1191 sk->sk_gso_max_size = dst->dev->gso_max_size;
1195 EXPORT_SYMBOL_GPL(sk_setup_caps);
1197 void __init sk_init(void)
1199 if (num_physpages <= 4096) {
1200 sysctl_wmem_max = 32767;
1201 sysctl_rmem_max = 32767;
1202 sysctl_wmem_default = 32767;
1203 sysctl_rmem_default = 32767;
1204 } else if (num_physpages >= 131072) {
1205 sysctl_wmem_max = 131071;
1206 sysctl_rmem_max = 131071;
1211 * Simple resource managers for sockets.
1216 * Write buffer destructor automatically called from kfree_skb.
1218 void sock_wfree(struct sk_buff *skb)
1220 struct sock *sk = skb->sk;
1223 /* In case it might be waiting for more memory. */
1224 res = atomic_sub_return(skb->truesize, &sk->sk_wmem_alloc);
1225 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1226 sk->sk_write_space(sk);
1228 * if sk_wmem_alloc reached 0, we are last user and should
1229 * free this sock, as sk_free() call could not do it.
1234 EXPORT_SYMBOL(sock_wfree);
1237 * Read buffer destructor automatically called from kfree_skb.
1239 void sock_rfree(struct sk_buff *skb)
1241 struct sock *sk = skb->sk;
1243 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1244 sk_mem_uncharge(skb->sk, skb->truesize);
1246 EXPORT_SYMBOL(sock_rfree);
1249 int sock_i_uid(struct sock *sk)
1253 read_lock(&sk->sk_callback_lock);
1254 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1255 read_unlock(&sk->sk_callback_lock);
1258 EXPORT_SYMBOL(sock_i_uid);
1260 unsigned long sock_i_ino(struct sock *sk)
1264 read_lock(&sk->sk_callback_lock);
1265 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1266 read_unlock(&sk->sk_callback_lock);
1269 EXPORT_SYMBOL(sock_i_ino);
1272 * Allocate a skb from the socket's send buffer.
1274 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1277 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1278 struct sk_buff *skb = alloc_skb(size, priority);
1280 skb_set_owner_w(skb, sk);
1286 EXPORT_SYMBOL(sock_wmalloc);
1289 * Allocate a skb from the socket's receive buffer.
1291 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1294 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1295 struct sk_buff *skb = alloc_skb(size, priority);
1297 skb_set_owner_r(skb, sk);
1305 * Allocate a memory block from the socket's option memory buffer.
1307 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1309 if ((unsigned)size <= sysctl_optmem_max &&
1310 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1312 /* First do the add, to avoid the race if kmalloc
1315 atomic_add(size, &sk->sk_omem_alloc);
1316 mem = kmalloc(size, priority);
1319 atomic_sub(size, &sk->sk_omem_alloc);
1323 EXPORT_SYMBOL(sock_kmalloc);
1326 * Free an option memory block.
1328 void sock_kfree_s(struct sock *sk, void *mem, int size)
1331 atomic_sub(size, &sk->sk_omem_alloc);
1333 EXPORT_SYMBOL(sock_kfree_s);
1335 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1336 I think, these locks should be removed for datagram sockets.
1338 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1342 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1346 if (signal_pending(current))
1348 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1349 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1350 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1352 if (sk->sk_shutdown & SEND_SHUTDOWN)
1356 timeo = schedule_timeout(timeo);
1358 finish_wait(sk->sk_sleep, &wait);
1364 * Generic send/receive buffer handlers
1367 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1368 unsigned long data_len, int noblock,
1371 struct sk_buff *skb;
1376 gfp_mask = sk->sk_allocation;
1377 if (gfp_mask & __GFP_WAIT)
1378 gfp_mask |= __GFP_REPEAT;
1380 timeo = sock_sndtimeo(sk, noblock);
1382 err = sock_error(sk);
1387 if (sk->sk_shutdown & SEND_SHUTDOWN)
1390 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1391 skb = alloc_skb(header_len, gfp_mask);
1396 /* No pages, we're done... */
1400 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1401 skb->truesize += data_len;
1402 skb_shinfo(skb)->nr_frags = npages;
1403 for (i = 0; i < npages; i++) {
1407 page = alloc_pages(sk->sk_allocation, 0);
1410 skb_shinfo(skb)->nr_frags = i;
1415 frag = &skb_shinfo(skb)->frags[i];
1417 frag->page_offset = 0;
1418 frag->size = (data_len >= PAGE_SIZE ?
1421 data_len -= PAGE_SIZE;
1424 /* Full success... */
1430 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1431 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1435 if (signal_pending(current))
1437 timeo = sock_wait_for_wmem(sk, timeo);
1440 skb_set_owner_w(skb, sk);
1444 err = sock_intr_errno(timeo);
1449 EXPORT_SYMBOL(sock_alloc_send_pskb);
1451 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1452 int noblock, int *errcode)
1454 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1456 EXPORT_SYMBOL(sock_alloc_send_skb);
1458 static void __lock_sock(struct sock *sk)
1463 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1464 TASK_UNINTERRUPTIBLE);
1465 spin_unlock_bh(&sk->sk_lock.slock);
1467 spin_lock_bh(&sk->sk_lock.slock);
1468 if (!sock_owned_by_user(sk))
1471 finish_wait(&sk->sk_lock.wq, &wait);
1474 static void __release_sock(struct sock *sk)
1476 struct sk_buff *skb = sk->sk_backlog.head;
1479 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1483 struct sk_buff *next = skb->next;
1486 sk_backlog_rcv(sk, skb);
1489 * We are in process context here with softirqs
1490 * disabled, use cond_resched_softirq() to preempt.
1491 * This is safe to do because we've taken the backlog
1494 cond_resched_softirq();
1497 } while (skb != NULL);
1500 } while ((skb = sk->sk_backlog.head) != NULL);
1504 * sk_wait_data - wait for data to arrive at sk_receive_queue
1505 * @sk: sock to wait on
1506 * @timeo: for how long
1508 * Now socket state including sk->sk_err is changed only under lock,
1509 * hence we may omit checks after joining wait queue.
1510 * We check receive queue before schedule() only as optimization;
1511 * it is very likely that release_sock() added new data.
1513 int sk_wait_data(struct sock *sk, long *timeo)
1518 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1519 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1520 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1521 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1522 finish_wait(sk->sk_sleep, &wait);
1525 EXPORT_SYMBOL(sk_wait_data);
1528 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1530 * @size: memory size to allocate
1531 * @kind: allocation type
1533 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1534 * rmem allocation. This function assumes that protocols which have
1535 * memory_pressure use sk_wmem_queued as write buffer accounting.
1537 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1539 struct proto *prot = sk->sk_prot;
1540 int amt = sk_mem_pages(size);
1543 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1544 allocated = atomic_add_return(amt, prot->memory_allocated);
1547 if (allocated <= prot->sysctl_mem[0]) {
1548 if (prot->memory_pressure && *prot->memory_pressure)
1549 *prot->memory_pressure = 0;
1553 /* Under pressure. */
1554 if (allocated > prot->sysctl_mem[1])
1555 if (prot->enter_memory_pressure)
1556 prot->enter_memory_pressure(sk);
1558 /* Over hard limit. */
1559 if (allocated > prot->sysctl_mem[2])
1560 goto suppress_allocation;
1562 /* guarantee minimum buffer size under pressure */
1563 if (kind == SK_MEM_RECV) {
1564 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1566 } else { /* SK_MEM_SEND */
1567 if (sk->sk_type == SOCK_STREAM) {
1568 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1570 } else if (atomic_read(&sk->sk_wmem_alloc) <
1571 prot->sysctl_wmem[0])
1575 if (prot->memory_pressure) {
1578 if (!*prot->memory_pressure)
1580 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1581 if (prot->sysctl_mem[2] > alloc *
1582 sk_mem_pages(sk->sk_wmem_queued +
1583 atomic_read(&sk->sk_rmem_alloc) +
1584 sk->sk_forward_alloc))
1588 suppress_allocation:
1590 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1591 sk_stream_moderate_sndbuf(sk);
1593 /* Fail only if socket is _under_ its sndbuf.
1594 * In this case we cannot block, so that we have to fail.
1596 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1600 /* Alas. Undo changes. */
1601 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1602 atomic_sub(amt, prot->memory_allocated);
1605 EXPORT_SYMBOL(__sk_mem_schedule);
1608 * __sk_reclaim - reclaim memory_allocated
1611 void __sk_mem_reclaim(struct sock *sk)
1613 struct proto *prot = sk->sk_prot;
1615 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1616 prot->memory_allocated);
1617 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1619 if (prot->memory_pressure && *prot->memory_pressure &&
1620 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1621 *prot->memory_pressure = 0;
1623 EXPORT_SYMBOL(__sk_mem_reclaim);
1627 * Set of default routines for initialising struct proto_ops when
1628 * the protocol does not support a particular function. In certain
1629 * cases where it makes no sense for a protocol to have a "do nothing"
1630 * function, some default processing is provided.
1633 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1637 EXPORT_SYMBOL(sock_no_bind);
1639 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1644 EXPORT_SYMBOL(sock_no_connect);
1646 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1650 EXPORT_SYMBOL(sock_no_socketpair);
1652 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1656 EXPORT_SYMBOL(sock_no_accept);
1658 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1663 EXPORT_SYMBOL(sock_no_getname);
1665 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1669 EXPORT_SYMBOL(sock_no_poll);
1671 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1675 EXPORT_SYMBOL(sock_no_ioctl);
1677 int sock_no_listen(struct socket *sock, int backlog)
1681 EXPORT_SYMBOL(sock_no_listen);
1683 int sock_no_shutdown(struct socket *sock, int how)
1687 EXPORT_SYMBOL(sock_no_shutdown);
1689 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1690 char __user *optval, int optlen)
1694 EXPORT_SYMBOL(sock_no_setsockopt);
1696 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1697 char __user *optval, int __user *optlen)
1701 EXPORT_SYMBOL(sock_no_getsockopt);
1703 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1708 EXPORT_SYMBOL(sock_no_sendmsg);
1710 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1711 size_t len, int flags)
1715 EXPORT_SYMBOL(sock_no_recvmsg);
1717 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1719 /* Mirror missing mmap method error code */
1722 EXPORT_SYMBOL(sock_no_mmap);
1724 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1727 struct msghdr msg = {.msg_flags = flags};
1729 char *kaddr = kmap(page);
1730 iov.iov_base = kaddr + offset;
1732 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1736 EXPORT_SYMBOL(sock_no_sendpage);
1739 * Default Socket Callbacks
1742 static void sock_def_wakeup(struct sock *sk)
1744 read_lock(&sk->sk_callback_lock);
1745 if (sk_has_sleeper(sk))
1746 wake_up_interruptible_all(sk->sk_sleep);
1747 read_unlock(&sk->sk_callback_lock);
1750 static void sock_def_error_report(struct sock *sk)
1752 read_lock(&sk->sk_callback_lock);
1753 if (sk_has_sleeper(sk))
1754 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1755 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1756 read_unlock(&sk->sk_callback_lock);
1759 static void sock_def_readable(struct sock *sk, int len)
1761 read_lock(&sk->sk_callback_lock);
1762 if (sk_has_sleeper(sk))
1763 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1764 POLLRDNORM | POLLRDBAND);
1765 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1766 read_unlock(&sk->sk_callback_lock);
1769 static void sock_def_write_space(struct sock *sk)
1771 read_lock(&sk->sk_callback_lock);
1773 /* Do not wake up a writer until he can make "significant"
1776 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1777 if (sk_has_sleeper(sk))
1778 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1779 POLLWRNORM | POLLWRBAND);
1781 /* Should agree with poll, otherwise some programs break */
1782 if (sock_writeable(sk))
1783 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1786 read_unlock(&sk->sk_callback_lock);
1789 static void sock_def_destruct(struct sock *sk)
1791 kfree(sk->sk_protinfo);
1794 void sk_send_sigurg(struct sock *sk)
1796 if (sk->sk_socket && sk->sk_socket->file)
1797 if (send_sigurg(&sk->sk_socket->file->f_owner))
1798 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1800 EXPORT_SYMBOL(sk_send_sigurg);
1802 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1803 unsigned long expires)
1805 if (!mod_timer(timer, expires))
1808 EXPORT_SYMBOL(sk_reset_timer);
1810 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1812 if (timer_pending(timer) && del_timer(timer))
1815 EXPORT_SYMBOL(sk_stop_timer);
1817 void sock_init_data(struct socket *sock, struct sock *sk)
1819 skb_queue_head_init(&sk->sk_receive_queue);
1820 skb_queue_head_init(&sk->sk_write_queue);
1821 skb_queue_head_init(&sk->sk_error_queue);
1822 #ifdef CONFIG_NET_DMA
1823 skb_queue_head_init(&sk->sk_async_wait_queue);
1826 sk->sk_send_head = NULL;
1828 init_timer(&sk->sk_timer);
1830 sk->sk_allocation = GFP_KERNEL;
1831 sk->sk_rcvbuf = sysctl_rmem_default;
1832 sk->sk_sndbuf = sysctl_wmem_default;
1833 sk->sk_state = TCP_CLOSE;
1834 sk_set_socket(sk, sock);
1836 sock_set_flag(sk, SOCK_ZAPPED);
1839 sk->sk_type = sock->type;
1840 sk->sk_sleep = &sock->wait;
1843 sk->sk_sleep = NULL;
1845 rwlock_init(&sk->sk_dst_lock);
1846 rwlock_init(&sk->sk_callback_lock);
1847 lockdep_set_class_and_name(&sk->sk_callback_lock,
1848 af_callback_keys + sk->sk_family,
1849 af_family_clock_key_strings[sk->sk_family]);
1851 sk->sk_state_change = sock_def_wakeup;
1852 sk->sk_data_ready = sock_def_readable;
1853 sk->sk_write_space = sock_def_write_space;
1854 sk->sk_error_report = sock_def_error_report;
1855 sk->sk_destruct = sock_def_destruct;
1857 sk->sk_sndmsg_page = NULL;
1858 sk->sk_sndmsg_off = 0;
1860 sk->sk_peercred.pid = 0;
1861 sk->sk_peercred.uid = -1;
1862 sk->sk_peercred.gid = -1;
1863 sk->sk_write_pending = 0;
1864 sk->sk_rcvlowat = 1;
1865 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1866 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1868 sk->sk_stamp = ktime_set(-1L, 0);
1871 * Before updating sk_refcnt, we must commit prior changes to memory
1872 * (Documentation/RCU/rculist_nulls.txt for details)
1875 atomic_set(&sk->sk_refcnt, 1);
1876 atomic_set(&sk->sk_drops, 0);
1878 EXPORT_SYMBOL(sock_init_data);
1880 void lock_sock_nested(struct sock *sk, int subclass)
1883 spin_lock_bh(&sk->sk_lock.slock);
1884 if (sk->sk_lock.owned)
1886 sk->sk_lock.owned = 1;
1887 spin_unlock(&sk->sk_lock.slock);
1889 * The sk_lock has mutex_lock() semantics here:
1891 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1894 EXPORT_SYMBOL(lock_sock_nested);
1896 void release_sock(struct sock *sk)
1899 * The sk_lock has mutex_unlock() semantics:
1901 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1903 spin_lock_bh(&sk->sk_lock.slock);
1904 if (sk->sk_backlog.tail)
1906 sk->sk_lock.owned = 0;
1907 if (waitqueue_active(&sk->sk_lock.wq))
1908 wake_up(&sk->sk_lock.wq);
1909 spin_unlock_bh(&sk->sk_lock.slock);
1911 EXPORT_SYMBOL(release_sock);
1913 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1916 if (!sock_flag(sk, SOCK_TIMESTAMP))
1917 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1918 tv = ktime_to_timeval(sk->sk_stamp);
1919 if (tv.tv_sec == -1)
1921 if (tv.tv_sec == 0) {
1922 sk->sk_stamp = ktime_get_real();
1923 tv = ktime_to_timeval(sk->sk_stamp);
1925 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1927 EXPORT_SYMBOL(sock_get_timestamp);
1929 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1932 if (!sock_flag(sk, SOCK_TIMESTAMP))
1933 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1934 ts = ktime_to_timespec(sk->sk_stamp);
1935 if (ts.tv_sec == -1)
1937 if (ts.tv_sec == 0) {
1938 sk->sk_stamp = ktime_get_real();
1939 ts = ktime_to_timespec(sk->sk_stamp);
1941 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1943 EXPORT_SYMBOL(sock_get_timestampns);
1945 void sock_enable_timestamp(struct sock *sk, int flag)
1947 if (!sock_flag(sk, flag)) {
1948 sock_set_flag(sk, flag);
1950 * we just set one of the two flags which require net
1951 * time stamping, but time stamping might have been on
1952 * already because of the other one
1955 flag == SOCK_TIMESTAMP ?
1956 SOCK_TIMESTAMPING_RX_SOFTWARE :
1958 net_enable_timestamp();
1963 * Get a socket option on an socket.
1965 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1966 * asynchronous errors should be reported by getsockopt. We assume
1967 * this means if you specify SO_ERROR (otherwise whats the point of it).
1969 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1970 char __user *optval, int __user *optlen)
1972 struct sock *sk = sock->sk;
1974 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1976 EXPORT_SYMBOL(sock_common_getsockopt);
1978 #ifdef CONFIG_COMPAT
1979 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1980 char __user *optval, int __user *optlen)
1982 struct sock *sk = sock->sk;
1984 if (sk->sk_prot->compat_getsockopt != NULL)
1985 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1987 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1989 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1992 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1993 struct msghdr *msg, size_t size, int flags)
1995 struct sock *sk = sock->sk;
1999 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2000 flags & ~MSG_DONTWAIT, &addr_len);
2002 msg->msg_namelen = addr_len;
2005 EXPORT_SYMBOL(sock_common_recvmsg);
2008 * Set socket options on an inet socket.
2010 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2011 char __user *optval, int optlen)
2013 struct sock *sk = sock->sk;
2015 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2017 EXPORT_SYMBOL(sock_common_setsockopt);
2019 #ifdef CONFIG_COMPAT
2020 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2021 char __user *optval, int optlen)
2023 struct sock *sk = sock->sk;
2025 if (sk->sk_prot->compat_setsockopt != NULL)
2026 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2028 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2030 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2033 void sk_common_release(struct sock *sk)
2035 if (sk->sk_prot->destroy)
2036 sk->sk_prot->destroy(sk);
2039 * Observation: when sock_common_release is called, processes have
2040 * no access to socket. But net still has.
2041 * Step one, detach it from networking:
2043 * A. Remove from hash tables.
2046 sk->sk_prot->unhash(sk);
2049 * In this point socket cannot receive new packets, but it is possible
2050 * that some packets are in flight because some CPU runs receiver and
2051 * did hash table lookup before we unhashed socket. They will achieve
2052 * receive queue and will be purged by socket destructor.
2054 * Also we still have packets pending on receive queue and probably,
2055 * our own packets waiting in device queues. sock_destroy will drain
2056 * receive queue, but transmitted packets will delay socket destruction
2057 * until the last reference will be released.
2062 xfrm_sk_free_policy(sk);
2064 sk_refcnt_debug_release(sk);
2067 EXPORT_SYMBOL(sk_common_release);
2069 static DEFINE_RWLOCK(proto_list_lock);
2070 static LIST_HEAD(proto_list);
2072 #ifdef CONFIG_PROC_FS
2073 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2075 int val[PROTO_INUSE_NR];
2078 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2080 #ifdef CONFIG_NET_NS
2081 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2083 int cpu = smp_processor_id();
2084 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2086 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2088 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2090 int cpu, idx = prot->inuse_idx;
2093 for_each_possible_cpu(cpu)
2094 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2096 return res >= 0 ? res : 0;
2098 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2100 static int sock_inuse_init_net(struct net *net)
2102 net->core.inuse = alloc_percpu(struct prot_inuse);
2103 return net->core.inuse ? 0 : -ENOMEM;
2106 static void sock_inuse_exit_net(struct net *net)
2108 free_percpu(net->core.inuse);
2111 static struct pernet_operations net_inuse_ops = {
2112 .init = sock_inuse_init_net,
2113 .exit = sock_inuse_exit_net,
2116 static __init int net_inuse_init(void)
2118 if (register_pernet_subsys(&net_inuse_ops))
2119 panic("Cannot initialize net inuse counters");
2124 core_initcall(net_inuse_init);
2126 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2128 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2130 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2132 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2134 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2136 int cpu, idx = prot->inuse_idx;
2139 for_each_possible_cpu(cpu)
2140 res += per_cpu(prot_inuse, cpu).val[idx];
2142 return res >= 0 ? res : 0;
2144 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2147 static void assign_proto_idx(struct proto *prot)
2149 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2151 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2152 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2156 set_bit(prot->inuse_idx, proto_inuse_idx);
2159 static void release_proto_idx(struct proto *prot)
2161 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2162 clear_bit(prot->inuse_idx, proto_inuse_idx);
2165 static inline void assign_proto_idx(struct proto *prot)
2169 static inline void release_proto_idx(struct proto *prot)
2174 int proto_register(struct proto *prot, int alloc_slab)
2177 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2178 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2181 if (prot->slab == NULL) {
2182 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2187 if (prot->rsk_prot != NULL) {
2188 static const char mask[] = "request_sock_%s";
2190 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2191 if (prot->rsk_prot->slab_name == NULL)
2192 goto out_free_sock_slab;
2194 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2195 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2196 prot->rsk_prot->obj_size, 0,
2197 SLAB_HWCACHE_ALIGN, NULL);
2199 if (prot->rsk_prot->slab == NULL) {
2200 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2202 goto out_free_request_sock_slab_name;
2206 if (prot->twsk_prot != NULL) {
2207 static const char mask[] = "tw_sock_%s";
2209 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2211 if (prot->twsk_prot->twsk_slab_name == NULL)
2212 goto out_free_request_sock_slab;
2214 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2215 prot->twsk_prot->twsk_slab =
2216 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2217 prot->twsk_prot->twsk_obj_size,
2219 SLAB_HWCACHE_ALIGN |
2222 if (prot->twsk_prot->twsk_slab == NULL)
2223 goto out_free_timewait_sock_slab_name;
2227 write_lock(&proto_list_lock);
2228 list_add(&prot->node, &proto_list);
2229 assign_proto_idx(prot);
2230 write_unlock(&proto_list_lock);
2233 out_free_timewait_sock_slab_name:
2234 kfree(prot->twsk_prot->twsk_slab_name);
2235 out_free_request_sock_slab:
2236 if (prot->rsk_prot && prot->rsk_prot->slab) {
2237 kmem_cache_destroy(prot->rsk_prot->slab);
2238 prot->rsk_prot->slab = NULL;
2240 out_free_request_sock_slab_name:
2241 kfree(prot->rsk_prot->slab_name);
2243 kmem_cache_destroy(prot->slab);
2248 EXPORT_SYMBOL(proto_register);
2250 void proto_unregister(struct proto *prot)
2252 write_lock(&proto_list_lock);
2253 release_proto_idx(prot);
2254 list_del(&prot->node);
2255 write_unlock(&proto_list_lock);
2257 if (prot->slab != NULL) {
2258 kmem_cache_destroy(prot->slab);
2262 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2263 kmem_cache_destroy(prot->rsk_prot->slab);
2264 kfree(prot->rsk_prot->slab_name);
2265 prot->rsk_prot->slab = NULL;
2268 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2269 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2270 kfree(prot->twsk_prot->twsk_slab_name);
2271 prot->twsk_prot->twsk_slab = NULL;
2274 EXPORT_SYMBOL(proto_unregister);
2276 #ifdef CONFIG_PROC_FS
2277 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2278 __acquires(proto_list_lock)
2280 read_lock(&proto_list_lock);
2281 return seq_list_start_head(&proto_list, *pos);
2284 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2286 return seq_list_next(v, &proto_list, pos);
2289 static void proto_seq_stop(struct seq_file *seq, void *v)
2290 __releases(proto_list_lock)
2292 read_unlock(&proto_list_lock);
2295 static char proto_method_implemented(const void *method)
2297 return method == NULL ? 'n' : 'y';
2300 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2302 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2303 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2306 sock_prot_inuse_get(seq_file_net(seq), proto),
2307 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2308 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2310 proto->slab == NULL ? "no" : "yes",
2311 module_name(proto->owner),
2312 proto_method_implemented(proto->close),
2313 proto_method_implemented(proto->connect),
2314 proto_method_implemented(proto->disconnect),
2315 proto_method_implemented(proto->accept),
2316 proto_method_implemented(proto->ioctl),
2317 proto_method_implemented(proto->init),
2318 proto_method_implemented(proto->destroy),
2319 proto_method_implemented(proto->shutdown),
2320 proto_method_implemented(proto->setsockopt),
2321 proto_method_implemented(proto->getsockopt),
2322 proto_method_implemented(proto->sendmsg),
2323 proto_method_implemented(proto->recvmsg),
2324 proto_method_implemented(proto->sendpage),
2325 proto_method_implemented(proto->bind),
2326 proto_method_implemented(proto->backlog_rcv),
2327 proto_method_implemented(proto->hash),
2328 proto_method_implemented(proto->unhash),
2329 proto_method_implemented(proto->get_port),
2330 proto_method_implemented(proto->enter_memory_pressure));
2333 static int proto_seq_show(struct seq_file *seq, void *v)
2335 if (v == &proto_list)
2336 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2345 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2347 proto_seq_printf(seq, list_entry(v, struct proto, node));
2351 static const struct seq_operations proto_seq_ops = {
2352 .start = proto_seq_start,
2353 .next = proto_seq_next,
2354 .stop = proto_seq_stop,
2355 .show = proto_seq_show,
2358 static int proto_seq_open(struct inode *inode, struct file *file)
2360 return seq_open_net(inode, file, &proto_seq_ops,
2361 sizeof(struct seq_net_private));
2364 static const struct file_operations proto_seq_fops = {
2365 .owner = THIS_MODULE,
2366 .open = proto_seq_open,
2368 .llseek = seq_lseek,
2369 .release = seq_release_net,
2372 static __net_init int proto_init_net(struct net *net)
2374 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2380 static __net_exit void proto_exit_net(struct net *net)
2382 proc_net_remove(net, "protocols");
2386 static __net_initdata struct pernet_operations proto_net_ops = {
2387 .init = proto_init_net,
2388 .exit = proto_exit_net,
2391 static int __init proto_init(void)
2393 return register_pernet_subsys(&proto_net_ops);
2396 subsys_initcall(proto_init);
2398 #endif /* PROC_FS */