sched: batch sleeper bonus
[usb.git] / net / core / sock.c
blobcfed7d42c4850525fffb56f88a65ba9f011ba9f7
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.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
12 * Authors: Ross Biro
13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Alan Cox, <A.Cox@swansea.ac.uk>
17 * Fixes:
18 * Alan Cox : Numerous verify_area() problems
19 * Alan Cox : Connecting on a connecting socket
20 * now returns an error for tcp.
21 * Alan Cox : sock->protocol is set correctly.
22 * and is not sometimes left as 0.
23 * Alan Cox : connect handles icmp errors on a
24 * connect properly. Unfortunately there
25 * is a restart syscall nasty there. I
26 * can't match BSD without hacking the C
27 * library. Ideas urgently sought!
28 * Alan Cox : Disallow bind() to addresses that are
29 * not ours - especially broadcast ones!!
30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
32 * instead they leave that for the DESTROY timer.
33 * Alan Cox : Clean up error flag in accept
34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
35 * was buggy. Put a remove_sock() in the handler
36 * for memory when we hit 0. Also altered the timer
37 * code. The ACK stuff can wait and needs major
38 * TCP layer surgery.
39 * Alan Cox : Fixed TCP ack bug, removed remove sock
40 * and fixed timer/inet_bh race.
41 * Alan Cox : Added zapped flag for TCP
42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
47 * Rick Sladkey : Relaxed UDP rules for matching packets.
48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
49 * Pauline Middelink : identd support
50 * Alan Cox : Fixed connect() taking signals I think.
51 * Alan Cox : SO_LINGER supported
52 * Alan Cox : Error reporting fixes
53 * Anonymous : inet_create tidied up (sk->reuse setting)
54 * Alan Cox : inet sockets don't set sk->type!
55 * Alan Cox : Split socket option code
56 * Alan Cox : Callbacks
57 * Alan Cox : Nagle flag for Charles & Johannes stuff
58 * Alex : Removed restriction on inet fioctl
59 * Alan Cox : Splitting INET from NET core
60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
62 * Alan Cox : Split IP from generic code
63 * Alan Cox : New kfree_skbmem()
64 * Alan Cox : Make SO_DEBUG superuser only.
65 * Alan Cox : Allow anyone to clear SO_DEBUG
66 * (compatibility fix)
67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
68 * Alan Cox : Allocator for a socket is settable.
69 * Alan Cox : SO_ERROR includes soft errors.
70 * Alan Cox : Allow NULL arguments on some SO_ opts
71 * Alan Cox : Generic socket allocation to make hooks
72 * easier (suggested by Craig Metz).
73 * Michael Pall : SO_ERROR returns positive errno again
74 * Steve Whitehouse: Added default destructor to free
75 * protocol private data.
76 * Steve Whitehouse: Added various other default routines
77 * common to several socket families.
78 * Chris Evans : Call suser() check last on F_SETOWN
79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
81 * Andi Kleen : Fix write_space callback
82 * Chris Evans : Security fixes - signedness again
83 * Arnaldo C. Melo : cleanups, use skb_queue_purge
85 * To Fix:
88 * This program is free software; you can redistribute it and/or
89 * modify it under the terms of the GNU General Public License
90 * as published by the Free Software Foundation; either version
91 * 2 of the License, or (at your option) any later version.
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
98 #include <linux/in.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
116 #include <asm/uaccess.h>
117 #include <asm/system.h>
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/request_sock.h>
123 #include <net/sock.h>
124 #include <net/xfrm.h>
125 #include <linux/ipsec.h>
127 #include <linux/filter.h>
129 #ifdef CONFIG_INET
130 #include <net/tcp.h>
131 #endif
134 * Each address family might have different locking rules, so we have
135 * one slock key per address family:
137 static struct lock_class_key af_family_keys[AF_MAX];
138 static struct lock_class_key af_family_slock_keys[AF_MAX];
140 #ifdef CONFIG_DEBUG_LOCK_ALLOC
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):
146 static const char *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-21" , "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-29" ,
157 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
158 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
160 static const char *af_family_slock_key_strings[AF_MAX+1] = {
161 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
162 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
163 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
164 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
165 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
166 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
167 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
168 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
169 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
170 "slock-27" , "slock-28" , "slock-29" ,
171 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
172 "slock-AF_RXRPC" , "slock-AF_MAX"
174 static const char *af_family_clock_key_strings[AF_MAX+1] = {
175 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
176 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
177 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
178 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
179 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
180 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
181 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
182 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
183 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
184 "clock-27" , "clock-28" , "clock-29" ,
185 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
186 "clock-AF_RXRPC" , "clock-AF_MAX"
188 #endif
191 * sk_callback_lock locking rules are per-address-family,
192 * so split the lock classes by using a per-AF key:
194 static struct lock_class_key af_callback_keys[AF_MAX];
196 /* Take into consideration the size of the struct sk_buff overhead in the
197 * determination of these values, since that is non-constant across
198 * platforms. This makes socket queueing behavior and performance
199 * not depend upon such differences.
201 #define _SK_MEM_PACKETS 256
202 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
203 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
204 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
206 /* Run time adjustable parameters. */
207 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
208 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
209 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
210 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
212 /* Maximal space eaten by iovec or ancilliary data plus some space */
213 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
215 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
217 struct timeval tv;
219 if (optlen < sizeof(tv))
220 return -EINVAL;
221 if (copy_from_user(&tv, optval, sizeof(tv)))
222 return -EFAULT;
223 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
224 return -EDOM;
226 if (tv.tv_sec < 0) {
227 static int warned __read_mostly;
229 *timeo_p = 0;
230 if (warned < 10 && net_ratelimit())
231 warned++;
232 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
233 "tries to set negative timeout\n",
234 current->comm, current->pid);
235 return 0;
237 *timeo_p = MAX_SCHEDULE_TIMEOUT;
238 if (tv.tv_sec == 0 && tv.tv_usec == 0)
239 return 0;
240 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
241 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
242 return 0;
245 static void sock_warn_obsolete_bsdism(const char *name)
247 static int warned;
248 static char warncomm[TASK_COMM_LEN];
249 if (strcmp(warncomm, current->comm) && warned < 5) {
250 strcpy(warncomm, current->comm);
251 printk(KERN_WARNING "process `%s' is using obsolete "
252 "%s SO_BSDCOMPAT\n", warncomm, name);
253 warned++;
257 static void sock_disable_timestamp(struct sock *sk)
259 if (sock_flag(sk, SOCK_TIMESTAMP)) {
260 sock_reset_flag(sk, SOCK_TIMESTAMP);
261 net_disable_timestamp();
266 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
268 int err = 0;
269 int skb_len;
271 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
272 number of warnings when compiling with -W --ANK
274 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
275 (unsigned)sk->sk_rcvbuf) {
276 err = -ENOMEM;
277 goto out;
280 err = sk_filter(sk, skb);
281 if (err)
282 goto out;
284 skb->dev = NULL;
285 skb_set_owner_r(skb, sk);
287 /* Cache the SKB length before we tack it onto the receive
288 * queue. Once it is added it no longer belongs to us and
289 * may be freed by other threads of control pulling packets
290 * from the queue.
292 skb_len = skb->len;
294 skb_queue_tail(&sk->sk_receive_queue, skb);
296 if (!sock_flag(sk, SOCK_DEAD))
297 sk->sk_data_ready(sk, skb_len);
298 out:
299 return err;
301 EXPORT_SYMBOL(sock_queue_rcv_skb);
303 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
305 int rc = NET_RX_SUCCESS;
307 if (sk_filter(sk, skb))
308 goto discard_and_relse;
310 skb->dev = NULL;
312 if (nested)
313 bh_lock_sock_nested(sk);
314 else
315 bh_lock_sock(sk);
316 if (!sock_owned_by_user(sk)) {
318 * trylock + unlock semantics:
320 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
322 rc = sk->sk_backlog_rcv(sk, skb);
324 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
325 } else
326 sk_add_backlog(sk, skb);
327 bh_unlock_sock(sk);
328 out:
329 sock_put(sk);
330 return rc;
331 discard_and_relse:
332 kfree_skb(skb);
333 goto out;
335 EXPORT_SYMBOL(sk_receive_skb);
337 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
339 struct dst_entry *dst = sk->sk_dst_cache;
341 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
342 sk->sk_dst_cache = NULL;
343 dst_release(dst);
344 return NULL;
347 return dst;
349 EXPORT_SYMBOL(__sk_dst_check);
351 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
353 struct dst_entry *dst = sk_dst_get(sk);
355 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
356 sk_dst_reset(sk);
357 dst_release(dst);
358 return NULL;
361 return dst;
363 EXPORT_SYMBOL(sk_dst_check);
366 * This is meant for all protocols to use and covers goings on
367 * at the socket level. Everything here is generic.
370 int sock_setsockopt(struct socket *sock, int level, int optname,
371 char __user *optval, int optlen)
373 struct sock *sk=sock->sk;
374 struct sk_filter *filter;
375 int val;
376 int valbool;
377 struct linger ling;
378 int ret = 0;
381 * Options without arguments
384 #ifdef SO_DONTLINGER /* Compatibility item... */
385 if (optname == SO_DONTLINGER) {
386 lock_sock(sk);
387 sock_reset_flag(sk, SOCK_LINGER);
388 release_sock(sk);
389 return 0;
391 #endif
393 if (optlen < sizeof(int))
394 return -EINVAL;
396 if (get_user(val, (int __user *)optval))
397 return -EFAULT;
399 valbool = val?1:0;
401 lock_sock(sk);
403 switch(optname) {
404 case SO_DEBUG:
405 if (val && !capable(CAP_NET_ADMIN)) {
406 ret = -EACCES;
408 else if (valbool)
409 sock_set_flag(sk, SOCK_DBG);
410 else
411 sock_reset_flag(sk, SOCK_DBG);
412 break;
413 case SO_REUSEADDR:
414 sk->sk_reuse = valbool;
415 break;
416 case SO_TYPE:
417 case SO_ERROR:
418 ret = -ENOPROTOOPT;
419 break;
420 case SO_DONTROUTE:
421 if (valbool)
422 sock_set_flag(sk, SOCK_LOCALROUTE);
423 else
424 sock_reset_flag(sk, SOCK_LOCALROUTE);
425 break;
426 case SO_BROADCAST:
427 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
428 break;
429 case SO_SNDBUF:
430 /* Don't error on this BSD doesn't and if you think
431 about it this is right. Otherwise apps have to
432 play 'guess the biggest size' games. RCVBUF/SNDBUF
433 are treated in BSD as hints */
435 if (val > sysctl_wmem_max)
436 val = sysctl_wmem_max;
437 set_sndbuf:
438 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
439 if ((val * 2) < SOCK_MIN_SNDBUF)
440 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
441 else
442 sk->sk_sndbuf = val * 2;
445 * Wake up sending tasks if we
446 * upped the value.
448 sk->sk_write_space(sk);
449 break;
451 case SO_SNDBUFFORCE:
452 if (!capable(CAP_NET_ADMIN)) {
453 ret = -EPERM;
454 break;
456 goto set_sndbuf;
458 case SO_RCVBUF:
459 /* Don't error on this BSD doesn't and if you think
460 about it this is right. Otherwise apps have to
461 play 'guess the biggest size' games. RCVBUF/SNDBUF
462 are treated in BSD as hints */
464 if (val > sysctl_rmem_max)
465 val = sysctl_rmem_max;
466 set_rcvbuf:
467 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
469 * We double it on the way in to account for
470 * "struct sk_buff" etc. overhead. Applications
471 * assume that the SO_RCVBUF setting they make will
472 * allow that much actual data to be received on that
473 * socket.
475 * Applications are unaware that "struct sk_buff" and
476 * other overheads allocate from the receive buffer
477 * during socket buffer allocation.
479 * And after considering the possible alternatives,
480 * returning the value we actually used in getsockopt
481 * is the most desirable behavior.
483 if ((val * 2) < SOCK_MIN_RCVBUF)
484 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
485 else
486 sk->sk_rcvbuf = val * 2;
487 break;
489 case SO_RCVBUFFORCE:
490 if (!capable(CAP_NET_ADMIN)) {
491 ret = -EPERM;
492 break;
494 goto set_rcvbuf;
496 case SO_KEEPALIVE:
497 #ifdef CONFIG_INET
498 if (sk->sk_protocol == IPPROTO_TCP)
499 tcp_set_keepalive(sk, valbool);
500 #endif
501 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
502 break;
504 case SO_OOBINLINE:
505 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
506 break;
508 case SO_NO_CHECK:
509 sk->sk_no_check = valbool;
510 break;
512 case SO_PRIORITY:
513 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
514 sk->sk_priority = val;
515 else
516 ret = -EPERM;
517 break;
519 case SO_LINGER:
520 if (optlen < sizeof(ling)) {
521 ret = -EINVAL; /* 1003.1g */
522 break;
524 if (copy_from_user(&ling,optval,sizeof(ling))) {
525 ret = -EFAULT;
526 break;
528 if (!ling.l_onoff)
529 sock_reset_flag(sk, SOCK_LINGER);
530 else {
531 #if (BITS_PER_LONG == 32)
532 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
533 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
534 else
535 #endif
536 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
537 sock_set_flag(sk, SOCK_LINGER);
539 break;
541 case SO_BSDCOMPAT:
542 sock_warn_obsolete_bsdism("setsockopt");
543 break;
545 case SO_PASSCRED:
546 if (valbool)
547 set_bit(SOCK_PASSCRED, &sock->flags);
548 else
549 clear_bit(SOCK_PASSCRED, &sock->flags);
550 break;
552 case SO_TIMESTAMP:
553 case SO_TIMESTAMPNS:
554 if (valbool) {
555 if (optname == SO_TIMESTAMP)
556 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
557 else
558 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
559 sock_set_flag(sk, SOCK_RCVTSTAMP);
560 sock_enable_timestamp(sk);
561 } else {
562 sock_reset_flag(sk, SOCK_RCVTSTAMP);
563 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
565 break;
567 case SO_RCVLOWAT:
568 if (val < 0)
569 val = INT_MAX;
570 sk->sk_rcvlowat = val ? : 1;
571 break;
573 case SO_RCVTIMEO:
574 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
575 break;
577 case SO_SNDTIMEO:
578 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
579 break;
581 #ifdef CONFIG_NETDEVICES
582 case SO_BINDTODEVICE:
584 char devname[IFNAMSIZ];
586 /* Sorry... */
587 if (!capable(CAP_NET_RAW)) {
588 ret = -EPERM;
589 break;
592 /* Bind this socket to a particular device like "eth0",
593 * as specified in the passed interface name. If the
594 * name is "" or the option length is zero the socket
595 * is not bound.
598 if (!valbool) {
599 sk->sk_bound_dev_if = 0;
600 } else {
601 if (optlen > IFNAMSIZ - 1)
602 optlen = IFNAMSIZ - 1;
603 memset(devname, 0, sizeof(devname));
604 if (copy_from_user(devname, optval, optlen)) {
605 ret = -EFAULT;
606 break;
609 /* Remove any cached route for this socket. */
610 sk_dst_reset(sk);
612 if (devname[0] == '\0') {
613 sk->sk_bound_dev_if = 0;
614 } else {
615 struct net_device *dev = dev_get_by_name(devname);
616 if (!dev) {
617 ret = -ENODEV;
618 break;
620 sk->sk_bound_dev_if = dev->ifindex;
621 dev_put(dev);
624 break;
626 #endif
629 case SO_ATTACH_FILTER:
630 ret = -EINVAL;
631 if (optlen == sizeof(struct sock_fprog)) {
632 struct sock_fprog fprog;
634 ret = -EFAULT;
635 if (copy_from_user(&fprog, optval, sizeof(fprog)))
636 break;
638 ret = sk_attach_filter(&fprog, sk);
640 break;
642 case SO_DETACH_FILTER:
643 rcu_read_lock_bh();
644 filter = rcu_dereference(sk->sk_filter);
645 if (filter) {
646 rcu_assign_pointer(sk->sk_filter, NULL);
647 sk_filter_release(sk, filter);
648 rcu_read_unlock_bh();
649 break;
651 rcu_read_unlock_bh();
652 ret = -ENONET;
653 break;
655 case SO_PASSSEC:
656 if (valbool)
657 set_bit(SOCK_PASSSEC, &sock->flags);
658 else
659 clear_bit(SOCK_PASSSEC, &sock->flags);
660 break;
662 /* We implement the SO_SNDLOWAT etc to
663 not be settable (1003.1g 5.3) */
664 default:
665 ret = -ENOPROTOOPT;
666 break;
668 release_sock(sk);
669 return ret;
673 int sock_getsockopt(struct socket *sock, int level, int optname,
674 char __user *optval, int __user *optlen)
676 struct sock *sk = sock->sk;
678 union {
679 int val;
680 struct linger ling;
681 struct timeval tm;
682 } v;
684 unsigned int lv = sizeof(int);
685 int len;
687 if (get_user(len, optlen))
688 return -EFAULT;
689 if (len < 0)
690 return -EINVAL;
692 switch(optname) {
693 case SO_DEBUG:
694 v.val = sock_flag(sk, SOCK_DBG);
695 break;
697 case SO_DONTROUTE:
698 v.val = sock_flag(sk, SOCK_LOCALROUTE);
699 break;
701 case SO_BROADCAST:
702 v.val = !!sock_flag(sk, SOCK_BROADCAST);
703 break;
705 case SO_SNDBUF:
706 v.val = sk->sk_sndbuf;
707 break;
709 case SO_RCVBUF:
710 v.val = sk->sk_rcvbuf;
711 break;
713 case SO_REUSEADDR:
714 v.val = sk->sk_reuse;
715 break;
717 case SO_KEEPALIVE:
718 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
719 break;
721 case SO_TYPE:
722 v.val = sk->sk_type;
723 break;
725 case SO_ERROR:
726 v.val = -sock_error(sk);
727 if (v.val==0)
728 v.val = xchg(&sk->sk_err_soft, 0);
729 break;
731 case SO_OOBINLINE:
732 v.val = !!sock_flag(sk, SOCK_URGINLINE);
733 break;
735 case SO_NO_CHECK:
736 v.val = sk->sk_no_check;
737 break;
739 case SO_PRIORITY:
740 v.val = sk->sk_priority;
741 break;
743 case SO_LINGER:
744 lv = sizeof(v.ling);
745 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
746 v.ling.l_linger = sk->sk_lingertime / HZ;
747 break;
749 case SO_BSDCOMPAT:
750 sock_warn_obsolete_bsdism("getsockopt");
751 break;
753 case SO_TIMESTAMP:
754 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
755 !sock_flag(sk, SOCK_RCVTSTAMPNS);
756 break;
758 case SO_TIMESTAMPNS:
759 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
760 break;
762 case SO_RCVTIMEO:
763 lv=sizeof(struct timeval);
764 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
765 v.tm.tv_sec = 0;
766 v.tm.tv_usec = 0;
767 } else {
768 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
769 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
771 break;
773 case SO_SNDTIMEO:
774 lv=sizeof(struct timeval);
775 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
776 v.tm.tv_sec = 0;
777 v.tm.tv_usec = 0;
778 } else {
779 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
780 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
782 break;
784 case SO_RCVLOWAT:
785 v.val = sk->sk_rcvlowat;
786 break;
788 case SO_SNDLOWAT:
789 v.val=1;
790 break;
792 case SO_PASSCRED:
793 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
794 break;
796 case SO_PEERCRED:
797 if (len > sizeof(sk->sk_peercred))
798 len = sizeof(sk->sk_peercred);
799 if (copy_to_user(optval, &sk->sk_peercred, len))
800 return -EFAULT;
801 goto lenout;
803 case SO_PEERNAME:
805 char address[128];
807 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
808 return -ENOTCONN;
809 if (lv < len)
810 return -EINVAL;
811 if (copy_to_user(optval, address, len))
812 return -EFAULT;
813 goto lenout;
816 /* Dubious BSD thing... Probably nobody even uses it, but
817 * the UNIX standard wants it for whatever reason... -DaveM
819 case SO_ACCEPTCONN:
820 v.val = sk->sk_state == TCP_LISTEN;
821 break;
823 case SO_PASSSEC:
824 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
825 break;
827 case SO_PEERSEC:
828 return security_socket_getpeersec_stream(sock, optval, optlen, len);
830 default:
831 return -ENOPROTOOPT;
834 if (len > lv)
835 len = lv;
836 if (copy_to_user(optval, &v, len))
837 return -EFAULT;
838 lenout:
839 if (put_user(len, optlen))
840 return -EFAULT;
841 return 0;
845 * Initialize an sk_lock.
847 * (We also register the sk_lock with the lock validator.)
849 static inline void sock_lock_init(struct sock *sk)
851 sock_lock_init_class_and_name(sk,
852 af_family_slock_key_strings[sk->sk_family],
853 af_family_slock_keys + sk->sk_family,
854 af_family_key_strings[sk->sk_family],
855 af_family_keys + sk->sk_family);
859 * sk_alloc - All socket objects are allocated here
860 * @family: protocol family
861 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
862 * @prot: struct proto associated with this new sock instance
863 * @zero_it: if we should zero the newly allocated sock
865 struct sock *sk_alloc(int family, gfp_t priority,
866 struct proto *prot, int zero_it)
868 struct sock *sk = NULL;
869 struct kmem_cache *slab = prot->slab;
871 if (slab != NULL)
872 sk = kmem_cache_alloc(slab, priority);
873 else
874 sk = kmalloc(prot->obj_size, priority);
876 if (sk) {
877 if (zero_it) {
878 memset(sk, 0, prot->obj_size);
879 sk->sk_family = family;
881 * See comment in struct sock definition to understand
882 * why we need sk_prot_creator -acme
884 sk->sk_prot = sk->sk_prot_creator = prot;
885 sock_lock_init(sk);
888 if (security_sk_alloc(sk, family, priority))
889 goto out_free;
891 if (!try_module_get(prot->owner))
892 goto out_free;
894 return sk;
896 out_free:
897 if (slab != NULL)
898 kmem_cache_free(slab, sk);
899 else
900 kfree(sk);
901 return NULL;
904 void sk_free(struct sock *sk)
906 struct sk_filter *filter;
907 struct module *owner = sk->sk_prot_creator->owner;
909 if (sk->sk_destruct)
910 sk->sk_destruct(sk);
912 filter = rcu_dereference(sk->sk_filter);
913 if (filter) {
914 sk_filter_release(sk, filter);
915 rcu_assign_pointer(sk->sk_filter, NULL);
918 sock_disable_timestamp(sk);
920 if (atomic_read(&sk->sk_omem_alloc))
921 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
922 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
924 security_sk_free(sk);
925 if (sk->sk_prot_creator->slab != NULL)
926 kmem_cache_free(sk->sk_prot_creator->slab, sk);
927 else
928 kfree(sk);
929 module_put(owner);
932 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
934 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
936 if (newsk != NULL) {
937 struct sk_filter *filter;
939 sock_copy(newsk, sk);
941 /* SANITY */
942 sk_node_init(&newsk->sk_node);
943 sock_lock_init(newsk);
944 bh_lock_sock(newsk);
945 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
947 atomic_set(&newsk->sk_rmem_alloc, 0);
948 atomic_set(&newsk->sk_wmem_alloc, 0);
949 atomic_set(&newsk->sk_omem_alloc, 0);
950 skb_queue_head_init(&newsk->sk_receive_queue);
951 skb_queue_head_init(&newsk->sk_write_queue);
952 #ifdef CONFIG_NET_DMA
953 skb_queue_head_init(&newsk->sk_async_wait_queue);
954 #endif
956 rwlock_init(&newsk->sk_dst_lock);
957 rwlock_init(&newsk->sk_callback_lock);
958 lockdep_set_class_and_name(&newsk->sk_callback_lock,
959 af_callback_keys + newsk->sk_family,
960 af_family_clock_key_strings[newsk->sk_family]);
962 newsk->sk_dst_cache = NULL;
963 newsk->sk_wmem_queued = 0;
964 newsk->sk_forward_alloc = 0;
965 newsk->sk_send_head = NULL;
966 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
968 sock_reset_flag(newsk, SOCK_DONE);
969 skb_queue_head_init(&newsk->sk_error_queue);
971 filter = newsk->sk_filter;
972 if (filter != NULL)
973 sk_filter_charge(newsk, filter);
975 if (unlikely(xfrm_sk_clone_policy(newsk))) {
976 /* It is still raw copy of parent, so invalidate
977 * destructor and make plain sk_free() */
978 newsk->sk_destruct = NULL;
979 sk_free(newsk);
980 newsk = NULL;
981 goto out;
984 newsk->sk_err = 0;
985 newsk->sk_priority = 0;
986 atomic_set(&newsk->sk_refcnt, 2);
989 * Increment the counter in the same struct proto as the master
990 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
991 * is the same as sk->sk_prot->socks, as this field was copied
992 * with memcpy).
994 * This _changes_ the previous behaviour, where
995 * tcp_create_openreq_child always was incrementing the
996 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
997 * to be taken into account in all callers. -acme
999 sk_refcnt_debug_inc(newsk);
1000 newsk->sk_socket = NULL;
1001 newsk->sk_sleep = NULL;
1003 if (newsk->sk_prot->sockets_allocated)
1004 atomic_inc(newsk->sk_prot->sockets_allocated);
1006 out:
1007 return newsk;
1010 EXPORT_SYMBOL_GPL(sk_clone);
1012 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1014 __sk_dst_set(sk, dst);
1015 sk->sk_route_caps = dst->dev->features;
1016 if (sk->sk_route_caps & NETIF_F_GSO)
1017 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1018 if (sk_can_gso(sk)) {
1019 if (dst->header_len)
1020 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1021 else
1022 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1025 EXPORT_SYMBOL_GPL(sk_setup_caps);
1027 void __init sk_init(void)
1029 if (num_physpages <= 4096) {
1030 sysctl_wmem_max = 32767;
1031 sysctl_rmem_max = 32767;
1032 sysctl_wmem_default = 32767;
1033 sysctl_rmem_default = 32767;
1034 } else if (num_physpages >= 131072) {
1035 sysctl_wmem_max = 131071;
1036 sysctl_rmem_max = 131071;
1041 * Simple resource managers for sockets.
1046 * Write buffer destructor automatically called from kfree_skb.
1048 void sock_wfree(struct sk_buff *skb)
1050 struct sock *sk = skb->sk;
1052 /* In case it might be waiting for more memory. */
1053 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1054 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1055 sk->sk_write_space(sk);
1056 sock_put(sk);
1060 * Read buffer destructor automatically called from kfree_skb.
1062 void sock_rfree(struct sk_buff *skb)
1064 struct sock *sk = skb->sk;
1066 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1070 int sock_i_uid(struct sock *sk)
1072 int uid;
1074 read_lock(&sk->sk_callback_lock);
1075 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1076 read_unlock(&sk->sk_callback_lock);
1077 return uid;
1080 unsigned long sock_i_ino(struct sock *sk)
1082 unsigned long ino;
1084 read_lock(&sk->sk_callback_lock);
1085 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1086 read_unlock(&sk->sk_callback_lock);
1087 return ino;
1091 * Allocate a skb from the socket's send buffer.
1093 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1094 gfp_t priority)
1096 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1097 struct sk_buff * skb = alloc_skb(size, priority);
1098 if (skb) {
1099 skb_set_owner_w(skb, sk);
1100 return skb;
1103 return NULL;
1107 * Allocate a skb from the socket's receive buffer.
1109 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1110 gfp_t priority)
1112 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1113 struct sk_buff *skb = alloc_skb(size, priority);
1114 if (skb) {
1115 skb_set_owner_r(skb, sk);
1116 return skb;
1119 return NULL;
1123 * Allocate a memory block from the socket's option memory buffer.
1125 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1127 if ((unsigned)size <= sysctl_optmem_max &&
1128 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1129 void *mem;
1130 /* First do the add, to avoid the race if kmalloc
1131 * might sleep.
1133 atomic_add(size, &sk->sk_omem_alloc);
1134 mem = kmalloc(size, priority);
1135 if (mem)
1136 return mem;
1137 atomic_sub(size, &sk->sk_omem_alloc);
1139 return NULL;
1143 * Free an option memory block.
1145 void sock_kfree_s(struct sock *sk, void *mem, int size)
1147 kfree(mem);
1148 atomic_sub(size, &sk->sk_omem_alloc);
1151 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1152 I think, these locks should be removed for datagram sockets.
1154 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1156 DEFINE_WAIT(wait);
1158 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1159 for (;;) {
1160 if (!timeo)
1161 break;
1162 if (signal_pending(current))
1163 break;
1164 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1165 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1166 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1167 break;
1168 if (sk->sk_shutdown & SEND_SHUTDOWN)
1169 break;
1170 if (sk->sk_err)
1171 break;
1172 timeo = schedule_timeout(timeo);
1174 finish_wait(sk->sk_sleep, &wait);
1175 return timeo;
1180 * Generic send/receive buffer handlers
1183 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1184 unsigned long header_len,
1185 unsigned long data_len,
1186 int noblock, int *errcode)
1188 struct sk_buff *skb;
1189 gfp_t gfp_mask;
1190 long timeo;
1191 int err;
1193 gfp_mask = sk->sk_allocation;
1194 if (gfp_mask & __GFP_WAIT)
1195 gfp_mask |= __GFP_REPEAT;
1197 timeo = sock_sndtimeo(sk, noblock);
1198 while (1) {
1199 err = sock_error(sk);
1200 if (err != 0)
1201 goto failure;
1203 err = -EPIPE;
1204 if (sk->sk_shutdown & SEND_SHUTDOWN)
1205 goto failure;
1207 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1208 skb = alloc_skb(header_len, gfp_mask);
1209 if (skb) {
1210 int npages;
1211 int i;
1213 /* No pages, we're done... */
1214 if (!data_len)
1215 break;
1217 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1218 skb->truesize += data_len;
1219 skb_shinfo(skb)->nr_frags = npages;
1220 for (i = 0; i < npages; i++) {
1221 struct page *page;
1222 skb_frag_t *frag;
1224 page = alloc_pages(sk->sk_allocation, 0);
1225 if (!page) {
1226 err = -ENOBUFS;
1227 skb_shinfo(skb)->nr_frags = i;
1228 kfree_skb(skb);
1229 goto failure;
1232 frag = &skb_shinfo(skb)->frags[i];
1233 frag->page = page;
1234 frag->page_offset = 0;
1235 frag->size = (data_len >= PAGE_SIZE ?
1236 PAGE_SIZE :
1237 data_len);
1238 data_len -= PAGE_SIZE;
1241 /* Full success... */
1242 break;
1244 err = -ENOBUFS;
1245 goto failure;
1247 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1248 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1249 err = -EAGAIN;
1250 if (!timeo)
1251 goto failure;
1252 if (signal_pending(current))
1253 goto interrupted;
1254 timeo = sock_wait_for_wmem(sk, timeo);
1257 skb_set_owner_w(skb, sk);
1258 return skb;
1260 interrupted:
1261 err = sock_intr_errno(timeo);
1262 failure:
1263 *errcode = err;
1264 return NULL;
1267 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1268 int noblock, int *errcode)
1270 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1273 static void __lock_sock(struct sock *sk)
1275 DEFINE_WAIT(wait);
1277 for (;;) {
1278 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1279 TASK_UNINTERRUPTIBLE);
1280 spin_unlock_bh(&sk->sk_lock.slock);
1281 schedule();
1282 spin_lock_bh(&sk->sk_lock.slock);
1283 if (!sock_owned_by_user(sk))
1284 break;
1286 finish_wait(&sk->sk_lock.wq, &wait);
1289 static void __release_sock(struct sock *sk)
1291 struct sk_buff *skb = sk->sk_backlog.head;
1293 do {
1294 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1295 bh_unlock_sock(sk);
1297 do {
1298 struct sk_buff *next = skb->next;
1300 skb->next = NULL;
1301 sk->sk_backlog_rcv(sk, skb);
1304 * We are in process context here with softirqs
1305 * disabled, use cond_resched_softirq() to preempt.
1306 * This is safe to do because we've taken the backlog
1307 * queue private:
1309 cond_resched_softirq();
1311 skb = next;
1312 } while (skb != NULL);
1314 bh_lock_sock(sk);
1315 } while ((skb = sk->sk_backlog.head) != NULL);
1319 * sk_wait_data - wait for data to arrive at sk_receive_queue
1320 * @sk: sock to wait on
1321 * @timeo: for how long
1323 * Now socket state including sk->sk_err is changed only under lock,
1324 * hence we may omit checks after joining wait queue.
1325 * We check receive queue before schedule() only as optimization;
1326 * it is very likely that release_sock() added new data.
1328 int sk_wait_data(struct sock *sk, long *timeo)
1330 int rc;
1331 DEFINE_WAIT(wait);
1333 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1334 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1335 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1336 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1337 finish_wait(sk->sk_sleep, &wait);
1338 return rc;
1341 EXPORT_SYMBOL(sk_wait_data);
1344 * Set of default routines for initialising struct proto_ops when
1345 * the protocol does not support a particular function. In certain
1346 * cases where it makes no sense for a protocol to have a "do nothing"
1347 * function, some default processing is provided.
1350 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1352 return -EOPNOTSUPP;
1355 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1356 int len, int flags)
1358 return -EOPNOTSUPP;
1361 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1363 return -EOPNOTSUPP;
1366 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1368 return -EOPNOTSUPP;
1371 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1372 int *len, int peer)
1374 return -EOPNOTSUPP;
1377 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1379 return 0;
1382 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1384 return -EOPNOTSUPP;
1387 int sock_no_listen(struct socket *sock, int backlog)
1389 return -EOPNOTSUPP;
1392 int sock_no_shutdown(struct socket *sock, int how)
1394 return -EOPNOTSUPP;
1397 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1398 char __user *optval, int optlen)
1400 return -EOPNOTSUPP;
1403 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1404 char __user *optval, int __user *optlen)
1406 return -EOPNOTSUPP;
1409 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1410 size_t len)
1412 return -EOPNOTSUPP;
1415 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1416 size_t len, int flags)
1418 return -EOPNOTSUPP;
1421 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1423 /* Mirror missing mmap method error code */
1424 return -ENODEV;
1427 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1429 ssize_t res;
1430 struct msghdr msg = {.msg_flags = flags};
1431 struct kvec iov;
1432 char *kaddr = kmap(page);
1433 iov.iov_base = kaddr + offset;
1434 iov.iov_len = size;
1435 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1436 kunmap(page);
1437 return res;
1441 * Default Socket Callbacks
1444 static void sock_def_wakeup(struct sock *sk)
1446 read_lock(&sk->sk_callback_lock);
1447 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1448 wake_up_interruptible_all(sk->sk_sleep);
1449 read_unlock(&sk->sk_callback_lock);
1452 static void sock_def_error_report(struct sock *sk)
1454 read_lock(&sk->sk_callback_lock);
1455 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1456 wake_up_interruptible(sk->sk_sleep);
1457 sk_wake_async(sk,0,POLL_ERR);
1458 read_unlock(&sk->sk_callback_lock);
1461 static void sock_def_readable(struct sock *sk, int len)
1463 read_lock(&sk->sk_callback_lock);
1464 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1465 wake_up_interruptible(sk->sk_sleep);
1466 sk_wake_async(sk,1,POLL_IN);
1467 read_unlock(&sk->sk_callback_lock);
1470 static void sock_def_write_space(struct sock *sk)
1472 read_lock(&sk->sk_callback_lock);
1474 /* Do not wake up a writer until he can make "significant"
1475 * progress. --DaveM
1477 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1478 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1479 wake_up_interruptible(sk->sk_sleep);
1481 /* Should agree with poll, otherwise some programs break */
1482 if (sock_writeable(sk))
1483 sk_wake_async(sk, 2, POLL_OUT);
1486 read_unlock(&sk->sk_callback_lock);
1489 static void sock_def_destruct(struct sock *sk)
1491 kfree(sk->sk_protinfo);
1494 void sk_send_sigurg(struct sock *sk)
1496 if (sk->sk_socket && sk->sk_socket->file)
1497 if (send_sigurg(&sk->sk_socket->file->f_owner))
1498 sk_wake_async(sk, 3, POLL_PRI);
1501 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1502 unsigned long expires)
1504 if (!mod_timer(timer, expires))
1505 sock_hold(sk);
1508 EXPORT_SYMBOL(sk_reset_timer);
1510 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1512 if (timer_pending(timer) && del_timer(timer))
1513 __sock_put(sk);
1516 EXPORT_SYMBOL(sk_stop_timer);
1518 void sock_init_data(struct socket *sock, struct sock *sk)
1520 skb_queue_head_init(&sk->sk_receive_queue);
1521 skb_queue_head_init(&sk->sk_write_queue);
1522 skb_queue_head_init(&sk->sk_error_queue);
1523 #ifdef CONFIG_NET_DMA
1524 skb_queue_head_init(&sk->sk_async_wait_queue);
1525 #endif
1527 sk->sk_send_head = NULL;
1529 init_timer(&sk->sk_timer);
1531 sk->sk_allocation = GFP_KERNEL;
1532 sk->sk_rcvbuf = sysctl_rmem_default;
1533 sk->sk_sndbuf = sysctl_wmem_default;
1534 sk->sk_state = TCP_CLOSE;
1535 sk->sk_socket = sock;
1537 sock_set_flag(sk, SOCK_ZAPPED);
1539 if (sock) {
1540 sk->sk_type = sock->type;
1541 sk->sk_sleep = &sock->wait;
1542 sock->sk = sk;
1543 } else
1544 sk->sk_sleep = NULL;
1546 rwlock_init(&sk->sk_dst_lock);
1547 rwlock_init(&sk->sk_callback_lock);
1548 lockdep_set_class_and_name(&sk->sk_callback_lock,
1549 af_callback_keys + sk->sk_family,
1550 af_family_clock_key_strings[sk->sk_family]);
1552 sk->sk_state_change = sock_def_wakeup;
1553 sk->sk_data_ready = sock_def_readable;
1554 sk->sk_write_space = sock_def_write_space;
1555 sk->sk_error_report = sock_def_error_report;
1556 sk->sk_destruct = sock_def_destruct;
1558 sk->sk_sndmsg_page = NULL;
1559 sk->sk_sndmsg_off = 0;
1561 sk->sk_peercred.pid = 0;
1562 sk->sk_peercred.uid = -1;
1563 sk->sk_peercred.gid = -1;
1564 sk->sk_write_pending = 0;
1565 sk->sk_rcvlowat = 1;
1566 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1567 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1569 sk->sk_stamp = ktime_set(-1L, -1L);
1571 atomic_set(&sk->sk_refcnt, 1);
1574 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1576 might_sleep();
1577 spin_lock_bh(&sk->sk_lock.slock);
1578 if (sk->sk_lock.owner)
1579 __lock_sock(sk);
1580 sk->sk_lock.owner = (void *)1;
1581 spin_unlock(&sk->sk_lock.slock);
1583 * The sk_lock has mutex_lock() semantics here:
1585 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1586 local_bh_enable();
1589 EXPORT_SYMBOL(lock_sock_nested);
1591 void fastcall release_sock(struct sock *sk)
1594 * The sk_lock has mutex_unlock() semantics:
1596 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1598 spin_lock_bh(&sk->sk_lock.slock);
1599 if (sk->sk_backlog.tail)
1600 __release_sock(sk);
1601 sk->sk_lock.owner = NULL;
1602 if (waitqueue_active(&sk->sk_lock.wq))
1603 wake_up(&sk->sk_lock.wq);
1604 spin_unlock_bh(&sk->sk_lock.slock);
1606 EXPORT_SYMBOL(release_sock);
1608 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1610 struct timeval tv;
1611 if (!sock_flag(sk, SOCK_TIMESTAMP))
1612 sock_enable_timestamp(sk);
1613 tv = ktime_to_timeval(sk->sk_stamp);
1614 if (tv.tv_sec == -1)
1615 return -ENOENT;
1616 if (tv.tv_sec == 0) {
1617 sk->sk_stamp = ktime_get_real();
1618 tv = ktime_to_timeval(sk->sk_stamp);
1620 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1622 EXPORT_SYMBOL(sock_get_timestamp);
1624 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1626 struct timespec ts;
1627 if (!sock_flag(sk, SOCK_TIMESTAMP))
1628 sock_enable_timestamp(sk);
1629 ts = ktime_to_timespec(sk->sk_stamp);
1630 if (ts.tv_sec == -1)
1631 return -ENOENT;
1632 if (ts.tv_sec == 0) {
1633 sk->sk_stamp = ktime_get_real();
1634 ts = ktime_to_timespec(sk->sk_stamp);
1636 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1638 EXPORT_SYMBOL(sock_get_timestampns);
1640 void sock_enable_timestamp(struct sock *sk)
1642 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1643 sock_set_flag(sk, SOCK_TIMESTAMP);
1644 net_enable_timestamp();
1647 EXPORT_SYMBOL(sock_enable_timestamp);
1650 * Get a socket option on an socket.
1652 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1653 * asynchronous errors should be reported by getsockopt. We assume
1654 * this means if you specify SO_ERROR (otherwise whats the point of it).
1656 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1657 char __user *optval, int __user *optlen)
1659 struct sock *sk = sock->sk;
1661 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1664 EXPORT_SYMBOL(sock_common_getsockopt);
1666 #ifdef CONFIG_COMPAT
1667 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1668 char __user *optval, int __user *optlen)
1670 struct sock *sk = sock->sk;
1672 if (sk->sk_prot->compat_getsockopt != NULL)
1673 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1674 optval, optlen);
1675 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1677 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1678 #endif
1680 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1681 struct msghdr *msg, size_t size, int flags)
1683 struct sock *sk = sock->sk;
1684 int addr_len = 0;
1685 int err;
1687 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1688 flags & ~MSG_DONTWAIT, &addr_len);
1689 if (err >= 0)
1690 msg->msg_namelen = addr_len;
1691 return err;
1694 EXPORT_SYMBOL(sock_common_recvmsg);
1697 * Set socket options on an inet socket.
1699 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1700 char __user *optval, int optlen)
1702 struct sock *sk = sock->sk;
1704 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1707 EXPORT_SYMBOL(sock_common_setsockopt);
1709 #ifdef CONFIG_COMPAT
1710 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1711 char __user *optval, int optlen)
1713 struct sock *sk = sock->sk;
1715 if (sk->sk_prot->compat_setsockopt != NULL)
1716 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1717 optval, optlen);
1718 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1720 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1721 #endif
1723 void sk_common_release(struct sock *sk)
1725 if (sk->sk_prot->destroy)
1726 sk->sk_prot->destroy(sk);
1729 * Observation: when sock_common_release is called, processes have
1730 * no access to socket. But net still has.
1731 * Step one, detach it from networking:
1733 * A. Remove from hash tables.
1736 sk->sk_prot->unhash(sk);
1739 * In this point socket cannot receive new packets, but it is possible
1740 * that some packets are in flight because some CPU runs receiver and
1741 * did hash table lookup before we unhashed socket. They will achieve
1742 * receive queue and will be purged by socket destructor.
1744 * Also we still have packets pending on receive queue and probably,
1745 * our own packets waiting in device queues. sock_destroy will drain
1746 * receive queue, but transmitted packets will delay socket destruction
1747 * until the last reference will be released.
1750 sock_orphan(sk);
1752 xfrm_sk_free_policy(sk);
1754 sk_refcnt_debug_release(sk);
1755 sock_put(sk);
1758 EXPORT_SYMBOL(sk_common_release);
1760 static DEFINE_RWLOCK(proto_list_lock);
1761 static LIST_HEAD(proto_list);
1763 int proto_register(struct proto *prot, int alloc_slab)
1765 char *request_sock_slab_name = NULL;
1766 char *timewait_sock_slab_name;
1767 int rc = -ENOBUFS;
1769 if (alloc_slab) {
1770 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1771 SLAB_HWCACHE_ALIGN, NULL);
1773 if (prot->slab == NULL) {
1774 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1775 prot->name);
1776 goto out;
1779 if (prot->rsk_prot != NULL) {
1780 static const char mask[] = "request_sock_%s";
1782 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1783 if (request_sock_slab_name == NULL)
1784 goto out_free_sock_slab;
1786 sprintf(request_sock_slab_name, mask, prot->name);
1787 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1788 prot->rsk_prot->obj_size, 0,
1789 SLAB_HWCACHE_ALIGN, NULL);
1791 if (prot->rsk_prot->slab == NULL) {
1792 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1793 prot->name);
1794 goto out_free_request_sock_slab_name;
1798 if (prot->twsk_prot != NULL) {
1799 static const char mask[] = "tw_sock_%s";
1801 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1803 if (timewait_sock_slab_name == NULL)
1804 goto out_free_request_sock_slab;
1806 sprintf(timewait_sock_slab_name, mask, prot->name);
1807 prot->twsk_prot->twsk_slab =
1808 kmem_cache_create(timewait_sock_slab_name,
1809 prot->twsk_prot->twsk_obj_size,
1810 0, SLAB_HWCACHE_ALIGN,
1811 NULL);
1812 if (prot->twsk_prot->twsk_slab == NULL)
1813 goto out_free_timewait_sock_slab_name;
1817 write_lock(&proto_list_lock);
1818 list_add(&prot->node, &proto_list);
1819 write_unlock(&proto_list_lock);
1820 rc = 0;
1821 out:
1822 return rc;
1823 out_free_timewait_sock_slab_name:
1824 kfree(timewait_sock_slab_name);
1825 out_free_request_sock_slab:
1826 if (prot->rsk_prot && prot->rsk_prot->slab) {
1827 kmem_cache_destroy(prot->rsk_prot->slab);
1828 prot->rsk_prot->slab = NULL;
1830 out_free_request_sock_slab_name:
1831 kfree(request_sock_slab_name);
1832 out_free_sock_slab:
1833 kmem_cache_destroy(prot->slab);
1834 prot->slab = NULL;
1835 goto out;
1838 EXPORT_SYMBOL(proto_register);
1840 void proto_unregister(struct proto *prot)
1842 write_lock(&proto_list_lock);
1843 list_del(&prot->node);
1844 write_unlock(&proto_list_lock);
1846 if (prot->slab != NULL) {
1847 kmem_cache_destroy(prot->slab);
1848 prot->slab = NULL;
1851 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1852 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1854 kmem_cache_destroy(prot->rsk_prot->slab);
1855 kfree(name);
1856 prot->rsk_prot->slab = NULL;
1859 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1860 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1862 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1863 kfree(name);
1864 prot->twsk_prot->twsk_slab = NULL;
1868 EXPORT_SYMBOL(proto_unregister);
1870 #ifdef CONFIG_PROC_FS
1871 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1873 read_lock(&proto_list_lock);
1874 return seq_list_start_head(&proto_list, *pos);
1877 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1879 return seq_list_next(v, &proto_list, pos);
1882 static void proto_seq_stop(struct seq_file *seq, void *v)
1884 read_unlock(&proto_list_lock);
1887 static char proto_method_implemented(const void *method)
1889 return method == NULL ? 'n' : 'y';
1892 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1894 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1895 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1896 proto->name,
1897 proto->obj_size,
1898 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1899 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1900 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1901 proto->max_header,
1902 proto->slab == NULL ? "no" : "yes",
1903 module_name(proto->owner),
1904 proto_method_implemented(proto->close),
1905 proto_method_implemented(proto->connect),
1906 proto_method_implemented(proto->disconnect),
1907 proto_method_implemented(proto->accept),
1908 proto_method_implemented(proto->ioctl),
1909 proto_method_implemented(proto->init),
1910 proto_method_implemented(proto->destroy),
1911 proto_method_implemented(proto->shutdown),
1912 proto_method_implemented(proto->setsockopt),
1913 proto_method_implemented(proto->getsockopt),
1914 proto_method_implemented(proto->sendmsg),
1915 proto_method_implemented(proto->recvmsg),
1916 proto_method_implemented(proto->sendpage),
1917 proto_method_implemented(proto->bind),
1918 proto_method_implemented(proto->backlog_rcv),
1919 proto_method_implemented(proto->hash),
1920 proto_method_implemented(proto->unhash),
1921 proto_method_implemented(proto->get_port),
1922 proto_method_implemented(proto->enter_memory_pressure));
1925 static int proto_seq_show(struct seq_file *seq, void *v)
1927 if (v == &proto_list)
1928 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1929 "protocol",
1930 "size",
1931 "sockets",
1932 "memory",
1933 "press",
1934 "maxhdr",
1935 "slab",
1936 "module",
1937 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1938 else
1939 proto_seq_printf(seq, list_entry(v, struct proto, node));
1940 return 0;
1943 static const struct seq_operations proto_seq_ops = {
1944 .start = proto_seq_start,
1945 .next = proto_seq_next,
1946 .stop = proto_seq_stop,
1947 .show = proto_seq_show,
1950 static int proto_seq_open(struct inode *inode, struct file *file)
1952 return seq_open(file, &proto_seq_ops);
1955 static const struct file_operations proto_seq_fops = {
1956 .owner = THIS_MODULE,
1957 .open = proto_seq_open,
1958 .read = seq_read,
1959 .llseek = seq_lseek,
1960 .release = seq_release,
1963 static int __init proto_init(void)
1965 /* register /proc/net/protocols */
1966 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1969 subsys_initcall(proto_init);
1971 #endif /* PROC_FS */
1973 EXPORT_SYMBOL(sk_alloc);
1974 EXPORT_SYMBOL(sk_free);
1975 EXPORT_SYMBOL(sk_send_sigurg);
1976 EXPORT_SYMBOL(sock_alloc_send_skb);
1977 EXPORT_SYMBOL(sock_init_data);
1978 EXPORT_SYMBOL(sock_kfree_s);
1979 EXPORT_SYMBOL(sock_kmalloc);
1980 EXPORT_SYMBOL(sock_no_accept);
1981 EXPORT_SYMBOL(sock_no_bind);
1982 EXPORT_SYMBOL(sock_no_connect);
1983 EXPORT_SYMBOL(sock_no_getname);
1984 EXPORT_SYMBOL(sock_no_getsockopt);
1985 EXPORT_SYMBOL(sock_no_ioctl);
1986 EXPORT_SYMBOL(sock_no_listen);
1987 EXPORT_SYMBOL(sock_no_mmap);
1988 EXPORT_SYMBOL(sock_no_poll);
1989 EXPORT_SYMBOL(sock_no_recvmsg);
1990 EXPORT_SYMBOL(sock_no_sendmsg);
1991 EXPORT_SYMBOL(sock_no_sendpage);
1992 EXPORT_SYMBOL(sock_no_setsockopt);
1993 EXPORT_SYMBOL(sock_no_shutdown);
1994 EXPORT_SYMBOL(sock_no_socketpair);
1995 EXPORT_SYMBOL(sock_rfree);
1996 EXPORT_SYMBOL(sock_setsockopt);
1997 EXPORT_SYMBOL(sock_wfree);
1998 EXPORT_SYMBOL(sock_wmalloc);
1999 EXPORT_SYMBOL(sock_i_uid);
2000 EXPORT_SYMBOL(sock_i_ino);
2001 EXPORT_SYMBOL(sysctl_optmem_max);
2002 #ifdef CONFIG_SYSCTL
2003 EXPORT_SYMBOL(sysctl_rmem_max);
2004 EXPORT_SYMBOL(sysctl_wmem_max);
2005 #endif