?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
proc: move /proc/vmstat boilerplate to mm/vmstat.c
[linux-2.6/zen-sources.git] / net / core / sock.c
blob5e2a3132a8c934d5aae50f8b359ebf28c6b65d8c
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 *
85 *
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
90 */
91
92 #include <linux/capability.h>
93 #include <linux/errno.h>
94 #include <linux/types.h>
95 #include <linux/socket.h>
96 #include <linux/in.h>
97 #include <linux/kernel.h>
98 #include <linux/module.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/mm.h>
107 #include <linux/slab.h>
108 #include <linux/interrupt.h>
109 #include <linux/poll.h>
110 #include <linux/tcp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113
114 #include <asm/uaccess.h>
115 #include <asm/system.h>
116
117 #include <linux/netdevice.h>
118 #include <net/protocol.h>
119 #include <linux/skbuff.h>
120 #include <net/net_namespace.h>
121 #include <net/request_sock.h>
122 #include <net/sock.h>
123 #include <net/xfrm.h>
124 #include <linux/ipsec.h>
125
126 #include <linux/filter.h>
127
128 #ifdef CONFIG_INET
129 #include <net/tcp.h>
130 #endif
131
132 /*
133 * Each address family might have different locking rules, so we have
134 * one slock key per address family:
135 */
136 static struct lock_class_key af_family_keys[AF_MAX];
137 static struct lock_class_key af_family_slock_keys[AF_MAX];
138
139 #ifdef CONFIG_DEBUG_LOCK_ALLOC
140 /*
141 * Make lock validator output more readable. (we pre-construct these
142 * strings build-time, so that runtime initialization of socket
143 * locks is fast):
144 */
145 static const char *af_family_key_strings[AF_MAX+1] = {
146 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
147 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
148 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
149 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
150 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
151 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
152 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
153 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
154 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
155 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
156 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
157 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
158 "sk_lock-AF_MAX"
159 };
160 static const char *af_family_slock_key_strings[AF_MAX+1] = {
161 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
162 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
163 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
164 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
165 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
166 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
167 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
168 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
169 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
170 "slock-27" , "slock-28" , "slock-AF_CAN" ,
171 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
172 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
173 "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-AF_CAN" ,
186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
187 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
188 "clock-AF_MAX"
189 };
190 #endif
191
192 /*
193 * sk_callback_lock locking rules are per-address-family,
194 * so split the lock classes by using a per-AF key:
195 */
196 static struct lock_class_key af_callback_keys[AF_MAX];
197
198 /* Take into consideration the size of the struct sk_buff overhead in the
199 * determination of these values, since that is non-constant across
200 * platforms. This makes socket queueing behavior and performance
201 * not depend upon such differences.
202 */
203 #define _SK_MEM_PACKETS 256
204 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
205 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
206 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
207
208 /* Run time adjustable parameters. */
209 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
210 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
211 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
212 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
213
214 /* Maximal space eaten by iovec or ancilliary data plus some space */
215 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
216
217 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
218 {
219 struct timeval tv;
220
221 if (optlen < sizeof(tv))
222 return -EINVAL;
223 if (copy_from_user(&tv, optval, sizeof(tv)))
224 return -EFAULT;
225 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
226 return -EDOM;
227
228 if (tv.tv_sec < 0) {
229 static int warned __read_mostly;
230
231 *timeo_p = 0;
232 if (warned < 10 && net_ratelimit()) {
233 warned++;
234 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
235 "tries to set negative timeout\n",
236 current->comm, task_pid_nr(current));
237 }
238 return 0;
239 }
240 *timeo_p = MAX_SCHEDULE_TIMEOUT;
241 if (tv.tv_sec == 0 && tv.tv_usec == 0)
242 return 0;
243 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
244 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
245 return 0;
246 }
247
248 static void sock_warn_obsolete_bsdism(const char *name)
249 {
250 static int warned;
251 static char warncomm[TASK_COMM_LEN];
252 if (strcmp(warncomm, current->comm) && warned < 5) {
253 strcpy(warncomm, current->comm);
254 printk(KERN_WARNING "process `%s' is using obsolete "
255 "%s SO_BSDCOMPAT\n", warncomm, name);
256 warned++;
257 }
258 }
259
260 static void sock_disable_timestamp(struct sock *sk)
261 {
262 if (sock_flag(sk, SOCK_TIMESTAMP)) {
263 sock_reset_flag(sk, SOCK_TIMESTAMP);
264 net_disable_timestamp();
265 }
266 }
267
268
269 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
270 {
271 int err = 0;
272 int skb_len;
273
274 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
275 number of warnings when compiling with -W --ANK
276 */
277 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
278 (unsigned)sk->sk_rcvbuf) {
279 err = -ENOMEM;
280 goto out;
281 }
282
283 err = sk_filter(sk, skb);
284 if (err)
285 goto out;
286
287 if (!sk_rmem_schedule(sk, skb->truesize)) {
288 err = -ENOBUFS;
289 goto out;
290 }
291
292 skb->dev = NULL;
293 skb_set_owner_r(skb, sk);
294
295 /* Cache the SKB length before we tack it onto the receive
296 * queue. Once it is added it no longer belongs to us and
297 * may be freed by other threads of control pulling packets
298 * from the queue.
299 */
300 skb_len = skb->len;
301
302 skb_queue_tail(&sk->sk_receive_queue, skb);
303
304 if (!sock_flag(sk, SOCK_DEAD))
305 sk->sk_data_ready(sk, skb_len);
306 out:
307 return err;
308 }
309 EXPORT_SYMBOL(sock_queue_rcv_skb);
310
311 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
312 {
313 int rc = NET_RX_SUCCESS;
314
315 if (sk_filter(sk, skb))
316 goto discard_and_relse;
317
318 skb->dev = NULL;
319
320 if (nested)
321 bh_lock_sock_nested(sk);
322 else
323 bh_lock_sock(sk);
324 if (!sock_owned_by_user(sk)) {
325 /*
326 * trylock + unlock semantics:
327 */
328 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
329
330 rc = sk_backlog_rcv(sk, skb);
331
332 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
333 } else
334 sk_add_backlog(sk, skb);
335 bh_unlock_sock(sk);
336 out:
337 sock_put(sk);
338 return rc;
339 discard_and_relse:
340 kfree_skb(skb);
341 goto out;
342 }
343 EXPORT_SYMBOL(sk_receive_skb);
344
345 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
346 {
347 struct dst_entry *dst = sk->sk_dst_cache;
348
349 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
350 sk->sk_dst_cache = NULL;
351 dst_release(dst);
352 return NULL;
353 }
354
355 return dst;
356 }
357 EXPORT_SYMBOL(__sk_dst_check);
358
359 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
360 {
361 struct dst_entry *dst = sk_dst_get(sk);
362
363 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
364 sk_dst_reset(sk);
365 dst_release(dst);
366 return NULL;
367 }
368
369 return dst;
370 }
371 EXPORT_SYMBOL(sk_dst_check);
372
373 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
374 {
375 int ret = -ENOPROTOOPT;
376 #ifdef CONFIG_NETDEVICES
377 struct net *net = sock_net(sk);
378 char devname[IFNAMSIZ];
379 int index;
380
381 /* Sorry... */
382 ret = -EPERM;
383 if (!capable(CAP_NET_RAW))
384 goto out;
385
386 ret = -EINVAL;
387 if (optlen < 0)
388 goto out;
389
390 /* Bind this socket to a particular device like "eth0",
391 * as specified in the passed interface name. If the
392 * name is "" or the option length is zero the socket
393 * is not bound.
394 */
395 if (optlen > IFNAMSIZ - 1)
396 optlen = IFNAMSIZ - 1;
397 memset(devname, 0, sizeof(devname));
398
399 ret = -EFAULT;
400 if (copy_from_user(devname, optval, optlen))
401 goto out;
402
403 if (devname[0] == '\0') {
404 index = 0;
405 } else {
406 struct net_device *dev = dev_get_by_name(net, devname);
407
408 ret = -ENODEV;
409 if (!dev)
410 goto out;
411
412 index = dev->ifindex;
413 dev_put(dev);
414 }
415
416 lock_sock(sk);
417 sk->sk_bound_dev_if = index;
418 sk_dst_reset(sk);
419 release_sock(sk);
420
421 ret = 0;
422
423 out:
424 #endif
425
426 return ret;
427 }
428
429 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
430 {
431 if (valbool)
432 sock_set_flag(sk, bit);
433 else
434 sock_reset_flag(sk, bit);
435 }
436
437 /*
438 * This is meant for all protocols to use and covers goings on
439 * at the socket level. Everything here is generic.
440 */
441
442 int sock_setsockopt(struct socket *sock, int level, int optname,
443 char __user *optval, int optlen)
444 {
445 struct sock *sk=sock->sk;
446 int val;
447 int valbool;
448 struct linger ling;
449 int ret = 0;
450
451 /*
452 * Options without arguments
453 */
454
455 if (optname == SO_BINDTODEVICE)
456 return sock_bindtodevice(sk, optval, optlen);
457
458 if (optlen < sizeof(int))
459 return -EINVAL;
460
461 if (get_user(val, (int __user *)optval))
462 return -EFAULT;
463
464 valbool = val?1:0;
465
466 lock_sock(sk);
467
468 switch(optname) {
469 case SO_DEBUG:
470 if (val && !capable(CAP_NET_ADMIN)) {
471 ret = -EACCES;
472 } else
473 sock_valbool_flag(sk, SOCK_DBG, valbool);
474 break;
475 case SO_REUSEADDR:
476 sk->sk_reuse = valbool;
477 break;
478 case SO_TYPE:
479 case SO_ERROR:
480 ret = -ENOPROTOOPT;
481 break;
482 case SO_DONTROUTE:
483 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
484 break;
485 case SO_BROADCAST:
486 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
487 break;
488 case SO_SNDBUF:
489 /* Don't error on this BSD doesn't and if you think
490 about it this is right. Otherwise apps have to
491 play 'guess the biggest size' games. RCVBUF/SNDBUF
492 are treated in BSD as hints */
493
494 if (val > sysctl_wmem_max)
495 val = sysctl_wmem_max;
496 set_sndbuf:
497 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
498 if ((val * 2) < SOCK_MIN_SNDBUF)
499 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
500 else
501 sk->sk_sndbuf = val * 2;
502
503 /*
504 * Wake up sending tasks if we
505 * upped the value.
506 */
507 sk->sk_write_space(sk);
508 break;
509
510 case SO_SNDBUFFORCE:
511 if (!capable(CAP_NET_ADMIN)) {
512 ret = -EPERM;
513 break;
514 }
515 goto set_sndbuf;
516
517 case SO_RCVBUF:
518 /* Don't error on this BSD doesn't and if you think
519 about it this is right. Otherwise apps have to
520 play 'guess the biggest size' games. RCVBUF/SNDBUF
521 are treated in BSD as hints */
522
523 if (val > sysctl_rmem_max)
524 val = sysctl_rmem_max;
525 set_rcvbuf:
526 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
527 /*
528 * We double it on the way in to account for
529 * "struct sk_buff" etc. overhead. Applications
530 * assume that the SO_RCVBUF setting they make will
531 * allow that much actual data to be received on that
532 * socket.
533 *
534 * Applications are unaware that "struct sk_buff" and
535 * other overheads allocate from the receive buffer
536 * during socket buffer allocation.
537 *
538 * And after considering the possible alternatives,
539 * returning the value we actually used in getsockopt
540 * is the most desirable behavior.
541 */
542 if ((val * 2) < SOCK_MIN_RCVBUF)
543 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
544 else
545 sk->sk_rcvbuf = val * 2;
546 break;
547
548 case SO_RCVBUFFORCE:
549 if (!capable(CAP_NET_ADMIN)) {
550 ret = -EPERM;
551 break;
552 }
553 goto set_rcvbuf;
554
555 case SO_KEEPALIVE:
556 #ifdef CONFIG_INET
557 if (sk->sk_protocol == IPPROTO_TCP)
558 tcp_set_keepalive(sk, valbool);
559 #endif
560 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
561 break;
562
563 case SO_OOBINLINE:
564 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
565 break;
566
567 case SO_NO_CHECK:
568 sk->sk_no_check = valbool;
569 break;
570
571 case SO_PRIORITY:
572 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
573 sk->sk_priority = val;
574 else
575 ret = -EPERM;
576 break;
577
578 case SO_LINGER:
579 if (optlen < sizeof(ling)) {
580 ret = -EINVAL; /* 1003.1g */
581 break;
582 }
583 if (copy_from_user(&ling,optval,sizeof(ling))) {
584 ret = -EFAULT;
585 break;
586 }
587 if (!ling.l_onoff)
588 sock_reset_flag(sk, SOCK_LINGER);
589 else {
590 #if (BITS_PER_LONG == 32)
591 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
592 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
593 else
594 #endif
595 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
596 sock_set_flag(sk, SOCK_LINGER);
597 }
598 break;
599
600 case SO_BSDCOMPAT:
601 sock_warn_obsolete_bsdism("setsockopt");
602 break;
603
604 case SO_PASSCRED:
605 if (valbool)
606 set_bit(SOCK_PASSCRED, &sock->flags);
607 else
608 clear_bit(SOCK_PASSCRED, &sock->flags);
609 break;
610
611 case SO_TIMESTAMP:
612 case SO_TIMESTAMPNS:
613 if (valbool) {
614 if (optname == SO_TIMESTAMP)
615 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
616 else
617 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
618 sock_set_flag(sk, SOCK_RCVTSTAMP);
619 sock_enable_timestamp(sk);
620 } else {
621 sock_reset_flag(sk, SOCK_RCVTSTAMP);
622 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
623 }
624 break;
625
626 case SO_RCVLOWAT:
627 if (val < 0)
628 val = INT_MAX;
629 sk->sk_rcvlowat = val ? : 1;
630 break;
631
632 case SO_RCVTIMEO:
633 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
634 break;
635
636 case SO_SNDTIMEO:
637 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
638 break;
639
640 case SO_ATTACH_FILTER:
641 ret = -EINVAL;
642 if (optlen == sizeof(struct sock_fprog)) {
643 struct sock_fprog fprog;
644
645 ret = -EFAULT;
646 if (copy_from_user(&fprog, optval, sizeof(fprog)))
647 break;
648
649 ret = sk_attach_filter(&fprog, sk);
650 }
651 break;
652
653 case SO_DETACH_FILTER:
654 ret = sk_detach_filter(sk);
655 break;
656
657 case SO_PASSSEC:
658 if (valbool)
659 set_bit(SOCK_PASSSEC, &sock->flags);
660 else
661 clear_bit(SOCK_PASSSEC, &sock->flags);
662 break;
663 case SO_MARK:
664 if (!capable(CAP_NET_ADMIN))
665 ret = -EPERM;
666 else {
667 sk->sk_mark = val;
668 }
669 break;
670
671 /* We implement the SO_SNDLOWAT etc to
672 not be settable (1003.1g 5.3) */
673 default:
674 ret = -ENOPROTOOPT;
675 break;
676 }
677 release_sock(sk);
678 return ret;
679 }
680
681
682 int sock_getsockopt(struct socket *sock, int level, int optname,
683 char __user *optval, int __user *optlen)
684 {
685 struct sock *sk = sock->sk;
686
687 union {
688 int val;
689 struct linger ling;
690 struct timeval tm;
691 } v;
692
693 unsigned int lv = sizeof(int);
694 int len;
695
696 if (get_user(len, optlen))
697 return -EFAULT;
698 if (len < 0)
699 return -EINVAL;
700
701 switch(optname) {
702 case SO_DEBUG:
703 v.val = sock_flag(sk, SOCK_DBG);
704 break;
705
706 case SO_DONTROUTE:
707 v.val = sock_flag(sk, SOCK_LOCALROUTE);
708 break;
709
710 case SO_BROADCAST:
711 v.val = !!sock_flag(sk, SOCK_BROADCAST);
712 break;
713
714 case SO_SNDBUF:
715 v.val = sk->sk_sndbuf;
716 break;
717
718 case SO_RCVBUF:
719 v.val = sk->sk_rcvbuf;
720 break;
721
722 case SO_REUSEADDR:
723 v.val = sk->sk_reuse;
724 break;
725
726 case SO_KEEPALIVE:
727 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
728 break;
729
730 case SO_TYPE:
731 v.val = sk->sk_type;
732 break;
733
734 case SO_ERROR:
735 v.val = -sock_error(sk);
736 if (v.val==0)
737 v.val = xchg(&sk->sk_err_soft, 0);
738 break;
739
740 case SO_OOBINLINE:
741 v.val = !!sock_flag(sk, SOCK_URGINLINE);
742 break;
743
744 case SO_NO_CHECK:
745 v.val = sk->sk_no_check;
746 break;
747
748 case SO_PRIORITY:
749 v.val = sk->sk_priority;
750 break;
751
752 case SO_LINGER:
753 lv = sizeof(v.ling);
754 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
755 v.ling.l_linger = sk->sk_lingertime / HZ;
756 break;
757
758 case SO_BSDCOMPAT:
759 sock_warn_obsolete_bsdism("getsockopt");
760 break;
761
762 case SO_TIMESTAMP:
763 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
764 !sock_flag(sk, SOCK_RCVTSTAMPNS);
765 break;
766
767 case SO_TIMESTAMPNS:
768 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
769 break;
770
771 case SO_RCVTIMEO:
772 lv=sizeof(struct timeval);
773 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
774 v.tm.tv_sec = 0;
775 v.tm.tv_usec = 0;
776 } else {
777 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
778 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
779 }
780 break;
781
782 case SO_SNDTIMEO:
783 lv=sizeof(struct timeval);
784 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
785 v.tm.tv_sec = 0;
786 v.tm.tv_usec = 0;
787 } else {
788 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
789 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
790 }
791 break;
792
793 case SO_RCVLOWAT:
794 v.val = sk->sk_rcvlowat;
795 break;
796
797 case SO_SNDLOWAT:
798 v.val=1;
799 break;
800
801 case SO_PASSCRED:
802 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
803 break;
804
805 case SO_PEERCRED:
806 if (len > sizeof(sk->sk_peercred))
807 len = sizeof(sk->sk_peercred);
808 if (copy_to_user(optval, &sk->sk_peercred, len))
809 return -EFAULT;
810 goto lenout;
811
812 case SO_PEERNAME:
813 {
814 char address[128];
815
816 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
817 return -ENOTCONN;
818 if (lv < len)
819 return -EINVAL;
820 if (copy_to_user(optval, address, len))
821 return -EFAULT;
822 goto lenout;
823 }
824
825 /* Dubious BSD thing... Probably nobody even uses it, but
826 * the UNIX standard wants it for whatever reason... -DaveM
827 */
828 case SO_ACCEPTCONN:
829 v.val = sk->sk_state == TCP_LISTEN;
830 break;
831
832 case SO_PASSSEC:
833 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
834 break;
835
836 case SO_PEERSEC:
837 return security_socket_getpeersec_stream(sock, optval, optlen, len);
838
839 case SO_MARK:
840 v.val = sk->sk_mark;
841 break;
842
843 default:
844 return -ENOPROTOOPT;
845 }
846
847 if (len > lv)
848 len = lv;
849 if (copy_to_user(optval, &v, len))
850 return -EFAULT;
851 lenout:
852 if (put_user(len, optlen))
853 return -EFAULT;
854 return 0;
855 }
856
857 /*
858 * Initialize an sk_lock.
859 *
860 * (We also register the sk_lock with the lock validator.)
861 */
862 static inline void sock_lock_init(struct sock *sk)
863 {
864 sock_lock_init_class_and_name(sk,
865 af_family_slock_key_strings[sk->sk_family],
866 af_family_slock_keys + sk->sk_family,
867 af_family_key_strings[sk->sk_family],
868 af_family_keys + sk->sk_family);
869 }
870
871 static void sock_copy(struct sock *nsk, const struct sock *osk)
872 {
873 #ifdef CONFIG_SECURITY_NETWORK
874 void *sptr = nsk->sk_security;
875 #endif
876
877 memcpy(nsk, osk, osk->sk_prot->obj_size);
878 #ifdef CONFIG_SECURITY_NETWORK
879 nsk->sk_security = sptr;
880 security_sk_clone(osk, nsk);
881 #endif
882 }
883
884 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
885 int family)
886 {
887 struct sock *sk;
888 struct kmem_cache *slab;
889
890 slab = prot->slab;
891 if (slab != NULL)
892 sk = kmem_cache_alloc(slab, priority);
893 else
894 sk = kmalloc(prot->obj_size, priority);
895
896 if (sk != NULL) {
897 if (security_sk_alloc(sk, family, priority))
898 goto out_free;
899
900 if (!try_module_get(prot->owner))
901 goto out_free_sec;
902 }
903
904 return sk;
905
906 out_free_sec:
907 security_sk_free(sk);
908 out_free:
909 if (slab != NULL)
910 kmem_cache_free(slab, sk);
911 else
912 kfree(sk);
913 return NULL;
914 }
915
916 static void sk_prot_free(struct proto *prot, struct sock *sk)
917 {
918 struct kmem_cache *slab;
919 struct module *owner;
920
921 owner = prot->owner;
922 slab = prot->slab;
923
924 security_sk_free(sk);
925 if (slab != NULL)
926 kmem_cache_free(slab, sk);
927 else
928 kfree(sk);
929 module_put(owner);
930 }
931
932 /**
933 * sk_alloc - All socket objects are allocated here
934 * @net: the applicable net namespace
935 * @family: protocol family
936 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
937 * @prot: struct proto associated with this new sock instance
938 */
939 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
940 struct proto *prot)
941 {
942 struct sock *sk;
943
944 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
945 if (sk) {
946 sk->sk_family = family;
947 /*
948 * See comment in struct sock definition to understand
949 * why we need sk_prot_creator -acme
950 */
951 sk->sk_prot = sk->sk_prot_creator = prot;
952 sock_lock_init(sk);
953 sock_net_set(sk, get_net(net));
954 }
955
956 return sk;
957 }
958
959 void sk_free(struct sock *sk)
960 {
961 struct sk_filter *filter;
962
963 if (sk->sk_destruct)
964 sk->sk_destruct(sk);
965
966 filter = rcu_dereference(sk->sk_filter);
967 if (filter) {
968 sk_filter_uncharge(sk, filter);
969 rcu_assign_pointer(sk->sk_filter, NULL);
970 }
971
972 sock_disable_timestamp(sk);
973
974 if (atomic_read(&sk->sk_omem_alloc))
975 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
976 __func__, atomic_read(&sk->sk_omem_alloc));
977
978 put_net(sock_net(sk));
979 sk_prot_free(sk->sk_prot_creator, sk);
980 }
981
982 /*
983 * Last sock_put should drop referrence to sk->sk_net. It has already
984 * been dropped in sk_change_net. Taking referrence to stopping namespace
985 * is not an option.
986 * Take referrence to a socket to remove it from hash _alive_ and after that
987 * destroy it in the context of init_net.
988 */
989 void sk_release_kernel(struct sock *sk)
990 {
991 if (sk == NULL || sk->sk_socket == NULL)
992 return;
993
994 sock_hold(sk);
995 sock_release(sk->sk_socket);
996 release_net(sock_net(sk));
997 sock_net_set(sk, get_net(&init_net));
998 sock_put(sk);
999 }
1000 EXPORT_SYMBOL(sk_release_kernel);
1001
1002 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1003 {
1004 struct sock *newsk;
1005
1006 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1007 if (newsk != NULL) {
1008 struct sk_filter *filter;
1009
1010 sock_copy(newsk, sk);
1011
1012 /* SANITY */
1013 get_net(sock_net(newsk));
1014 sk_node_init(&newsk->sk_node);
1015 sock_lock_init(newsk);
1016 bh_lock_sock(newsk);
1017 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1018
1019 atomic_set(&newsk->sk_rmem_alloc, 0);
1020 atomic_set(&newsk->sk_wmem_alloc, 0);
1021 atomic_set(&newsk->sk_omem_alloc, 0);
1022 skb_queue_head_init(&newsk->sk_receive_queue);
1023 skb_queue_head_init(&newsk->sk_write_queue);
1024 #ifdef CONFIG_NET_DMA
1025 skb_queue_head_init(&newsk->sk_async_wait_queue);
1026 #endif
1027
1028 rwlock_init(&newsk->sk_dst_lock);
1029 rwlock_init(&newsk->sk_callback_lock);
1030 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1031 af_callback_keys + newsk->sk_family,
1032 af_family_clock_key_strings[newsk->sk_family]);
1033
1034 newsk->sk_dst_cache = NULL;
1035 newsk->sk_wmem_queued = 0;
1036 newsk->sk_forward_alloc = 0;
1037 newsk->sk_send_head = NULL;
1038 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1039
1040 sock_reset_flag(newsk, SOCK_DONE);
1041 skb_queue_head_init(&newsk->sk_error_queue);
1042
1043 filter = newsk->sk_filter;
1044 if (filter != NULL)
1045 sk_filter_charge(newsk, filter);
1046
1047 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1048 /* It is still raw copy of parent, so invalidate
1049 * destructor and make plain sk_free() */
1050 newsk->sk_destruct = NULL;
1051 sk_free(newsk);
1052 newsk = NULL;
1053 goto out;
1054 }
1055
1056 newsk->sk_err = 0;
1057 newsk->sk_priority = 0;
1058 atomic_set(&newsk->sk_refcnt, 2);
1059
1060 /*
1061 * Increment the counter in the same struct proto as the master
1062 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1063 * is the same as sk->sk_prot->socks, as this field was copied
1064 * with memcpy).
1065 *
1066 * This _changes_ the previous behaviour, where
1067 * tcp_create_openreq_child always was incrementing the
1068 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1069 * to be taken into account in all callers. -acme
1070 */
1071 sk_refcnt_debug_inc(newsk);
1072 sk_set_socket(newsk, NULL);
1073 newsk->sk_sleep = NULL;
1074
1075 if (newsk->sk_prot->sockets_allocated)
1076 atomic_inc(newsk->sk_prot->sockets_allocated);
1077 }
1078 out:
1079 return newsk;
1080 }
1081
1082 EXPORT_SYMBOL_GPL(sk_clone);
1083
1084 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1085 {
1086 __sk_dst_set(sk, dst);
1087 sk->sk_route_caps = dst->dev->features;
1088 if (sk->sk_route_caps & NETIF_F_GSO)
1089 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1090 if (sk_can_gso(sk)) {
1091 if (dst->header_len) {
1092 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1093 } else {
1094 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1095 sk->sk_gso_max_size = dst->dev->gso_max_size;
1096 }
1097 }
1098 }
1099 EXPORT_SYMBOL_GPL(sk_setup_caps);
1100
1101 void __init sk_init(void)
1102 {
1103 if (num_physpages <= 4096) {
1104 sysctl_wmem_max = 32767;
1105 sysctl_rmem_max = 32767;
1106 sysctl_wmem_default = 32767;
1107 sysctl_rmem_default = 32767;
1108 } else if (num_physpages >= 131072) {
1109 sysctl_wmem_max = 131071;
1110 sysctl_rmem_max = 131071;
1111 }
1112 }
1113
1114 /*
1115 * Simple resource managers for sockets.
1116 */
1117
1118
1119 /*
1120 * Write buffer destructor automatically called from kfree_skb.
1121 */
1122 void sock_wfree(struct sk_buff *skb)
1123 {
1124 struct sock *sk = skb->sk;
1125
1126 /* In case it might be waiting for more memory. */
1127 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1128 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1129 sk->sk_write_space(sk);
1130 sock_put(sk);
1131 }
1132
1133 /*
1134 * Read buffer destructor automatically called from kfree_skb.
1135 */
1136 void sock_rfree(struct sk_buff *skb)
1137 {
1138 struct sock *sk = skb->sk;
1139
1140 skb_truesize_check(skb);
1141 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1142 sk_mem_uncharge(skb->sk, skb->truesize);
1143 }
1144
1145
1146 int sock_i_uid(struct sock *sk)
1147 {
1148 int uid;
1149
1150 read_lock(&sk->sk_callback_lock);
1151 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1152 read_unlock(&sk->sk_callback_lock);
1153 return uid;
1154 }
1155
1156 unsigned long sock_i_ino(struct sock *sk)
1157 {
1158 unsigned long ino;
1159
1160 read_lock(&sk->sk_callback_lock);
1161 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1162 read_unlock(&sk->sk_callback_lock);
1163 return ino;
1164 }
1165
1166 /*
1167 * Allocate a skb from the socket's send buffer.
1168 */
1169 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1170 gfp_t priority)
1171 {
1172 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1173 struct sk_buff * skb = alloc_skb(size, priority);
1174 if (skb) {
1175 skb_set_owner_w(skb, sk);
1176 return skb;
1177 }
1178 }
1179 return NULL;
1180 }
1181
1182 /*
1183 * Allocate a skb from the socket's receive buffer.
1184 */
1185 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1186 gfp_t priority)
1187 {
1188 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1189 struct sk_buff *skb = alloc_skb(size, priority);
1190 if (skb) {
1191 skb_set_owner_r(skb, sk);
1192 return skb;
1193 }
1194 }
1195 return NULL;
1196 }
1197
1198 /*
1199 * Allocate a memory block from the socket's option memory buffer.
1200 */
1201 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1202 {
1203 if ((unsigned)size <= sysctl_optmem_max &&
1204 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1205 void *mem;
1206 /* First do the add, to avoid the race if kmalloc
1207 * might sleep.
1208 */
1209 atomic_add(size, &sk->sk_omem_alloc);
1210 mem = kmalloc(size, priority);
1211 if (mem)
1212 return mem;
1213 atomic_sub(size, &sk->sk_omem_alloc);
1214 }
1215 return NULL;
1216 }
1217
1218 /*
1219 * Free an option memory block.
1220 */
1221 void sock_kfree_s(struct sock *sk, void *mem, int size)
1222 {
1223 kfree(mem);
1224 atomic_sub(size, &sk->sk_omem_alloc);
1225 }
1226
1227 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1228 I think, these locks should be removed for datagram sockets.
1229 */
1230 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1231 {
1232 DEFINE_WAIT(wait);
1233
1234 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1235 for (;;) {
1236 if (!timeo)
1237 break;
1238 if (signal_pending(current))
1239 break;
1240 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1241 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1242 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1243 break;
1244 if (sk->sk_shutdown & SEND_SHUTDOWN)
1245 break;
1246 if (sk->sk_err)
1247 break;
1248 timeo = schedule_timeout(timeo);
1249 }
1250 finish_wait(sk->sk_sleep, &wait);
1251 return timeo;
1252 }
1253
1254
1255 /*
1256 * Generic send/receive buffer handlers
1257 */
1258
1259 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1260 unsigned long header_len,
1261 unsigned long data_len,
1262 int noblock, int *errcode)
1263 {
1264 struct sk_buff *skb;
1265 gfp_t gfp_mask;
1266 long timeo;
1267 int err;
1268
1269 gfp_mask = sk->sk_allocation;
1270 if (gfp_mask & __GFP_WAIT)
1271 gfp_mask |= __GFP_REPEAT;
1272
1273 timeo = sock_sndtimeo(sk, noblock);
1274 while (1) {
1275 err = sock_error(sk);
1276 if (err != 0)
1277 goto failure;
1278
1279 err = -EPIPE;
1280 if (sk->sk_shutdown & SEND_SHUTDOWN)
1281 goto failure;
1282
1283 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1284 skb = alloc_skb(header_len, gfp_mask);
1285 if (skb) {
1286 int npages;
1287 int i;
1288
1289 /* No pages, we're done... */
1290 if (!data_len)
1291 break;
1292
1293 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1294 skb->truesize += data_len;
1295 skb_shinfo(skb)->nr_frags = npages;
1296 for (i = 0; i < npages; i++) {
1297 struct page *page;
1298 skb_frag_t *frag;
1299
1300 page = alloc_pages(sk->sk_allocation, 0);
1301 if (!page) {
1302 err = -ENOBUFS;
1303 skb_shinfo(skb)->nr_frags = i;
1304 kfree_skb(skb);
1305 goto failure;
1306 }
1307
1308 frag = &skb_shinfo(skb)->frags[i];
1309 frag->page = page;
1310 frag->page_offset = 0;
1311 frag->size = (data_len >= PAGE_SIZE ?
1312 PAGE_SIZE :
1313 data_len);
1314 data_len -= PAGE_SIZE;
1315 }
1316
1317 /* Full success... */
1318 break;
1319 }
1320 err = -ENOBUFS;
1321 goto failure;
1322 }
1323 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1324 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1325 err = -EAGAIN;
1326 if (!timeo)
1327 goto failure;
1328 if (signal_pending(current))
1329 goto interrupted;
1330 timeo = sock_wait_for_wmem(sk, timeo);
1331 }
1332
1333 skb_set_owner_w(skb, sk);
1334 return skb;
1335
1336 interrupted:
1337 err = sock_intr_errno(timeo);
1338 failure:
1339 *errcode = err;
1340 return NULL;
1341 }
1342
1343 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1344 int noblock, int *errcode)
1345 {
1346 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1347 }
1348
1349 static void __lock_sock(struct sock *sk)
1350 {
1351 DEFINE_WAIT(wait);
1352
1353 for (;;) {
1354 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1355 TASK_UNINTERRUPTIBLE);
1356 spin_unlock_bh(&sk->sk_lock.slock);
1357 schedule();
1358 spin_lock_bh(&sk->sk_lock.slock);
1359 if (!sock_owned_by_user(sk))
1360 break;
1361 }
1362 finish_wait(&sk->sk_lock.wq, &wait);
1363 }
1364
1365 static void __release_sock(struct sock *sk)
1366 {
1367 struct sk_buff *skb = sk->sk_backlog.head;
1368
1369 do {
1370 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1371 bh_unlock_sock(sk);
1372
1373 do {
1374 struct sk_buff *next = skb->next;
1375
1376 skb->next = NULL;
1377 sk_backlog_rcv(sk, skb);
1378
1379 /*
1380 * We are in process context here with softirqs
1381 * disabled, use cond_resched_softirq() to preempt.
1382 * This is safe to do because we've taken the backlog
1383 * queue private:
1384 */
1385 cond_resched_softirq();
1386
1387 skb = next;
1388 } while (skb != NULL);
1389
1390 bh_lock_sock(sk);
1391 } while ((skb = sk->sk_backlog.head) != NULL);
1392 }
1393
1394 /**
1395 * sk_wait_data - wait for data to arrive at sk_receive_queue
1396 * @sk: sock to wait on
1397 * @timeo: for how long
1398 *
1399 * Now socket state including sk->sk_err is changed only under lock,
1400 * hence we may omit checks after joining wait queue.
1401 * We check receive queue before schedule() only as optimization;
1402 * it is very likely that release_sock() added new data.
1403 */
1404 int sk_wait_data(struct sock *sk, long *timeo)
1405 {
1406 int rc;
1407 DEFINE_WAIT(wait);
1408
1409 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1410 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1411 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1412 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1413 finish_wait(sk->sk_sleep, &wait);
1414 return rc;
1415 }
1416
1417 EXPORT_SYMBOL(sk_wait_data);
1418
1419 /**
1420 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1421 * @sk: socket
1422 * @size: memory size to allocate
1423 * @kind: allocation type
1424 *
1425 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1426 * rmem allocation. This function assumes that protocols which have
1427 * memory_pressure use sk_wmem_queued as write buffer accounting.
1428 */
1429 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1430 {
1431 struct proto *prot = sk->sk_prot;
1432 int amt = sk_mem_pages(size);
1433 int allocated;
1434
1435 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1436 allocated = atomic_add_return(amt, prot->memory_allocated);
1437
1438 /* Under limit. */
1439 if (allocated <= prot->sysctl_mem[0]) {
1440 if (prot->memory_pressure && *prot->memory_pressure)
1441 *prot->memory_pressure = 0;
1442 return 1;
1443 }
1444
1445 /* Under pressure. */
1446 if (allocated > prot->sysctl_mem[1])
1447 if (prot->enter_memory_pressure)
1448 prot->enter_memory_pressure(sk);
1449
1450 /* Over hard limit. */
1451 if (allocated > prot->sysctl_mem[2])
1452 goto suppress_allocation;
1453
1454 /* guarantee minimum buffer size under pressure */
1455 if (kind == SK_MEM_RECV) {
1456 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1457 return 1;
1458 } else { /* SK_MEM_SEND */
1459 if (sk->sk_type == SOCK_STREAM) {
1460 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1461 return 1;
1462 } else if (atomic_read(&sk->sk_wmem_alloc) <
1463 prot->sysctl_wmem[0])
1464 return 1;
1465 }
1466
1467 if (prot->memory_pressure) {
1468 if (!*prot->memory_pressure ||
1469 prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) *
1470 sk_mem_pages(sk->sk_wmem_queued +
1471 atomic_read(&sk->sk_rmem_alloc) +
1472 sk->sk_forward_alloc))
1473 return 1;
1474 }
1475
1476 suppress_allocation:
1477
1478 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1479 sk_stream_moderate_sndbuf(sk);
1480
1481 /* Fail only if socket is _under_ its sndbuf.
1482 * In this case we cannot block, so that we have to fail.
1483 */
1484 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1485 return 1;
1486 }
1487
1488 /* Alas. Undo changes. */
1489 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1490 atomic_sub(amt, prot->memory_allocated);
1491 return 0;
1492 }
1493
1494 EXPORT_SYMBOL(__sk_mem_schedule);
1495
1496 /**
1497 * __sk_reclaim - reclaim memory_allocated
1498 * @sk: socket
1499 */
1500 void __sk_mem_reclaim(struct sock *sk)
1501 {
1502 struct proto *prot = sk->sk_prot;
1503
1504 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1505 prot->memory_allocated);
1506 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1507
1508 if (prot->memory_pressure && *prot->memory_pressure &&
1509 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1510 *prot->memory_pressure = 0;
1511 }
1512
1513 EXPORT_SYMBOL(__sk_mem_reclaim);
1514
1515
1516 /*
1517 * Set of default routines for initialising struct proto_ops when
1518 * the protocol does not support a particular function. In certain
1519 * cases where it makes no sense for a protocol to have a "do nothing"
1520 * function, some default processing is provided.
1521 */
1522
1523 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1524 {
1525 return -EOPNOTSUPP;
1526 }
1527
1528 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1529 int len, int flags)
1530 {
1531 return -EOPNOTSUPP;
1532 }
1533
1534 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1535 {
1536 return -EOPNOTSUPP;
1537 }
1538
1539 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1540 {
1541 return -EOPNOTSUPP;
1542 }
1543
1544 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1545 int *len, int peer)
1546 {
1547 return -EOPNOTSUPP;
1548 }
1549
1550 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1551 {
1552 return 0;
1553 }
1554
1555 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1556 {
1557 return -EOPNOTSUPP;
1558 }
1559
1560 int sock_no_listen(struct socket *sock, int backlog)
1561 {
1562 return -EOPNOTSUPP;
1563 }
1564
1565 int sock_no_shutdown(struct socket *sock, int how)
1566 {
1567 return -EOPNOTSUPP;
1568 }
1569
1570 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1571 char __user *optval, int optlen)
1572 {
1573 return -EOPNOTSUPP;
1574 }
1575
1576 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1577 char __user *optval, int __user *optlen)
1578 {
1579 return -EOPNOTSUPP;
1580 }
1581
1582 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1583 size_t len)
1584 {
1585 return -EOPNOTSUPP;
1586 }
1587
1588 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1589 size_t len, int flags)
1590 {
1591 return -EOPNOTSUPP;
1592 }
1593
1594 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1595 {
1596 /* Mirror missing mmap method error code */
1597 return -ENODEV;
1598 }
1599
1600 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1601 {
1602 ssize_t res;
1603 struct msghdr msg = {.msg_flags = flags};
1604 struct kvec iov;
1605 char *kaddr = kmap(page);
1606 iov.iov_base = kaddr + offset;
1607 iov.iov_len = size;
1608 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1609 kunmap(page);
1610 return res;
1611 }
1612
1613 /*
1614 * Default Socket Callbacks
1615 */
1616
1617 static void sock_def_wakeup(struct sock *sk)
1618 {
1619 read_lock(&sk->sk_callback_lock);
1620 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1621 wake_up_interruptible_all(sk->sk_sleep);
1622 read_unlock(&sk->sk_callback_lock);
1623 }
1624
1625 static void sock_def_error_report(struct sock *sk)
1626 {
1627 read_lock(&sk->sk_callback_lock);
1628 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1629 wake_up_interruptible(sk->sk_sleep);
1630 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1631 read_unlock(&sk->sk_callback_lock);
1632 }
1633
1634 static void sock_def_readable(struct sock *sk, int len)
1635 {
1636 read_lock(&sk->sk_callback_lock);
1637 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1638 wake_up_interruptible_sync(sk->sk_sleep);
1639 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1640 read_unlock(&sk->sk_callback_lock);
1641 }
1642
1643 static void sock_def_write_space(struct sock *sk)
1644 {
1645 read_lock(&sk->sk_callback_lock);
1646
1647 /* Do not wake up a writer until he can make "significant"
1648 * progress. --DaveM
1649 */
1650 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1651 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1652 wake_up_interruptible_sync(sk->sk_sleep);
1653
1654 /* Should agree with poll, otherwise some programs break */
1655 if (sock_writeable(sk))
1656 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1657 }
1658
1659 read_unlock(&sk->sk_callback_lock);
1660 }
1661
1662 static void sock_def_destruct(struct sock *sk)
1663 {
1664 kfree(sk->sk_protinfo);
1665 }
1666
1667 void sk_send_sigurg(struct sock *sk)
1668 {
1669 if (sk->sk_socket && sk->sk_socket->file)
1670 if (send_sigurg(&sk->sk_socket->file->f_owner))
1671 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1672 }
1673
1674 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1675 unsigned long expires)
1676 {
1677 if (!mod_timer(timer, expires))
1678 sock_hold(sk);
1679 }
1680
1681 EXPORT_SYMBOL(sk_reset_timer);
1682
1683 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1684 {
1685 if (timer_pending(timer) && del_timer(timer))
1686 __sock_put(sk);
1687 }
1688
1689 EXPORT_SYMBOL(sk_stop_timer);
1690
1691 void sock_init_data(struct socket *sock, struct sock *sk)
1692 {
1693 skb_queue_head_init(&sk->sk_receive_queue);
1694 skb_queue_head_init(&sk->sk_write_queue);
1695 skb_queue_head_init(&sk->sk_error_queue);
1696 #ifdef CONFIG_NET_DMA
1697 skb_queue_head_init(&sk->sk_async_wait_queue);
1698 #endif
1699
1700 sk->sk_send_head = NULL;
1701
1702 init_timer(&sk->sk_timer);
1703
1704 sk->sk_allocation = GFP_KERNEL;
1705 sk->sk_rcvbuf = sysctl_rmem_default;
1706 sk->sk_sndbuf = sysctl_wmem_default;
1707 sk->sk_state = TCP_CLOSE;
1708 sk_set_socket(sk, sock);
1709
1710 sock_set_flag(sk, SOCK_ZAPPED);
1711
1712 if (sock) {
1713 sk->sk_type = sock->type;
1714 sk->sk_sleep = &sock->wait;
1715 sock->sk = sk;
1716 } else
1717 sk->sk_sleep = NULL;
1718
1719 rwlock_init(&sk->sk_dst_lock);
1720 rwlock_init(&sk->sk_callback_lock);
1721 lockdep_set_class_and_name(&sk->sk_callback_lock,
1722 af_callback_keys + sk->sk_family,
1723 af_family_clock_key_strings[sk->sk_family]);
1724
1725 sk->sk_state_change = sock_def_wakeup;
1726 sk->sk_data_ready = sock_def_readable;
1727 sk->sk_write_space = sock_def_write_space;
1728 sk->sk_error_report = sock_def_error_report;
1729 sk->sk_destruct = sock_def_destruct;
1730
1731 sk->sk_sndmsg_page = NULL;
1732 sk->sk_sndmsg_off = 0;
1733
1734 sk->sk_peercred.pid = 0;
1735 sk->sk_peercred.uid = -1;
1736 sk->sk_peercred.gid = -1;
1737 sk->sk_write_pending = 0;
1738 sk->sk_rcvlowat = 1;
1739 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1740 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1741
1742 sk->sk_stamp = ktime_set(-1L, 0);
1743
1744 atomic_set(&sk->sk_refcnt, 1);
1745 atomic_set(&sk->sk_drops, 0);
1746 }
1747
1748 void lock_sock_nested(struct sock *sk, int subclass)
1749 {
1750 might_sleep();
1751 spin_lock_bh(&sk->sk_lock.slock);
1752 if (sk->sk_lock.owned)
1753 __lock_sock(sk);
1754 sk->sk_lock.owned = 1;
1755 spin_unlock(&sk->sk_lock.slock);
1756 /*
1757 * The sk_lock has mutex_lock() semantics here:
1758 */
1759 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1760 local_bh_enable();
1761 }
1762
1763 EXPORT_SYMBOL(lock_sock_nested);
1764
1765 void release_sock(struct sock *sk)
1766 {
1767 /*
1768 * The sk_lock has mutex_unlock() semantics:
1769 */
1770 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1771
1772 spin_lock_bh(&sk->sk_lock.slock);
1773 if (sk->sk_backlog.tail)
1774 __release_sock(sk);
1775 sk->sk_lock.owned = 0;
1776 if (waitqueue_active(&sk->sk_lock.wq))
1777 wake_up(&sk->sk_lock.wq);
1778 spin_unlock_bh(&sk->sk_lock.slock);
1779 }
1780 EXPORT_SYMBOL(release_sock);
1781
1782 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1783 {
1784 struct timeval tv;
1785 if (!sock_flag(sk, SOCK_TIMESTAMP))
1786 sock_enable_timestamp(sk);
1787 tv = ktime_to_timeval(sk->sk_stamp);
1788 if (tv.tv_sec == -1)
1789 return -ENOENT;
1790 if (tv.tv_sec == 0) {
1791 sk->sk_stamp = ktime_get_real();
1792 tv = ktime_to_timeval(sk->sk_stamp);
1793 }
1794 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1795 }
1796 EXPORT_SYMBOL(sock_get_timestamp);
1797
1798 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1799 {
1800 struct timespec ts;
1801 if (!sock_flag(sk, SOCK_TIMESTAMP))
1802 sock_enable_timestamp(sk);
1803 ts = ktime_to_timespec(sk->sk_stamp);
1804 if (ts.tv_sec == -1)
1805 return -ENOENT;
1806 if (ts.tv_sec == 0) {
1807 sk->sk_stamp = ktime_get_real();
1808 ts = ktime_to_timespec(sk->sk_stamp);
1809 }
1810 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1811 }
1812 EXPORT_SYMBOL(sock_get_timestampns);
1813
1814 void sock_enable_timestamp(struct sock *sk)
1815 {
1816 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1817 sock_set_flag(sk, SOCK_TIMESTAMP);
1818 net_enable_timestamp();
1819 }
1820 }
1821
1822 /*
1823 * Get a socket option on an socket.
1824 *
1825 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1826 * asynchronous errors should be reported by getsockopt. We assume
1827 * this means if you specify SO_ERROR (otherwise whats the point of it).
1828 */
1829 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1830 char __user *optval, int __user *optlen)
1831 {
1832 struct sock *sk = sock->sk;
1833
1834 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1835 }
1836
1837 EXPORT_SYMBOL(sock_common_getsockopt);
1838
1839 #ifdef CONFIG_COMPAT
1840 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1841 char __user *optval, int __user *optlen)
1842 {
1843 struct sock *sk = sock->sk;
1844
1845 if (sk->sk_prot->compat_getsockopt != NULL)
1846 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1847 optval, optlen);
1848 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1849 }
1850 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1851 #endif
1852
1853 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1854 struct msghdr *msg, size_t size, int flags)
1855 {
1856 struct sock *sk = sock->sk;
1857 int addr_len = 0;
1858 int err;
1859
1860 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1861 flags & ~MSG_DONTWAIT, &addr_len);
1862 if (err >= 0)
1863 msg->msg_namelen = addr_len;
1864 return err;
1865 }
1866
1867 EXPORT_SYMBOL(sock_common_recvmsg);
1868
1869 /*
1870 * Set socket options on an inet socket.
1871 */
1872 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1873 char __user *optval, int optlen)
1874 {
1875 struct sock *sk = sock->sk;
1876
1877 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1878 }
1879
1880 EXPORT_SYMBOL(sock_common_setsockopt);
1881
1882 #ifdef CONFIG_COMPAT
1883 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1884 char __user *optval, int optlen)
1885 {
1886 struct sock *sk = sock->sk;
1887
1888 if (sk->sk_prot->compat_setsockopt != NULL)
1889 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1890 optval, optlen);
1891 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1892 }
1893 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1894 #endif
1895
1896 void sk_common_release(struct sock *sk)
1897 {
1898 if (sk->sk_prot->destroy)
1899 sk->sk_prot->destroy(sk);
1900
1901 /*
1902 * Observation: when sock_common_release is called, processes have
1903 * no access to socket. But net still has.
1904 * Step one, detach it from networking:
1905 *
1906 * A. Remove from hash tables.
1907 */
1908
1909 sk->sk_prot->unhash(sk);
1910
1911 /*
1912 * In this point socket cannot receive new packets, but it is possible
1913 * that some packets are in flight because some CPU runs receiver and
1914 * did hash table lookup before we unhashed socket. They will achieve
1915 * receive queue and will be purged by socket destructor.
1916 *
1917 * Also we still have packets pending on receive queue and probably,
1918 * our own packets waiting in device queues. sock_destroy will drain
1919 * receive queue, but transmitted packets will delay socket destruction
1920 * until the last reference will be released.
1921 */
1922
1923 sock_orphan(sk);
1924
1925 xfrm_sk_free_policy(sk);
1926
1927 sk_refcnt_debug_release(sk);
1928 sock_put(sk);
1929 }
1930
1931 EXPORT_SYMBOL(sk_common_release);
1932
1933 static DEFINE_RWLOCK(proto_list_lock);
1934 static LIST_HEAD(proto_list);
1935
1936 #ifdef CONFIG_PROC_FS
1937 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1938 struct prot_inuse {
1939 int val[PROTO_INUSE_NR];
1940 };
1941
1942 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
1943
1944 #ifdef CONFIG_NET_NS
1945 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1946 {
1947 int cpu = smp_processor_id();
1948 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1949 }
1950 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1951
1952 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1953 {
1954 int cpu, idx = prot->inuse_idx;
1955 int res = 0;
1956
1957 for_each_possible_cpu(cpu)
1958 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1959
1960 return res >= 0 ? res : 0;
1961 }
1962 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1963
1964 static int sock_inuse_init_net(struct net *net)
1965 {
1966 net->core.inuse = alloc_percpu(struct prot_inuse);
1967 return net->core.inuse ? 0 : -ENOMEM;
1968 }
1969
1970 static void sock_inuse_exit_net(struct net *net)
1971 {
1972 free_percpu(net->core.inuse);
1973 }
1974
1975 static struct pernet_operations net_inuse_ops = {
1976 .init = sock_inuse_init_net,
1977 .exit = sock_inuse_exit_net,
1978 };
1979
1980 static __init int net_inuse_init(void)
1981 {
1982 if (register_pernet_subsys(&net_inuse_ops))
1983 panic("Cannot initialize net inuse counters");
1984
1985 return 0;
1986 }
1987
1988 core_initcall(net_inuse_init);
1989 #else
1990 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1991
1992 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1993 {
1994 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
1995 }
1996 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1997
1998 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1999 {
2000 int cpu, idx = prot->inuse_idx;
2001 int res = 0;
2002
2003 for_each_possible_cpu(cpu)
2004 res += per_cpu(prot_inuse, cpu).val[idx];
2005
2006 return res >= 0 ? res : 0;
2007 }
2008 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2009 #endif
2010
2011 static void assign_proto_idx(struct proto *prot)
2012 {
2013 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2014
2015 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2016 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2017 return;
2018 }
2019
2020 set_bit(prot->inuse_idx, proto_inuse_idx);
2021 }
2022
2023 static void release_proto_idx(struct proto *prot)
2024 {
2025 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2026 clear_bit(prot->inuse_idx, proto_inuse_idx);
2027 }
2028 #else
2029 static inline void assign_proto_idx(struct proto *prot)
2030 {
2031 }
2032
2033 static inline void release_proto_idx(struct proto *prot)
2034 {
2035 }
2036 #endif
2037
2038 int proto_register(struct proto *prot, int alloc_slab)
2039 {
2040 char *request_sock_slab_name = NULL;
2041 char *timewait_sock_slab_name;
2042
2043 if (alloc_slab) {
2044 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2045 SLAB_HWCACHE_ALIGN, NULL);
2046
2047 if (prot->slab == NULL) {
2048 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2049 prot->name);
2050 goto out;
2051 }
2052
2053 if (prot->rsk_prot != NULL) {
2054 static const char mask[] = "request_sock_%s";
2055
2056 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2057 if (request_sock_slab_name == NULL)
2058 goto out_free_sock_slab;
2059
2060 sprintf(request_sock_slab_name, mask, prot->name);
2061 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
2062 prot->rsk_prot->obj_size, 0,
2063 SLAB_HWCACHE_ALIGN, NULL);
2064
2065 if (prot->rsk_prot->slab == NULL) {
2066 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2067 prot->name);
2068 goto out_free_request_sock_slab_name;
2069 }
2070 }
2071
2072 if (prot->twsk_prot != NULL) {
2073 static const char mask[] = "tw_sock_%s";
2074
2075 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2076
2077 if (timewait_sock_slab_name == NULL)
2078 goto out_free_request_sock_slab;
2079
2080 sprintf(timewait_sock_slab_name, mask, prot->name);
2081 prot->twsk_prot->twsk_slab =
2082 kmem_cache_create(timewait_sock_slab_name,
2083 prot->twsk_prot->twsk_obj_size,
2084 0, SLAB_HWCACHE_ALIGN,
2085 NULL);
2086 if (prot->twsk_prot->twsk_slab == NULL)
2087 goto out_free_timewait_sock_slab_name;
2088 }
2089 }
2090
2091 write_lock(&proto_list_lock);
2092 list_add(&prot->node, &proto_list);
2093 assign_proto_idx(prot);
2094 write_unlock(&proto_list_lock);
2095 return 0;
2096
2097 out_free_timewait_sock_slab_name:
2098 kfree(timewait_sock_slab_name);
2099 out_free_request_sock_slab:
2100 if (prot->rsk_prot && prot->rsk_prot->slab) {
2101 kmem_cache_destroy(prot->rsk_prot->slab);
2102 prot->rsk_prot->slab = NULL;
2103 }
2104 out_free_request_sock_slab_name:
2105 kfree(request_sock_slab_name);
2106 out_free_sock_slab:
2107 kmem_cache_destroy(prot->slab);
2108 prot->slab = NULL;
2109 out:
2110 return -ENOBUFS;
2111 }
2112
2113 EXPORT_SYMBOL(proto_register);
2114
2115 void proto_unregister(struct proto *prot)
2116 {
2117 write_lock(&proto_list_lock);
2118 release_proto_idx(prot);
2119 list_del(&prot->node);
2120 write_unlock(&proto_list_lock);
2121
2122 if (prot->slab != NULL) {
2123 kmem_cache_destroy(prot->slab);
2124 prot->slab = NULL;
2125 }
2126
2127 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2128 const char *name = kmem_cache_name(prot->rsk_prot->slab);
2129
2130 kmem_cache_destroy(prot->rsk_prot->slab);
2131 kfree(name);
2132 prot->rsk_prot->slab = NULL;
2133 }
2134
2135 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2136 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
2137
2138 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2139 kfree(name);
2140 prot->twsk_prot->twsk_slab = NULL;
2141 }
2142 }
2143
2144 EXPORT_SYMBOL(proto_unregister);
2145
2146 #ifdef CONFIG_PROC_FS
2147 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2148 __acquires(proto_list_lock)
2149 {
2150 read_lock(&proto_list_lock);
2151 return seq_list_start_head(&proto_list, *pos);
2152 }
2153
2154 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2155 {
2156 return seq_list_next(v, &proto_list, pos);
2157 }
2158
2159 static void proto_seq_stop(struct seq_file *seq, void *v)
2160 __releases(proto_list_lock)
2161 {
2162 read_unlock(&proto_list_lock);
2163 }
2164
2165 static char proto_method_implemented(const void *method)
2166 {
2167 return method == NULL ? 'n' : 'y';
2168 }
2169
2170 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2171 {
2172 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2173 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2174 proto->name,
2175 proto->obj_size,
2176 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
2177 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2178 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2179 proto->max_header,
2180 proto->slab == NULL ? "no" : "yes",
2181 module_name(proto->owner),
2182 proto_method_implemented(proto->close),
2183 proto_method_implemented(proto->connect),
2184 proto_method_implemented(proto->disconnect),
2185 proto_method_implemented(proto->accept),
2186 proto_method_implemented(proto->ioctl),
2187 proto_method_implemented(proto->init),
2188 proto_method_implemented(proto->destroy),
2189 proto_method_implemented(proto->shutdown),
2190 proto_method_implemented(proto->setsockopt),
2191 proto_method_implemented(proto->getsockopt),
2192 proto_method_implemented(proto->sendmsg),
2193 proto_method_implemented(proto->recvmsg),
2194 proto_method_implemented(proto->sendpage),
2195 proto_method_implemented(proto->bind),
2196 proto_method_implemented(proto->backlog_rcv),
2197 proto_method_implemented(proto->hash),
2198 proto_method_implemented(proto->unhash),
2199 proto_method_implemented(proto->get_port),
2200 proto_method_implemented(proto->enter_memory_pressure));
2201 }
2202
2203 static int proto_seq_show(struct seq_file *seq, void *v)
2204 {
2205 if (v == &proto_list)
2206 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2207 "protocol",
2208 "size",
2209 "sockets",
2210 "memory",
2211 "press",
2212 "maxhdr",
2213 "slab",
2214 "module",
2215 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2216 else
2217 proto_seq_printf(seq, list_entry(v, struct proto, node));
2218 return 0;
2219 }
2220
2221 static const struct seq_operations proto_seq_ops = {
2222 .start = proto_seq_start,
2223 .next = proto_seq_next,
2224 .stop = proto_seq_stop,
2225 .show = proto_seq_show,
2226 };
2227
2228 static int proto_seq_open(struct inode *inode, struct file *file)
2229 {
2230 return seq_open(file, &proto_seq_ops);
2231 }
2232
2233 static const struct file_operations proto_seq_fops = {
2234 .owner = THIS_MODULE,
2235 .open = proto_seq_open,
2236 .read = seq_read,
2237 .llseek = seq_lseek,
2238 .release = seq_release,
2239 };
2240
2241 static int __init proto_init(void)
2242 {
2243 /* register /proc/net/protocols */
2244 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
2245 }
2246
2247 subsys_initcall(proto_init);
2248
2249 #endif /* PROC_FS */
2250
2251 EXPORT_SYMBOL(sk_alloc);
2252 EXPORT_SYMBOL(sk_free);
2253 EXPORT_SYMBOL(sk_send_sigurg);
2254 EXPORT_SYMBOL(sock_alloc_send_skb);
2255 EXPORT_SYMBOL(sock_init_data);
2256 EXPORT_SYMBOL(sock_kfree_s);
2257 EXPORT_SYMBOL(sock_kmalloc);
2258 EXPORT_SYMBOL(sock_no_accept);
2259 EXPORT_SYMBOL(sock_no_bind);
2260 EXPORT_SYMBOL(sock_no_connect);
2261 EXPORT_SYMBOL(sock_no_getname);
2262 EXPORT_SYMBOL(sock_no_getsockopt);
2263 EXPORT_SYMBOL(sock_no_ioctl);
2264 EXPORT_SYMBOL(sock_no_listen);
2265 EXPORT_SYMBOL(sock_no_mmap);
2266 EXPORT_SYMBOL(sock_no_poll);
2267 EXPORT_SYMBOL(sock_no_recvmsg);
2268 EXPORT_SYMBOL(sock_no_sendmsg);
2269 EXPORT_SYMBOL(sock_no_sendpage);
2270 EXPORT_SYMBOL(sock_no_setsockopt);
2271 EXPORT_SYMBOL(sock_no_shutdown);
2272 EXPORT_SYMBOL(sock_no_socketpair);
2273 EXPORT_SYMBOL(sock_rfree);
2274 EXPORT_SYMBOL(sock_setsockopt);
2275 EXPORT_SYMBOL(sock_wfree);
2276 EXPORT_SYMBOL(sock_wmalloc);
2277 EXPORT_SYMBOL(sock_i_uid);
2278 EXPORT_SYMBOL(sock_i_ino);
2279 EXPORT_SYMBOL(sysctl_optmem_max);
Stable & featurefull patchset based on Linus's git branch