sysctl: ipv4 remove binary sysctl paths where they are broken
[linux-2.6/x86.git] / net / core / sock.c
blobd45ecdccc6a153ec1ffccb1430e6085a3dd42f7d
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/net_namespace.h>
123 #include <net/request_sock.h>
124 #include <net/sock.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
128 #include <linux/filter.h>
130 #ifdef CONFIG_INET
131 #include <net/tcp.h>
132 #endif
135 * Each address family might have different locking rules, so we have
136 * one slock key per address family:
138 static struct lock_class_key af_family_keys[AF_MAX];
139 static struct lock_class_key af_family_slock_keys[AF_MAX];
141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
143 * Make lock validator output more readable. (we pre-construct these
144 * strings build-time, so that runtime initialization of socket
145 * locks is fast):
147 static const char *af_family_key_strings[AF_MAX+1] = {
148 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
149 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
150 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
151 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
152 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
153 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
154 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
155 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
157 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" ,
158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
161 static const char *af_family_slock_key_strings[AF_MAX+1] = {
162 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
163 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
164 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
165 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
166 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
167 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
168 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
169 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
170 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
171 "slock-27" , "slock-28" , "slock-29" ,
172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
173 "slock-AF_RXRPC" , "slock-AF_MAX"
175 static const char *af_family_clock_key_strings[AF_MAX+1] = {
176 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
177 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
178 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
179 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
180 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
181 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
182 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
183 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
184 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
185 "clock-27" , "clock-28" , "clock-29" ,
186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
187 "clock-AF_RXRPC" , "clock-AF_MAX"
189 #endif
192 * sk_callback_lock locking rules are per-address-family,
193 * so split the lock classes by using a per-AF key:
195 static struct lock_class_key af_callback_keys[AF_MAX];
197 /* Take into consideration the size of the struct sk_buff overhead in the
198 * determination of these values, since that is non-constant across
199 * platforms. This makes socket queueing behavior and performance
200 * not depend upon such differences.
202 #define _SK_MEM_PACKETS 256
203 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
204 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
207 /* Run time adjustable parameters. */
208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
213 /* Maximal space eaten by iovec or ancilliary data plus some space */
214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
218 struct timeval tv;
220 if (optlen < sizeof(tv))
221 return -EINVAL;
222 if (copy_from_user(&tv, optval, sizeof(tv)))
223 return -EFAULT;
224 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
225 return -EDOM;
227 if (tv.tv_sec < 0) {
228 static int warned __read_mostly;
230 *timeo_p = 0;
231 if (warned < 10 && net_ratelimit())
232 warned++;
233 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
234 "tries to set negative timeout\n",
235 current->comm, current->pid);
236 return 0;
238 *timeo_p = MAX_SCHEDULE_TIMEOUT;
239 if (tv.tv_sec == 0 && tv.tv_usec == 0)
240 return 0;
241 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
242 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
243 return 0;
246 static void sock_warn_obsolete_bsdism(const char *name)
248 static int warned;
249 static char warncomm[TASK_COMM_LEN];
250 if (strcmp(warncomm, current->comm) && warned < 5) {
251 strcpy(warncomm, current->comm);
252 printk(KERN_WARNING "process `%s' is using obsolete "
253 "%s SO_BSDCOMPAT\n", warncomm, name);
254 warned++;
258 static void sock_disable_timestamp(struct sock *sk)
260 if (sock_flag(sk, SOCK_TIMESTAMP)) {
261 sock_reset_flag(sk, SOCK_TIMESTAMP);
262 net_disable_timestamp();
267 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
269 int err = 0;
270 int skb_len;
272 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
273 number of warnings when compiling with -W --ANK
275 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
276 (unsigned)sk->sk_rcvbuf) {
277 err = -ENOMEM;
278 goto out;
281 err = sk_filter(sk, skb);
282 if (err)
283 goto out;
285 skb->dev = NULL;
286 skb_set_owner_r(skb, sk);
288 /* Cache the SKB length before we tack it onto the receive
289 * queue. Once it is added it no longer belongs to us and
290 * may be freed by other threads of control pulling packets
291 * from the queue.
293 skb_len = skb->len;
295 skb_queue_tail(&sk->sk_receive_queue, skb);
297 if (!sock_flag(sk, SOCK_DEAD))
298 sk->sk_data_ready(sk, skb_len);
299 out:
300 return err;
302 EXPORT_SYMBOL(sock_queue_rcv_skb);
304 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
306 int rc = NET_RX_SUCCESS;
308 if (sk_filter(sk, skb))
309 goto discard_and_relse;
311 skb->dev = NULL;
313 if (nested)
314 bh_lock_sock_nested(sk);
315 else
316 bh_lock_sock(sk);
317 if (!sock_owned_by_user(sk)) {
319 * trylock + unlock semantics:
321 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
323 rc = sk->sk_backlog_rcv(sk, skb);
325 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
326 } else
327 sk_add_backlog(sk, skb);
328 bh_unlock_sock(sk);
329 out:
330 sock_put(sk);
331 return rc;
332 discard_and_relse:
333 kfree_skb(skb);
334 goto out;
336 EXPORT_SYMBOL(sk_receive_skb);
338 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
340 struct dst_entry *dst = sk->sk_dst_cache;
342 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
343 sk->sk_dst_cache = NULL;
344 dst_release(dst);
345 return NULL;
348 return dst;
350 EXPORT_SYMBOL(__sk_dst_check);
352 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
354 struct dst_entry *dst = sk_dst_get(sk);
356 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
357 sk_dst_reset(sk);
358 dst_release(dst);
359 return NULL;
362 return dst;
364 EXPORT_SYMBOL(sk_dst_check);
366 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
368 int ret = -ENOPROTOOPT;
369 #ifdef CONFIG_NETDEVICES
370 struct net *net = sk->sk_net;
371 char devname[IFNAMSIZ];
372 int index;
374 /* Sorry... */
375 ret = -EPERM;
376 if (!capable(CAP_NET_RAW))
377 goto out;
379 ret = -EINVAL;
380 if (optlen < 0)
381 goto out;
383 /* Bind this socket to a particular device like "eth0",
384 * as specified in the passed interface name. If the
385 * name is "" or the option length is zero the socket
386 * is not bound.
388 if (optlen > IFNAMSIZ - 1)
389 optlen = IFNAMSIZ - 1;
390 memset(devname, 0, sizeof(devname));
392 ret = -EFAULT;
393 if (copy_from_user(devname, optval, optlen))
394 goto out;
396 if (devname[0] == '\0') {
397 index = 0;
398 } else {
399 struct net_device *dev = dev_get_by_name(net, devname);
401 ret = -ENODEV;
402 if (!dev)
403 goto out;
405 index = dev->ifindex;
406 dev_put(dev);
409 lock_sock(sk);
410 sk->sk_bound_dev_if = index;
411 sk_dst_reset(sk);
412 release_sock(sk);
414 ret = 0;
416 out:
417 #endif
419 return ret;
423 * This is meant for all protocols to use and covers goings on
424 * at the socket level. Everything here is generic.
427 int sock_setsockopt(struct socket *sock, int level, int optname,
428 char __user *optval, int optlen)
430 struct sock *sk=sock->sk;
431 struct sk_filter *filter;
432 int val;
433 int valbool;
434 struct linger ling;
435 int ret = 0;
438 * Options without arguments
441 #ifdef SO_DONTLINGER /* Compatibility item... */
442 if (optname == SO_DONTLINGER) {
443 lock_sock(sk);
444 sock_reset_flag(sk, SOCK_LINGER);
445 release_sock(sk);
446 return 0;
448 #endif
450 if (optname == SO_BINDTODEVICE)
451 return sock_bindtodevice(sk, optval, optlen);
453 if (optlen < sizeof(int))
454 return -EINVAL;
456 if (get_user(val, (int __user *)optval))
457 return -EFAULT;
459 valbool = val?1:0;
461 lock_sock(sk);
463 switch(optname) {
464 case SO_DEBUG:
465 if (val && !capable(CAP_NET_ADMIN)) {
466 ret = -EACCES;
468 else if (valbool)
469 sock_set_flag(sk, SOCK_DBG);
470 else
471 sock_reset_flag(sk, SOCK_DBG);
472 break;
473 case SO_REUSEADDR:
474 sk->sk_reuse = valbool;
475 break;
476 case SO_TYPE:
477 case SO_ERROR:
478 ret = -ENOPROTOOPT;
479 break;
480 case SO_DONTROUTE:
481 if (valbool)
482 sock_set_flag(sk, SOCK_LOCALROUTE);
483 else
484 sock_reset_flag(sk, SOCK_LOCALROUTE);
485 break;
486 case SO_BROADCAST:
487 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
488 break;
489 case SO_SNDBUF:
490 /* Don't error on this BSD doesn't and if you think
491 about it this is right. Otherwise apps have to
492 play 'guess the biggest size' games. RCVBUF/SNDBUF
493 are treated in BSD as hints */
495 if (val > sysctl_wmem_max)
496 val = sysctl_wmem_max;
497 set_sndbuf:
498 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
499 if ((val * 2) < SOCK_MIN_SNDBUF)
500 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
501 else
502 sk->sk_sndbuf = val * 2;
505 * Wake up sending tasks if we
506 * upped the value.
508 sk->sk_write_space(sk);
509 break;
511 case SO_SNDBUFFORCE:
512 if (!capable(CAP_NET_ADMIN)) {
513 ret = -EPERM;
514 break;
516 goto set_sndbuf;
518 case SO_RCVBUF:
519 /* Don't error on this BSD doesn't and if you think
520 about it this is right. Otherwise apps have to
521 play 'guess the biggest size' games. RCVBUF/SNDBUF
522 are treated in BSD as hints */
524 if (val > sysctl_rmem_max)
525 val = sysctl_rmem_max;
526 set_rcvbuf:
527 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
529 * We double it on the way in to account for
530 * "struct sk_buff" etc. overhead. Applications
531 * assume that the SO_RCVBUF setting they make will
532 * allow that much actual data to be received on that
533 * socket.
535 * Applications are unaware that "struct sk_buff" and
536 * other overheads allocate from the receive buffer
537 * during socket buffer allocation.
539 * And after considering the possible alternatives,
540 * returning the value we actually used in getsockopt
541 * is the most desirable behavior.
543 if ((val * 2) < SOCK_MIN_RCVBUF)
544 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
545 else
546 sk->sk_rcvbuf = val * 2;
547 break;
549 case SO_RCVBUFFORCE:
550 if (!capable(CAP_NET_ADMIN)) {
551 ret = -EPERM;
552 break;
554 goto set_rcvbuf;
556 case SO_KEEPALIVE:
557 #ifdef CONFIG_INET
558 if (sk->sk_protocol == IPPROTO_TCP)
559 tcp_set_keepalive(sk, valbool);
560 #endif
561 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
562 break;
564 case SO_OOBINLINE:
565 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
566 break;
568 case SO_NO_CHECK:
569 sk->sk_no_check = valbool;
570 break;
572 case SO_PRIORITY:
573 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
574 sk->sk_priority = val;
575 else
576 ret = -EPERM;
577 break;
579 case SO_LINGER:
580 if (optlen < sizeof(ling)) {
581 ret = -EINVAL; /* 1003.1g */
582 break;
584 if (copy_from_user(&ling,optval,sizeof(ling))) {
585 ret = -EFAULT;
586 break;
588 if (!ling.l_onoff)
589 sock_reset_flag(sk, SOCK_LINGER);
590 else {
591 #if (BITS_PER_LONG == 32)
592 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
593 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
594 else
595 #endif
596 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
597 sock_set_flag(sk, SOCK_LINGER);
599 break;
601 case SO_BSDCOMPAT:
602 sock_warn_obsolete_bsdism("setsockopt");
603 break;
605 case SO_PASSCRED:
606 if (valbool)
607 set_bit(SOCK_PASSCRED, &sock->flags);
608 else
609 clear_bit(SOCK_PASSCRED, &sock->flags);
610 break;
612 case SO_TIMESTAMP:
613 case SO_TIMESTAMPNS:
614 if (valbool) {
615 if (optname == SO_TIMESTAMP)
616 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
617 else
618 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
619 sock_set_flag(sk, SOCK_RCVTSTAMP);
620 sock_enable_timestamp(sk);
621 } else {
622 sock_reset_flag(sk, SOCK_RCVTSTAMP);
623 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
625 break;
627 case SO_RCVLOWAT:
628 if (val < 0)
629 val = INT_MAX;
630 sk->sk_rcvlowat = val ? : 1;
631 break;
633 case SO_RCVTIMEO:
634 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
635 break;
637 case SO_SNDTIMEO:
638 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
639 break;
641 case SO_ATTACH_FILTER:
642 ret = -EINVAL;
643 if (optlen == sizeof(struct sock_fprog)) {
644 struct sock_fprog fprog;
646 ret = -EFAULT;
647 if (copy_from_user(&fprog, optval, sizeof(fprog)))
648 break;
650 ret = sk_attach_filter(&fprog, sk);
652 break;
654 case SO_DETACH_FILTER:
655 rcu_read_lock_bh();
656 filter = rcu_dereference(sk->sk_filter);
657 if (filter) {
658 rcu_assign_pointer(sk->sk_filter, NULL);
659 sk_filter_release(sk, filter);
660 rcu_read_unlock_bh();
661 break;
663 rcu_read_unlock_bh();
664 ret = -ENONET;
665 break;
667 case SO_PASSSEC:
668 if (valbool)
669 set_bit(SOCK_PASSSEC, &sock->flags);
670 else
671 clear_bit(SOCK_PASSSEC, &sock->flags);
672 break;
674 /* We implement the SO_SNDLOWAT etc to
675 not be settable (1003.1g 5.3) */
676 default:
677 ret = -ENOPROTOOPT;
678 break;
680 release_sock(sk);
681 return ret;
685 int sock_getsockopt(struct socket *sock, int level, int optname,
686 char __user *optval, int __user *optlen)
688 struct sock *sk = sock->sk;
690 union {
691 int val;
692 struct linger ling;
693 struct timeval tm;
694 } v;
696 unsigned int lv = sizeof(int);
697 int len;
699 if (get_user(len, optlen))
700 return -EFAULT;
701 if (len < 0)
702 return -EINVAL;
704 switch(optname) {
705 case SO_DEBUG:
706 v.val = sock_flag(sk, SOCK_DBG);
707 break;
709 case SO_DONTROUTE:
710 v.val = sock_flag(sk, SOCK_LOCALROUTE);
711 break;
713 case SO_BROADCAST:
714 v.val = !!sock_flag(sk, SOCK_BROADCAST);
715 break;
717 case SO_SNDBUF:
718 v.val = sk->sk_sndbuf;
719 break;
721 case SO_RCVBUF:
722 v.val = sk->sk_rcvbuf;
723 break;
725 case SO_REUSEADDR:
726 v.val = sk->sk_reuse;
727 break;
729 case SO_KEEPALIVE:
730 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
731 break;
733 case SO_TYPE:
734 v.val = sk->sk_type;
735 break;
737 case SO_ERROR:
738 v.val = -sock_error(sk);
739 if (v.val==0)
740 v.val = xchg(&sk->sk_err_soft, 0);
741 break;
743 case SO_OOBINLINE:
744 v.val = !!sock_flag(sk, SOCK_URGINLINE);
745 break;
747 case SO_NO_CHECK:
748 v.val = sk->sk_no_check;
749 break;
751 case SO_PRIORITY:
752 v.val = sk->sk_priority;
753 break;
755 case SO_LINGER:
756 lv = sizeof(v.ling);
757 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
758 v.ling.l_linger = sk->sk_lingertime / HZ;
759 break;
761 case SO_BSDCOMPAT:
762 sock_warn_obsolete_bsdism("getsockopt");
763 break;
765 case SO_TIMESTAMP:
766 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
767 !sock_flag(sk, SOCK_RCVTSTAMPNS);
768 break;
770 case SO_TIMESTAMPNS:
771 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
772 break;
774 case SO_RCVTIMEO:
775 lv=sizeof(struct timeval);
776 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
777 v.tm.tv_sec = 0;
778 v.tm.tv_usec = 0;
779 } else {
780 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
781 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
783 break;
785 case SO_SNDTIMEO:
786 lv=sizeof(struct timeval);
787 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
788 v.tm.tv_sec = 0;
789 v.tm.tv_usec = 0;
790 } else {
791 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
792 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
794 break;
796 case SO_RCVLOWAT:
797 v.val = sk->sk_rcvlowat;
798 break;
800 case SO_SNDLOWAT:
801 v.val=1;
802 break;
804 case SO_PASSCRED:
805 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
806 break;
808 case SO_PEERCRED:
809 if (len > sizeof(sk->sk_peercred))
810 len = sizeof(sk->sk_peercred);
811 if (copy_to_user(optval, &sk->sk_peercred, len))
812 return -EFAULT;
813 goto lenout;
815 case SO_PEERNAME:
817 char address[128];
819 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
820 return -ENOTCONN;
821 if (lv < len)
822 return -EINVAL;
823 if (copy_to_user(optval, address, len))
824 return -EFAULT;
825 goto lenout;
828 /* Dubious BSD thing... Probably nobody even uses it, but
829 * the UNIX standard wants it for whatever reason... -DaveM
831 case SO_ACCEPTCONN:
832 v.val = sk->sk_state == TCP_LISTEN;
833 break;
835 case SO_PASSSEC:
836 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
837 break;
839 case SO_PEERSEC:
840 return security_socket_getpeersec_stream(sock, optval, optlen, len);
842 default:
843 return -ENOPROTOOPT;
846 if (len > lv)
847 len = lv;
848 if (copy_to_user(optval, &v, len))
849 return -EFAULT;
850 lenout:
851 if (put_user(len, optlen))
852 return -EFAULT;
853 return 0;
857 * Initialize an sk_lock.
859 * (We also register the sk_lock with the lock validator.)
861 static inline void sock_lock_init(struct sock *sk)
863 sock_lock_init_class_and_name(sk,
864 af_family_slock_key_strings[sk->sk_family],
865 af_family_slock_keys + sk->sk_family,
866 af_family_key_strings[sk->sk_family],
867 af_family_keys + sk->sk_family);
871 * sk_alloc - All socket objects are allocated here
872 * @net: the applicable net namespace
873 * @family: protocol family
874 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
875 * @prot: struct proto associated with this new sock instance
876 * @zero_it: if we should zero the newly allocated sock
878 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
879 struct proto *prot, int zero_it)
881 struct sock *sk = NULL;
882 struct kmem_cache *slab = prot->slab;
884 if (slab != NULL)
885 sk = kmem_cache_alloc(slab, priority);
886 else
887 sk = kmalloc(prot->obj_size, priority);
889 if (sk) {
890 if (zero_it) {
891 memset(sk, 0, prot->obj_size);
892 sk->sk_family = family;
894 * See comment in struct sock definition to understand
895 * why we need sk_prot_creator -acme
897 sk->sk_prot = sk->sk_prot_creator = prot;
898 sock_lock_init(sk);
899 sk->sk_net = get_net(net);
902 if (security_sk_alloc(sk, family, priority))
903 goto out_free;
905 if (!try_module_get(prot->owner))
906 goto out_free;
908 return sk;
910 out_free:
911 if (slab != NULL)
912 kmem_cache_free(slab, sk);
913 else
914 kfree(sk);
915 return NULL;
918 void sk_free(struct sock *sk)
920 struct sk_filter *filter;
921 struct module *owner = sk->sk_prot_creator->owner;
923 if (sk->sk_destruct)
924 sk->sk_destruct(sk);
926 filter = rcu_dereference(sk->sk_filter);
927 if (filter) {
928 sk_filter_release(sk, filter);
929 rcu_assign_pointer(sk->sk_filter, NULL);
932 sock_disable_timestamp(sk);
934 if (atomic_read(&sk->sk_omem_alloc))
935 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
936 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
938 security_sk_free(sk);
939 put_net(sk->sk_net);
940 if (sk->sk_prot_creator->slab != NULL)
941 kmem_cache_free(sk->sk_prot_creator->slab, sk);
942 else
943 kfree(sk);
944 module_put(owner);
947 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
949 struct sock *newsk = sk_alloc(sk->sk_net, sk->sk_family, priority, sk->sk_prot, 0);
951 if (newsk != NULL) {
952 struct sk_filter *filter;
954 sock_copy(newsk, sk);
956 /* SANITY */
957 sk_node_init(&newsk->sk_node);
958 sock_lock_init(newsk);
959 bh_lock_sock(newsk);
960 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
962 atomic_set(&newsk->sk_rmem_alloc, 0);
963 atomic_set(&newsk->sk_wmem_alloc, 0);
964 atomic_set(&newsk->sk_omem_alloc, 0);
965 skb_queue_head_init(&newsk->sk_receive_queue);
966 skb_queue_head_init(&newsk->sk_write_queue);
967 #ifdef CONFIG_NET_DMA
968 skb_queue_head_init(&newsk->sk_async_wait_queue);
969 #endif
971 rwlock_init(&newsk->sk_dst_lock);
972 rwlock_init(&newsk->sk_callback_lock);
973 lockdep_set_class_and_name(&newsk->sk_callback_lock,
974 af_callback_keys + newsk->sk_family,
975 af_family_clock_key_strings[newsk->sk_family]);
977 newsk->sk_dst_cache = NULL;
978 newsk->sk_wmem_queued = 0;
979 newsk->sk_forward_alloc = 0;
980 newsk->sk_send_head = NULL;
981 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
983 sock_reset_flag(newsk, SOCK_DONE);
984 skb_queue_head_init(&newsk->sk_error_queue);
986 filter = newsk->sk_filter;
987 if (filter != NULL)
988 sk_filter_charge(newsk, filter);
990 if (unlikely(xfrm_sk_clone_policy(newsk))) {
991 /* It is still raw copy of parent, so invalidate
992 * destructor and make plain sk_free() */
993 newsk->sk_destruct = NULL;
994 sk_free(newsk);
995 newsk = NULL;
996 goto out;
999 newsk->sk_err = 0;
1000 newsk->sk_priority = 0;
1001 atomic_set(&newsk->sk_refcnt, 2);
1004 * Increment the counter in the same struct proto as the master
1005 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1006 * is the same as sk->sk_prot->socks, as this field was copied
1007 * with memcpy).
1009 * This _changes_ the previous behaviour, where
1010 * tcp_create_openreq_child always was incrementing the
1011 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1012 * to be taken into account in all callers. -acme
1014 sk_refcnt_debug_inc(newsk);
1015 newsk->sk_socket = NULL;
1016 newsk->sk_sleep = NULL;
1018 if (newsk->sk_prot->sockets_allocated)
1019 atomic_inc(newsk->sk_prot->sockets_allocated);
1021 out:
1022 return newsk;
1025 EXPORT_SYMBOL_GPL(sk_clone);
1027 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1029 __sk_dst_set(sk, dst);
1030 sk->sk_route_caps = dst->dev->features;
1031 if (sk->sk_route_caps & NETIF_F_GSO)
1032 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1033 if (sk_can_gso(sk)) {
1034 if (dst->header_len)
1035 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1036 else
1037 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1040 EXPORT_SYMBOL_GPL(sk_setup_caps);
1042 void __init sk_init(void)
1044 if (num_physpages <= 4096) {
1045 sysctl_wmem_max = 32767;
1046 sysctl_rmem_max = 32767;
1047 sysctl_wmem_default = 32767;
1048 sysctl_rmem_default = 32767;
1049 } else if (num_physpages >= 131072) {
1050 sysctl_wmem_max = 131071;
1051 sysctl_rmem_max = 131071;
1056 * Simple resource managers for sockets.
1061 * Write buffer destructor automatically called from kfree_skb.
1063 void sock_wfree(struct sk_buff *skb)
1065 struct sock *sk = skb->sk;
1067 /* In case it might be waiting for more memory. */
1068 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1069 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1070 sk->sk_write_space(sk);
1071 sock_put(sk);
1075 * Read buffer destructor automatically called from kfree_skb.
1077 void sock_rfree(struct sk_buff *skb)
1079 struct sock *sk = skb->sk;
1081 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1085 int sock_i_uid(struct sock *sk)
1087 int uid;
1089 read_lock(&sk->sk_callback_lock);
1090 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1091 read_unlock(&sk->sk_callback_lock);
1092 return uid;
1095 unsigned long sock_i_ino(struct sock *sk)
1097 unsigned long ino;
1099 read_lock(&sk->sk_callback_lock);
1100 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1101 read_unlock(&sk->sk_callback_lock);
1102 return ino;
1106 * Allocate a skb from the socket's send buffer.
1108 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1109 gfp_t priority)
1111 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1112 struct sk_buff * skb = alloc_skb(size, priority);
1113 if (skb) {
1114 skb_set_owner_w(skb, sk);
1115 return skb;
1118 return NULL;
1122 * Allocate a skb from the socket's receive buffer.
1124 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1125 gfp_t priority)
1127 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1128 struct sk_buff *skb = alloc_skb(size, priority);
1129 if (skb) {
1130 skb_set_owner_r(skb, sk);
1131 return skb;
1134 return NULL;
1138 * Allocate a memory block from the socket's option memory buffer.
1140 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1142 if ((unsigned)size <= sysctl_optmem_max &&
1143 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1144 void *mem;
1145 /* First do the add, to avoid the race if kmalloc
1146 * might sleep.
1148 atomic_add(size, &sk->sk_omem_alloc);
1149 mem = kmalloc(size, priority);
1150 if (mem)
1151 return mem;
1152 atomic_sub(size, &sk->sk_omem_alloc);
1154 return NULL;
1158 * Free an option memory block.
1160 void sock_kfree_s(struct sock *sk, void *mem, int size)
1162 kfree(mem);
1163 atomic_sub(size, &sk->sk_omem_alloc);
1166 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1167 I think, these locks should be removed for datagram sockets.
1169 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1171 DEFINE_WAIT(wait);
1173 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1174 for (;;) {
1175 if (!timeo)
1176 break;
1177 if (signal_pending(current))
1178 break;
1179 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1180 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1181 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1182 break;
1183 if (sk->sk_shutdown & SEND_SHUTDOWN)
1184 break;
1185 if (sk->sk_err)
1186 break;
1187 timeo = schedule_timeout(timeo);
1189 finish_wait(sk->sk_sleep, &wait);
1190 return timeo;
1195 * Generic send/receive buffer handlers
1198 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1199 unsigned long header_len,
1200 unsigned long data_len,
1201 int noblock, int *errcode)
1203 struct sk_buff *skb;
1204 gfp_t gfp_mask;
1205 long timeo;
1206 int err;
1208 gfp_mask = sk->sk_allocation;
1209 if (gfp_mask & __GFP_WAIT)
1210 gfp_mask |= __GFP_REPEAT;
1212 timeo = sock_sndtimeo(sk, noblock);
1213 while (1) {
1214 err = sock_error(sk);
1215 if (err != 0)
1216 goto failure;
1218 err = -EPIPE;
1219 if (sk->sk_shutdown & SEND_SHUTDOWN)
1220 goto failure;
1222 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1223 skb = alloc_skb(header_len, gfp_mask);
1224 if (skb) {
1225 int npages;
1226 int i;
1228 /* No pages, we're done... */
1229 if (!data_len)
1230 break;
1232 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1233 skb->truesize += data_len;
1234 skb_shinfo(skb)->nr_frags = npages;
1235 for (i = 0; i < npages; i++) {
1236 struct page *page;
1237 skb_frag_t *frag;
1239 page = alloc_pages(sk->sk_allocation, 0);
1240 if (!page) {
1241 err = -ENOBUFS;
1242 skb_shinfo(skb)->nr_frags = i;
1243 kfree_skb(skb);
1244 goto failure;
1247 frag = &skb_shinfo(skb)->frags[i];
1248 frag->page = page;
1249 frag->page_offset = 0;
1250 frag->size = (data_len >= PAGE_SIZE ?
1251 PAGE_SIZE :
1252 data_len);
1253 data_len -= PAGE_SIZE;
1256 /* Full success... */
1257 break;
1259 err = -ENOBUFS;
1260 goto failure;
1262 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1263 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1264 err = -EAGAIN;
1265 if (!timeo)
1266 goto failure;
1267 if (signal_pending(current))
1268 goto interrupted;
1269 timeo = sock_wait_for_wmem(sk, timeo);
1272 skb_set_owner_w(skb, sk);
1273 return skb;
1275 interrupted:
1276 err = sock_intr_errno(timeo);
1277 failure:
1278 *errcode = err;
1279 return NULL;
1282 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1283 int noblock, int *errcode)
1285 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1288 static void __lock_sock(struct sock *sk)
1290 DEFINE_WAIT(wait);
1292 for (;;) {
1293 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1294 TASK_UNINTERRUPTIBLE);
1295 spin_unlock_bh(&sk->sk_lock.slock);
1296 schedule();
1297 spin_lock_bh(&sk->sk_lock.slock);
1298 if (!sock_owned_by_user(sk))
1299 break;
1301 finish_wait(&sk->sk_lock.wq, &wait);
1304 static void __release_sock(struct sock *sk)
1306 struct sk_buff *skb = sk->sk_backlog.head;
1308 do {
1309 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1310 bh_unlock_sock(sk);
1312 do {
1313 struct sk_buff *next = skb->next;
1315 skb->next = NULL;
1316 sk->sk_backlog_rcv(sk, skb);
1319 * We are in process context here with softirqs
1320 * disabled, use cond_resched_softirq() to preempt.
1321 * This is safe to do because we've taken the backlog
1322 * queue private:
1324 cond_resched_softirq();
1326 skb = next;
1327 } while (skb != NULL);
1329 bh_lock_sock(sk);
1330 } while ((skb = sk->sk_backlog.head) != NULL);
1334 * sk_wait_data - wait for data to arrive at sk_receive_queue
1335 * @sk: sock to wait on
1336 * @timeo: for how long
1338 * Now socket state including sk->sk_err is changed only under lock,
1339 * hence we may omit checks after joining wait queue.
1340 * We check receive queue before schedule() only as optimization;
1341 * it is very likely that release_sock() added new data.
1343 int sk_wait_data(struct sock *sk, long *timeo)
1345 int rc;
1346 DEFINE_WAIT(wait);
1348 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1349 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1350 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1351 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1352 finish_wait(sk->sk_sleep, &wait);
1353 return rc;
1356 EXPORT_SYMBOL(sk_wait_data);
1359 * Set of default routines for initialising struct proto_ops when
1360 * the protocol does not support a particular function. In certain
1361 * cases where it makes no sense for a protocol to have a "do nothing"
1362 * function, some default processing is provided.
1365 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1367 return -EOPNOTSUPP;
1370 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1371 int len, int flags)
1373 return -EOPNOTSUPP;
1376 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1378 return -EOPNOTSUPP;
1381 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1383 return -EOPNOTSUPP;
1386 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1387 int *len, int peer)
1389 return -EOPNOTSUPP;
1392 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1394 return 0;
1397 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1399 return -EOPNOTSUPP;
1402 int sock_no_listen(struct socket *sock, int backlog)
1404 return -EOPNOTSUPP;
1407 int sock_no_shutdown(struct socket *sock, int how)
1409 return -EOPNOTSUPP;
1412 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1413 char __user *optval, int optlen)
1415 return -EOPNOTSUPP;
1418 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1419 char __user *optval, int __user *optlen)
1421 return -EOPNOTSUPP;
1424 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1425 size_t len)
1427 return -EOPNOTSUPP;
1430 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1431 size_t len, int flags)
1433 return -EOPNOTSUPP;
1436 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1438 /* Mirror missing mmap method error code */
1439 return -ENODEV;
1442 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1444 ssize_t res;
1445 struct msghdr msg = {.msg_flags = flags};
1446 struct kvec iov;
1447 char *kaddr = kmap(page);
1448 iov.iov_base = kaddr + offset;
1449 iov.iov_len = size;
1450 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1451 kunmap(page);
1452 return res;
1456 * Default Socket Callbacks
1459 static void sock_def_wakeup(struct sock *sk)
1461 read_lock(&sk->sk_callback_lock);
1462 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1463 wake_up_interruptible_all(sk->sk_sleep);
1464 read_unlock(&sk->sk_callback_lock);
1467 static void sock_def_error_report(struct sock *sk)
1469 read_lock(&sk->sk_callback_lock);
1470 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1471 wake_up_interruptible(sk->sk_sleep);
1472 sk_wake_async(sk,0,POLL_ERR);
1473 read_unlock(&sk->sk_callback_lock);
1476 static void sock_def_readable(struct sock *sk, int len)
1478 read_lock(&sk->sk_callback_lock);
1479 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1480 wake_up_interruptible(sk->sk_sleep);
1481 sk_wake_async(sk,1,POLL_IN);
1482 read_unlock(&sk->sk_callback_lock);
1485 static void sock_def_write_space(struct sock *sk)
1487 read_lock(&sk->sk_callback_lock);
1489 /* Do not wake up a writer until he can make "significant"
1490 * progress. --DaveM
1492 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1493 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1494 wake_up_interruptible(sk->sk_sleep);
1496 /* Should agree with poll, otherwise some programs break */
1497 if (sock_writeable(sk))
1498 sk_wake_async(sk, 2, POLL_OUT);
1501 read_unlock(&sk->sk_callback_lock);
1504 static void sock_def_destruct(struct sock *sk)
1506 kfree(sk->sk_protinfo);
1509 void sk_send_sigurg(struct sock *sk)
1511 if (sk->sk_socket && sk->sk_socket->file)
1512 if (send_sigurg(&sk->sk_socket->file->f_owner))
1513 sk_wake_async(sk, 3, POLL_PRI);
1516 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1517 unsigned long expires)
1519 if (!mod_timer(timer, expires))
1520 sock_hold(sk);
1523 EXPORT_SYMBOL(sk_reset_timer);
1525 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1527 if (timer_pending(timer) && del_timer(timer))
1528 __sock_put(sk);
1531 EXPORT_SYMBOL(sk_stop_timer);
1533 void sock_init_data(struct socket *sock, struct sock *sk)
1535 skb_queue_head_init(&sk->sk_receive_queue);
1536 skb_queue_head_init(&sk->sk_write_queue);
1537 skb_queue_head_init(&sk->sk_error_queue);
1538 #ifdef CONFIG_NET_DMA
1539 skb_queue_head_init(&sk->sk_async_wait_queue);
1540 #endif
1542 sk->sk_send_head = NULL;
1544 init_timer(&sk->sk_timer);
1546 sk->sk_allocation = GFP_KERNEL;
1547 sk->sk_rcvbuf = sysctl_rmem_default;
1548 sk->sk_sndbuf = sysctl_wmem_default;
1549 sk->sk_state = TCP_CLOSE;
1550 sk->sk_socket = sock;
1552 sock_set_flag(sk, SOCK_ZAPPED);
1554 if (sock) {
1555 sk->sk_type = sock->type;
1556 sk->sk_sleep = &sock->wait;
1557 sock->sk = sk;
1558 } else
1559 sk->sk_sleep = NULL;
1561 rwlock_init(&sk->sk_dst_lock);
1562 rwlock_init(&sk->sk_callback_lock);
1563 lockdep_set_class_and_name(&sk->sk_callback_lock,
1564 af_callback_keys + sk->sk_family,
1565 af_family_clock_key_strings[sk->sk_family]);
1567 sk->sk_state_change = sock_def_wakeup;
1568 sk->sk_data_ready = sock_def_readable;
1569 sk->sk_write_space = sock_def_write_space;
1570 sk->sk_error_report = sock_def_error_report;
1571 sk->sk_destruct = sock_def_destruct;
1573 sk->sk_sndmsg_page = NULL;
1574 sk->sk_sndmsg_off = 0;
1576 sk->sk_peercred.pid = 0;
1577 sk->sk_peercred.uid = -1;
1578 sk->sk_peercred.gid = -1;
1579 sk->sk_write_pending = 0;
1580 sk->sk_rcvlowat = 1;
1581 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1582 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1584 sk->sk_stamp = ktime_set(-1L, -1L);
1586 atomic_set(&sk->sk_refcnt, 1);
1589 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1591 might_sleep();
1592 spin_lock_bh(&sk->sk_lock.slock);
1593 if (sk->sk_lock.owned)
1594 __lock_sock(sk);
1595 sk->sk_lock.owned = 1;
1596 spin_unlock(&sk->sk_lock.slock);
1598 * The sk_lock has mutex_lock() semantics here:
1600 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1601 local_bh_enable();
1604 EXPORT_SYMBOL(lock_sock_nested);
1606 void fastcall release_sock(struct sock *sk)
1609 * The sk_lock has mutex_unlock() semantics:
1611 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1613 spin_lock_bh(&sk->sk_lock.slock);
1614 if (sk->sk_backlog.tail)
1615 __release_sock(sk);
1616 sk->sk_lock.owned = 0;
1617 if (waitqueue_active(&sk->sk_lock.wq))
1618 wake_up(&sk->sk_lock.wq);
1619 spin_unlock_bh(&sk->sk_lock.slock);
1621 EXPORT_SYMBOL(release_sock);
1623 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1625 struct timeval tv;
1626 if (!sock_flag(sk, SOCK_TIMESTAMP))
1627 sock_enable_timestamp(sk);
1628 tv = ktime_to_timeval(sk->sk_stamp);
1629 if (tv.tv_sec == -1)
1630 return -ENOENT;
1631 if (tv.tv_sec == 0) {
1632 sk->sk_stamp = ktime_get_real();
1633 tv = ktime_to_timeval(sk->sk_stamp);
1635 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1637 EXPORT_SYMBOL(sock_get_timestamp);
1639 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1641 struct timespec ts;
1642 if (!sock_flag(sk, SOCK_TIMESTAMP))
1643 sock_enable_timestamp(sk);
1644 ts = ktime_to_timespec(sk->sk_stamp);
1645 if (ts.tv_sec == -1)
1646 return -ENOENT;
1647 if (ts.tv_sec == 0) {
1648 sk->sk_stamp = ktime_get_real();
1649 ts = ktime_to_timespec(sk->sk_stamp);
1651 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1653 EXPORT_SYMBOL(sock_get_timestampns);
1655 void sock_enable_timestamp(struct sock *sk)
1657 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1658 sock_set_flag(sk, SOCK_TIMESTAMP);
1659 net_enable_timestamp();
1662 EXPORT_SYMBOL(sock_enable_timestamp);
1665 * Get a socket option on an socket.
1667 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1668 * asynchronous errors should be reported by getsockopt. We assume
1669 * this means if you specify SO_ERROR (otherwise whats the point of it).
1671 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1672 char __user *optval, int __user *optlen)
1674 struct sock *sk = sock->sk;
1676 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1679 EXPORT_SYMBOL(sock_common_getsockopt);
1681 #ifdef CONFIG_COMPAT
1682 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1683 char __user *optval, int __user *optlen)
1685 struct sock *sk = sock->sk;
1687 if (sk->sk_prot->compat_getsockopt != NULL)
1688 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1689 optval, optlen);
1690 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1692 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1693 #endif
1695 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1696 struct msghdr *msg, size_t size, int flags)
1698 struct sock *sk = sock->sk;
1699 int addr_len = 0;
1700 int err;
1702 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1703 flags & ~MSG_DONTWAIT, &addr_len);
1704 if (err >= 0)
1705 msg->msg_namelen = addr_len;
1706 return err;
1709 EXPORT_SYMBOL(sock_common_recvmsg);
1712 * Set socket options on an inet socket.
1714 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1715 char __user *optval, int optlen)
1717 struct sock *sk = sock->sk;
1719 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1722 EXPORT_SYMBOL(sock_common_setsockopt);
1724 #ifdef CONFIG_COMPAT
1725 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1726 char __user *optval, int optlen)
1728 struct sock *sk = sock->sk;
1730 if (sk->sk_prot->compat_setsockopt != NULL)
1731 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1732 optval, optlen);
1733 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1735 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1736 #endif
1738 void sk_common_release(struct sock *sk)
1740 if (sk->sk_prot->destroy)
1741 sk->sk_prot->destroy(sk);
1744 * Observation: when sock_common_release is called, processes have
1745 * no access to socket. But net still has.
1746 * Step one, detach it from networking:
1748 * A. Remove from hash tables.
1751 sk->sk_prot->unhash(sk);
1754 * In this point socket cannot receive new packets, but it is possible
1755 * that some packets are in flight because some CPU runs receiver and
1756 * did hash table lookup before we unhashed socket. They will achieve
1757 * receive queue and will be purged by socket destructor.
1759 * Also we still have packets pending on receive queue and probably,
1760 * our own packets waiting in device queues. sock_destroy will drain
1761 * receive queue, but transmitted packets will delay socket destruction
1762 * until the last reference will be released.
1765 sock_orphan(sk);
1767 xfrm_sk_free_policy(sk);
1769 sk_refcnt_debug_release(sk);
1770 sock_put(sk);
1773 EXPORT_SYMBOL(sk_common_release);
1775 static DEFINE_RWLOCK(proto_list_lock);
1776 static LIST_HEAD(proto_list);
1778 int proto_register(struct proto *prot, int alloc_slab)
1780 char *request_sock_slab_name = NULL;
1781 char *timewait_sock_slab_name;
1782 int rc = -ENOBUFS;
1784 if (alloc_slab) {
1785 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1786 SLAB_HWCACHE_ALIGN, NULL);
1788 if (prot->slab == NULL) {
1789 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1790 prot->name);
1791 goto out;
1794 if (prot->rsk_prot != NULL) {
1795 static const char mask[] = "request_sock_%s";
1797 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1798 if (request_sock_slab_name == NULL)
1799 goto out_free_sock_slab;
1801 sprintf(request_sock_slab_name, mask, prot->name);
1802 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1803 prot->rsk_prot->obj_size, 0,
1804 SLAB_HWCACHE_ALIGN, NULL);
1806 if (prot->rsk_prot->slab == NULL) {
1807 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1808 prot->name);
1809 goto out_free_request_sock_slab_name;
1813 if (prot->twsk_prot != NULL) {
1814 static const char mask[] = "tw_sock_%s";
1816 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1818 if (timewait_sock_slab_name == NULL)
1819 goto out_free_request_sock_slab;
1821 sprintf(timewait_sock_slab_name, mask, prot->name);
1822 prot->twsk_prot->twsk_slab =
1823 kmem_cache_create(timewait_sock_slab_name,
1824 prot->twsk_prot->twsk_obj_size,
1825 0, SLAB_HWCACHE_ALIGN,
1826 NULL);
1827 if (prot->twsk_prot->twsk_slab == NULL)
1828 goto out_free_timewait_sock_slab_name;
1832 write_lock(&proto_list_lock);
1833 list_add(&prot->node, &proto_list);
1834 write_unlock(&proto_list_lock);
1835 rc = 0;
1836 out:
1837 return rc;
1838 out_free_timewait_sock_slab_name:
1839 kfree(timewait_sock_slab_name);
1840 out_free_request_sock_slab:
1841 if (prot->rsk_prot && prot->rsk_prot->slab) {
1842 kmem_cache_destroy(prot->rsk_prot->slab);
1843 prot->rsk_prot->slab = NULL;
1845 out_free_request_sock_slab_name:
1846 kfree(request_sock_slab_name);
1847 out_free_sock_slab:
1848 kmem_cache_destroy(prot->slab);
1849 prot->slab = NULL;
1850 goto out;
1853 EXPORT_SYMBOL(proto_register);
1855 void proto_unregister(struct proto *prot)
1857 write_lock(&proto_list_lock);
1858 list_del(&prot->node);
1859 write_unlock(&proto_list_lock);
1861 if (prot->slab != NULL) {
1862 kmem_cache_destroy(prot->slab);
1863 prot->slab = NULL;
1866 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1867 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1869 kmem_cache_destroy(prot->rsk_prot->slab);
1870 kfree(name);
1871 prot->rsk_prot->slab = NULL;
1874 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1875 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1877 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1878 kfree(name);
1879 prot->twsk_prot->twsk_slab = NULL;
1883 EXPORT_SYMBOL(proto_unregister);
1885 #ifdef CONFIG_PROC_FS
1886 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1888 read_lock(&proto_list_lock);
1889 return seq_list_start_head(&proto_list, *pos);
1892 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1894 return seq_list_next(v, &proto_list, pos);
1897 static void proto_seq_stop(struct seq_file *seq, void *v)
1899 read_unlock(&proto_list_lock);
1902 static char proto_method_implemented(const void *method)
1904 return method == NULL ? 'n' : 'y';
1907 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1909 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1910 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1911 proto->name,
1912 proto->obj_size,
1913 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1914 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1915 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1916 proto->max_header,
1917 proto->slab == NULL ? "no" : "yes",
1918 module_name(proto->owner),
1919 proto_method_implemented(proto->close),
1920 proto_method_implemented(proto->connect),
1921 proto_method_implemented(proto->disconnect),
1922 proto_method_implemented(proto->accept),
1923 proto_method_implemented(proto->ioctl),
1924 proto_method_implemented(proto->init),
1925 proto_method_implemented(proto->destroy),
1926 proto_method_implemented(proto->shutdown),
1927 proto_method_implemented(proto->setsockopt),
1928 proto_method_implemented(proto->getsockopt),
1929 proto_method_implemented(proto->sendmsg),
1930 proto_method_implemented(proto->recvmsg),
1931 proto_method_implemented(proto->sendpage),
1932 proto_method_implemented(proto->bind),
1933 proto_method_implemented(proto->backlog_rcv),
1934 proto_method_implemented(proto->hash),
1935 proto_method_implemented(proto->unhash),
1936 proto_method_implemented(proto->get_port),
1937 proto_method_implemented(proto->enter_memory_pressure));
1940 static int proto_seq_show(struct seq_file *seq, void *v)
1942 if (v == &proto_list)
1943 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1944 "protocol",
1945 "size",
1946 "sockets",
1947 "memory",
1948 "press",
1949 "maxhdr",
1950 "slab",
1951 "module",
1952 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1953 else
1954 proto_seq_printf(seq, list_entry(v, struct proto, node));
1955 return 0;
1958 static const struct seq_operations proto_seq_ops = {
1959 .start = proto_seq_start,
1960 .next = proto_seq_next,
1961 .stop = proto_seq_stop,
1962 .show = proto_seq_show,
1965 static int proto_seq_open(struct inode *inode, struct file *file)
1967 return seq_open(file, &proto_seq_ops);
1970 static const struct file_operations proto_seq_fops = {
1971 .owner = THIS_MODULE,
1972 .open = proto_seq_open,
1973 .read = seq_read,
1974 .llseek = seq_lseek,
1975 .release = seq_release,
1978 static int __init proto_init(void)
1980 /* register /proc/net/protocols */
1981 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1984 subsys_initcall(proto_init);
1986 #endif /* PROC_FS */
1988 EXPORT_SYMBOL(sk_alloc);
1989 EXPORT_SYMBOL(sk_free);
1990 EXPORT_SYMBOL(sk_send_sigurg);
1991 EXPORT_SYMBOL(sock_alloc_send_skb);
1992 EXPORT_SYMBOL(sock_init_data);
1993 EXPORT_SYMBOL(sock_kfree_s);
1994 EXPORT_SYMBOL(sock_kmalloc);
1995 EXPORT_SYMBOL(sock_no_accept);
1996 EXPORT_SYMBOL(sock_no_bind);
1997 EXPORT_SYMBOL(sock_no_connect);
1998 EXPORT_SYMBOL(sock_no_getname);
1999 EXPORT_SYMBOL(sock_no_getsockopt);
2000 EXPORT_SYMBOL(sock_no_ioctl);
2001 EXPORT_SYMBOL(sock_no_listen);
2002 EXPORT_SYMBOL(sock_no_mmap);
2003 EXPORT_SYMBOL(sock_no_poll);
2004 EXPORT_SYMBOL(sock_no_recvmsg);
2005 EXPORT_SYMBOL(sock_no_sendmsg);
2006 EXPORT_SYMBOL(sock_no_sendpage);
2007 EXPORT_SYMBOL(sock_no_setsockopt);
2008 EXPORT_SYMBOL(sock_no_shutdown);
2009 EXPORT_SYMBOL(sock_no_socketpair);
2010 EXPORT_SYMBOL(sock_rfree);
2011 EXPORT_SYMBOL(sock_setsockopt);
2012 EXPORT_SYMBOL(sock_wfree);
2013 EXPORT_SYMBOL(sock_wmalloc);
2014 EXPORT_SYMBOL(sock_i_uid);
2015 EXPORT_SYMBOL(sock_i_ino);
2016 EXPORT_SYMBOL(sysctl_optmem_max);
2017 #ifdef CONFIG_SYSCTL
2018 EXPORT_SYMBOL(sysctl_rmem_max);
2019 EXPORT_SYMBOL(sysctl_wmem_max);
2020 #endif