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>
113 #include <linux/reserve.h>
115 #include <asm/uaccess.h>
116 #include <asm/system.h>
118 #include <linux/netdevice.h>
119 #include <net/protocol.h>
120 #include <linux/skbuff.h>
121 #include <net/net_namespace.h>
122 #include <net/request_sock.h>
123 #include <net/sock.h>
124 #include <linux/net_tstamp.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
128 #include <linux/filter.h>
135 * Each address family might have different locking rules, so we have
136 * one slock key per address family:
138 static struct lock_class_key af_family_keys[AF_MAX];
139 static struct lock_class_key af_family_slock_keys[AF_MAX];
142 * Make lock validator output more readable. (we pre-construct these
143 * strings build-time, so that runtime initialization of socket
146 static const char *const af_family_key_strings[AF_MAX+1] = {
147 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
148 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
149 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
150 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
151 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
152 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
153 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
154 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
155 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
156 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
157 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
158 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
159 "sk_lock-AF_IEEE802154",
162 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
163 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
164 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
165 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
166 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
167 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
168 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
169 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
170 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
171 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
172 "slock-27" , "slock-28" , "slock-AF_CAN" ,
173 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
174 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
175 "slock-AF_IEEE802154",
178 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
179 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
180 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
181 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
182 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
183 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
184 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
185 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
186 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
187 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
188 "clock-27" , "clock-28" , "clock-AF_CAN" ,
189 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
190 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
191 "clock-AF_IEEE802154",
196 * sk_callback_lock locking rules are per-address-family,
197 * so split the lock classes by using a per-AF key:
199 static struct lock_class_key af_callback_keys[AF_MAX];
201 /* Take into consideration the size of the struct sk_buff overhead in the
202 * determination of these values, since that is non-constant across
203 * platforms. This makes socket queueing behavior and performance
204 * not depend upon such differences.
206 #define _SK_MEM_PACKETS 256
207 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
208 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
209 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
211 /* Run time adjustable parameters. */
212 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
213 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
214 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
215 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
217 /* Maximal space eaten by iovec or ancilliary data plus some space */
218 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
219 EXPORT_SYMBOL(sysctl_optmem_max);
221 static struct mem_reserve net_reserve;
222 struct mem_reserve net_rx_reserve;
223 EXPORT_SYMBOL_GPL(net_rx_reserve); /* modular ipv6 only */
224 struct mem_reserve net_skb_reserve;
225 EXPORT_SYMBOL_GPL(net_skb_reserve); /* modular ipv6 only */
226 static struct mem_reserve net_tx_reserve;
227 static struct mem_reserve net_tx_pages;
230 static DEFINE_MUTEX(memalloc_socks_lock);
234 * sk_adjust_memalloc - adjust the global memalloc reserve for critical RX
235 * @socks: number of new %SOCK_MEMALLOC sockets
236 * @tx_resserve_pages: number of pages to (un)reserve for TX
238 * This function adjusts the memalloc reserve based on system demand.
239 * The RX reserve is a limit, and only added once, not for each socket.
242 * @tx_reserve_pages is an upper-bound of memory used for TX hence
243 * we need not account the pages like we do for RX pages.
245 int sk_adjust_memalloc(int socks, long tx_reserve_pages)
249 mutex_lock(&memalloc_socks_lock);
250 err = mem_reserve_pages_add(&net_tx_pages, tx_reserve_pages);
255 * either socks is positive and we need to check for 0 -> !0
256 * transition and connect the reserve tree when we observe it.
258 if (!memalloc_socks && socks > 0) {
259 err = mem_reserve_connect(&net_reserve, &mem_reserve_root);
262 * if we failed to connect the tree, undo the tx
263 * reserve so that failure has no side effects.
265 mem_reserve_pages_add(&net_tx_pages, -tx_reserve_pages);
269 memalloc_socks += socks;
271 * or socks is negative and we must observe the !0 -> 0 transition
272 * and disconnect the reserve tree.
274 if (!memalloc_socks && socks)
275 mem_reserve_disconnect(&net_reserve);
278 mutex_unlock(&memalloc_socks_lock);
282 EXPORT_SYMBOL_GPL(sk_adjust_memalloc);
285 * sk_set_memalloc - sets %SOCK_MEMALLOC
286 * @sk: socket to set it on
288 * Set %SOCK_MEMALLOC on a socket and increase the memalloc reserve
291 int sk_set_memalloc(struct sock *sk)
293 int set = sock_flag(sk, SOCK_MEMALLOC);
296 int err = sk_adjust_memalloc(1, 0);
300 sock_set_flag(sk, SOCK_MEMALLOC);
301 sk->sk_allocation |= __GFP_MEMALLOC;
305 EXPORT_SYMBOL_GPL(sk_set_memalloc);
307 int sk_clear_memalloc(struct sock *sk)
309 int set = sock_flag(sk, SOCK_MEMALLOC);
311 sk_adjust_memalloc(-1, 0);
312 sock_reset_flag(sk, SOCK_MEMALLOC);
313 sk->sk_allocation &= ~__GFP_MEMALLOC;
317 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
319 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
322 unsigned long pflags = current->flags;
324 /* these should have been dropped before queueing */
325 BUG_ON(!sk_has_memalloc(sk));
327 current->flags |= PF_MEMALLOC;
328 ret = sk->sk_backlog_rcv(sk, skb);
329 tsk_restore_flags(current, pflags, PF_MEMALLOC);
333 EXPORT_SYMBOL(__sk_backlog_rcv);
336 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
340 if (optlen < sizeof(tv))
342 if (copy_from_user(&tv, optval, sizeof(tv)))
344 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
348 static int warned __read_mostly;
351 if (warned < 10 && net_ratelimit()) {
353 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
354 "tries to set negative timeout\n",
355 current->comm, task_pid_nr(current));
359 *timeo_p = MAX_SCHEDULE_TIMEOUT;
360 if (tv.tv_sec == 0 && tv.tv_usec == 0)
362 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
363 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
367 static void sock_warn_obsolete_bsdism(const char *name)
370 static char warncomm[TASK_COMM_LEN];
371 if (strcmp(warncomm, current->comm) && warned < 5) {
372 strcpy(warncomm, current->comm);
373 printk(KERN_WARNING "process `%s' is using obsolete "
374 "%s SO_BSDCOMPAT\n", warncomm, name);
379 static void sock_disable_timestamp(struct sock *sk, int flag)
381 if (sock_flag(sk, flag)) {
382 sock_reset_flag(sk, flag);
383 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
384 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
385 net_disable_timestamp();
391 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
396 struct sk_buff_head *list = &sk->sk_receive_queue;
398 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
399 number of warnings when compiling with -W --ANK
401 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
402 (unsigned)sk->sk_rcvbuf) {
403 atomic_inc(&sk->sk_drops);
407 err = sk_filter(sk, skb);
411 if (!sk_rmem_schedule(sk, skb)) {
412 atomic_inc(&sk->sk_drops);
417 skb_set_owner_r(skb, sk);
419 /* Cache the SKB length before we tack it onto the receive
420 * queue. Once it is added it no longer belongs to us and
421 * may be freed by other threads of control pulling packets
426 spin_lock_irqsave(&list->lock, flags);
427 skb->dropcount = atomic_read(&sk->sk_drops);
428 __skb_queue_tail(list, skb);
429 spin_unlock_irqrestore(&list->lock, flags);
431 if (!sock_flag(sk, SOCK_DEAD))
432 sk->sk_data_ready(sk, skb_len);
435 EXPORT_SYMBOL(sock_queue_rcv_skb);
437 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
439 int rc = NET_RX_SUCCESS;
441 if (sk_filter(sk, skb))
442 goto discard_and_relse;
447 bh_lock_sock_nested(sk);
450 if (!sock_owned_by_user(sk)) {
452 * trylock + unlock semantics:
454 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
456 rc = sk_backlog_rcv(sk, skb);
458 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
460 sk_add_backlog(sk, skb);
469 EXPORT_SYMBOL(sk_receive_skb);
471 void sk_reset_txq(struct sock *sk)
473 sk_tx_queue_clear(sk);
475 EXPORT_SYMBOL(sk_reset_txq);
477 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
479 struct dst_entry *dst = sk->sk_dst_cache;
481 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
482 sk_tx_queue_clear(sk);
483 sk->sk_dst_cache = NULL;
490 EXPORT_SYMBOL(__sk_dst_check);
492 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
494 struct dst_entry *dst = sk_dst_get(sk);
496 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
504 EXPORT_SYMBOL(sk_dst_check);
506 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
508 int ret = -ENOPROTOOPT;
509 #ifdef CONFIG_NETDEVICES
510 struct net *net = sock_net(sk);
511 char devname[IFNAMSIZ];
516 if (!capable(CAP_NET_RAW))
523 /* Bind this socket to a particular device like "eth0",
524 * as specified in the passed interface name. If the
525 * name is "" or the option length is zero the socket
528 if (optlen > IFNAMSIZ - 1)
529 optlen = IFNAMSIZ - 1;
530 memset(devname, 0, sizeof(devname));
533 if (copy_from_user(devname, optval, optlen))
537 if (devname[0] != '\0') {
538 struct net_device *dev;
541 dev = dev_get_by_name_rcu(net, devname);
543 index = dev->ifindex;
551 sk->sk_bound_dev_if = index;
563 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
566 sock_set_flag(sk, bit);
568 sock_reset_flag(sk, bit);
572 * This is meant for all protocols to use and covers goings on
573 * at the socket level. Everything here is generic.
576 int sock_setsockopt(struct socket *sock, int level, int optname,
577 char __user *optval, unsigned int optlen)
579 struct sock *sk = sock->sk;
586 * Options without arguments
589 if (optname == SO_BINDTODEVICE)
590 return sock_bindtodevice(sk, optval, optlen);
592 if (optlen < sizeof(int))
595 if (get_user(val, (int __user *)optval))
598 valbool = val ? 1 : 0;
604 if (val && !capable(CAP_NET_ADMIN))
607 sock_valbool_flag(sk, SOCK_DBG, valbool);
610 sk->sk_reuse = valbool;
619 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
622 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
625 /* Don't error on this BSD doesn't and if you think
626 about it this is right. Otherwise apps have to
627 play 'guess the biggest size' games. RCVBUF/SNDBUF
628 are treated in BSD as hints */
630 if (val > sysctl_wmem_max)
631 val = sysctl_wmem_max;
633 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
634 if ((val * 2) < SOCK_MIN_SNDBUF)
635 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
637 sk->sk_sndbuf = val * 2;
640 * Wake up sending tasks if we
643 sk->sk_write_space(sk);
647 if (!capable(CAP_NET_ADMIN)) {
654 /* Don't error on this BSD doesn't and if you think
655 about it this is right. Otherwise apps have to
656 play 'guess the biggest size' games. RCVBUF/SNDBUF
657 are treated in BSD as hints */
659 if (val > sysctl_rmem_max)
660 val = sysctl_rmem_max;
662 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
664 * We double it on the way in to account for
665 * "struct sk_buff" etc. overhead. Applications
666 * assume that the SO_RCVBUF setting they make will
667 * allow that much actual data to be received on that
670 * Applications are unaware that "struct sk_buff" and
671 * other overheads allocate from the receive buffer
672 * during socket buffer allocation.
674 * And after considering the possible alternatives,
675 * returning the value we actually used in getsockopt
676 * is the most desirable behavior.
678 if ((val * 2) < SOCK_MIN_RCVBUF)
679 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
681 sk->sk_rcvbuf = val * 2;
685 if (!capable(CAP_NET_ADMIN)) {
693 if (sk->sk_protocol == IPPROTO_TCP)
694 tcp_set_keepalive(sk, valbool);
696 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
700 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
704 sk->sk_no_check = valbool;
708 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
709 sk->sk_priority = val;
715 if (optlen < sizeof(ling)) {
716 ret = -EINVAL; /* 1003.1g */
719 if (copy_from_user(&ling, optval, sizeof(ling))) {
724 sock_reset_flag(sk, SOCK_LINGER);
726 #if (BITS_PER_LONG == 32)
727 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
728 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
731 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
732 sock_set_flag(sk, SOCK_LINGER);
737 sock_warn_obsolete_bsdism("setsockopt");
742 set_bit(SOCK_PASSCRED, &sock->flags);
744 clear_bit(SOCK_PASSCRED, &sock->flags);
750 if (optname == SO_TIMESTAMP)
751 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
753 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
754 sock_set_flag(sk, SOCK_RCVTSTAMP);
755 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
757 sock_reset_flag(sk, SOCK_RCVTSTAMP);
758 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
762 case SO_TIMESTAMPING:
763 if (val & ~SOF_TIMESTAMPING_MASK) {
767 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
768 val & SOF_TIMESTAMPING_TX_HARDWARE);
769 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
770 val & SOF_TIMESTAMPING_TX_SOFTWARE);
771 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
772 val & SOF_TIMESTAMPING_RX_HARDWARE);
773 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
774 sock_enable_timestamp(sk,
775 SOCK_TIMESTAMPING_RX_SOFTWARE);
777 sock_disable_timestamp(sk,
778 SOCK_TIMESTAMPING_RX_SOFTWARE);
779 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
780 val & SOF_TIMESTAMPING_SOFTWARE);
781 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
782 val & SOF_TIMESTAMPING_SYS_HARDWARE);
783 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
784 val & SOF_TIMESTAMPING_RAW_HARDWARE);
790 sk->sk_rcvlowat = val ? : 1;
794 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
798 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
801 case SO_ATTACH_FILTER:
803 if (optlen == sizeof(struct sock_fprog)) {
804 struct sock_fprog fprog;
807 if (copy_from_user(&fprog, optval, sizeof(fprog)))
810 ret = sk_attach_filter(&fprog, sk);
814 case SO_DETACH_FILTER:
815 ret = sk_detach_filter(sk);
820 set_bit(SOCK_PASSSEC, &sock->flags);
822 clear_bit(SOCK_PASSSEC, &sock->flags);
825 if (!capable(CAP_NET_ADMIN))
831 /* We implement the SO_SNDLOWAT etc to
832 not be settable (1003.1g 5.3) */
835 sock_set_flag(sk, SOCK_RXQ_OVFL);
837 sock_reset_flag(sk, SOCK_RXQ_OVFL);
846 EXPORT_SYMBOL(sock_setsockopt);
849 int sock_getsockopt(struct socket *sock, int level, int optname,
850 char __user *optval, int __user *optlen)
852 struct sock *sk = sock->sk;
860 unsigned int lv = sizeof(int);
863 if (get_user(len, optlen))
868 memset(&v, 0, sizeof(v));
872 v.val = sock_flag(sk, SOCK_DBG);
876 v.val = sock_flag(sk, SOCK_LOCALROUTE);
880 v.val = !!sock_flag(sk, SOCK_BROADCAST);
884 v.val = sk->sk_sndbuf;
888 v.val = sk->sk_rcvbuf;
892 v.val = sk->sk_reuse;
896 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
904 v.val = sk->sk_protocol;
908 v.val = sk->sk_family;
912 v.val = -sock_error(sk);
914 v.val = xchg(&sk->sk_err_soft, 0);
918 v.val = !!sock_flag(sk, SOCK_URGINLINE);
922 v.val = sk->sk_no_check;
926 v.val = sk->sk_priority;
931 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
932 v.ling.l_linger = sk->sk_lingertime / HZ;
936 sock_warn_obsolete_bsdism("getsockopt");
940 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
941 !sock_flag(sk, SOCK_RCVTSTAMPNS);
945 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
948 case SO_TIMESTAMPING:
950 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
951 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
952 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
953 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
954 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
955 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
956 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
957 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
958 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
959 v.val |= SOF_TIMESTAMPING_SOFTWARE;
960 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
961 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
962 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
963 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
967 lv = sizeof(struct timeval);
968 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
972 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
973 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
978 lv = sizeof(struct timeval);
979 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
983 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
984 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
989 v.val = sk->sk_rcvlowat;
997 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
1001 if (len > sizeof(sk->sk_peercred))
1002 len = sizeof(sk->sk_peercred);
1003 if (copy_to_user(optval, &sk->sk_peercred, len))
1011 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1015 if (copy_to_user(optval, address, len))
1020 /* Dubious BSD thing... Probably nobody even uses it, but
1021 * the UNIX standard wants it for whatever reason... -DaveM
1024 v.val = sk->sk_state == TCP_LISTEN;
1028 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
1032 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1035 v.val = sk->sk_mark;
1039 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
1043 return -ENOPROTOOPT;
1048 if (copy_to_user(optval, &v, len))
1051 if (put_user(len, optlen))
1057 * Initialize an sk_lock.
1059 * (We also register the sk_lock with the lock validator.)
1061 static inline void sock_lock_init(struct sock *sk)
1063 sock_lock_init_class_and_name(sk,
1064 af_family_slock_key_strings[sk->sk_family],
1065 af_family_slock_keys + sk->sk_family,
1066 af_family_key_strings[sk->sk_family],
1067 af_family_keys + sk->sk_family);
1071 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1072 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1074 static void sock_copy(struct sock *nsk, const struct sock *osk)
1076 #ifdef CONFIG_SECURITY_NETWORK
1077 void *sptr = nsk->sk_security;
1079 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
1080 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
1081 sizeof(osk->sk_tx_queue_mapping));
1082 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
1083 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
1084 #ifdef CONFIG_SECURITY_NETWORK
1085 nsk->sk_security = sptr;
1086 security_sk_clone(osk, nsk);
1090 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1094 struct kmem_cache *slab;
1098 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1101 if (priority & __GFP_ZERO) {
1103 * caches using SLAB_DESTROY_BY_RCU should let
1104 * sk_node.next un-modified. Special care is taken
1105 * when initializing object to zero.
1107 if (offsetof(struct sock, sk_node.next) != 0)
1108 memset(sk, 0, offsetof(struct sock, sk_node.next));
1109 memset(&sk->sk_node.pprev, 0,
1110 prot->obj_size - offsetof(struct sock,
1115 sk = kmalloc(prot->obj_size, priority);
1118 kmemcheck_annotate_bitfield(sk, flags);
1120 if (security_sk_alloc(sk, family, priority))
1123 if (!try_module_get(prot->owner))
1125 sk_tx_queue_clear(sk);
1131 security_sk_free(sk);
1134 kmem_cache_free(slab, sk);
1140 static void sk_prot_free(struct proto *prot, struct sock *sk)
1142 struct kmem_cache *slab;
1143 struct module *owner;
1145 owner = prot->owner;
1148 security_sk_free(sk);
1150 kmem_cache_free(slab, sk);
1157 * sk_alloc - All socket objects are allocated here
1158 * @net: the applicable net namespace
1159 * @family: protocol family
1160 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1161 * @prot: struct proto associated with this new sock instance
1163 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1168 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1170 sk->sk_family = family;
1172 * See comment in struct sock definition to understand
1173 * why we need sk_prot_creator -acme
1175 sk->sk_prot = sk->sk_prot_creator = prot;
1177 sock_net_set(sk, get_net(net));
1178 atomic_set(&sk->sk_wmem_alloc, 1);
1183 EXPORT_SYMBOL(sk_alloc);
1185 static void __sk_free(struct sock *sk)
1187 struct sk_filter *filter;
1189 sk_clear_memalloc(sk);
1190 if (sk->sk_destruct)
1191 sk->sk_destruct(sk);
1193 filter = rcu_dereference(sk->sk_filter);
1195 sk_filter_uncharge(sk, filter);
1196 rcu_assign_pointer(sk->sk_filter, NULL);
1199 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1200 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1202 if (atomic_read(&sk->sk_omem_alloc))
1203 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1204 __func__, atomic_read(&sk->sk_omem_alloc));
1206 put_net(sock_net(sk));
1207 sk_prot_free(sk->sk_prot_creator, sk);
1210 void sk_free(struct sock *sk)
1213 * We substract one from sk_wmem_alloc and can know if
1214 * some packets are still in some tx queue.
1215 * If not null, sock_wfree() will call __sk_free(sk) later
1217 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1220 EXPORT_SYMBOL(sk_free);
1223 * Last sock_put should drop referrence to sk->sk_net. It has already
1224 * been dropped in sk_change_net. Taking referrence to stopping namespace
1226 * Take referrence to a socket to remove it from hash _alive_ and after that
1227 * destroy it in the context of init_net.
1229 void sk_release_kernel(struct sock *sk)
1231 if (sk == NULL || sk->sk_socket == NULL)
1235 sock_release(sk->sk_socket);
1236 release_net(sock_net(sk));
1237 sock_net_set(sk, get_net(&init_net));
1240 EXPORT_SYMBOL(sk_release_kernel);
1242 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1246 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1247 if (newsk != NULL) {
1248 struct sk_filter *filter;
1250 sock_copy(newsk, sk);
1253 get_net(sock_net(newsk));
1254 sk_node_init(&newsk->sk_node);
1255 sock_lock_init(newsk);
1256 bh_lock_sock(newsk);
1257 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1259 atomic_set(&newsk->sk_rmem_alloc, 0);
1261 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1263 atomic_set(&newsk->sk_wmem_alloc, 1);
1264 atomic_set(&newsk->sk_omem_alloc, 0);
1265 skb_queue_head_init(&newsk->sk_receive_queue);
1266 skb_queue_head_init(&newsk->sk_write_queue);
1267 #ifdef CONFIG_NET_DMA
1268 skb_queue_head_init(&newsk->sk_async_wait_queue);
1271 rwlock_init(&newsk->sk_dst_lock);
1272 rwlock_init(&newsk->sk_callback_lock);
1273 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1274 af_callback_keys + newsk->sk_family,
1275 af_family_clock_key_strings[newsk->sk_family]);
1277 newsk->sk_dst_cache = NULL;
1278 newsk->sk_wmem_queued = 0;
1279 newsk->sk_forward_alloc = 0;
1280 newsk->sk_send_head = NULL;
1281 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1283 sock_reset_flag(newsk, SOCK_DONE);
1284 skb_queue_head_init(&newsk->sk_error_queue);
1286 filter = newsk->sk_filter;
1288 sk_filter_charge(newsk, filter);
1290 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1291 /* It is still raw copy of parent, so invalidate
1292 * destructor and make plain sk_free() */
1293 newsk->sk_destruct = NULL;
1300 newsk->sk_priority = 0;
1302 * Before updating sk_refcnt, we must commit prior changes to memory
1303 * (Documentation/RCU/rculist_nulls.txt for details)
1306 atomic_set(&newsk->sk_refcnt, 2);
1309 * Increment the counter in the same struct proto as the master
1310 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1311 * is the same as sk->sk_prot->socks, as this field was copied
1314 * This _changes_ the previous behaviour, where
1315 * tcp_create_openreq_child always was incrementing the
1316 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1317 * to be taken into account in all callers. -acme
1319 sk_refcnt_debug_inc(newsk);
1320 sk_set_socket(newsk, NULL);
1321 newsk->sk_sleep = NULL;
1323 if (newsk->sk_prot->sockets_allocated)
1324 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1326 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1327 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1328 net_enable_timestamp();
1333 EXPORT_SYMBOL_GPL(sk_clone);
1335 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1337 __sk_dst_set(sk, dst);
1338 sk->sk_route_caps = dst->dev->features;
1339 if (sk->sk_route_caps & NETIF_F_GSO)
1340 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1341 if (sk_can_gso(sk)) {
1342 if (dst->header_len) {
1343 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1345 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1346 sk->sk_gso_max_size = dst->dev->gso_max_size;
1350 EXPORT_SYMBOL_GPL(sk_setup_caps);
1352 void __init sk_init(void)
1354 if (totalram_pages <= 4096) {
1355 sysctl_wmem_max = 32767;
1356 sysctl_rmem_max = 32767;
1357 sysctl_wmem_default = 32767;
1358 sysctl_rmem_default = 32767;
1359 } else if (totalram_pages >= 131072) {
1360 sysctl_wmem_max = 131071;
1361 sysctl_rmem_max = 131071;
1364 mem_reserve_init(&net_reserve, "total network reserve", NULL);
1365 mem_reserve_init(&net_rx_reserve, "network RX reserve", &net_reserve);
1366 mem_reserve_init(&net_skb_reserve, "SKB data reserve", &net_rx_reserve);
1367 mem_reserve_init(&net_tx_reserve, "network TX reserve", &net_reserve);
1368 mem_reserve_init(&net_tx_pages, "protocol TX pages", &net_tx_reserve);
1372 * Simple resource managers for sockets.
1377 * Write buffer destructor automatically called from kfree_skb.
1379 void sock_wfree(struct sk_buff *skb)
1381 struct sock *sk = skb->sk;
1382 unsigned int len = skb->truesize;
1384 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1386 * Keep a reference on sk_wmem_alloc, this will be released
1387 * after sk_write_space() call
1389 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1390 sk->sk_write_space(sk);
1394 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1395 * could not do because of in-flight packets
1397 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1400 EXPORT_SYMBOL(sock_wfree);
1403 * Read buffer destructor automatically called from kfree_skb.
1405 void sock_rfree(struct sk_buff *skb)
1407 struct sock *sk = skb->sk;
1409 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1410 sk_mem_uncharge(skb->sk, skb->truesize);
1412 EXPORT_SYMBOL(sock_rfree);
1415 int sock_i_uid(struct sock *sk)
1419 read_lock(&sk->sk_callback_lock);
1420 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1421 read_unlock(&sk->sk_callback_lock);
1424 EXPORT_SYMBOL(sock_i_uid);
1426 unsigned long sock_i_ino(struct sock *sk)
1430 read_lock(&sk->sk_callback_lock);
1431 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1432 read_unlock(&sk->sk_callback_lock);
1435 EXPORT_SYMBOL(sock_i_ino);
1438 * Allocate a skb from the socket's send buffer.
1440 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1443 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1444 struct sk_buff *skb = alloc_skb(size, priority);
1446 skb_set_owner_w(skb, sk);
1452 EXPORT_SYMBOL(sock_wmalloc);
1455 * Allocate a skb from the socket's receive buffer.
1457 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1460 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1461 struct sk_buff *skb = alloc_skb(size, priority);
1463 skb_set_owner_r(skb, sk);
1471 * Allocate a memory block from the socket's option memory buffer.
1473 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1475 if ((unsigned)size <= sysctl_optmem_max &&
1476 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1478 /* First do the add, to avoid the race if kmalloc
1481 atomic_add(size, &sk->sk_omem_alloc);
1482 mem = kmalloc(size, priority);
1485 atomic_sub(size, &sk->sk_omem_alloc);
1489 EXPORT_SYMBOL(sock_kmalloc);
1492 * Free an option memory block.
1494 void sock_kfree_s(struct sock *sk, void *mem, int size)
1497 atomic_sub(size, &sk->sk_omem_alloc);
1499 EXPORT_SYMBOL(sock_kfree_s);
1501 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1502 I think, these locks should be removed for datagram sockets.
1504 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1508 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1512 if (signal_pending(current))
1514 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1515 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1516 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1518 if (sk->sk_shutdown & SEND_SHUTDOWN)
1522 timeo = schedule_timeout(timeo);
1524 finish_wait(sk->sk_sleep, &wait);
1530 * Generic send/receive buffer handlers
1533 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1534 unsigned long data_len, int noblock,
1537 struct sk_buff *skb;
1542 gfp_mask = sk->sk_allocation;
1543 if (gfp_mask & __GFP_WAIT)
1544 gfp_mask |= __GFP_REPEAT;
1546 timeo = sock_sndtimeo(sk, noblock);
1548 err = sock_error(sk);
1553 if (sk->sk_shutdown & SEND_SHUTDOWN)
1556 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1557 skb = alloc_skb(header_len, gfp_mask);
1562 /* No pages, we're done... */
1566 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1567 skb->truesize += data_len;
1568 skb_shinfo(skb)->nr_frags = npages;
1569 for (i = 0; i < npages; i++) {
1573 page = alloc_pages(sk->sk_allocation, 0);
1576 skb_shinfo(skb)->nr_frags = i;
1581 frag = &skb_shinfo(skb)->frags[i];
1583 frag->page_offset = 0;
1584 frag->size = (data_len >= PAGE_SIZE ?
1587 data_len -= PAGE_SIZE;
1590 /* Full success... */
1596 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1597 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1601 if (signal_pending(current))
1603 timeo = sock_wait_for_wmem(sk, timeo);
1606 skb_set_owner_w(skb, sk);
1610 err = sock_intr_errno(timeo);
1615 EXPORT_SYMBOL(sock_alloc_send_pskb);
1617 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1618 int noblock, int *errcode)
1620 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1622 EXPORT_SYMBOL(sock_alloc_send_skb);
1624 static void __lock_sock(struct sock *sk)
1629 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1630 TASK_UNINTERRUPTIBLE);
1631 spin_unlock_bh(&sk->sk_lock.slock);
1633 spin_lock_bh(&sk->sk_lock.slock);
1634 if (!sock_owned_by_user(sk))
1637 finish_wait(&sk->sk_lock.wq, &wait);
1640 static void __release_sock(struct sock *sk)
1642 struct sk_buff *skb = sk->sk_backlog.head;
1645 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1649 struct sk_buff *next = skb->next;
1652 sk_backlog_rcv(sk, skb);
1655 * We are in process context here with softirqs
1656 * disabled, use cond_resched_softirq() to preempt.
1657 * This is safe to do because we've taken the backlog
1660 cond_resched_softirq();
1663 } while (skb != NULL);
1666 } while ((skb = sk->sk_backlog.head) != NULL);
1670 * sk_wait_data - wait for data to arrive at sk_receive_queue
1671 * @sk: sock to wait on
1672 * @timeo: for how long
1674 * Now socket state including sk->sk_err is changed only under lock,
1675 * hence we may omit checks after joining wait queue.
1676 * We check receive queue before schedule() only as optimization;
1677 * it is very likely that release_sock() added new data.
1679 int sk_wait_data(struct sock *sk, long *timeo)
1684 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1685 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1686 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1687 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1688 finish_wait(sk->sk_sleep, &wait);
1691 EXPORT_SYMBOL(sk_wait_data);
1694 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1696 * @size: memory size to allocate
1697 * @kind: allocation type
1699 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1700 * rmem allocation. This function assumes that protocols which have
1701 * memory_pressure use sk_wmem_queued as write buffer accounting.
1703 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1705 struct proto *prot = sk->sk_prot;
1706 int amt = sk_mem_pages(size);
1709 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1710 allocated = atomic_add_return(amt, prot->memory_allocated);
1713 if (allocated <= prot->sysctl_mem[0]) {
1714 if (prot->memory_pressure && *prot->memory_pressure)
1715 *prot->memory_pressure = 0;
1719 /* Under pressure. */
1720 if (allocated > prot->sysctl_mem[1])
1721 if (prot->enter_memory_pressure)
1722 prot->enter_memory_pressure(sk);
1724 /* Over hard limit. */
1725 if (allocated > prot->sysctl_mem[2])
1726 goto suppress_allocation;
1728 /* guarantee minimum buffer size under pressure */
1729 if (kind == SK_MEM_RECV) {
1730 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1732 } else { /* SK_MEM_SEND */
1733 if (sk->sk_type == SOCK_STREAM) {
1734 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1736 } else if (atomic_read(&sk->sk_wmem_alloc) <
1737 prot->sysctl_wmem[0])
1741 if (prot->memory_pressure) {
1744 if (!*prot->memory_pressure)
1746 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1747 if (prot->sysctl_mem[2] > alloc *
1748 sk_mem_pages(sk->sk_wmem_queued +
1749 atomic_read(&sk->sk_rmem_alloc) +
1750 sk->sk_forward_alloc))
1754 suppress_allocation:
1756 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1757 sk_stream_moderate_sndbuf(sk);
1759 /* Fail only if socket is _under_ its sndbuf.
1760 * In this case we cannot block, so that we have to fail.
1762 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1766 /* Alas. Undo changes. */
1767 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1768 atomic_sub(amt, prot->memory_allocated);
1771 EXPORT_SYMBOL(__sk_mem_schedule);
1774 * __sk_reclaim - reclaim memory_allocated
1777 void __sk_mem_reclaim(struct sock *sk)
1779 struct proto *prot = sk->sk_prot;
1781 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1782 prot->memory_allocated);
1783 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1785 if (prot->memory_pressure && *prot->memory_pressure &&
1786 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1787 *prot->memory_pressure = 0;
1789 EXPORT_SYMBOL(__sk_mem_reclaim);
1793 * Set of default routines for initialising struct proto_ops when
1794 * the protocol does not support a particular function. In certain
1795 * cases where it makes no sense for a protocol to have a "do nothing"
1796 * function, some default processing is provided.
1799 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1803 EXPORT_SYMBOL(sock_no_bind);
1805 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1810 EXPORT_SYMBOL(sock_no_connect);
1812 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1816 EXPORT_SYMBOL(sock_no_socketpair);
1818 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1822 EXPORT_SYMBOL(sock_no_accept);
1824 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1829 EXPORT_SYMBOL(sock_no_getname);
1831 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1835 EXPORT_SYMBOL(sock_no_poll);
1837 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1841 EXPORT_SYMBOL(sock_no_ioctl);
1843 int sock_no_listen(struct socket *sock, int backlog)
1847 EXPORT_SYMBOL(sock_no_listen);
1849 int sock_no_shutdown(struct socket *sock, int how)
1853 EXPORT_SYMBOL(sock_no_shutdown);
1855 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1856 char __user *optval, unsigned int optlen)
1860 EXPORT_SYMBOL(sock_no_setsockopt);
1862 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1863 char __user *optval, int __user *optlen)
1867 EXPORT_SYMBOL(sock_no_getsockopt);
1869 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1874 EXPORT_SYMBOL(sock_no_sendmsg);
1876 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1877 size_t len, int flags)
1881 EXPORT_SYMBOL(sock_no_recvmsg);
1883 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1885 /* Mirror missing mmap method error code */
1888 EXPORT_SYMBOL(sock_no_mmap);
1890 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1893 struct msghdr msg = {.msg_flags = flags};
1895 char *kaddr = kmap(page);
1896 iov.iov_base = kaddr + offset;
1898 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1902 EXPORT_SYMBOL(sock_no_sendpage);
1905 * Default Socket Callbacks
1908 static void sock_def_wakeup(struct sock *sk)
1910 read_lock(&sk->sk_callback_lock);
1911 if (sk_has_sleeper(sk))
1912 wake_up_interruptible_all(sk->sk_sleep);
1913 read_unlock(&sk->sk_callback_lock);
1916 static void sock_def_error_report(struct sock *sk)
1918 read_lock(&sk->sk_callback_lock);
1919 if (sk_has_sleeper(sk))
1920 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1921 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1922 read_unlock(&sk->sk_callback_lock);
1925 static void sock_def_readable(struct sock *sk, int len)
1927 read_lock(&sk->sk_callback_lock);
1928 if (sk_has_sleeper(sk))
1929 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1930 POLLRDNORM | POLLRDBAND);
1931 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1932 read_unlock(&sk->sk_callback_lock);
1935 static void sock_def_write_space(struct sock *sk)
1937 read_lock(&sk->sk_callback_lock);
1939 /* Do not wake up a writer until he can make "significant"
1942 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1943 if (sk_has_sleeper(sk))
1944 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1945 POLLWRNORM | POLLWRBAND);
1947 /* Should agree with poll, otherwise some programs break */
1948 if (sock_writeable(sk))
1949 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1952 read_unlock(&sk->sk_callback_lock);
1955 static void sock_def_destruct(struct sock *sk)
1957 kfree(sk->sk_protinfo);
1960 void sk_send_sigurg(struct sock *sk)
1962 if (sk->sk_socket && sk->sk_socket->file)
1963 if (send_sigurg(&sk->sk_socket->file->f_owner))
1964 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1966 EXPORT_SYMBOL(sk_send_sigurg);
1968 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1969 unsigned long expires)
1971 if (!mod_timer(timer, expires))
1974 EXPORT_SYMBOL(sk_reset_timer);
1976 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1978 if (timer_pending(timer) && del_timer(timer))
1981 EXPORT_SYMBOL(sk_stop_timer);
1983 void sock_init_data(struct socket *sock, struct sock *sk)
1985 skb_queue_head_init(&sk->sk_receive_queue);
1986 skb_queue_head_init(&sk->sk_write_queue);
1987 skb_queue_head_init(&sk->sk_error_queue);
1988 #ifdef CONFIG_NET_DMA
1989 skb_queue_head_init(&sk->sk_async_wait_queue);
1992 sk->sk_send_head = NULL;
1994 init_timer(&sk->sk_timer);
1996 sk->sk_allocation = GFP_KERNEL;
1997 sk->sk_rcvbuf = sysctl_rmem_default;
1998 sk->sk_sndbuf = sysctl_wmem_default;
1999 sk->sk_state = TCP_CLOSE;
2000 sk_set_socket(sk, sock);
2002 sock_set_flag(sk, SOCK_ZAPPED);
2005 sk->sk_type = sock->type;
2006 sk->sk_sleep = &sock->wait;
2009 sk->sk_sleep = NULL;
2011 rwlock_init(&sk->sk_dst_lock);
2012 rwlock_init(&sk->sk_callback_lock);
2013 lockdep_set_class_and_name(&sk->sk_callback_lock,
2014 af_callback_keys + sk->sk_family,
2015 af_family_clock_key_strings[sk->sk_family]);
2017 sk->sk_state_change = sock_def_wakeup;
2018 sk->sk_data_ready = sock_def_readable;
2019 sk->sk_write_space = sock_def_write_space;
2020 sk->sk_error_report = sock_def_error_report;
2021 sk->sk_destruct = sock_def_destruct;
2023 sk->sk_sndmsg_page = NULL;
2024 sk->sk_sndmsg_off = 0;
2026 sk->sk_peercred.pid = 0;
2027 sk->sk_peercred.uid = -1;
2028 sk->sk_peercred.gid = -1;
2029 sk->sk_write_pending = 0;
2030 sk->sk_rcvlowat = 1;
2031 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2032 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2034 sk->sk_stamp = ktime_set(-1L, 0);
2037 * Before updating sk_refcnt, we must commit prior changes to memory
2038 * (Documentation/RCU/rculist_nulls.txt for details)
2041 atomic_set(&sk->sk_refcnt, 1);
2042 atomic_set(&sk->sk_drops, 0);
2044 EXPORT_SYMBOL(sock_init_data);
2046 void lock_sock_nested(struct sock *sk, int subclass)
2049 spin_lock_bh(&sk->sk_lock.slock);
2050 if (sk->sk_lock.owned)
2052 sk->sk_lock.owned = 1;
2053 spin_unlock(&sk->sk_lock.slock);
2055 * The sk_lock has mutex_lock() semantics here:
2057 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2060 EXPORT_SYMBOL(lock_sock_nested);
2062 void release_sock(struct sock *sk)
2065 * The sk_lock has mutex_unlock() semantics:
2067 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2069 spin_lock_bh(&sk->sk_lock.slock);
2070 if (sk->sk_backlog.tail)
2072 sk->sk_lock.owned = 0;
2073 if (waitqueue_active(&sk->sk_lock.wq))
2074 wake_up(&sk->sk_lock.wq);
2075 spin_unlock_bh(&sk->sk_lock.slock);
2077 EXPORT_SYMBOL(release_sock);
2079 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2082 if (!sock_flag(sk, SOCK_TIMESTAMP))
2083 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2084 tv = ktime_to_timeval(sk->sk_stamp);
2085 if (tv.tv_sec == -1)
2087 if (tv.tv_sec == 0) {
2088 sk->sk_stamp = ktime_get_real();
2089 tv = ktime_to_timeval(sk->sk_stamp);
2091 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2093 EXPORT_SYMBOL(sock_get_timestamp);
2095 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2098 if (!sock_flag(sk, SOCK_TIMESTAMP))
2099 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2100 ts = ktime_to_timespec(sk->sk_stamp);
2101 if (ts.tv_sec == -1)
2103 if (ts.tv_sec == 0) {
2104 sk->sk_stamp = ktime_get_real();
2105 ts = ktime_to_timespec(sk->sk_stamp);
2107 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2109 EXPORT_SYMBOL(sock_get_timestampns);
2111 void sock_enable_timestamp(struct sock *sk, int flag)
2113 if (!sock_flag(sk, flag)) {
2114 sock_set_flag(sk, flag);
2116 * we just set one of the two flags which require net
2117 * time stamping, but time stamping might have been on
2118 * already because of the other one
2121 flag == SOCK_TIMESTAMP ?
2122 SOCK_TIMESTAMPING_RX_SOFTWARE :
2124 net_enable_timestamp();
2129 * Get a socket option on an socket.
2131 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2132 * asynchronous errors should be reported by getsockopt. We assume
2133 * this means if you specify SO_ERROR (otherwise whats the point of it).
2135 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2136 char __user *optval, int __user *optlen)
2138 struct sock *sk = sock->sk;
2140 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2142 EXPORT_SYMBOL(sock_common_getsockopt);
2144 #ifdef CONFIG_COMPAT
2145 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2146 char __user *optval, int __user *optlen)
2148 struct sock *sk = sock->sk;
2150 if (sk->sk_prot->compat_getsockopt != NULL)
2151 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2153 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2155 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2158 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2159 struct msghdr *msg, size_t size, int flags)
2161 struct sock *sk = sock->sk;
2165 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2166 flags & ~MSG_DONTWAIT, &addr_len);
2168 msg->msg_namelen = addr_len;
2171 EXPORT_SYMBOL(sock_common_recvmsg);
2174 * Set socket options on an inet socket.
2176 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2177 char __user *optval, unsigned int optlen)
2179 struct sock *sk = sock->sk;
2181 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2183 EXPORT_SYMBOL(sock_common_setsockopt);
2185 #ifdef CONFIG_COMPAT
2186 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2187 char __user *optval, unsigned int optlen)
2189 struct sock *sk = sock->sk;
2191 if (sk->sk_prot->compat_setsockopt != NULL)
2192 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2194 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2196 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2199 void sk_common_release(struct sock *sk)
2201 if (sk->sk_prot->destroy)
2202 sk->sk_prot->destroy(sk);
2205 * Observation: when sock_common_release is called, processes have
2206 * no access to socket. But net still has.
2207 * Step one, detach it from networking:
2209 * A. Remove from hash tables.
2212 sk->sk_prot->unhash(sk);
2215 * In this point socket cannot receive new packets, but it is possible
2216 * that some packets are in flight because some CPU runs receiver and
2217 * did hash table lookup before we unhashed socket. They will achieve
2218 * receive queue and will be purged by socket destructor.
2220 * Also we still have packets pending on receive queue and probably,
2221 * our own packets waiting in device queues. sock_destroy will drain
2222 * receive queue, but transmitted packets will delay socket destruction
2223 * until the last reference will be released.
2228 xfrm_sk_free_policy(sk);
2230 sk_refcnt_debug_release(sk);
2233 EXPORT_SYMBOL(sk_common_release);
2235 static DEFINE_RWLOCK(proto_list_lock);
2236 static LIST_HEAD(proto_list);
2238 #ifdef CONFIG_PROC_FS
2239 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2241 int val[PROTO_INUSE_NR];
2244 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2246 #ifdef CONFIG_NET_NS
2247 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2249 int cpu = smp_processor_id();
2250 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2252 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2254 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2256 int cpu, idx = prot->inuse_idx;
2259 for_each_possible_cpu(cpu)
2260 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2262 return res >= 0 ? res : 0;
2264 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2266 static int sock_inuse_init_net(struct net *net)
2268 net->core.inuse = alloc_percpu(struct prot_inuse);
2269 return net->core.inuse ? 0 : -ENOMEM;
2272 static void sock_inuse_exit_net(struct net *net)
2274 free_percpu(net->core.inuse);
2277 static struct pernet_operations net_inuse_ops = {
2278 .init = sock_inuse_init_net,
2279 .exit = sock_inuse_exit_net,
2282 static __init int net_inuse_init(void)
2284 if (register_pernet_subsys(&net_inuse_ops))
2285 panic("Cannot initialize net inuse counters");
2290 core_initcall(net_inuse_init);
2292 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2294 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2296 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2298 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2300 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2302 int cpu, idx = prot->inuse_idx;
2305 for_each_possible_cpu(cpu)
2306 res += per_cpu(prot_inuse, cpu).val[idx];
2308 return res >= 0 ? res : 0;
2310 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2313 static void assign_proto_idx(struct proto *prot)
2315 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2317 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2318 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2322 set_bit(prot->inuse_idx, proto_inuse_idx);
2325 static void release_proto_idx(struct proto *prot)
2327 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2328 clear_bit(prot->inuse_idx, proto_inuse_idx);
2331 static inline void assign_proto_idx(struct proto *prot)
2335 static inline void release_proto_idx(struct proto *prot)
2340 int proto_register(struct proto *prot, int alloc_slab)
2343 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2344 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2347 if (prot->slab == NULL) {
2348 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2353 if (prot->rsk_prot != NULL) {
2354 static const char mask[] = "request_sock_%s";
2356 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2357 if (prot->rsk_prot->slab_name == NULL)
2358 goto out_free_sock_slab;
2360 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2361 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2362 prot->rsk_prot->obj_size, 0,
2363 SLAB_HWCACHE_ALIGN, NULL);
2365 if (prot->rsk_prot->slab == NULL) {
2366 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2368 goto out_free_request_sock_slab_name;
2372 if (prot->twsk_prot != NULL) {
2373 static const char mask[] = "tw_sock_%s";
2375 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2377 if (prot->twsk_prot->twsk_slab_name == NULL)
2378 goto out_free_request_sock_slab;
2380 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2381 prot->twsk_prot->twsk_slab =
2382 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2383 prot->twsk_prot->twsk_obj_size,
2385 SLAB_HWCACHE_ALIGN |
2388 if (prot->twsk_prot->twsk_slab == NULL)
2389 goto out_free_timewait_sock_slab_name;
2393 write_lock(&proto_list_lock);
2394 list_add(&prot->node, &proto_list);
2395 assign_proto_idx(prot);
2396 write_unlock(&proto_list_lock);
2399 out_free_timewait_sock_slab_name:
2400 kfree(prot->twsk_prot->twsk_slab_name);
2401 out_free_request_sock_slab:
2402 if (prot->rsk_prot && prot->rsk_prot->slab) {
2403 kmem_cache_destroy(prot->rsk_prot->slab);
2404 prot->rsk_prot->slab = NULL;
2406 out_free_request_sock_slab_name:
2407 kfree(prot->rsk_prot->slab_name);
2409 kmem_cache_destroy(prot->slab);
2414 EXPORT_SYMBOL(proto_register);
2416 void proto_unregister(struct proto *prot)
2418 write_lock(&proto_list_lock);
2419 release_proto_idx(prot);
2420 list_del(&prot->node);
2421 write_unlock(&proto_list_lock);
2423 if (prot->slab != NULL) {
2424 kmem_cache_destroy(prot->slab);
2428 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2429 kmem_cache_destroy(prot->rsk_prot->slab);
2430 kfree(prot->rsk_prot->slab_name);
2431 prot->rsk_prot->slab = NULL;
2434 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2435 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2436 kfree(prot->twsk_prot->twsk_slab_name);
2437 prot->twsk_prot->twsk_slab = NULL;
2440 EXPORT_SYMBOL(proto_unregister);
2442 #ifdef CONFIG_PROC_FS
2443 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2444 __acquires(proto_list_lock)
2446 read_lock(&proto_list_lock);
2447 return seq_list_start_head(&proto_list, *pos);
2450 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2452 return seq_list_next(v, &proto_list, pos);
2455 static void proto_seq_stop(struct seq_file *seq, void *v)
2456 __releases(proto_list_lock)
2458 read_unlock(&proto_list_lock);
2461 static char proto_method_implemented(const void *method)
2463 return method == NULL ? 'n' : 'y';
2466 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2468 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2469 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2472 sock_prot_inuse_get(seq_file_net(seq), proto),
2473 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2474 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2476 proto->slab == NULL ? "no" : "yes",
2477 module_name(proto->owner),
2478 proto_method_implemented(proto->close),
2479 proto_method_implemented(proto->connect),
2480 proto_method_implemented(proto->disconnect),
2481 proto_method_implemented(proto->accept),
2482 proto_method_implemented(proto->ioctl),
2483 proto_method_implemented(proto->init),
2484 proto_method_implemented(proto->destroy),
2485 proto_method_implemented(proto->shutdown),
2486 proto_method_implemented(proto->setsockopt),
2487 proto_method_implemented(proto->getsockopt),
2488 proto_method_implemented(proto->sendmsg),
2489 proto_method_implemented(proto->recvmsg),
2490 proto_method_implemented(proto->sendpage),
2491 proto_method_implemented(proto->bind),
2492 proto_method_implemented(proto->backlog_rcv),
2493 proto_method_implemented(proto->hash),
2494 proto_method_implemented(proto->unhash),
2495 proto_method_implemented(proto->get_port),
2496 proto_method_implemented(proto->enter_memory_pressure));
2499 static int proto_seq_show(struct seq_file *seq, void *v)
2501 if (v == &proto_list)
2502 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2511 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2513 proto_seq_printf(seq, list_entry(v, struct proto, node));
2517 static const struct seq_operations proto_seq_ops = {
2518 .start = proto_seq_start,
2519 .next = proto_seq_next,
2520 .stop = proto_seq_stop,
2521 .show = proto_seq_show,
2524 static int proto_seq_open(struct inode *inode, struct file *file)
2526 return seq_open_net(inode, file, &proto_seq_ops,
2527 sizeof(struct seq_net_private));
2530 static const struct file_operations proto_seq_fops = {
2531 .owner = THIS_MODULE,
2532 .open = proto_seq_open,
2534 .llseek = seq_lseek,
2535 .release = seq_release_net,
2538 static __net_init int proto_init_net(struct net *net)
2540 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2546 static __net_exit void proto_exit_net(struct net *net)
2548 proc_net_remove(net, "protocols");
2552 static __net_initdata struct pernet_operations proto_net_ops = {
2553 .init = proto_init_net,
2554 .exit = proto_exit_net,
2557 static int __init proto_init(void)
2559 return register_pernet_subsys(&proto_net_ops);
2562 subsys_initcall(proto_init);
2564 #endif /* PROC_FS */