sock.c revision 8a74ad60a546b13bd1096b2a61a7a5c6fd9ae17c
1/*
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.
5 *
6 *		Generic socket support routines. Memory allocators, socket lock/release
7 *		handler for protocols to use and generic option handler.
8 *
9 *
10 * Authors:	Ross Biro
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>
14 *
15 * Fixes:
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
36 *					TCP layer surgery.
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
64 *					(compatibility fix)
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
82 *
83 * To Fix:
84 *
85 *
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.
90 */
91
92#include <linux/capability.h>
93#include <linux/errno.h>
94#include <linux/types.h>
95#include <linux/socket.h>
96#include <linux/in.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
114#include <asm/uaccess.h>
115#include <asm/system.h>
116
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>
126#include <net/cls_cgroup.h>
127
128#include <linux/filter.h>
129
130#ifdef CONFIG_INET
131#include <net/tcp.h>
132#endif
133
134/*
135 * Each address family might have different locking rules, so we have
136 * one slock key per address family:
137 */
138static struct lock_class_key af_family_keys[AF_MAX];
139static struct lock_class_key af_family_slock_keys[AF_MAX];
140
141/*
142 * Make lock validator output more readable. (we pre-construct these
143 * strings build-time, so that runtime initialization of socket
144 * locks is fast):
145 */
146static 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",
160  "sk_lock-AF_MAX"
161};
162static 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",
176  "slock-AF_MAX"
177};
178static 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",
192  "clock-AF_MAX"
193};
194
195/*
196 * sk_callback_lock locking rules are per-address-family,
197 * so split the lock classes by using a per-AF key:
198 */
199static struct lock_class_key af_callback_keys[AF_MAX];
200
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.
205 */
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)
210
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;
216
217/* Maximal space eaten by iovec or ancilliary data plus some space */
218int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
219EXPORT_SYMBOL(sysctl_optmem_max);
220
221#if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
222int net_cls_subsys_id = -1;
223EXPORT_SYMBOL_GPL(net_cls_subsys_id);
224#endif
225
226static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
227{
228	struct timeval tv;
229
230	if (optlen < sizeof(tv))
231		return -EINVAL;
232	if (copy_from_user(&tv, optval, sizeof(tv)))
233		return -EFAULT;
234	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
235		return -EDOM;
236
237	if (tv.tv_sec < 0) {
238		static int warned __read_mostly;
239
240		*timeo_p = 0;
241		if (warned < 10 && net_ratelimit()) {
242			warned++;
243			printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
244			       "tries to set negative timeout\n",
245				current->comm, task_pid_nr(current));
246		}
247		return 0;
248	}
249	*timeo_p = MAX_SCHEDULE_TIMEOUT;
250	if (tv.tv_sec == 0 && tv.tv_usec == 0)
251		return 0;
252	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
253		*timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
254	return 0;
255}
256
257static void sock_warn_obsolete_bsdism(const char *name)
258{
259	static int warned;
260	static char warncomm[TASK_COMM_LEN];
261	if (strcmp(warncomm, current->comm) && warned < 5) {
262		strcpy(warncomm,  current->comm);
263		printk(KERN_WARNING "process `%s' is using obsolete "
264		       "%s SO_BSDCOMPAT\n", warncomm, name);
265		warned++;
266	}
267}
268
269static void sock_disable_timestamp(struct sock *sk, int flag)
270{
271	if (sock_flag(sk, flag)) {
272		sock_reset_flag(sk, flag);
273		if (!sock_flag(sk, SOCK_TIMESTAMP) &&
274		    !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
275			net_disable_timestamp();
276		}
277	}
278}
279
280
281int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
282{
283	int err;
284	int skb_len;
285	unsigned long flags;
286	struct sk_buff_head *list = &sk->sk_receive_queue;
287
288	/* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
289	   number of warnings when compiling with -W --ANK
290	 */
291	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
292	    (unsigned)sk->sk_rcvbuf) {
293		atomic_inc(&sk->sk_drops);
294		return -ENOMEM;
295	}
296
297	err = sk_filter(sk, skb);
298	if (err)
299		return err;
300
301	if (!sk_rmem_schedule(sk, skb->truesize)) {
302		atomic_inc(&sk->sk_drops);
303		return -ENOBUFS;
304	}
305
306	skb->dev = NULL;
307	skb_set_owner_r(skb, sk);
308
309	/* Cache the SKB length before we tack it onto the receive
310	 * queue.  Once it is added it no longer belongs to us and
311	 * may be freed by other threads of control pulling packets
312	 * from the queue.
313	 */
314	skb_len = skb->len;
315
316	/* we escape from rcu protected region, make sure we dont leak
317	 * a norefcounted dst
318	 */
319	skb_dst_force(skb);
320
321	spin_lock_irqsave(&list->lock, flags);
322	skb->dropcount = atomic_read(&sk->sk_drops);
323	__skb_queue_tail(list, skb);
324	spin_unlock_irqrestore(&list->lock, flags);
325
326	if (!sock_flag(sk, SOCK_DEAD))
327		sk->sk_data_ready(sk, skb_len);
328	return 0;
329}
330EXPORT_SYMBOL(sock_queue_rcv_skb);
331
332int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
333{
334	int rc = NET_RX_SUCCESS;
335
336	if (sk_filter(sk, skb))
337		goto discard_and_relse;
338
339	skb->dev = NULL;
340
341	if (sk_rcvqueues_full(sk, skb)) {
342		atomic_inc(&sk->sk_drops);
343		goto discard_and_relse;
344	}
345	if (nested)
346		bh_lock_sock_nested(sk);
347	else
348		bh_lock_sock(sk);
349	if (!sock_owned_by_user(sk)) {
350		/*
351		 * trylock + unlock semantics:
352		 */
353		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
354
355		rc = sk_backlog_rcv(sk, skb);
356
357		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
358	} else if (sk_add_backlog(sk, skb)) {
359		bh_unlock_sock(sk);
360		atomic_inc(&sk->sk_drops);
361		goto discard_and_relse;
362	}
363
364	bh_unlock_sock(sk);
365out:
366	sock_put(sk);
367	return rc;
368discard_and_relse:
369	kfree_skb(skb);
370	goto out;
371}
372EXPORT_SYMBOL(sk_receive_skb);
373
374void sk_reset_txq(struct sock *sk)
375{
376	sk_tx_queue_clear(sk);
377}
378EXPORT_SYMBOL(sk_reset_txq);
379
380struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
381{
382	struct dst_entry *dst = __sk_dst_get(sk);
383
384	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
385		sk_tx_queue_clear(sk);
386		rcu_assign_pointer(sk->sk_dst_cache, NULL);
387		dst_release(dst);
388		return NULL;
389	}
390
391	return dst;
392}
393EXPORT_SYMBOL(__sk_dst_check);
394
395struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
396{
397	struct dst_entry *dst = sk_dst_get(sk);
398
399	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
400		sk_dst_reset(sk);
401		dst_release(dst);
402		return NULL;
403	}
404
405	return dst;
406}
407EXPORT_SYMBOL(sk_dst_check);
408
409static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
410{
411	int ret = -ENOPROTOOPT;
412#ifdef CONFIG_NETDEVICES
413	struct net *net = sock_net(sk);
414	char devname[IFNAMSIZ];
415	int index;
416
417	/* Sorry... */
418	ret = -EPERM;
419	if (!capable(CAP_NET_RAW))
420		goto out;
421
422	ret = -EINVAL;
423	if (optlen < 0)
424		goto out;
425
426	/* Bind this socket to a particular device like "eth0",
427	 * as specified in the passed interface name. If the
428	 * name is "" or the option length is zero the socket
429	 * is not bound.
430	 */
431	if (optlen > IFNAMSIZ - 1)
432		optlen = IFNAMSIZ - 1;
433	memset(devname, 0, sizeof(devname));
434
435	ret = -EFAULT;
436	if (copy_from_user(devname, optval, optlen))
437		goto out;
438
439	index = 0;
440	if (devname[0] != '\0') {
441		struct net_device *dev;
442
443		rcu_read_lock();
444		dev = dev_get_by_name_rcu(net, devname);
445		if (dev)
446			index = dev->ifindex;
447		rcu_read_unlock();
448		ret = -ENODEV;
449		if (!dev)
450			goto out;
451	}
452
453	lock_sock(sk);
454	sk->sk_bound_dev_if = index;
455	sk_dst_reset(sk);
456	release_sock(sk);
457
458	ret = 0;
459
460out:
461#endif
462
463	return ret;
464}
465
466static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
467{
468	if (valbool)
469		sock_set_flag(sk, bit);
470	else
471		sock_reset_flag(sk, bit);
472}
473
474/*
475 *	This is meant for all protocols to use and covers goings on
476 *	at the socket level. Everything here is generic.
477 */
478
479int sock_setsockopt(struct socket *sock, int level, int optname,
480		    char __user *optval, unsigned int optlen)
481{
482	struct sock *sk = sock->sk;
483	int val;
484	int valbool;
485	struct linger ling;
486	int ret = 0;
487
488	/*
489	 *	Options without arguments
490	 */
491
492	if (optname == SO_BINDTODEVICE)
493		return sock_bindtodevice(sk, optval, optlen);
494
495	if (optlen < sizeof(int))
496		return -EINVAL;
497
498	if (get_user(val, (int __user *)optval))
499		return -EFAULT;
500
501	valbool = val ? 1 : 0;
502
503	lock_sock(sk);
504
505	switch (optname) {
506	case SO_DEBUG:
507		if (val && !capable(CAP_NET_ADMIN))
508			ret = -EACCES;
509		else
510			sock_valbool_flag(sk, SOCK_DBG, valbool);
511		break;
512	case SO_REUSEADDR:
513		sk->sk_reuse = valbool;
514		break;
515	case SO_TYPE:
516	case SO_PROTOCOL:
517	case SO_DOMAIN:
518	case SO_ERROR:
519		ret = -ENOPROTOOPT;
520		break;
521	case SO_DONTROUTE:
522		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
523		break;
524	case SO_BROADCAST:
525		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
526		break;
527	case SO_SNDBUF:
528		/* Don't error on this BSD doesn't and if you think
529		   about it this is right. Otherwise apps have to
530		   play 'guess the biggest size' games. RCVBUF/SNDBUF
531		   are treated in BSD as hints */
532
533		if (val > sysctl_wmem_max)
534			val = sysctl_wmem_max;
535set_sndbuf:
536		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
537		if ((val * 2) < SOCK_MIN_SNDBUF)
538			sk->sk_sndbuf = SOCK_MIN_SNDBUF;
539		else
540			sk->sk_sndbuf = val * 2;
541
542		/*
543		 *	Wake up sending tasks if we
544		 *	upped the value.
545		 */
546		sk->sk_write_space(sk);
547		break;
548
549	case SO_SNDBUFFORCE:
550		if (!capable(CAP_NET_ADMIN)) {
551			ret = -EPERM;
552			break;
553		}
554		goto set_sndbuf;
555
556	case SO_RCVBUF:
557		/* Don't error on this BSD doesn't and if you think
558		   about it this is right. Otherwise apps have to
559		   play 'guess the biggest size' games. RCVBUF/SNDBUF
560		   are treated in BSD as hints */
561
562		if (val > sysctl_rmem_max)
563			val = sysctl_rmem_max;
564set_rcvbuf:
565		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
566		/*
567		 * We double it on the way in to account for
568		 * "struct sk_buff" etc. overhead.   Applications
569		 * assume that the SO_RCVBUF setting they make will
570		 * allow that much actual data to be received on that
571		 * socket.
572		 *
573		 * Applications are unaware that "struct sk_buff" and
574		 * other overheads allocate from the receive buffer
575		 * during socket buffer allocation.
576		 *
577		 * And after considering the possible alternatives,
578		 * returning the value we actually used in getsockopt
579		 * is the most desirable behavior.
580		 */
581		if ((val * 2) < SOCK_MIN_RCVBUF)
582			sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
583		else
584			sk->sk_rcvbuf = val * 2;
585		break;
586
587	case SO_RCVBUFFORCE:
588		if (!capable(CAP_NET_ADMIN)) {
589			ret = -EPERM;
590			break;
591		}
592		goto set_rcvbuf;
593
594	case SO_KEEPALIVE:
595#ifdef CONFIG_INET
596		if (sk->sk_protocol == IPPROTO_TCP)
597			tcp_set_keepalive(sk, valbool);
598#endif
599		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
600		break;
601
602	case SO_OOBINLINE:
603		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
604		break;
605
606	case SO_NO_CHECK:
607		sk->sk_no_check = valbool;
608		break;
609
610	case SO_PRIORITY:
611		if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
612			sk->sk_priority = val;
613		else
614			ret = -EPERM;
615		break;
616
617	case SO_LINGER:
618		if (optlen < sizeof(ling)) {
619			ret = -EINVAL;	/* 1003.1g */
620			break;
621		}
622		if (copy_from_user(&ling, optval, sizeof(ling))) {
623			ret = -EFAULT;
624			break;
625		}
626		if (!ling.l_onoff)
627			sock_reset_flag(sk, SOCK_LINGER);
628		else {
629#if (BITS_PER_LONG == 32)
630			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
631				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
632			else
633#endif
634				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
635			sock_set_flag(sk, SOCK_LINGER);
636		}
637		break;
638
639	case SO_BSDCOMPAT:
640		sock_warn_obsolete_bsdism("setsockopt");
641		break;
642
643	case SO_PASSCRED:
644		if (valbool)
645			set_bit(SOCK_PASSCRED, &sock->flags);
646		else
647			clear_bit(SOCK_PASSCRED, &sock->flags);
648		break;
649
650	case SO_TIMESTAMP:
651	case SO_TIMESTAMPNS:
652		if (valbool)  {
653			if (optname == SO_TIMESTAMP)
654				sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
655			else
656				sock_set_flag(sk, SOCK_RCVTSTAMPNS);
657			sock_set_flag(sk, SOCK_RCVTSTAMP);
658			sock_enable_timestamp(sk, SOCK_TIMESTAMP);
659		} else {
660			sock_reset_flag(sk, SOCK_RCVTSTAMP);
661			sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
662		}
663		break;
664
665	case SO_TIMESTAMPING:
666		if (val & ~SOF_TIMESTAMPING_MASK) {
667			ret = -EINVAL;
668			break;
669		}
670		sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
671				  val & SOF_TIMESTAMPING_TX_HARDWARE);
672		sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
673				  val & SOF_TIMESTAMPING_TX_SOFTWARE);
674		sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
675				  val & SOF_TIMESTAMPING_RX_HARDWARE);
676		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
677			sock_enable_timestamp(sk,
678					      SOCK_TIMESTAMPING_RX_SOFTWARE);
679		else
680			sock_disable_timestamp(sk,
681					       SOCK_TIMESTAMPING_RX_SOFTWARE);
682		sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
683				  val & SOF_TIMESTAMPING_SOFTWARE);
684		sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
685				  val & SOF_TIMESTAMPING_SYS_HARDWARE);
686		sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
687				  val & SOF_TIMESTAMPING_RAW_HARDWARE);
688		break;
689
690	case SO_RCVLOWAT:
691		if (val < 0)
692			val = INT_MAX;
693		sk->sk_rcvlowat = val ? : 1;
694		break;
695
696	case SO_RCVTIMEO:
697		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
698		break;
699
700	case SO_SNDTIMEO:
701		ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
702		break;
703
704	case SO_ATTACH_FILTER:
705		ret = -EINVAL;
706		if (optlen == sizeof(struct sock_fprog)) {
707			struct sock_fprog fprog;
708
709			ret = -EFAULT;
710			if (copy_from_user(&fprog, optval, sizeof(fprog)))
711				break;
712
713			ret = sk_attach_filter(&fprog, sk);
714		}
715		break;
716
717	case SO_DETACH_FILTER:
718		ret = sk_detach_filter(sk);
719		break;
720
721	case SO_PASSSEC:
722		if (valbool)
723			set_bit(SOCK_PASSSEC, &sock->flags);
724		else
725			clear_bit(SOCK_PASSSEC, &sock->flags);
726		break;
727	case SO_MARK:
728		if (!capable(CAP_NET_ADMIN))
729			ret = -EPERM;
730		else
731			sk->sk_mark = val;
732		break;
733
734		/* We implement the SO_SNDLOWAT etc to
735		   not be settable (1003.1g 5.3) */
736	case SO_RXQ_OVFL:
737		if (valbool)
738			sock_set_flag(sk, SOCK_RXQ_OVFL);
739		else
740			sock_reset_flag(sk, SOCK_RXQ_OVFL);
741		break;
742	default:
743		ret = -ENOPROTOOPT;
744		break;
745	}
746	release_sock(sk);
747	return ret;
748}
749EXPORT_SYMBOL(sock_setsockopt);
750
751
752int sock_getsockopt(struct socket *sock, int level, int optname,
753		    char __user *optval, int __user *optlen)
754{
755	struct sock *sk = sock->sk;
756
757	union {
758		int val;
759		struct linger ling;
760		struct timeval tm;
761	} v;
762
763	int lv = sizeof(int);
764	int len;
765
766	if (get_user(len, optlen))
767		return -EFAULT;
768	if (len < 0)
769		return -EINVAL;
770
771	memset(&v, 0, sizeof(v));
772
773	switch (optname) {
774	case SO_DEBUG:
775		v.val = sock_flag(sk, SOCK_DBG);
776		break;
777
778	case SO_DONTROUTE:
779		v.val = sock_flag(sk, SOCK_LOCALROUTE);
780		break;
781
782	case SO_BROADCAST:
783		v.val = !!sock_flag(sk, SOCK_BROADCAST);
784		break;
785
786	case SO_SNDBUF:
787		v.val = sk->sk_sndbuf;
788		break;
789
790	case SO_RCVBUF:
791		v.val = sk->sk_rcvbuf;
792		break;
793
794	case SO_REUSEADDR:
795		v.val = sk->sk_reuse;
796		break;
797
798	case SO_KEEPALIVE:
799		v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
800		break;
801
802	case SO_TYPE:
803		v.val = sk->sk_type;
804		break;
805
806	case SO_PROTOCOL:
807		v.val = sk->sk_protocol;
808		break;
809
810	case SO_DOMAIN:
811		v.val = sk->sk_family;
812		break;
813
814	case SO_ERROR:
815		v.val = -sock_error(sk);
816		if (v.val == 0)
817			v.val = xchg(&sk->sk_err_soft, 0);
818		break;
819
820	case SO_OOBINLINE:
821		v.val = !!sock_flag(sk, SOCK_URGINLINE);
822		break;
823
824	case SO_NO_CHECK:
825		v.val = sk->sk_no_check;
826		break;
827
828	case SO_PRIORITY:
829		v.val = sk->sk_priority;
830		break;
831
832	case SO_LINGER:
833		lv		= sizeof(v.ling);
834		v.ling.l_onoff	= !!sock_flag(sk, SOCK_LINGER);
835		v.ling.l_linger	= sk->sk_lingertime / HZ;
836		break;
837
838	case SO_BSDCOMPAT:
839		sock_warn_obsolete_bsdism("getsockopt");
840		break;
841
842	case SO_TIMESTAMP:
843		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
844				!sock_flag(sk, SOCK_RCVTSTAMPNS);
845		break;
846
847	case SO_TIMESTAMPNS:
848		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
849		break;
850
851	case SO_TIMESTAMPING:
852		v.val = 0;
853		if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
854			v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
855		if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
856			v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
857		if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
858			v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
859		if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
860			v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
861		if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
862			v.val |= SOF_TIMESTAMPING_SOFTWARE;
863		if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
864			v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
865		if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
866			v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
867		break;
868
869	case SO_RCVTIMEO:
870		lv = sizeof(struct timeval);
871		if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
872			v.tm.tv_sec = 0;
873			v.tm.tv_usec = 0;
874		} else {
875			v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
876			v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
877		}
878		break;
879
880	case SO_SNDTIMEO:
881		lv = sizeof(struct timeval);
882		if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
883			v.tm.tv_sec = 0;
884			v.tm.tv_usec = 0;
885		} else {
886			v.tm.tv_sec = sk->sk_sndtimeo / HZ;
887			v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
888		}
889		break;
890
891	case SO_RCVLOWAT:
892		v.val = sk->sk_rcvlowat;
893		break;
894
895	case SO_SNDLOWAT:
896		v.val = 1;
897		break;
898
899	case SO_PASSCRED:
900		v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
901		break;
902
903	case SO_PEERCRED:
904		if (len > sizeof(sk->sk_peercred))
905			len = sizeof(sk->sk_peercred);
906		if (copy_to_user(optval, &sk->sk_peercred, len))
907			return -EFAULT;
908		goto lenout;
909
910	case SO_PEERNAME:
911	{
912		char address[128];
913
914		if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
915			return -ENOTCONN;
916		if (lv < len)
917			return -EINVAL;
918		if (copy_to_user(optval, address, len))
919			return -EFAULT;
920		goto lenout;
921	}
922
923	/* Dubious BSD thing... Probably nobody even uses it, but
924	 * the UNIX standard wants it for whatever reason... -DaveM
925	 */
926	case SO_ACCEPTCONN:
927		v.val = sk->sk_state == TCP_LISTEN;
928		break;
929
930	case SO_PASSSEC:
931		v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
932		break;
933
934	case SO_PEERSEC:
935		return security_socket_getpeersec_stream(sock, optval, optlen, len);
936
937	case SO_MARK:
938		v.val = sk->sk_mark;
939		break;
940
941	case SO_RXQ_OVFL:
942		v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
943		break;
944
945	default:
946		return -ENOPROTOOPT;
947	}
948
949	if (len > lv)
950		len = lv;
951	if (copy_to_user(optval, &v, len))
952		return -EFAULT;
953lenout:
954	if (put_user(len, optlen))
955		return -EFAULT;
956	return 0;
957}
958
959/*
960 * Initialize an sk_lock.
961 *
962 * (We also register the sk_lock with the lock validator.)
963 */
964static inline void sock_lock_init(struct sock *sk)
965{
966	sock_lock_init_class_and_name(sk,
967			af_family_slock_key_strings[sk->sk_family],
968			af_family_slock_keys + sk->sk_family,
969			af_family_key_strings[sk->sk_family],
970			af_family_keys + sk->sk_family);
971}
972
973/*
974 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
975 * even temporarly, because of RCU lookups. sk_node should also be left as is.
976 */
977static void sock_copy(struct sock *nsk, const struct sock *osk)
978{
979#ifdef CONFIG_SECURITY_NETWORK
980	void *sptr = nsk->sk_security;
981#endif
982	BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
983		     sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
984		     sizeof(osk->sk_tx_queue_mapping));
985	memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
986	       osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
987#ifdef CONFIG_SECURITY_NETWORK
988	nsk->sk_security = sptr;
989	security_sk_clone(osk, nsk);
990#endif
991}
992
993static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
994		int family)
995{
996	struct sock *sk;
997	struct kmem_cache *slab;
998
999	slab = prot->slab;
1000	if (slab != NULL) {
1001		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1002		if (!sk)
1003			return sk;
1004		if (priority & __GFP_ZERO) {
1005			/*
1006			 * caches using SLAB_DESTROY_BY_RCU should let
1007			 * sk_node.next un-modified. Special care is taken
1008			 * when initializing object to zero.
1009			 */
1010			if (offsetof(struct sock, sk_node.next) != 0)
1011				memset(sk, 0, offsetof(struct sock, sk_node.next));
1012			memset(&sk->sk_node.pprev, 0,
1013			       prot->obj_size - offsetof(struct sock,
1014							 sk_node.pprev));
1015		}
1016	}
1017	else
1018		sk = kmalloc(prot->obj_size, priority);
1019
1020	if (sk != NULL) {
1021		kmemcheck_annotate_bitfield(sk, flags);
1022
1023		if (security_sk_alloc(sk, family, priority))
1024			goto out_free;
1025
1026		if (!try_module_get(prot->owner))
1027			goto out_free_sec;
1028		sk_tx_queue_clear(sk);
1029	}
1030
1031	return sk;
1032
1033out_free_sec:
1034	security_sk_free(sk);
1035out_free:
1036	if (slab != NULL)
1037		kmem_cache_free(slab, sk);
1038	else
1039		kfree(sk);
1040	return NULL;
1041}
1042
1043static void sk_prot_free(struct proto *prot, struct sock *sk)
1044{
1045	struct kmem_cache *slab;
1046	struct module *owner;
1047
1048	owner = prot->owner;
1049	slab = prot->slab;
1050
1051	security_sk_free(sk);
1052	if (slab != NULL)
1053		kmem_cache_free(slab, sk);
1054	else
1055		kfree(sk);
1056	module_put(owner);
1057}
1058
1059#ifdef CONFIG_CGROUPS
1060void sock_update_classid(struct sock *sk)
1061{
1062	u32 classid = task_cls_classid(current);
1063
1064	if (classid && classid != sk->sk_classid)
1065		sk->sk_classid = classid;
1066}
1067EXPORT_SYMBOL(sock_update_classid);
1068#endif
1069
1070/**
1071 *	sk_alloc - All socket objects are allocated here
1072 *	@net: the applicable net namespace
1073 *	@family: protocol family
1074 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1075 *	@prot: struct proto associated with this new sock instance
1076 */
1077struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1078		      struct proto *prot)
1079{
1080	struct sock *sk;
1081
1082	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1083	if (sk) {
1084		sk->sk_family = family;
1085		/*
1086		 * See comment in struct sock definition to understand
1087		 * why we need sk_prot_creator -acme
1088		 */
1089		sk->sk_prot = sk->sk_prot_creator = prot;
1090		sock_lock_init(sk);
1091		sock_net_set(sk, get_net(net));
1092		atomic_set(&sk->sk_wmem_alloc, 1);
1093
1094		sock_update_classid(sk);
1095	}
1096
1097	return sk;
1098}
1099EXPORT_SYMBOL(sk_alloc);
1100
1101static void __sk_free(struct sock *sk)
1102{
1103	struct sk_filter *filter;
1104
1105	if (sk->sk_destruct)
1106		sk->sk_destruct(sk);
1107
1108	filter = rcu_dereference_check(sk->sk_filter,
1109				       atomic_read(&sk->sk_wmem_alloc) == 0);
1110	if (filter) {
1111		sk_filter_uncharge(sk, filter);
1112		rcu_assign_pointer(sk->sk_filter, NULL);
1113	}
1114
1115	sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1116	sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1117
1118	if (atomic_read(&sk->sk_omem_alloc))
1119		printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1120		       __func__, atomic_read(&sk->sk_omem_alloc));
1121
1122	put_net(sock_net(sk));
1123	sk_prot_free(sk->sk_prot_creator, sk);
1124}
1125
1126void sk_free(struct sock *sk)
1127{
1128	/*
1129	 * We substract one from sk_wmem_alloc and can know if
1130	 * some packets are still in some tx queue.
1131	 * If not null, sock_wfree() will call __sk_free(sk) later
1132	 */
1133	if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1134		__sk_free(sk);
1135}
1136EXPORT_SYMBOL(sk_free);
1137
1138/*
1139 * Last sock_put should drop referrence to sk->sk_net. It has already
1140 * been dropped in sk_change_net. Taking referrence to stopping namespace
1141 * is not an option.
1142 * Take referrence to a socket to remove it from hash _alive_ and after that
1143 * destroy it in the context of init_net.
1144 */
1145void sk_release_kernel(struct sock *sk)
1146{
1147	if (sk == NULL || sk->sk_socket == NULL)
1148		return;
1149
1150	sock_hold(sk);
1151	sock_release(sk->sk_socket);
1152	release_net(sock_net(sk));
1153	sock_net_set(sk, get_net(&init_net));
1154	sock_put(sk);
1155}
1156EXPORT_SYMBOL(sk_release_kernel);
1157
1158struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1159{
1160	struct sock *newsk;
1161
1162	newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1163	if (newsk != NULL) {
1164		struct sk_filter *filter;
1165
1166		sock_copy(newsk, sk);
1167
1168		/* SANITY */
1169		get_net(sock_net(newsk));
1170		sk_node_init(&newsk->sk_node);
1171		sock_lock_init(newsk);
1172		bh_lock_sock(newsk);
1173		newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
1174		newsk->sk_backlog.len = 0;
1175
1176		atomic_set(&newsk->sk_rmem_alloc, 0);
1177		/*
1178		 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1179		 */
1180		atomic_set(&newsk->sk_wmem_alloc, 1);
1181		atomic_set(&newsk->sk_omem_alloc, 0);
1182		skb_queue_head_init(&newsk->sk_receive_queue);
1183		skb_queue_head_init(&newsk->sk_write_queue);
1184#ifdef CONFIG_NET_DMA
1185		skb_queue_head_init(&newsk->sk_async_wait_queue);
1186#endif
1187
1188		spin_lock_init(&newsk->sk_dst_lock);
1189		rwlock_init(&newsk->sk_callback_lock);
1190		lockdep_set_class_and_name(&newsk->sk_callback_lock,
1191				af_callback_keys + newsk->sk_family,
1192				af_family_clock_key_strings[newsk->sk_family]);
1193
1194		newsk->sk_dst_cache	= NULL;
1195		newsk->sk_wmem_queued	= 0;
1196		newsk->sk_forward_alloc = 0;
1197		newsk->sk_send_head	= NULL;
1198		newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1199
1200		sock_reset_flag(newsk, SOCK_DONE);
1201		skb_queue_head_init(&newsk->sk_error_queue);
1202
1203		filter = newsk->sk_filter;
1204		if (filter != NULL)
1205			sk_filter_charge(newsk, filter);
1206
1207		if (unlikely(xfrm_sk_clone_policy(newsk))) {
1208			/* It is still raw copy of parent, so invalidate
1209			 * destructor and make plain sk_free() */
1210			newsk->sk_destruct = NULL;
1211			sk_free(newsk);
1212			newsk = NULL;
1213			goto out;
1214		}
1215
1216		newsk->sk_err	   = 0;
1217		newsk->sk_priority = 0;
1218		/*
1219		 * Before updating sk_refcnt, we must commit prior changes to memory
1220		 * (Documentation/RCU/rculist_nulls.txt for details)
1221		 */
1222		smp_wmb();
1223		atomic_set(&newsk->sk_refcnt, 2);
1224
1225		/*
1226		 * Increment the counter in the same struct proto as the master
1227		 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1228		 * is the same as sk->sk_prot->socks, as this field was copied
1229		 * with memcpy).
1230		 *
1231		 * This _changes_ the previous behaviour, where
1232		 * tcp_create_openreq_child always was incrementing the
1233		 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1234		 * to be taken into account in all callers. -acme
1235		 */
1236		sk_refcnt_debug_inc(newsk);
1237		sk_set_socket(newsk, NULL);
1238		newsk->sk_wq = NULL;
1239
1240		if (newsk->sk_prot->sockets_allocated)
1241			percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1242
1243		if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1244		    sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1245			net_enable_timestamp();
1246	}
1247out:
1248	return newsk;
1249}
1250EXPORT_SYMBOL_GPL(sk_clone);
1251
1252void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1253{
1254	__sk_dst_set(sk, dst);
1255	sk->sk_route_caps = dst->dev->features;
1256	if (sk->sk_route_caps & NETIF_F_GSO)
1257		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1258	sk->sk_route_caps &= ~sk->sk_route_nocaps;
1259	if (sk_can_gso(sk)) {
1260		if (dst->header_len) {
1261			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1262		} else {
1263			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1264			sk->sk_gso_max_size = dst->dev->gso_max_size;
1265		}
1266	}
1267}
1268EXPORT_SYMBOL_GPL(sk_setup_caps);
1269
1270void __init sk_init(void)
1271{
1272	if (totalram_pages <= 4096) {
1273		sysctl_wmem_max = 32767;
1274		sysctl_rmem_max = 32767;
1275		sysctl_wmem_default = 32767;
1276		sysctl_rmem_default = 32767;
1277	} else if (totalram_pages >= 131072) {
1278		sysctl_wmem_max = 131071;
1279		sysctl_rmem_max = 131071;
1280	}
1281}
1282
1283/*
1284 *	Simple resource managers for sockets.
1285 */
1286
1287
1288/*
1289 * Write buffer destructor automatically called from kfree_skb.
1290 */
1291void sock_wfree(struct sk_buff *skb)
1292{
1293	struct sock *sk = skb->sk;
1294	unsigned int len = skb->truesize;
1295
1296	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1297		/*
1298		 * Keep a reference on sk_wmem_alloc, this will be released
1299		 * after sk_write_space() call
1300		 */
1301		atomic_sub(len - 1, &sk->sk_wmem_alloc);
1302		sk->sk_write_space(sk);
1303		len = 1;
1304	}
1305	/*
1306	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1307	 * could not do because of in-flight packets
1308	 */
1309	if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1310		__sk_free(sk);
1311}
1312EXPORT_SYMBOL(sock_wfree);
1313
1314/*
1315 * Read buffer destructor automatically called from kfree_skb.
1316 */
1317void sock_rfree(struct sk_buff *skb)
1318{
1319	struct sock *sk = skb->sk;
1320
1321	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1322	sk_mem_uncharge(skb->sk, skb->truesize);
1323}
1324EXPORT_SYMBOL(sock_rfree);
1325
1326
1327int sock_i_uid(struct sock *sk)
1328{
1329	int uid;
1330
1331	read_lock(&sk->sk_callback_lock);
1332	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1333	read_unlock(&sk->sk_callback_lock);
1334	return uid;
1335}
1336EXPORT_SYMBOL(sock_i_uid);
1337
1338unsigned long sock_i_ino(struct sock *sk)
1339{
1340	unsigned long ino;
1341
1342	read_lock(&sk->sk_callback_lock);
1343	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1344	read_unlock(&sk->sk_callback_lock);
1345	return ino;
1346}
1347EXPORT_SYMBOL(sock_i_ino);
1348
1349/*
1350 * Allocate a skb from the socket's send buffer.
1351 */
1352struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1353			     gfp_t priority)
1354{
1355	if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1356		struct sk_buff *skb = alloc_skb(size, priority);
1357		if (skb) {
1358			skb_set_owner_w(skb, sk);
1359			return skb;
1360		}
1361	}
1362	return NULL;
1363}
1364EXPORT_SYMBOL(sock_wmalloc);
1365
1366/*
1367 * Allocate a skb from the socket's receive buffer.
1368 */
1369struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1370			     gfp_t priority)
1371{
1372	if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1373		struct sk_buff *skb = alloc_skb(size, priority);
1374		if (skb) {
1375			skb_set_owner_r(skb, sk);
1376			return skb;
1377		}
1378	}
1379	return NULL;
1380}
1381
1382/*
1383 * Allocate a memory block from the socket's option memory buffer.
1384 */
1385void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1386{
1387	if ((unsigned)size <= sysctl_optmem_max &&
1388	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1389		void *mem;
1390		/* First do the add, to avoid the race if kmalloc
1391		 * might sleep.
1392		 */
1393		atomic_add(size, &sk->sk_omem_alloc);
1394		mem = kmalloc(size, priority);
1395		if (mem)
1396			return mem;
1397		atomic_sub(size, &sk->sk_omem_alloc);
1398	}
1399	return NULL;
1400}
1401EXPORT_SYMBOL(sock_kmalloc);
1402
1403/*
1404 * Free an option memory block.
1405 */
1406void sock_kfree_s(struct sock *sk, void *mem, int size)
1407{
1408	kfree(mem);
1409	atomic_sub(size, &sk->sk_omem_alloc);
1410}
1411EXPORT_SYMBOL(sock_kfree_s);
1412
1413/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1414   I think, these locks should be removed for datagram sockets.
1415 */
1416static long sock_wait_for_wmem(struct sock *sk, long timeo)
1417{
1418	DEFINE_WAIT(wait);
1419
1420	clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1421	for (;;) {
1422		if (!timeo)
1423			break;
1424		if (signal_pending(current))
1425			break;
1426		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1427		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1428		if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1429			break;
1430		if (sk->sk_shutdown & SEND_SHUTDOWN)
1431			break;
1432		if (sk->sk_err)
1433			break;
1434		timeo = schedule_timeout(timeo);
1435	}
1436	finish_wait(sk_sleep(sk), &wait);
1437	return timeo;
1438}
1439
1440
1441/*
1442 *	Generic send/receive buffer handlers
1443 */
1444
1445struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1446				     unsigned long data_len, int noblock,
1447				     int *errcode)
1448{
1449	struct sk_buff *skb;
1450	gfp_t gfp_mask;
1451	long timeo;
1452	int err;
1453
1454	gfp_mask = sk->sk_allocation;
1455	if (gfp_mask & __GFP_WAIT)
1456		gfp_mask |= __GFP_REPEAT;
1457
1458	timeo = sock_sndtimeo(sk, noblock);
1459	while (1) {
1460		err = sock_error(sk);
1461		if (err != 0)
1462			goto failure;
1463
1464		err = -EPIPE;
1465		if (sk->sk_shutdown & SEND_SHUTDOWN)
1466			goto failure;
1467
1468		if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1469			skb = alloc_skb(header_len, gfp_mask);
1470			if (skb) {
1471				int npages;
1472				int i;
1473
1474				/* No pages, we're done... */
1475				if (!data_len)
1476					break;
1477
1478				npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1479				skb->truesize += data_len;
1480				skb_shinfo(skb)->nr_frags = npages;
1481				for (i = 0; i < npages; i++) {
1482					struct page *page;
1483					skb_frag_t *frag;
1484
1485					page = alloc_pages(sk->sk_allocation, 0);
1486					if (!page) {
1487						err = -ENOBUFS;
1488						skb_shinfo(skb)->nr_frags = i;
1489						kfree_skb(skb);
1490						goto failure;
1491					}
1492
1493					frag = &skb_shinfo(skb)->frags[i];
1494					frag->page = page;
1495					frag->page_offset = 0;
1496					frag->size = (data_len >= PAGE_SIZE ?
1497						      PAGE_SIZE :
1498						      data_len);
1499					data_len -= PAGE_SIZE;
1500				}
1501
1502				/* Full success... */
1503				break;
1504			}
1505			err = -ENOBUFS;
1506			goto failure;
1507		}
1508		set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1509		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1510		err = -EAGAIN;
1511		if (!timeo)
1512			goto failure;
1513		if (signal_pending(current))
1514			goto interrupted;
1515		timeo = sock_wait_for_wmem(sk, timeo);
1516	}
1517
1518	skb_set_owner_w(skb, sk);
1519	return skb;
1520
1521interrupted:
1522	err = sock_intr_errno(timeo);
1523failure:
1524	*errcode = err;
1525	return NULL;
1526}
1527EXPORT_SYMBOL(sock_alloc_send_pskb);
1528
1529struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1530				    int noblock, int *errcode)
1531{
1532	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1533}
1534EXPORT_SYMBOL(sock_alloc_send_skb);
1535
1536static void __lock_sock(struct sock *sk)
1537{
1538	DEFINE_WAIT(wait);
1539
1540	for (;;) {
1541		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1542					TASK_UNINTERRUPTIBLE);
1543		spin_unlock_bh(&sk->sk_lock.slock);
1544		schedule();
1545		spin_lock_bh(&sk->sk_lock.slock);
1546		if (!sock_owned_by_user(sk))
1547			break;
1548	}
1549	finish_wait(&sk->sk_lock.wq, &wait);
1550}
1551
1552static void __release_sock(struct sock *sk)
1553{
1554	struct sk_buff *skb = sk->sk_backlog.head;
1555
1556	do {
1557		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1558		bh_unlock_sock(sk);
1559
1560		do {
1561			struct sk_buff *next = skb->next;
1562
1563			WARN_ON_ONCE(skb_dst_is_noref(skb));
1564			skb->next = NULL;
1565			sk_backlog_rcv(sk, skb);
1566
1567			/*
1568			 * We are in process context here with softirqs
1569			 * disabled, use cond_resched_softirq() to preempt.
1570			 * This is safe to do because we've taken the backlog
1571			 * queue private:
1572			 */
1573			cond_resched_softirq();
1574
1575			skb = next;
1576		} while (skb != NULL);
1577
1578		bh_lock_sock(sk);
1579	} while ((skb = sk->sk_backlog.head) != NULL);
1580
1581	/*
1582	 * Doing the zeroing here guarantee we can not loop forever
1583	 * while a wild producer attempts to flood us.
1584	 */
1585	sk->sk_backlog.len = 0;
1586}
1587
1588/**
1589 * sk_wait_data - wait for data to arrive at sk_receive_queue
1590 * @sk:    sock to wait on
1591 * @timeo: for how long
1592 *
1593 * Now socket state including sk->sk_err is changed only under lock,
1594 * hence we may omit checks after joining wait queue.
1595 * We check receive queue before schedule() only as optimization;
1596 * it is very likely that release_sock() added new data.
1597 */
1598int sk_wait_data(struct sock *sk, long *timeo)
1599{
1600	int rc;
1601	DEFINE_WAIT(wait);
1602
1603	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1604	set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1605	rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1606	clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1607	finish_wait(sk_sleep(sk), &wait);
1608	return rc;
1609}
1610EXPORT_SYMBOL(sk_wait_data);
1611
1612/**
1613 *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1614 *	@sk: socket
1615 *	@size: memory size to allocate
1616 *	@kind: allocation type
1617 *
1618 *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1619 *	rmem allocation. This function assumes that protocols which have
1620 *	memory_pressure use sk_wmem_queued as write buffer accounting.
1621 */
1622int __sk_mem_schedule(struct sock *sk, int size, int kind)
1623{
1624	struct proto *prot = sk->sk_prot;
1625	int amt = sk_mem_pages(size);
1626	int allocated;
1627
1628	sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1629	allocated = atomic_add_return(amt, prot->memory_allocated);
1630
1631	/* Under limit. */
1632	if (allocated <= prot->sysctl_mem[0]) {
1633		if (prot->memory_pressure && *prot->memory_pressure)
1634			*prot->memory_pressure = 0;
1635		return 1;
1636	}
1637
1638	/* Under pressure. */
1639	if (allocated > prot->sysctl_mem[1])
1640		if (prot->enter_memory_pressure)
1641			prot->enter_memory_pressure(sk);
1642
1643	/* Over hard limit. */
1644	if (allocated > prot->sysctl_mem[2])
1645		goto suppress_allocation;
1646
1647	/* guarantee minimum buffer size under pressure */
1648	if (kind == SK_MEM_RECV) {
1649		if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1650			return 1;
1651	} else { /* SK_MEM_SEND */
1652		if (sk->sk_type == SOCK_STREAM) {
1653			if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1654				return 1;
1655		} else if (atomic_read(&sk->sk_wmem_alloc) <
1656			   prot->sysctl_wmem[0])
1657				return 1;
1658	}
1659
1660	if (prot->memory_pressure) {
1661		int alloc;
1662
1663		if (!*prot->memory_pressure)
1664			return 1;
1665		alloc = percpu_counter_read_positive(prot->sockets_allocated);
1666		if (prot->sysctl_mem[2] > alloc *
1667		    sk_mem_pages(sk->sk_wmem_queued +
1668				 atomic_read(&sk->sk_rmem_alloc) +
1669				 sk->sk_forward_alloc))
1670			return 1;
1671	}
1672
1673suppress_allocation:
1674
1675	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1676		sk_stream_moderate_sndbuf(sk);
1677
1678		/* Fail only if socket is _under_ its sndbuf.
1679		 * In this case we cannot block, so that we have to fail.
1680		 */
1681		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1682			return 1;
1683	}
1684
1685	/* Alas. Undo changes. */
1686	sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1687	atomic_sub(amt, prot->memory_allocated);
1688	return 0;
1689}
1690EXPORT_SYMBOL(__sk_mem_schedule);
1691
1692/**
1693 *	__sk_reclaim - reclaim memory_allocated
1694 *	@sk: socket
1695 */
1696void __sk_mem_reclaim(struct sock *sk)
1697{
1698	struct proto *prot = sk->sk_prot;
1699
1700	atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1701		   prot->memory_allocated);
1702	sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1703
1704	if (prot->memory_pressure && *prot->memory_pressure &&
1705	    (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1706		*prot->memory_pressure = 0;
1707}
1708EXPORT_SYMBOL(__sk_mem_reclaim);
1709
1710
1711/*
1712 * Set of default routines for initialising struct proto_ops when
1713 * the protocol does not support a particular function. In certain
1714 * cases where it makes no sense for a protocol to have a "do nothing"
1715 * function, some default processing is provided.
1716 */
1717
1718int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1719{
1720	return -EOPNOTSUPP;
1721}
1722EXPORT_SYMBOL(sock_no_bind);
1723
1724int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1725		    int len, int flags)
1726{
1727	return -EOPNOTSUPP;
1728}
1729EXPORT_SYMBOL(sock_no_connect);
1730
1731int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1732{
1733	return -EOPNOTSUPP;
1734}
1735EXPORT_SYMBOL(sock_no_socketpair);
1736
1737int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1738{
1739	return -EOPNOTSUPP;
1740}
1741EXPORT_SYMBOL(sock_no_accept);
1742
1743int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1744		    int *len, int peer)
1745{
1746	return -EOPNOTSUPP;
1747}
1748EXPORT_SYMBOL(sock_no_getname);
1749
1750unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1751{
1752	return 0;
1753}
1754EXPORT_SYMBOL(sock_no_poll);
1755
1756int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1757{
1758	return -EOPNOTSUPP;
1759}
1760EXPORT_SYMBOL(sock_no_ioctl);
1761
1762int sock_no_listen(struct socket *sock, int backlog)
1763{
1764	return -EOPNOTSUPP;
1765}
1766EXPORT_SYMBOL(sock_no_listen);
1767
1768int sock_no_shutdown(struct socket *sock, int how)
1769{
1770	return -EOPNOTSUPP;
1771}
1772EXPORT_SYMBOL(sock_no_shutdown);
1773
1774int sock_no_setsockopt(struct socket *sock, int level, int optname,
1775		    char __user *optval, unsigned int optlen)
1776{
1777	return -EOPNOTSUPP;
1778}
1779EXPORT_SYMBOL(sock_no_setsockopt);
1780
1781int sock_no_getsockopt(struct socket *sock, int level, int optname,
1782		    char __user *optval, int __user *optlen)
1783{
1784	return -EOPNOTSUPP;
1785}
1786EXPORT_SYMBOL(sock_no_getsockopt);
1787
1788int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1789		    size_t len)
1790{
1791	return -EOPNOTSUPP;
1792}
1793EXPORT_SYMBOL(sock_no_sendmsg);
1794
1795int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1796		    size_t len, int flags)
1797{
1798	return -EOPNOTSUPP;
1799}
1800EXPORT_SYMBOL(sock_no_recvmsg);
1801
1802int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1803{
1804	/* Mirror missing mmap method error code */
1805	return -ENODEV;
1806}
1807EXPORT_SYMBOL(sock_no_mmap);
1808
1809ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1810{
1811	ssize_t res;
1812	struct msghdr msg = {.msg_flags = flags};
1813	struct kvec iov;
1814	char *kaddr = kmap(page);
1815	iov.iov_base = kaddr + offset;
1816	iov.iov_len = size;
1817	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1818	kunmap(page);
1819	return res;
1820}
1821EXPORT_SYMBOL(sock_no_sendpage);
1822
1823/*
1824 *	Default Socket Callbacks
1825 */
1826
1827static void sock_def_wakeup(struct sock *sk)
1828{
1829	struct socket_wq *wq;
1830
1831	rcu_read_lock();
1832	wq = rcu_dereference(sk->sk_wq);
1833	if (wq_has_sleeper(wq))
1834		wake_up_interruptible_all(&wq->wait);
1835	rcu_read_unlock();
1836}
1837
1838static void sock_def_error_report(struct sock *sk)
1839{
1840	struct socket_wq *wq;
1841
1842	rcu_read_lock();
1843	wq = rcu_dereference(sk->sk_wq);
1844	if (wq_has_sleeper(wq))
1845		wake_up_interruptible_poll(&wq->wait, POLLERR);
1846	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1847	rcu_read_unlock();
1848}
1849
1850static void sock_def_readable(struct sock *sk, int len)
1851{
1852	struct socket_wq *wq;
1853
1854	rcu_read_lock();
1855	wq = rcu_dereference(sk->sk_wq);
1856	if (wq_has_sleeper(wq))
1857		wake_up_interruptible_sync_poll(&wq->wait, POLLIN |
1858						POLLRDNORM | POLLRDBAND);
1859	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1860	rcu_read_unlock();
1861}
1862
1863static void sock_def_write_space(struct sock *sk)
1864{
1865	struct socket_wq *wq;
1866
1867	rcu_read_lock();
1868
1869	/* Do not wake up a writer until he can make "significant"
1870	 * progress.  --DaveM
1871	 */
1872	if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1873		wq = rcu_dereference(sk->sk_wq);
1874		if (wq_has_sleeper(wq))
1875			wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1876						POLLWRNORM | POLLWRBAND);
1877
1878		/* Should agree with poll, otherwise some programs break */
1879		if (sock_writeable(sk))
1880			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1881	}
1882
1883	rcu_read_unlock();
1884}
1885
1886static void sock_def_destruct(struct sock *sk)
1887{
1888	kfree(sk->sk_protinfo);
1889}
1890
1891void sk_send_sigurg(struct sock *sk)
1892{
1893	if (sk->sk_socket && sk->sk_socket->file)
1894		if (send_sigurg(&sk->sk_socket->file->f_owner))
1895			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1896}
1897EXPORT_SYMBOL(sk_send_sigurg);
1898
1899void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1900		    unsigned long expires)
1901{
1902	if (!mod_timer(timer, expires))
1903		sock_hold(sk);
1904}
1905EXPORT_SYMBOL(sk_reset_timer);
1906
1907void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1908{
1909	if (timer_pending(timer) && del_timer(timer))
1910		__sock_put(sk);
1911}
1912EXPORT_SYMBOL(sk_stop_timer);
1913
1914void sock_init_data(struct socket *sock, struct sock *sk)
1915{
1916	skb_queue_head_init(&sk->sk_receive_queue);
1917	skb_queue_head_init(&sk->sk_write_queue);
1918	skb_queue_head_init(&sk->sk_error_queue);
1919#ifdef CONFIG_NET_DMA
1920	skb_queue_head_init(&sk->sk_async_wait_queue);
1921#endif
1922
1923	sk->sk_send_head	=	NULL;
1924
1925	init_timer(&sk->sk_timer);
1926
1927	sk->sk_allocation	=	GFP_KERNEL;
1928	sk->sk_rcvbuf		=	sysctl_rmem_default;
1929	sk->sk_sndbuf		=	sysctl_wmem_default;
1930	sk->sk_state		=	TCP_CLOSE;
1931	sk_set_socket(sk, sock);
1932
1933	sock_set_flag(sk, SOCK_ZAPPED);
1934
1935	if (sock) {
1936		sk->sk_type	=	sock->type;
1937		sk->sk_wq	=	sock->wq;
1938		sock->sk	=	sk;
1939	} else
1940		sk->sk_wq	=	NULL;
1941
1942	spin_lock_init(&sk->sk_dst_lock);
1943	rwlock_init(&sk->sk_callback_lock);
1944	lockdep_set_class_and_name(&sk->sk_callback_lock,
1945			af_callback_keys + sk->sk_family,
1946			af_family_clock_key_strings[sk->sk_family]);
1947
1948	sk->sk_state_change	=	sock_def_wakeup;
1949	sk->sk_data_ready	=	sock_def_readable;
1950	sk->sk_write_space	=	sock_def_write_space;
1951	sk->sk_error_report	=	sock_def_error_report;
1952	sk->sk_destruct		=	sock_def_destruct;
1953
1954	sk->sk_sndmsg_page	=	NULL;
1955	sk->sk_sndmsg_off	=	0;
1956
1957	sk->sk_peercred.pid 	=	0;
1958	sk->sk_peercred.uid	=	-1;
1959	sk->sk_peercred.gid	=	-1;
1960	sk->sk_write_pending	=	0;
1961	sk->sk_rcvlowat		=	1;
1962	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
1963	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
1964
1965	sk->sk_stamp = ktime_set(-1L, 0);
1966
1967	/*
1968	 * Before updating sk_refcnt, we must commit prior changes to memory
1969	 * (Documentation/RCU/rculist_nulls.txt for details)
1970	 */
1971	smp_wmb();
1972	atomic_set(&sk->sk_refcnt, 1);
1973	atomic_set(&sk->sk_drops, 0);
1974}
1975EXPORT_SYMBOL(sock_init_data);
1976
1977void lock_sock_nested(struct sock *sk, int subclass)
1978{
1979	might_sleep();
1980	spin_lock_bh(&sk->sk_lock.slock);
1981	if (sk->sk_lock.owned)
1982		__lock_sock(sk);
1983	sk->sk_lock.owned = 1;
1984	spin_unlock(&sk->sk_lock.slock);
1985	/*
1986	 * The sk_lock has mutex_lock() semantics here:
1987	 */
1988	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1989	local_bh_enable();
1990}
1991EXPORT_SYMBOL(lock_sock_nested);
1992
1993void release_sock(struct sock *sk)
1994{
1995	/*
1996	 * The sk_lock has mutex_unlock() semantics:
1997	 */
1998	mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1999
2000	spin_lock_bh(&sk->sk_lock.slock);
2001	if (sk->sk_backlog.tail)
2002		__release_sock(sk);
2003	sk->sk_lock.owned = 0;
2004	if (waitqueue_active(&sk->sk_lock.wq))
2005		wake_up(&sk->sk_lock.wq);
2006	spin_unlock_bh(&sk->sk_lock.slock);
2007}
2008EXPORT_SYMBOL(release_sock);
2009
2010/**
2011 * lock_sock_fast - fast version of lock_sock
2012 * @sk: socket
2013 *
2014 * This version should be used for very small section, where process wont block
2015 * return false if fast path is taken
2016 *   sk_lock.slock locked, owned = 0, BH disabled
2017 * return true if slow path is taken
2018 *   sk_lock.slock unlocked, owned = 1, BH enabled
2019 */
2020bool lock_sock_fast(struct sock *sk)
2021{
2022	might_sleep();
2023	spin_lock_bh(&sk->sk_lock.slock);
2024
2025	if (!sk->sk_lock.owned)
2026		/*
2027		 * Note : We must disable BH
2028		 */
2029		return false;
2030
2031	__lock_sock(sk);
2032	sk->sk_lock.owned = 1;
2033	spin_unlock(&sk->sk_lock.slock);
2034	/*
2035	 * The sk_lock has mutex_lock() semantics here:
2036	 */
2037	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2038	local_bh_enable();
2039	return true;
2040}
2041EXPORT_SYMBOL(lock_sock_fast);
2042
2043int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2044{
2045	struct timeval tv;
2046	if (!sock_flag(sk, SOCK_TIMESTAMP))
2047		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2048	tv = ktime_to_timeval(sk->sk_stamp);
2049	if (tv.tv_sec == -1)
2050		return -ENOENT;
2051	if (tv.tv_sec == 0) {
2052		sk->sk_stamp = ktime_get_real();
2053		tv = ktime_to_timeval(sk->sk_stamp);
2054	}
2055	return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2056}
2057EXPORT_SYMBOL(sock_get_timestamp);
2058
2059int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2060{
2061	struct timespec ts;
2062	if (!sock_flag(sk, SOCK_TIMESTAMP))
2063		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2064	ts = ktime_to_timespec(sk->sk_stamp);
2065	if (ts.tv_sec == -1)
2066		return -ENOENT;
2067	if (ts.tv_sec == 0) {
2068		sk->sk_stamp = ktime_get_real();
2069		ts = ktime_to_timespec(sk->sk_stamp);
2070	}
2071	return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2072}
2073EXPORT_SYMBOL(sock_get_timestampns);
2074
2075void sock_enable_timestamp(struct sock *sk, int flag)
2076{
2077	if (!sock_flag(sk, flag)) {
2078		sock_set_flag(sk, flag);
2079		/*
2080		 * we just set one of the two flags which require net
2081		 * time stamping, but time stamping might have been on
2082		 * already because of the other one
2083		 */
2084		if (!sock_flag(sk,
2085				flag == SOCK_TIMESTAMP ?
2086				SOCK_TIMESTAMPING_RX_SOFTWARE :
2087				SOCK_TIMESTAMP))
2088			net_enable_timestamp();
2089	}
2090}
2091
2092/*
2093 *	Get a socket option on an socket.
2094 *
2095 *	FIX: POSIX 1003.1g is very ambiguous here. It states that
2096 *	asynchronous errors should be reported by getsockopt. We assume
2097 *	this means if you specify SO_ERROR (otherwise whats the point of it).
2098 */
2099int sock_common_getsockopt(struct socket *sock, int level, int optname,
2100			   char __user *optval, int __user *optlen)
2101{
2102	struct sock *sk = sock->sk;
2103
2104	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2105}
2106EXPORT_SYMBOL(sock_common_getsockopt);
2107
2108#ifdef CONFIG_COMPAT
2109int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2110				  char __user *optval, int __user *optlen)
2111{
2112	struct sock *sk = sock->sk;
2113
2114	if (sk->sk_prot->compat_getsockopt != NULL)
2115		return sk->sk_prot->compat_getsockopt(sk, level, optname,
2116						      optval, optlen);
2117	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2118}
2119EXPORT_SYMBOL(compat_sock_common_getsockopt);
2120#endif
2121
2122int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2123			struct msghdr *msg, size_t size, int flags)
2124{
2125	struct sock *sk = sock->sk;
2126	int addr_len = 0;
2127	int err;
2128
2129	err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2130				   flags & ~MSG_DONTWAIT, &addr_len);
2131	if (err >= 0)
2132		msg->msg_namelen = addr_len;
2133	return err;
2134}
2135EXPORT_SYMBOL(sock_common_recvmsg);
2136
2137/*
2138 *	Set socket options on an inet socket.
2139 */
2140int sock_common_setsockopt(struct socket *sock, int level, int optname,
2141			   char __user *optval, unsigned int optlen)
2142{
2143	struct sock *sk = sock->sk;
2144
2145	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2146}
2147EXPORT_SYMBOL(sock_common_setsockopt);
2148
2149#ifdef CONFIG_COMPAT
2150int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2151				  char __user *optval, unsigned int optlen)
2152{
2153	struct sock *sk = sock->sk;
2154
2155	if (sk->sk_prot->compat_setsockopt != NULL)
2156		return sk->sk_prot->compat_setsockopt(sk, level, optname,
2157						      optval, optlen);
2158	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2159}
2160EXPORT_SYMBOL(compat_sock_common_setsockopt);
2161#endif
2162
2163void sk_common_release(struct sock *sk)
2164{
2165	if (sk->sk_prot->destroy)
2166		sk->sk_prot->destroy(sk);
2167
2168	/*
2169	 * Observation: when sock_common_release is called, processes have
2170	 * no access to socket. But net still has.
2171	 * Step one, detach it from networking:
2172	 *
2173	 * A. Remove from hash tables.
2174	 */
2175
2176	sk->sk_prot->unhash(sk);
2177
2178	/*
2179	 * In this point socket cannot receive new packets, but it is possible
2180	 * that some packets are in flight because some CPU runs receiver and
2181	 * did hash table lookup before we unhashed socket. They will achieve
2182	 * receive queue and will be purged by socket destructor.
2183	 *
2184	 * Also we still have packets pending on receive queue and probably,
2185	 * our own packets waiting in device queues. sock_destroy will drain
2186	 * receive queue, but transmitted packets will delay socket destruction
2187	 * until the last reference will be released.
2188	 */
2189
2190	sock_orphan(sk);
2191
2192	xfrm_sk_free_policy(sk);
2193
2194	sk_refcnt_debug_release(sk);
2195	sock_put(sk);
2196}
2197EXPORT_SYMBOL(sk_common_release);
2198
2199static DEFINE_RWLOCK(proto_list_lock);
2200static LIST_HEAD(proto_list);
2201
2202#ifdef CONFIG_PROC_FS
2203#define PROTO_INUSE_NR	64	/* should be enough for the first time */
2204struct prot_inuse {
2205	int val[PROTO_INUSE_NR];
2206};
2207
2208static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2209
2210#ifdef CONFIG_NET_NS
2211void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2212{
2213	int cpu = smp_processor_id();
2214	per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2215}
2216EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2217
2218int sock_prot_inuse_get(struct net *net, struct proto *prot)
2219{
2220	int cpu, idx = prot->inuse_idx;
2221	int res = 0;
2222
2223	for_each_possible_cpu(cpu)
2224		res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2225
2226	return res >= 0 ? res : 0;
2227}
2228EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2229
2230static int __net_init sock_inuse_init_net(struct net *net)
2231{
2232	net->core.inuse = alloc_percpu(struct prot_inuse);
2233	return net->core.inuse ? 0 : -ENOMEM;
2234}
2235
2236static void __net_exit sock_inuse_exit_net(struct net *net)
2237{
2238	free_percpu(net->core.inuse);
2239}
2240
2241static struct pernet_operations net_inuse_ops = {
2242	.init = sock_inuse_init_net,
2243	.exit = sock_inuse_exit_net,
2244};
2245
2246static __init int net_inuse_init(void)
2247{
2248	if (register_pernet_subsys(&net_inuse_ops))
2249		panic("Cannot initialize net inuse counters");
2250
2251	return 0;
2252}
2253
2254core_initcall(net_inuse_init);
2255#else
2256static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2257
2258void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2259{
2260	__get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2261}
2262EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2263
2264int sock_prot_inuse_get(struct net *net, struct proto *prot)
2265{
2266	int cpu, idx = prot->inuse_idx;
2267	int res = 0;
2268
2269	for_each_possible_cpu(cpu)
2270		res += per_cpu(prot_inuse, cpu).val[idx];
2271
2272	return res >= 0 ? res : 0;
2273}
2274EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2275#endif
2276
2277static void assign_proto_idx(struct proto *prot)
2278{
2279	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2280
2281	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2282		printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2283		return;
2284	}
2285
2286	set_bit(prot->inuse_idx, proto_inuse_idx);
2287}
2288
2289static void release_proto_idx(struct proto *prot)
2290{
2291	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2292		clear_bit(prot->inuse_idx, proto_inuse_idx);
2293}
2294#else
2295static inline void assign_proto_idx(struct proto *prot)
2296{
2297}
2298
2299static inline void release_proto_idx(struct proto *prot)
2300{
2301}
2302#endif
2303
2304int proto_register(struct proto *prot, int alloc_slab)
2305{
2306	if (alloc_slab) {
2307		prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2308					SLAB_HWCACHE_ALIGN | prot->slab_flags,
2309					NULL);
2310
2311		if (prot->slab == NULL) {
2312			printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2313			       prot->name);
2314			goto out;
2315		}
2316
2317		if (prot->rsk_prot != NULL) {
2318			prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2319			if (prot->rsk_prot->slab_name == NULL)
2320				goto out_free_sock_slab;
2321
2322			prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2323								 prot->rsk_prot->obj_size, 0,
2324								 SLAB_HWCACHE_ALIGN, NULL);
2325
2326			if (prot->rsk_prot->slab == NULL) {
2327				printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2328				       prot->name);
2329				goto out_free_request_sock_slab_name;
2330			}
2331		}
2332
2333		if (prot->twsk_prot != NULL) {
2334			prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2335
2336			if (prot->twsk_prot->twsk_slab_name == NULL)
2337				goto out_free_request_sock_slab;
2338
2339			prot->twsk_prot->twsk_slab =
2340				kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2341						  prot->twsk_prot->twsk_obj_size,
2342						  0,
2343						  SLAB_HWCACHE_ALIGN |
2344							prot->slab_flags,
2345						  NULL);
2346			if (prot->twsk_prot->twsk_slab == NULL)
2347				goto out_free_timewait_sock_slab_name;
2348		}
2349	}
2350
2351	write_lock(&proto_list_lock);
2352	list_add(&prot->node, &proto_list);
2353	assign_proto_idx(prot);
2354	write_unlock(&proto_list_lock);
2355	return 0;
2356
2357out_free_timewait_sock_slab_name:
2358	kfree(prot->twsk_prot->twsk_slab_name);
2359out_free_request_sock_slab:
2360	if (prot->rsk_prot && prot->rsk_prot->slab) {
2361		kmem_cache_destroy(prot->rsk_prot->slab);
2362		prot->rsk_prot->slab = NULL;
2363	}
2364out_free_request_sock_slab_name:
2365	if (prot->rsk_prot)
2366		kfree(prot->rsk_prot->slab_name);
2367out_free_sock_slab:
2368	kmem_cache_destroy(prot->slab);
2369	prot->slab = NULL;
2370out:
2371	return -ENOBUFS;
2372}
2373EXPORT_SYMBOL(proto_register);
2374
2375void proto_unregister(struct proto *prot)
2376{
2377	write_lock(&proto_list_lock);
2378	release_proto_idx(prot);
2379	list_del(&prot->node);
2380	write_unlock(&proto_list_lock);
2381
2382	if (prot->slab != NULL) {
2383		kmem_cache_destroy(prot->slab);
2384		prot->slab = NULL;
2385	}
2386
2387	if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2388		kmem_cache_destroy(prot->rsk_prot->slab);
2389		kfree(prot->rsk_prot->slab_name);
2390		prot->rsk_prot->slab = NULL;
2391	}
2392
2393	if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2394		kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2395		kfree(prot->twsk_prot->twsk_slab_name);
2396		prot->twsk_prot->twsk_slab = NULL;
2397	}
2398}
2399EXPORT_SYMBOL(proto_unregister);
2400
2401#ifdef CONFIG_PROC_FS
2402static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2403	__acquires(proto_list_lock)
2404{
2405	read_lock(&proto_list_lock);
2406	return seq_list_start_head(&proto_list, *pos);
2407}
2408
2409static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2410{
2411	return seq_list_next(v, &proto_list, pos);
2412}
2413
2414static void proto_seq_stop(struct seq_file *seq, void *v)
2415	__releases(proto_list_lock)
2416{
2417	read_unlock(&proto_list_lock);
2418}
2419
2420static char proto_method_implemented(const void *method)
2421{
2422	return method == NULL ? 'n' : 'y';
2423}
2424
2425static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2426{
2427	seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2428			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2429		   proto->name,
2430		   proto->obj_size,
2431		   sock_prot_inuse_get(seq_file_net(seq), proto),
2432		   proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2433		   proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2434		   proto->max_header,
2435		   proto->slab == NULL ? "no" : "yes",
2436		   module_name(proto->owner),
2437		   proto_method_implemented(proto->close),
2438		   proto_method_implemented(proto->connect),
2439		   proto_method_implemented(proto->disconnect),
2440		   proto_method_implemented(proto->accept),
2441		   proto_method_implemented(proto->ioctl),
2442		   proto_method_implemented(proto->init),
2443		   proto_method_implemented(proto->destroy),
2444		   proto_method_implemented(proto->shutdown),
2445		   proto_method_implemented(proto->setsockopt),
2446		   proto_method_implemented(proto->getsockopt),
2447		   proto_method_implemented(proto->sendmsg),
2448		   proto_method_implemented(proto->recvmsg),
2449		   proto_method_implemented(proto->sendpage),
2450		   proto_method_implemented(proto->bind),
2451		   proto_method_implemented(proto->backlog_rcv),
2452		   proto_method_implemented(proto->hash),
2453		   proto_method_implemented(proto->unhash),
2454		   proto_method_implemented(proto->get_port),
2455		   proto_method_implemented(proto->enter_memory_pressure));
2456}
2457
2458static int proto_seq_show(struct seq_file *seq, void *v)
2459{
2460	if (v == &proto_list)
2461		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2462			   "protocol",
2463			   "size",
2464			   "sockets",
2465			   "memory",
2466			   "press",
2467			   "maxhdr",
2468			   "slab",
2469			   "module",
2470			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2471	else
2472		proto_seq_printf(seq, list_entry(v, struct proto, node));
2473	return 0;
2474}
2475
2476static const struct seq_operations proto_seq_ops = {
2477	.start  = proto_seq_start,
2478	.next   = proto_seq_next,
2479	.stop   = proto_seq_stop,
2480	.show   = proto_seq_show,
2481};
2482
2483static int proto_seq_open(struct inode *inode, struct file *file)
2484{
2485	return seq_open_net(inode, file, &proto_seq_ops,
2486			    sizeof(struct seq_net_private));
2487}
2488
2489static const struct file_operations proto_seq_fops = {
2490	.owner		= THIS_MODULE,
2491	.open		= proto_seq_open,
2492	.read		= seq_read,
2493	.llseek		= seq_lseek,
2494	.release	= seq_release_net,
2495};
2496
2497static __net_init int proto_init_net(struct net *net)
2498{
2499	if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2500		return -ENOMEM;
2501
2502	return 0;
2503}
2504
2505static __net_exit void proto_exit_net(struct net *net)
2506{
2507	proc_net_remove(net, "protocols");
2508}
2509
2510
2511static __net_initdata struct pernet_operations proto_net_ops = {
2512	.init = proto_init_net,
2513	.exit = proto_exit_net,
2514};
2515
2516static int __init proto_init(void)
2517{
2518	return register_pernet_subsys(&proto_net_ops);
2519}
2520
2521subsys_initcall(proto_init);
2522
2523#endif /* PROC_FS */
2524