?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
net: Fix typo in net/core/sock.c.
[linux-2.6.git] / net / core / sock.c
blob88094cb09c06c0749fc768338d93bf8d4b5208c3
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
11 *
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>
16 *
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
84 *
85 * To Fix:
86 *
87 *
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.
92 */
93
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>
115
116 #include <asm/uaccess.h>
117 #include <asm/system.h>
118
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>
127
128 #include <linux/filter.h>
129
130 #ifdef CONFIG_INET
131 #include <net/tcp.h>
132 #endif
133
134 /*
135 * Each address family might have different locking rules, so we have
136 * one slock key per address family:
137 */
138 static struct lock_class_key af_family_keys[AF_MAX];
139 static struct lock_class_key af_family_slock_keys[AF_MAX];
140
141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
142 /*
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):
146 */
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-AF_CAN" ,
158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
160 };
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-AF_CAN" ,
172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
173 "slock-AF_RXRPC" , "slock-AF_MAX"
174 };
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"
188 };
189 #endif
190
191 /*
192 * sk_callback_lock locking rules are per-address-family,
193 * so split the lock classes by using a per-AF key:
194 */
195 static struct lock_class_key af_callback_keys[AF_MAX];
196
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.
201 */
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)
206
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;
212
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);
215
216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
217 {
218 struct timeval tv;
219
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;
226
227 if (tv.tv_sec < 0) {
228 static int warned __read_mostly;
229
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, task_pid_nr(current));
236 }
237 return 0;
238 }
239 *timeo_p = MAX_SCHEDULE_TIMEOUT;
240 if (tv.tv_sec == 0 && tv.tv_usec == 0)
241 return 0;
242 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
243 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
244 return 0;
245 }
246
247 static void sock_warn_obsolete_bsdism(const char *name)
248 {
249 static int warned;
250 static char warncomm[TASK_COMM_LEN];
251 if (strcmp(warncomm, current->comm) && warned < 5) {
252 strcpy(warncomm, current->comm);
253 printk(KERN_WARNING "process `%s' is using obsolete "
254 "%s SO_BSDCOMPAT\n", warncomm, name);
255 warned++;
256 }
257 }
258
259 static void sock_disable_timestamp(struct sock *sk)
260 {
261 if (sock_flag(sk, SOCK_TIMESTAMP)) {
262 sock_reset_flag(sk, SOCK_TIMESTAMP);
263 net_disable_timestamp();
264 }
265 }
266
267
268 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
269 {
270 int err = 0;
271 int skb_len;
272
273 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
274 number of warnings when compiling with -W --ANK
275 */
276 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
277 (unsigned)sk->sk_rcvbuf) {
278 err = -ENOMEM;
279 goto out;
280 }
281
282 err = sk_filter(sk, skb);
283 if (err)
284 goto out;
285
286 if (!sk_rmem_schedule(sk, skb->truesize)) {
287 err = -ENOBUFS;
288 goto out;
289 }
290
291 skb->dev = NULL;
292 skb_set_owner_r(skb, sk);
293
294 /* Cache the SKB length before we tack it onto the receive
295 * queue. Once it is added it no longer belongs to us and
296 * may be freed by other threads of control pulling packets
297 * from the queue.
298 */
299 skb_len = skb->len;
300
301 skb_queue_tail(&sk->sk_receive_queue, skb);
302
303 if (!sock_flag(sk, SOCK_DEAD))
304 sk->sk_data_ready(sk, skb_len);
305 out:
306 return err;
307 }
308 EXPORT_SYMBOL(sock_queue_rcv_skb);
309
310 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
311 {
312 int rc = NET_RX_SUCCESS;
313
314 if (sk_filter(sk, skb))
315 goto discard_and_relse;
316
317 skb->dev = NULL;
318
319 if (nested)
320 bh_lock_sock_nested(sk);
321 else
322 bh_lock_sock(sk);
323 if (!sock_owned_by_user(sk)) {
324 /*
325 * trylock + unlock semantics:
326 */
327 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
328
329 rc = sk->sk_backlog_rcv(sk, skb);
330
331 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
332 } else
333 sk_add_backlog(sk, skb);
334 bh_unlock_sock(sk);
335 out:
336 sock_put(sk);
337 return rc;
338 discard_and_relse:
339 kfree_skb(skb);
340 goto out;
341 }
342 EXPORT_SYMBOL(sk_receive_skb);
343
344 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
345 {
346 struct dst_entry *dst = sk->sk_dst_cache;
347
348 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
349 sk->sk_dst_cache = NULL;
350 dst_release(dst);
351 return NULL;
352 }
353
354 return dst;
355 }
356 EXPORT_SYMBOL(__sk_dst_check);
357
358 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
359 {
360 struct dst_entry *dst = sk_dst_get(sk);
361
362 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
363 sk_dst_reset(sk);
364 dst_release(dst);
365 return NULL;
366 }
367
368 return dst;
369 }
370 EXPORT_SYMBOL(sk_dst_check);
371
372 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
373 {
374 int ret = -ENOPROTOOPT;
375 #ifdef CONFIG_NETDEVICES
376 struct net *net = sock_net(sk);
377 char devname[IFNAMSIZ];
378 int index;
379
380 /* Sorry... */
381 ret = -EPERM;
382 if (!capable(CAP_NET_RAW))
383 goto out;
384
385 ret = -EINVAL;
386 if (optlen < 0)
387 goto out;
388
389 /* Bind this socket to a particular device like "eth0",
390 * as specified in the passed interface name. If the
391 * name is "" or the option length is zero the socket
392 * is not bound.
393 */
394 if (optlen > IFNAMSIZ - 1)
395 optlen = IFNAMSIZ - 1;
396 memset(devname, 0, sizeof(devname));
397
398 ret = -EFAULT;
399 if (copy_from_user(devname, optval, optlen))
400 goto out;
401
402 if (devname[0] == '\0') {
403 index = 0;
404 } else {
405 struct net_device *dev = dev_get_by_name(net, devname);
406
407 ret = -ENODEV;
408 if (!dev)
409 goto out;
410
411 index = dev->ifindex;
412 dev_put(dev);
413 }
414
415 lock_sock(sk);
416 sk->sk_bound_dev_if = index;
417 sk_dst_reset(sk);
418 release_sock(sk);
419
420 ret = 0;
421
422 out:
423 #endif
424
425 return ret;
426 }
427
428 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
429 {
430 if (valbool)
431 sock_set_flag(sk, bit);
432 else
433 sock_reset_flag(sk, bit);
434 }
435
436 /*
437 * This is meant for all protocols to use and covers goings on
438 * at the socket level. Everything here is generic.
439 */
440
441 int sock_setsockopt(struct socket *sock, int level, int optname,
442 char __user *optval, int optlen)
443 {
444 struct sock *sk=sock->sk;
445 int val;
446 int valbool;
447 struct linger ling;
448 int ret = 0;
449
450 /*
451 * Options without arguments
452 */
453
454 if (optname == SO_BINDTODEVICE)
455 return sock_bindtodevice(sk, optval, optlen);
456
457 if (optlen < sizeof(int))
458 return -EINVAL;
459
460 if (get_user(val, (int __user *)optval))
461 return -EFAULT;
462
463 valbool = val?1:0;
464
465 lock_sock(sk);
466
467 switch(optname) {
468 case SO_DEBUG:
469 if (val && !capable(CAP_NET_ADMIN)) {
470 ret = -EACCES;
471 } else
472 sock_valbool_flag(sk, SOCK_DBG, valbool);
473 break;
474 case SO_REUSEADDR:
475 sk->sk_reuse = valbool;
476 break;
477 case SO_TYPE:
478 case SO_ERROR:
479 ret = -ENOPROTOOPT;
480 break;
481 case SO_DONTROUTE:
482 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
483 break;
484 case SO_BROADCAST:
485 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
486 break;
487 case SO_SNDBUF:
488 /* Don't error on this BSD doesn't and if you think
489 about it this is right. Otherwise apps have to
490 play 'guess the biggest size' games. RCVBUF/SNDBUF
491 are treated in BSD as hints */
492
493 if (val > sysctl_wmem_max)
494 val = sysctl_wmem_max;
495 set_sndbuf:
496 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
497 if ((val * 2) < SOCK_MIN_SNDBUF)
498 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
499 else
500 sk->sk_sndbuf = val * 2;
501
502 /*
503 * Wake up sending tasks if we
504 * upped the value.
505 */
506 sk->sk_write_space(sk);
507 break;
508
509 case SO_SNDBUFFORCE:
510 if (!capable(CAP_NET_ADMIN)) {
511 ret = -EPERM;
512 break;
513 }
514 goto set_sndbuf;
515
516 case SO_RCVBUF:
517 /* Don't error on this BSD doesn't and if you think
518 about it this is right. Otherwise apps have to
519 play 'guess the biggest size' games. RCVBUF/SNDBUF
520 are treated in BSD as hints */
521
522 if (val > sysctl_rmem_max)
523 val = sysctl_rmem_max;
524 set_rcvbuf:
525 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
526 /*
527 * We double it on the way in to account for
528 * "struct sk_buff" etc. overhead. Applications
529 * assume that the SO_RCVBUF setting they make will
530 * allow that much actual data to be received on that
531 * socket.
532 *
533 * Applications are unaware that "struct sk_buff" and
534 * other overheads allocate from the receive buffer
535 * during socket buffer allocation.
536 *
537 * And after considering the possible alternatives,
538 * returning the value we actually used in getsockopt
539 * is the most desirable behavior.
540 */
541 if ((val * 2) < SOCK_MIN_RCVBUF)
542 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
543 else
544 sk->sk_rcvbuf = val * 2;
545 break;
546
547 case SO_RCVBUFFORCE:
548 if (!capable(CAP_NET_ADMIN)) {
549 ret = -EPERM;
550 break;
551 }
552 goto set_rcvbuf;
553
554 case SO_KEEPALIVE:
555 #ifdef CONFIG_INET
556 if (sk->sk_protocol == IPPROTO_TCP)
557 tcp_set_keepalive(sk, valbool);
558 #endif
559 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
560 break;
561
562 case SO_OOBINLINE:
563 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
564 break;
565
566 case SO_NO_CHECK:
567 sk->sk_no_check = valbool;
568 break;
569
570 case SO_PRIORITY:
571 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
572 sk->sk_priority = val;
573 else
574 ret = -EPERM;
575 break;
576
577 case SO_LINGER:
578 if (optlen < sizeof(ling)) {
579 ret = -EINVAL; /* 1003.1g */
580 break;
581 }
582 if (copy_from_user(&ling,optval,sizeof(ling))) {
583 ret = -EFAULT;
584 break;
585 }
586 if (!ling.l_onoff)
587 sock_reset_flag(sk, SOCK_LINGER);
588 else {
589 #if (BITS_PER_LONG == 32)
590 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
591 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
592 else
593 #endif
594 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
595 sock_set_flag(sk, SOCK_LINGER);
596 }
597 break;
598
599 case SO_BSDCOMPAT:
600 sock_warn_obsolete_bsdism("setsockopt");
601 break;
602
603 case SO_PASSCRED:
604 if (valbool)
605 set_bit(SOCK_PASSCRED, &sock->flags);
606 else
607 clear_bit(SOCK_PASSCRED, &sock->flags);
608 break;
609
610 case SO_TIMESTAMP:
611 case SO_TIMESTAMPNS:
612 if (valbool) {
613 if (optname == SO_TIMESTAMP)
614 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
615 else
616 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
617 sock_set_flag(sk, SOCK_RCVTSTAMP);
618 sock_enable_timestamp(sk);
619 } else {
620 sock_reset_flag(sk, SOCK_RCVTSTAMP);
621 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
622 }
623 break;
624
625 case SO_RCVLOWAT:
626 if (val < 0)
627 val = INT_MAX;
628 sk->sk_rcvlowat = val ? : 1;
629 break;
630
631 case SO_RCVTIMEO:
632 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
633 break;
634
635 case SO_SNDTIMEO:
636 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
637 break;
638
639 case SO_ATTACH_FILTER:
640 ret = -EINVAL;
641 if (optlen == sizeof(struct sock_fprog)) {
642 struct sock_fprog fprog;
643
644 ret = -EFAULT;
645 if (copy_from_user(&fprog, optval, sizeof(fprog)))
646 break;
647
648 ret = sk_attach_filter(&fprog, sk);
649 }
650 break;
651
652 case SO_DETACH_FILTER:
653 ret = sk_detach_filter(sk);
654 break;
655
656 case SO_PASSSEC:
657 if (valbool)
658 set_bit(SOCK_PASSSEC, &sock->flags);
659 else
660 clear_bit(SOCK_PASSSEC, &sock->flags);
661 break;
662 case SO_MARK:
663 if (!capable(CAP_NET_ADMIN))
664 ret = -EPERM;
665 else {
666 sk->sk_mark = val;
667 }
668 break;
669
670 /* We implement the SO_SNDLOWAT etc to
671 not be settable (1003.1g 5.3) */
672 default:
673 ret = -ENOPROTOOPT;
674 break;
675 }
676 release_sock(sk);
677 return ret;
678 }
679
680
681 int sock_getsockopt(struct socket *sock, int level, int optname,
682 char __user *optval, int __user *optlen)
683 {
684 struct sock *sk = sock->sk;
685
686 union {
687 int val;
688 struct linger ling;
689 struct timeval tm;
690 } v;
691
692 unsigned int lv = sizeof(int);
693 int len;
694
695 if (get_user(len, optlen))
696 return -EFAULT;
697 if (len < 0)
698 return -EINVAL;
699
700 switch(optname) {
701 case SO_DEBUG:
702 v.val = sock_flag(sk, SOCK_DBG);
703 break;
704
705 case SO_DONTROUTE:
706 v.val = sock_flag(sk, SOCK_LOCALROUTE);
707 break;
708
709 case SO_BROADCAST:
710 v.val = !!sock_flag(sk, SOCK_BROADCAST);
711 break;
712
713 case SO_SNDBUF:
714 v.val = sk->sk_sndbuf;
715 break;
716
717 case SO_RCVBUF:
718 v.val = sk->sk_rcvbuf;
719 break;
720
721 case SO_REUSEADDR:
722 v.val = sk->sk_reuse;
723 break;
724
725 case SO_KEEPALIVE:
726 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
727 break;
728
729 case SO_TYPE:
730 v.val = sk->sk_type;
731 break;
732
733 case SO_ERROR:
734 v.val = -sock_error(sk);
735 if (v.val==0)
736 v.val = xchg(&sk->sk_err_soft, 0);
737 break;
738
739 case SO_OOBINLINE:
740 v.val = !!sock_flag(sk, SOCK_URGINLINE);
741 break;
742
743 case SO_NO_CHECK:
744 v.val = sk->sk_no_check;
745 break;
746
747 case SO_PRIORITY:
748 v.val = sk->sk_priority;
749 break;
750
751 case SO_LINGER:
752 lv = sizeof(v.ling);
753 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
754 v.ling.l_linger = sk->sk_lingertime / HZ;
755 break;
756
757 case SO_BSDCOMPAT:
758 sock_warn_obsolete_bsdism("getsockopt");
759 break;
760
761 case SO_TIMESTAMP:
762 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
763 !sock_flag(sk, SOCK_RCVTSTAMPNS);
764 break;
765
766 case SO_TIMESTAMPNS:
767 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
768 break;
769
770 case SO_RCVTIMEO:
771 lv=sizeof(struct timeval);
772 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
773 v.tm.tv_sec = 0;
774 v.tm.tv_usec = 0;
775 } else {
776 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
777 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
778 }
779 break;
780
781 case SO_SNDTIMEO:
782 lv=sizeof(struct timeval);
783 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
784 v.tm.tv_sec = 0;
785 v.tm.tv_usec = 0;
786 } else {
787 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
788 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
789 }
790 break;
791
792 case SO_RCVLOWAT:
793 v.val = sk->sk_rcvlowat;
794 break;
795
796 case SO_SNDLOWAT:
797 v.val=1;
798 break;
799
800 case SO_PASSCRED:
801 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
802 break;
803
804 case SO_PEERCRED:
805 if (len > sizeof(sk->sk_peercred))
806 len = sizeof(sk->sk_peercred);
807 if (copy_to_user(optval, &sk->sk_peercred, len))
808 return -EFAULT;
809 goto lenout;
810
811 case SO_PEERNAME:
812 {
813 char address[128];
814
815 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
816 return -ENOTCONN;
817 if (lv < len)
818 return -EINVAL;
819 if (copy_to_user(optval, address, len))
820 return -EFAULT;
821 goto lenout;
822 }
823
824 /* Dubious BSD thing... Probably nobody even uses it, but
825 * the UNIX standard wants it for whatever reason... -DaveM
826 */
827 case SO_ACCEPTCONN:
828 v.val = sk->sk_state == TCP_LISTEN;
829 break;
830
831 case SO_PASSSEC:
832 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
833 break;
834
835 case SO_PEERSEC:
836 return security_socket_getpeersec_stream(sock, optval, optlen, len);
837
838 case SO_MARK:
839 v.val = sk->sk_mark;
840 break;
841
842 default:
843 return -ENOPROTOOPT;
844 }
845
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;
854 }
855
856 /*
857 * Initialize an sk_lock.
858 *
859 * (We also register the sk_lock with the lock validator.)
860 */
861 static inline void sock_lock_init(struct sock *sk)
862 {
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);
868 }
869
870 static void sock_copy(struct sock *nsk, const struct sock *osk)
871 {
872 #ifdef CONFIG_SECURITY_NETWORK
873 void *sptr = nsk->sk_security;
874 #endif
875
876 memcpy(nsk, osk, osk->sk_prot->obj_size);
877 #ifdef CONFIG_SECURITY_NETWORK
878 nsk->sk_security = sptr;
879 security_sk_clone(osk, nsk);
880 #endif
881 }
882
883 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
884 int family)
885 {
886 struct sock *sk;
887 struct kmem_cache *slab;
888
889 slab = prot->slab;
890 if (slab != NULL)
891 sk = kmem_cache_alloc(slab, priority);
892 else
893 sk = kmalloc(prot->obj_size, priority);
894
895 if (sk != NULL) {
896 if (security_sk_alloc(sk, family, priority))
897 goto out_free;
898
899 if (!try_module_get(prot->owner))
900 goto out_free_sec;
901 }
902
903 return sk;
904
905 out_free_sec:
906 security_sk_free(sk);
907 out_free:
908 if (slab != NULL)
909 kmem_cache_free(slab, sk);
910 else
911 kfree(sk);
912 return NULL;
913 }
914
915 static void sk_prot_free(struct proto *prot, struct sock *sk)
916 {
917 struct kmem_cache *slab;
918 struct module *owner;
919
920 owner = prot->owner;
921 slab = prot->slab;
922
923 security_sk_free(sk);
924 if (slab != NULL)
925 kmem_cache_free(slab, sk);
926 else
927 kfree(sk);
928 module_put(owner);
929 }
930
931 /**
932 * sk_alloc - All socket objects are allocated here
933 * @net: the applicable net namespace
934 * @family: protocol family
935 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
936 * @prot: struct proto associated with this new sock instance
937 */
938 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
939 struct proto *prot)
940 {
941 struct sock *sk;
942
943 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
944 if (sk) {
945 sk->sk_family = family;
946 /*
947 * See comment in struct sock definition to understand
948 * why we need sk_prot_creator -acme
949 */
950 sk->sk_prot = sk->sk_prot_creator = prot;
951 sock_lock_init(sk);
952 sock_net_set(sk, get_net(net));
953 }
954
955 return sk;
956 }
957
958 void sk_free(struct sock *sk)
959 {
960 struct sk_filter *filter;
961
962 if (sk->sk_destruct)
963 sk->sk_destruct(sk);
964
965 filter = rcu_dereference(sk->sk_filter);
966 if (filter) {
967 sk_filter_uncharge(sk, filter);
968 rcu_assign_pointer(sk->sk_filter, NULL);
969 }
970
971 sock_disable_timestamp(sk);
972
973 if (atomic_read(&sk->sk_omem_alloc))
974 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
975 __func__, atomic_read(&sk->sk_omem_alloc));
976
977 put_net(sock_net(sk));
978 sk_prot_free(sk->sk_prot_creator, sk);
979 }
980
981 /*
982 * Last sock_put should drop referrence to sk->sk_net. It has already
983 * been dropped in sk_change_net. Taking referrence to stopping namespace
984 * is not an option.
985 * Take referrence to a socket to remove it from hash _alive_ and after that
986 * destroy it in the context of init_net.
987 */
988 void sk_release_kernel(struct sock *sk)
989 {
990 if (sk == NULL || sk->sk_socket == NULL)
991 return;
992
993 sock_hold(sk);
994 sock_release(sk->sk_socket);
995 release_net(sock_net(sk));
996 sock_net_set(sk, get_net(&init_net));
997 sock_put(sk);
998 }
999 EXPORT_SYMBOL(sk_release_kernel);
1000
1001 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1002 {
1003 struct sock *newsk;
1004
1005 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1006 if (newsk != NULL) {
1007 struct sk_filter *filter;
1008
1009 sock_copy(newsk, sk);
1010
1011 /* SANITY */
1012 get_net(sock_net(newsk));
1013 sk_node_init(&newsk->sk_node);
1014 sock_lock_init(newsk);
1015 bh_lock_sock(newsk);
1016 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1017
1018 atomic_set(&newsk->sk_rmem_alloc, 0);
1019 atomic_set(&newsk->sk_wmem_alloc, 0);
1020 atomic_set(&newsk->sk_omem_alloc, 0);
1021 skb_queue_head_init(&newsk->sk_receive_queue);
1022 skb_queue_head_init(&newsk->sk_write_queue);
1023 #ifdef CONFIG_NET_DMA
1024 skb_queue_head_init(&newsk->sk_async_wait_queue);
1025 #endif
1026
1027 rwlock_init(&newsk->sk_dst_lock);
1028 rwlock_init(&newsk->sk_callback_lock);
1029 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1030 af_callback_keys + newsk->sk_family,
1031 af_family_clock_key_strings[newsk->sk_family]);
1032
1033 newsk->sk_dst_cache = NULL;
1034 newsk->sk_wmem_queued = 0;
1035 newsk->sk_forward_alloc = 0;
1036 newsk->sk_send_head = NULL;
1037 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1038
1039 sock_reset_flag(newsk, SOCK_DONE);
1040 skb_queue_head_init(&newsk->sk_error_queue);
1041
1042 filter = newsk->sk_filter;
1043 if (filter != NULL)
1044 sk_filter_charge(newsk, filter);
1045
1046 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1047 /* It is still raw copy of parent, so invalidate
1048 * destructor and make plain sk_free() */
1049 newsk->sk_destruct = NULL;
1050 sk_free(newsk);
1051 newsk = NULL;
1052 goto out;
1053 }
1054
1055 newsk->sk_err = 0;
1056 newsk->sk_priority = 0;
1057 atomic_set(&newsk->sk_refcnt, 2);
1058
1059 /*
1060 * Increment the counter in the same struct proto as the master
1061 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1062 * is the same as sk->sk_prot->socks, as this field was copied
1063 * with memcpy).
1064 *
1065 * This _changes_ the previous behaviour, where
1066 * tcp_create_openreq_child always was incrementing the
1067 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1068 * to be taken into account in all callers. -acme
1069 */
1070 sk_refcnt_debug_inc(newsk);
1071 newsk->sk_socket = NULL;
1072 newsk->sk_sleep = NULL;
1073
1074 if (newsk->sk_prot->sockets_allocated)
1075 atomic_inc(newsk->sk_prot->sockets_allocated);
1076 }
1077 out:
1078 return newsk;
1079 }
1080
1081 EXPORT_SYMBOL_GPL(sk_clone);
1082
1083 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1084 {
1085 __sk_dst_set(sk, dst);
1086 sk->sk_route_caps = dst->dev->features;
1087 if (sk->sk_route_caps & NETIF_F_GSO)
1088 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1089 if (sk_can_gso(sk)) {
1090 if (dst->header_len) {
1091 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1092 } else {
1093 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1094 sk->sk_gso_max_size = dst->dev->gso_max_size;
1095 }
1096 }
1097 }
1098 EXPORT_SYMBOL_GPL(sk_setup_caps);
1099
1100 void __init sk_init(void)
1101 {
1102 if (num_physpages <= 4096) {
1103 sysctl_wmem_max = 32767;
1104 sysctl_rmem_max = 32767;
1105 sysctl_wmem_default = 32767;
1106 sysctl_rmem_default = 32767;
1107 } else if (num_physpages >= 131072) {
1108 sysctl_wmem_max = 131071;
1109 sysctl_rmem_max = 131071;
1110 }
1111 }
1112
1113 /*
1114 * Simple resource managers for sockets.
1115 */
1116
1117
1118 /*
1119 * Write buffer destructor automatically called from kfree_skb.
1120 */
1121 void sock_wfree(struct sk_buff *skb)
1122 {
1123 struct sock *sk = skb->sk;
1124
1125 /* In case it might be waiting for more memory. */
1126 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1127 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1128 sk->sk_write_space(sk);
1129 sock_put(sk);
1130 }
1131
1132 /*
1133 * Read buffer destructor automatically called from kfree_skb.
1134 */
1135 void sock_rfree(struct sk_buff *skb)
1136 {
1137 struct sock *sk = skb->sk;
1138
1139 skb_truesize_check(skb);
1140 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1141 sk_mem_uncharge(skb->sk, skb->truesize);
1142 }
1143
1144
1145 int sock_i_uid(struct sock *sk)
1146 {
1147 int uid;
1148
1149 read_lock(&sk->sk_callback_lock);
1150 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1151 read_unlock(&sk->sk_callback_lock);
1152 return uid;
1153 }
1154
1155 unsigned long sock_i_ino(struct sock *sk)
1156 {
1157 unsigned long ino;
1158
1159 read_lock(&sk->sk_callback_lock);
1160 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1161 read_unlock(&sk->sk_callback_lock);
1162 return ino;
1163 }
1164
1165 /*
1166 * Allocate a skb from the socket's send buffer.
1167 */
1168 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1169 gfp_t priority)
1170 {
1171 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1172 struct sk_buff * skb = alloc_skb(size, priority);
1173 if (skb) {
1174 skb_set_owner_w(skb, sk);
1175 return skb;
1176 }
1177 }
1178 return NULL;
1179 }
1180
1181 /*
1182 * Allocate a skb from the socket's receive buffer.
1183 */
1184 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1185 gfp_t priority)
1186 {
1187 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1188 struct sk_buff *skb = alloc_skb(size, priority);
1189 if (skb) {
1190 skb_set_owner_r(skb, sk);
1191 return skb;
1192 }
1193 }
1194 return NULL;
1195 }
1196
1197 /*
1198 * Allocate a memory block from the socket's option memory buffer.
1199 */
1200 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1201 {
1202 if ((unsigned)size <= sysctl_optmem_max &&
1203 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1204 void *mem;
1205 /* First do the add, to avoid the race if kmalloc
1206 * might sleep.
1207 */
1208 atomic_add(size, &sk->sk_omem_alloc);
1209 mem = kmalloc(size, priority);
1210 if (mem)
1211 return mem;
1212 atomic_sub(size, &sk->sk_omem_alloc);
1213 }
1214 return NULL;
1215 }
1216
1217 /*
1218 * Free an option memory block.
1219 */
1220 void sock_kfree_s(struct sock *sk, void *mem, int size)
1221 {
1222 kfree(mem);
1223 atomic_sub(size, &sk->sk_omem_alloc);
1224 }
1225
1226 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1227 I think, these locks should be removed for datagram sockets.
1228 */
1229 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1230 {
1231 DEFINE_WAIT(wait);
1232
1233 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1234 for (;;) {
1235 if (!timeo)
1236 break;
1237 if (signal_pending(current))
1238 break;
1239 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1240 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1241 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1242 break;
1243 if (sk->sk_shutdown & SEND_SHUTDOWN)
1244 break;
1245 if (sk->sk_err)
1246 break;
1247 timeo = schedule_timeout(timeo);
1248 }
1249 finish_wait(sk->sk_sleep, &wait);
1250 return timeo;
1251 }
1252
1253
1254 /*
1255 * Generic send/receive buffer handlers
1256 */
1257
1258 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1259 unsigned long header_len,
1260 unsigned long data_len,
1261 int noblock, int *errcode)
1262 {
1263 struct sk_buff *skb;
1264 gfp_t gfp_mask;
1265 long timeo;
1266 int err;
1267
1268 gfp_mask = sk->sk_allocation;
1269 if (gfp_mask & __GFP_WAIT)
1270 gfp_mask |= __GFP_REPEAT;
1271
1272 timeo = sock_sndtimeo(sk, noblock);
1273 while (1) {
1274 err = sock_error(sk);
1275 if (err != 0)
1276 goto failure;
1277
1278 err = -EPIPE;
1279 if (sk->sk_shutdown & SEND_SHUTDOWN)
1280 goto failure;
1281
1282 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1283 skb = alloc_skb(header_len, gfp_mask);
1284 if (skb) {
1285 int npages;
1286 int i;
1287
1288 /* No pages, we're done... */
1289 if (!data_len)
1290 break;
1291
1292 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1293 skb->truesize += data_len;
1294 skb_shinfo(skb)->nr_frags = npages;
1295 for (i = 0; i < npages; i++) {
1296 struct page *page;
1297 skb_frag_t *frag;
1298
1299 page = alloc_pages(sk->sk_allocation, 0);
1300 if (!page) {
1301 err = -ENOBUFS;
1302 skb_shinfo(skb)->nr_frags = i;
1303 kfree_skb(skb);
1304 goto failure;
1305 }
1306
1307 frag = &skb_shinfo(skb)->frags[i];
1308 frag->page = page;
1309 frag->page_offset = 0;
1310 frag->size = (data_len >= PAGE_SIZE ?
1311 PAGE_SIZE :
1312 data_len);
1313 data_len -= PAGE_SIZE;
1314 }
1315
1316 /* Full success... */
1317 break;
1318 }
1319 err = -ENOBUFS;
1320 goto failure;
1321 }
1322 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1323 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1324 err = -EAGAIN;
1325 if (!timeo)
1326 goto failure;
1327 if (signal_pending(current))
1328 goto interrupted;
1329 timeo = sock_wait_for_wmem(sk, timeo);
1330 }
1331
1332 skb_set_owner_w(skb, sk);
1333 return skb;
1334
1335 interrupted:
1336 err = sock_intr_errno(timeo);
1337 failure:
1338 *errcode = err;
1339 return NULL;
1340 }
1341
1342 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1343 int noblock, int *errcode)
1344 {
1345 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1346 }
1347
1348 static void __lock_sock(struct sock *sk)
1349 {
1350 DEFINE_WAIT(wait);
1351
1352 for (;;) {
1353 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1354 TASK_UNINTERRUPTIBLE);
1355 spin_unlock_bh(&sk->sk_lock.slock);
1356 schedule();
1357 spin_lock_bh(&sk->sk_lock.slock);
1358 if (!sock_owned_by_user(sk))
1359 break;
1360 }
1361 finish_wait(&sk->sk_lock.wq, &wait);
1362 }
1363
1364 static void __release_sock(struct sock *sk)
1365 {
1366 struct sk_buff *skb = sk->sk_backlog.head;
1367
1368 do {
1369 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1370 bh_unlock_sock(sk);
1371
1372 do {
1373 struct sk_buff *next = skb->next;
1374
1375 skb->next = NULL;
1376 sk->sk_backlog_rcv(sk, skb);
1377
1378 /*
1379 * We are in process context here with softirqs
1380 * disabled, use cond_resched_softirq() to preempt.
1381 * This is safe to do because we've taken the backlog
1382 * queue private:
1383 */
1384 cond_resched_softirq();
1385
1386 skb = next;
1387 } while (skb != NULL);
1388
1389 bh_lock_sock(sk);
1390 } while ((skb = sk->sk_backlog.head) != NULL);
1391 }
1392
1393 /**
1394 * sk_wait_data - wait for data to arrive at sk_receive_queue
1395 * @sk: sock to wait on
1396 * @timeo: for how long
1397 *
1398 * Now socket state including sk->sk_err is changed only under lock,
1399 * hence we may omit checks after joining wait queue.
1400 * We check receive queue before schedule() only as optimization;
1401 * it is very likely that release_sock() added new data.
1402 */
1403 int sk_wait_data(struct sock *sk, long *timeo)
1404 {
1405 int rc;
1406 DEFINE_WAIT(wait);
1407
1408 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1409 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1410 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1411 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1412 finish_wait(sk->sk_sleep, &wait);
1413 return rc;
1414 }
1415
1416 EXPORT_SYMBOL(sk_wait_data);
1417
1418 /**
1419 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1420 * @sk: socket
1421 * @size: memory size to allocate
1422 * @kind: allocation type
1423 *
1424 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1425 * rmem allocation. This function assumes that protocols which have
1426 * memory_pressure use sk_wmem_queued as write buffer accounting.
1427 */
1428 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1429 {
1430 struct proto *prot = sk->sk_prot;
1431 int amt = sk_mem_pages(size);
1432 int allocated;
1433
1434 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1435 allocated = atomic_add_return(amt, prot->memory_allocated);
1436
1437 /* Under limit. */
1438 if (allocated <= prot->sysctl_mem[0]) {
1439 if (prot->memory_pressure && *prot->memory_pressure)
1440 *prot->memory_pressure = 0;
1441 return 1;
1442 }
1443
1444 /* Under pressure. */
1445 if (allocated > prot->sysctl_mem[1])
1446 if (prot->enter_memory_pressure)
1447 prot->enter_memory_pressure();
1448
1449 /* Over hard limit. */
1450 if (allocated > prot->sysctl_mem[2])
1451 goto suppress_allocation;
1452
1453 /* guarantee minimum buffer size under pressure */
1454 if (kind == SK_MEM_RECV) {
1455 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1456 return 1;
1457 } else { /* SK_MEM_SEND */
1458 if (sk->sk_type == SOCK_STREAM) {
1459 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1460 return 1;
1461 } else if (atomic_read(&sk->sk_wmem_alloc) <
1462 prot->sysctl_wmem[0])
1463 return 1;
1464 }
1465
1466 if (prot->memory_pressure) {
1467 if (!*prot->memory_pressure ||
1468 prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) *
1469 sk_mem_pages(sk->sk_wmem_queued +
1470 atomic_read(&sk->sk_rmem_alloc) +
1471 sk->sk_forward_alloc))
1472 return 1;
1473 }
1474
1475 suppress_allocation:
1476
1477 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1478 sk_stream_moderate_sndbuf(sk);
1479
1480 /* Fail only if socket is _under_ its sndbuf.
1481 * In this case we cannot block, so that we have to fail.
1482 */
1483 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1484 return 1;
1485 }
1486
1487 /* Alas. Undo changes. */
1488 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1489 atomic_sub(amt, prot->memory_allocated);
1490 return 0;
1491 }
1492
1493 EXPORT_SYMBOL(__sk_mem_schedule);
1494
1495 /**
1496 * __sk_reclaim - reclaim memory_allocated
1497 * @sk: socket
1498 */
1499 void __sk_mem_reclaim(struct sock *sk)
1500 {
1501 struct proto *prot = sk->sk_prot;
1502
1503 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1504 prot->memory_allocated);
1505 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1506
1507 if (prot->memory_pressure && *prot->memory_pressure &&
1508 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1509 *prot->memory_pressure = 0;
1510 }
1511
1512 EXPORT_SYMBOL(__sk_mem_reclaim);
1513
1514
1515 /*
1516 * Set of default routines for initialising struct proto_ops when
1517 * the protocol does not support a particular function. In certain
1518 * cases where it makes no sense for a protocol to have a "do nothing"
1519 * function, some default processing is provided.
1520 */
1521
1522 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1523 {
1524 return -EOPNOTSUPP;
1525 }
1526
1527 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1528 int len, int flags)
1529 {
1530 return -EOPNOTSUPP;
1531 }
1532
1533 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1534 {
1535 return -EOPNOTSUPP;
1536 }
1537
1538 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1539 {
1540 return -EOPNOTSUPP;
1541 }
1542
1543 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1544 int *len, int peer)
1545 {
1546 return -EOPNOTSUPP;
1547 }
1548
1549 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1550 {
1551 return 0;
1552 }
1553
1554 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1555 {
1556 return -EOPNOTSUPP;
1557 }
1558
1559 int sock_no_listen(struct socket *sock, int backlog)
1560 {
1561 return -EOPNOTSUPP;
1562 }
1563
1564 int sock_no_shutdown(struct socket *sock, int how)
1565 {
1566 return -EOPNOTSUPP;
1567 }
1568
1569 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1570 char __user *optval, int optlen)
1571 {
1572 return -EOPNOTSUPP;
1573 }
1574
1575 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1576 char __user *optval, int __user *optlen)
1577 {
1578 return -EOPNOTSUPP;
1579 }
1580
1581 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1582 size_t len)
1583 {
1584 return -EOPNOTSUPP;
1585 }
1586
1587 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1588 size_t len, int flags)
1589 {
1590 return -EOPNOTSUPP;
1591 }
1592
1593 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1594 {
1595 /* Mirror missing mmap method error code */
1596 return -ENODEV;
1597 }
1598
1599 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1600 {
1601 ssize_t res;
1602 struct msghdr msg = {.msg_flags = flags};
1603 struct kvec iov;
1604 char *kaddr = kmap(page);
1605 iov.iov_base = kaddr + offset;
1606 iov.iov_len = size;
1607 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1608 kunmap(page);
1609 return res;
1610 }
1611
1612 /*
1613 * Default Socket Callbacks
1614 */
1615
1616 static void sock_def_wakeup(struct sock *sk)
1617 {
1618 read_lock(&sk->sk_callback_lock);
1619 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1620 wake_up_interruptible_all(sk->sk_sleep);
1621 read_unlock(&sk->sk_callback_lock);
1622 }
1623
1624 static void sock_def_error_report(struct sock *sk)
1625 {
1626 read_lock(&sk->sk_callback_lock);
1627 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1628 wake_up_interruptible(sk->sk_sleep);
1629 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1630 read_unlock(&sk->sk_callback_lock);
1631 }
1632
1633 static void sock_def_readable(struct sock *sk, int len)
1634 {
1635 read_lock(&sk->sk_callback_lock);
1636 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1637 wake_up_interruptible_sync(sk->sk_sleep);
1638 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1639 read_unlock(&sk->sk_callback_lock);
1640 }
1641
1642 static void sock_def_write_space(struct sock *sk)
1643 {
1644 read_lock(&sk->sk_callback_lock);
1645
1646 /* Do not wake up a writer until he can make "significant"
1647 * progress. --DaveM
1648 */
1649 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1650 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1651 wake_up_interruptible_sync(sk->sk_sleep);
1652
1653 /* Should agree with poll, otherwise some programs break */
1654 if (sock_writeable(sk))
1655 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1656 }
1657
1658 read_unlock(&sk->sk_callback_lock);
1659 }
1660
1661 static void sock_def_destruct(struct sock *sk)
1662 {
1663 kfree(sk->sk_protinfo);
1664 }
1665
1666 void sk_send_sigurg(struct sock *sk)
1667 {
1668 if (sk->sk_socket && sk->sk_socket->file)
1669 if (send_sigurg(&sk->sk_socket->file->f_owner))
1670 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1671 }
1672
1673 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1674 unsigned long expires)
1675 {
1676 if (!mod_timer(timer, expires))
1677 sock_hold(sk);
1678 }
1679
1680 EXPORT_SYMBOL(sk_reset_timer);
1681
1682 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1683 {
1684 if (timer_pending(timer) && del_timer(timer))
1685 __sock_put(sk);
1686 }
1687
1688 EXPORT_SYMBOL(sk_stop_timer);
1689
1690 void sock_init_data(struct socket *sock, struct sock *sk)
1691 {
1692 skb_queue_head_init(&sk->sk_receive_queue);
1693 skb_queue_head_init(&sk->sk_write_queue);
1694 skb_queue_head_init(&sk->sk_error_queue);
1695 #ifdef CONFIG_NET_DMA
1696 skb_queue_head_init(&sk->sk_async_wait_queue);
1697 #endif
1698
1699 sk->sk_send_head = NULL;
1700
1701 init_timer(&sk->sk_timer);
1702
1703 sk->sk_allocation = GFP_KERNEL;
1704 sk->sk_rcvbuf = sysctl_rmem_default;
1705 sk->sk_sndbuf = sysctl_wmem_default;
1706 sk->sk_state = TCP_CLOSE;
1707 sk->sk_socket = sock;
1708
1709 sock_set_flag(sk, SOCK_ZAPPED);
1710
1711 if (sock) {
1712 sk->sk_type = sock->type;
1713 sk->sk_sleep = &sock->wait;
1714 sock->sk = sk;
1715 } else
1716 sk->sk_sleep = NULL;
1717
1718 rwlock_init(&sk->sk_dst_lock);
1719 rwlock_init(&sk->sk_callback_lock);
1720 lockdep_set_class_and_name(&sk->sk_callback_lock,
1721 af_callback_keys + sk->sk_family,
1722 af_family_clock_key_strings[sk->sk_family]);
1723
1724 sk->sk_state_change = sock_def_wakeup;
1725 sk->sk_data_ready = sock_def_readable;
1726 sk->sk_write_space = sock_def_write_space;
1727 sk->sk_error_report = sock_def_error_report;
1728 sk->sk_destruct = sock_def_destruct;
1729
1730 sk->sk_sndmsg_page = NULL;
1731 sk->sk_sndmsg_off = 0;
1732
1733 sk->sk_peercred.pid = 0;
1734 sk->sk_peercred.uid = -1;
1735 sk->sk_peercred.gid = -1;
1736 sk->sk_write_pending = 0;
1737 sk->sk_rcvlowat = 1;
1738 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1739 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1740
1741 sk->sk_stamp = ktime_set(-1L, 0);
1742
1743 atomic_set(&sk->sk_refcnt, 1);
1744 atomic_set(&sk->sk_drops, 0);
1745 }
1746
1747 void lock_sock_nested(struct sock *sk, int subclass)
1748 {
1749 might_sleep();
1750 spin_lock_bh(&sk->sk_lock.slock);
1751 if (sk->sk_lock.owned)
1752 __lock_sock(sk);
1753 sk->sk_lock.owned = 1;
1754 spin_unlock(&sk->sk_lock.slock);
1755 /*
1756 * The sk_lock has mutex_lock() semantics here:
1757 */
1758 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1759 local_bh_enable();
1760 }
1761
1762 EXPORT_SYMBOL(lock_sock_nested);
1763
1764 void release_sock(struct sock *sk)
1765 {
1766 /*
1767 * The sk_lock has mutex_unlock() semantics:
1768 */
1769 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1770
1771 spin_lock_bh(&sk->sk_lock.slock);
1772 if (sk->sk_backlog.tail)
1773 __release_sock(sk);
1774 sk->sk_lock.owned = 0;
1775 if (waitqueue_active(&sk->sk_lock.wq))
1776 wake_up(&sk->sk_lock.wq);
1777 spin_unlock_bh(&sk->sk_lock.slock);
1778 }
1779 EXPORT_SYMBOL(release_sock);
1780
1781 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1782 {
1783 struct timeval tv;
1784 if (!sock_flag(sk, SOCK_TIMESTAMP))
1785 sock_enable_timestamp(sk);
1786 tv = ktime_to_timeval(sk->sk_stamp);
1787 if (tv.tv_sec == -1)
1788 return -ENOENT;
1789 if (tv.tv_sec == 0) {
1790 sk->sk_stamp = ktime_get_real();
1791 tv = ktime_to_timeval(sk->sk_stamp);
1792 }
1793 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1794 }
1795 EXPORT_SYMBOL(sock_get_timestamp);
1796
1797 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1798 {
1799 struct timespec ts;
1800 if (!sock_flag(sk, SOCK_TIMESTAMP))
1801 sock_enable_timestamp(sk);
1802 ts = ktime_to_timespec(sk->sk_stamp);
1803 if (ts.tv_sec == -1)
1804 return -ENOENT;
1805 if (ts.tv_sec == 0) {
1806 sk->sk_stamp = ktime_get_real();
1807 ts = ktime_to_timespec(sk->sk_stamp);
1808 }
1809 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1810 }
1811 EXPORT_SYMBOL(sock_get_timestampns);
1812
1813 void sock_enable_timestamp(struct sock *sk)
1814 {
1815 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1816 sock_set_flag(sk, SOCK_TIMESTAMP);
1817 net_enable_timestamp();
1818 }
1819 }
1820
1821 /*
1822 * Get a socket option on an socket.
1823 *
1824 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1825 * asynchronous errors should be reported by getsockopt. We assume
1826 * this means if you specify SO_ERROR (otherwise whats the point of it).
1827 */
1828 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1829 char __user *optval, int __user *optlen)
1830 {
1831 struct sock *sk = sock->sk;
1832
1833 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1834 }
1835
1836 EXPORT_SYMBOL(sock_common_getsockopt);
1837
1838 #ifdef CONFIG_COMPAT
1839 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1840 char __user *optval, int __user *optlen)
1841 {
1842 struct sock *sk = sock->sk;
1843
1844 if (sk->sk_prot->compat_getsockopt != NULL)
1845 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1846 optval, optlen);
1847 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1848 }
1849 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1850 #endif
1851
1852 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1853 struct msghdr *msg, size_t size, int flags)
1854 {
1855 struct sock *sk = sock->sk;
1856 int addr_len = 0;
1857 int err;
1858
1859 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1860 flags & ~MSG_DONTWAIT, &addr_len);
1861 if (err >= 0)
1862 msg->msg_namelen = addr_len;
1863 return err;
1864 }
1865
1866 EXPORT_SYMBOL(sock_common_recvmsg);
1867
1868 /*
1869 * Set socket options on an inet socket.
1870 */
1871 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1872 char __user *optval, int optlen)
1873 {
1874 struct sock *sk = sock->sk;
1875
1876 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1877 }
1878
1879 EXPORT_SYMBOL(sock_common_setsockopt);
1880
1881 #ifdef CONFIG_COMPAT
1882 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1883 char __user *optval, int optlen)
1884 {
1885 struct sock *sk = sock->sk;
1886
1887 if (sk->sk_prot->compat_setsockopt != NULL)
1888 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1889 optval, optlen);
1890 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1891 }
1892 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1893 #endif
1894
1895 void sk_common_release(struct sock *sk)
1896 {
1897 if (sk->sk_prot->destroy)
1898 sk->sk_prot->destroy(sk);
1899
1900 /*
1901 * Observation: when sock_common_release is called, processes have
1902 * no access to socket. But net still has.
1903 * Step one, detach it from networking:
1904 *
1905 * A. Remove from hash tables.
1906 */
1907
1908 sk->sk_prot->unhash(sk);
1909
1910 /*
1911 * In this point socket cannot receive new packets, but it is possible
1912 * that some packets are in flight because some CPU runs receiver and
1913 * did hash table lookup before we unhashed socket. They will achieve
1914 * receive queue and will be purged by socket destructor.
1915 *
1916 * Also we still have packets pending on receive queue and probably,
1917 * our own packets waiting in device queues. sock_destroy will drain
1918 * receive queue, but transmitted packets will delay socket destruction
1919 * until the last reference will be released.
1920 */
1921
1922 sock_orphan(sk);
1923
1924 xfrm_sk_free_policy(sk);
1925
1926 sk_refcnt_debug_release(sk);
1927 sock_put(sk);
1928 }
1929
1930 EXPORT_SYMBOL(sk_common_release);
1931
1932 static DEFINE_RWLOCK(proto_list_lock);
1933 static LIST_HEAD(proto_list);
1934
1935 #ifdef CONFIG_PROC_FS
1936 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1937 struct prot_inuse {
1938 int val[PROTO_INUSE_NR];
1939 };
1940
1941 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
1942
1943 #ifdef CONFIG_NET_NS
1944 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1945 {
1946 int cpu = smp_processor_id();
1947 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1948 }
1949 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1950
1951 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1952 {
1953 int cpu, idx = prot->inuse_idx;
1954 int res = 0;
1955
1956 for_each_possible_cpu(cpu)
1957 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1958
1959 return res >= 0 ? res : 0;
1960 }
1961 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1962
1963 static int sock_inuse_init_net(struct net *net)
1964 {
1965 net->core.inuse = alloc_percpu(struct prot_inuse);
1966 return net->core.inuse ? 0 : -ENOMEM;
1967 }
1968
1969 static void sock_inuse_exit_net(struct net *net)
1970 {
1971 free_percpu(net->core.inuse);
1972 }
1973
1974 static struct pernet_operations net_inuse_ops = {
1975 .init = sock_inuse_init_net,
1976 .exit = sock_inuse_exit_net,
1977 };
1978
1979 static __init int net_inuse_init(void)
1980 {
1981 if (register_pernet_subsys(&net_inuse_ops))
1982 panic("Cannot initialize net inuse counters");
1983
1984 return 0;
1985 }
1986
1987 core_initcall(net_inuse_init);
1988 #else
1989 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1990
1991 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1992 {
1993 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
1994 }
1995 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1996
1997 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1998 {
1999 int cpu, idx = prot->inuse_idx;
2000 int res = 0;
2001
2002 for_each_possible_cpu(cpu)
2003 res += per_cpu(prot_inuse, cpu).val[idx];
2004
2005 return res >= 0 ? res : 0;
2006 }
2007 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2008 #endif
2009
2010 static void assign_proto_idx(struct proto *prot)
2011 {
2012 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2013
2014 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2015 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2016 return;
2017 }
2018
2019 set_bit(prot->inuse_idx, proto_inuse_idx);
2020 }
2021
2022 static void release_proto_idx(struct proto *prot)
2023 {
2024 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2025 clear_bit(prot->inuse_idx, proto_inuse_idx);
2026 }
2027 #else
2028 static inline void assign_proto_idx(struct proto *prot)
2029 {
2030 }
2031
2032 static inline void release_proto_idx(struct proto *prot)
2033 {
2034 }
2035 #endif
2036
2037 int proto_register(struct proto *prot, int alloc_slab)
2038 {
2039 char *request_sock_slab_name = NULL;
2040 char *timewait_sock_slab_name;
2041
2042 if (alloc_slab) {
2043 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2044 SLAB_HWCACHE_ALIGN, NULL);
2045
2046 if (prot->slab == NULL) {
2047 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2048 prot->name);
2049 goto out;
2050 }
2051
2052 if (prot->rsk_prot != NULL) {
2053 static const char mask[] = "request_sock_%s";
2054
2055 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2056 if (request_sock_slab_name == NULL)
2057 goto out_free_sock_slab;
2058
2059 sprintf(request_sock_slab_name, mask, prot->name);
2060 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
2061 prot->rsk_prot->obj_size, 0,
2062 SLAB_HWCACHE_ALIGN, NULL);
2063
2064 if (prot->rsk_prot->slab == NULL) {
2065 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2066 prot->name);
2067 goto out_free_request_sock_slab_name;
2068 }
2069 }
2070
2071 if (prot->twsk_prot != NULL) {
2072 static const char mask[] = "tw_sock_%s";
2073
2074 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2075
2076 if (timewait_sock_slab_name == NULL)
2077 goto out_free_request_sock_slab;
2078
2079 sprintf(timewait_sock_slab_name, mask, prot->name);
2080 prot->twsk_prot->twsk_slab =
2081 kmem_cache_create(timewait_sock_slab_name,
2082 prot->twsk_prot->twsk_obj_size,
2083 0, SLAB_HWCACHE_ALIGN,
2084 NULL);
2085 if (prot->twsk_prot->twsk_slab == NULL)
2086 goto out_free_timewait_sock_slab_name;
2087 }
2088 }
2089
2090 write_lock(&proto_list_lock);
2091 list_add(&prot->node, &proto_list);
2092 assign_proto_idx(prot);
2093 write_unlock(&proto_list_lock);
2094 return 0;
2095
2096 out_free_timewait_sock_slab_name:
2097 kfree(timewait_sock_slab_name);
2098 out_free_request_sock_slab:
2099 if (prot->rsk_prot && prot->rsk_prot->slab) {
2100 kmem_cache_destroy(prot->rsk_prot->slab);
2101 prot->rsk_prot->slab = NULL;
2102 }
2103 out_free_request_sock_slab_name:
2104 kfree(request_sock_slab_name);
2105 out_free_sock_slab:
2106 kmem_cache_destroy(prot->slab);
2107 prot->slab = NULL;
2108 out:
2109 return -ENOBUFS;
2110 }
2111
2112 EXPORT_SYMBOL(proto_register);
2113
2114 void proto_unregister(struct proto *prot)
2115 {
2116 write_lock(&proto_list_lock);
2117 release_proto_idx(prot);
2118 list_del(&prot->node);
2119 write_unlock(&proto_list_lock);
2120
2121 if (prot->slab != NULL) {
2122 kmem_cache_destroy(prot->slab);
2123 prot->slab = NULL;
2124 }
2125
2126 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2127 const char *name = kmem_cache_name(prot->rsk_prot->slab);
2128
2129 kmem_cache_destroy(prot->rsk_prot->slab);
2130 kfree(name);
2131 prot->rsk_prot->slab = NULL;
2132 }
2133
2134 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2135 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
2136
2137 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2138 kfree(name);
2139 prot->twsk_prot->twsk_slab = NULL;
2140 }
2141 }
2142
2143 EXPORT_SYMBOL(proto_unregister);
2144
2145 #ifdef CONFIG_PROC_FS
2146 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2147 __acquires(proto_list_lock)
2148 {
2149 read_lock(&proto_list_lock);
2150 return seq_list_start_head(&proto_list, *pos);
2151 }
2152
2153 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2154 {
2155 return seq_list_next(v, &proto_list, pos);
2156 }
2157
2158 static void proto_seq_stop(struct seq_file *seq, void *v)
2159 __releases(proto_list_lock)
2160 {
2161 read_unlock(&proto_list_lock);
2162 }
2163
2164 static char proto_method_implemented(const void *method)
2165 {
2166 return method == NULL ? 'n' : 'y';
2167 }
2168
2169 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2170 {
2171 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2172 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2173 proto->name,
2174 proto->obj_size,
2175 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
2176 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2177 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2178 proto->max_header,
2179 proto->slab == NULL ? "no" : "yes",
2180 module_name(proto->owner),
2181 proto_method_implemented(proto->close),
2182 proto_method_implemented(proto->connect),
2183 proto_method_implemented(proto->disconnect),
2184 proto_method_implemented(proto->accept),
2185 proto_method_implemented(proto->ioctl),
2186 proto_method_implemented(proto->init),
2187 proto_method_implemented(proto->destroy),
2188 proto_method_implemented(proto->shutdown),
2189 proto_method_implemented(proto->setsockopt),
2190 proto_method_implemented(proto->getsockopt),
2191 proto_method_implemented(proto->sendmsg),
2192 proto_method_implemented(proto->recvmsg),
2193 proto_method_implemented(proto->sendpage),
2194 proto_method_implemented(proto->bind),
2195 proto_method_implemented(proto->backlog_rcv),
2196 proto_method_implemented(proto->hash),
2197 proto_method_implemented(proto->unhash),
2198 proto_method_implemented(proto->get_port),
2199 proto_method_implemented(proto->enter_memory_pressure));
2200 }
2201
2202 static int proto_seq_show(struct seq_file *seq, void *v)
2203 {
2204 if (v == &proto_list)
2205 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2206 "protocol",
2207 "size",
2208 "sockets",
2209 "memory",
2210 "press",
2211 "maxhdr",
2212 "slab",
2213 "module",
2214 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2215 else
2216 proto_seq_printf(seq, list_entry(v, struct proto, node));
2217 return 0;
2218 }
2219
2220 static const struct seq_operations proto_seq_ops = {
2221 .start = proto_seq_start,
2222 .next = proto_seq_next,
2223 .stop = proto_seq_stop,
2224 .show = proto_seq_show,
2225 };
2226
2227 static int proto_seq_open(struct inode *inode, struct file *file)
2228 {
2229 return seq_open(file, &proto_seq_ops);
2230 }
2231
2232 static const struct file_operations proto_seq_fops = {
2233 .owner = THIS_MODULE,
2234 .open = proto_seq_open,
2235 .read = seq_read,
2236 .llseek = seq_lseek,
2237 .release = seq_release,
2238 };
2239
2240 static int __init proto_init(void)
2241 {
2242 /* register /proc/net/protocols */
2243 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
2244 }
2245
2246 subsys_initcall(proto_init);
2247
2248 #endif /* PROC_FS */
2249
2250 EXPORT_SYMBOL(sk_alloc);
2251 EXPORT_SYMBOL(sk_free);
2252 EXPORT_SYMBOL(sk_send_sigurg);
2253 EXPORT_SYMBOL(sock_alloc_send_skb);
2254 EXPORT_SYMBOL(sock_init_data);
2255 EXPORT_SYMBOL(sock_kfree_s);
2256 EXPORT_SYMBOL(sock_kmalloc);
2257 EXPORT_SYMBOL(sock_no_accept);
2258 EXPORT_SYMBOL(sock_no_bind);
2259 EXPORT_SYMBOL(sock_no_connect);
2260 EXPORT_SYMBOL(sock_no_getname);
2261 EXPORT_SYMBOL(sock_no_getsockopt);
2262 EXPORT_SYMBOL(sock_no_ioctl);
2263 EXPORT_SYMBOL(sock_no_listen);
2264 EXPORT_SYMBOL(sock_no_mmap);
2265 EXPORT_SYMBOL(sock_no_poll);
2266 EXPORT_SYMBOL(sock_no_recvmsg);
2267 EXPORT_SYMBOL(sock_no_sendmsg);
2268 EXPORT_SYMBOL(sock_no_sendpage);
2269 EXPORT_SYMBOL(sock_no_setsockopt);
2270 EXPORT_SYMBOL(sock_no_shutdown);
2271 EXPORT_SYMBOL(sock_no_socketpair);
2272 EXPORT_SYMBOL(sock_rfree);
2273 EXPORT_SYMBOL(sock_setsockopt);
2274 EXPORT_SYMBOL(sock_wfree);
2275 EXPORT_SYMBOL(sock_wmalloc);
2276 EXPORT_SYMBOL(sock_i_uid);
2277 EXPORT_SYMBOL(sock_i_ino);
2278 EXPORT_SYMBOL(sysctl_optmem_max);
Linus' kernel tree