?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
added 2.6.29.6 aldebaran kernel
[nao-ulib.git] / kernel / 2.6.29.6-aldebaran-rt / net / core / sock.c
blob08ec8b0b5eec536b3d75c5ffe2dfeaa344e8a160
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 /*
140 * Make lock validator output more readable. (we pre-construct these
141 * strings build-time, so that runtime initialization of socket
142 * locks is fast):
143 */
144 static const char *af_family_key_strings[AF_MAX+1] = {
145 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
146 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
147 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
148 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
149 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
150 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
151 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
152 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
153 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
154 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
155 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
156 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
157 "sk_lock-AF_MAX"
158 };
159 static const char *af_family_slock_key_strings[AF_MAX+1] = {
160 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
161 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
162 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
163 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
164 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
165 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
166 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
167 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
168 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
169 "slock-27" , "slock-28" , "slock-AF_CAN" ,
170 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
171 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
172 "slock-AF_MAX"
173 };
174 static const char *af_family_clock_key_strings[AF_MAX+1] = {
175 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
176 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
177 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
178 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
179 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
180 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
181 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
182 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
183 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
184 "clock-27" , "clock-28" , "clock-AF_CAN" ,
185 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
186 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
187 "clock-AF_MAX"
188 };
189
190 /*
191 * sk_callback_lock locking rules are per-address-family,
192 * so split the lock classes by using a per-AF key:
193 */
194 static struct lock_class_key af_callback_keys[AF_MAX];
195
196 /* Take into consideration the size of the struct sk_buff overhead in the
197 * determination of these values, since that is non-constant across
198 * platforms. This makes socket queueing behavior and performance
199 * not depend upon such differences.
200 */
201 #define _SK_MEM_PACKETS 256
202 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
203 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
204 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205
206 /* Run time adjustable parameters. */
207 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
208 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
209 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
210 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
211
212 /* Maximal space eaten by iovec or ancilliary data plus some space */
213 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
214
215 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
216 {
217 struct timeval tv;
218
219 if (optlen < sizeof(tv))
220 return -EINVAL;
221 if (copy_from_user(&tv, optval, sizeof(tv)))
222 return -EFAULT;
223 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
224 return -EDOM;
225
226 if (tv.tv_sec < 0) {
227 static int warned __read_mostly;
228
229 *timeo_p = 0;
230 if (warned < 10 && net_ratelimit()) {
231 warned++;
232 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
233 "tries to set negative timeout\n",
234 current->comm, task_pid_nr(current));
235 }
236 return 0;
237 }
238 *timeo_p = MAX_SCHEDULE_TIMEOUT;
239 if (tv.tv_sec == 0 && tv.tv_usec == 0)
240 return 0;
241 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
242 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
243 return 0;
244 }
245
246 static void sock_warn_obsolete_bsdism(const char *name)
247 {
248 static int warned;
249 static char warncomm[TASK_COMM_LEN];
250 if (strcmp(warncomm, current->comm) && warned < 5) {
251 strcpy(warncomm, current->comm);
252 printk(KERN_WARNING "process `%s' is using obsolete "
253 "%s SO_BSDCOMPAT\n", warncomm, name);
254 warned++;
255 }
256 }
257
258 static void sock_disable_timestamp(struct sock *sk)
259 {
260 if (sock_flag(sk, SOCK_TIMESTAMP)) {
261 sock_reset_flag(sk, SOCK_TIMESTAMP);
262 net_disable_timestamp();
263 }
264 }
265
266
267 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
268 {
269 int err = 0;
270 int skb_len;
271
272 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
273 number of warnings when compiling with -W --ANK
274 */
275 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
276 (unsigned)sk->sk_rcvbuf) {
277 err = -ENOMEM;
278 goto out;
279 }
280
281 err = sk_filter(sk, skb);
282 if (err)
283 goto out;
284
285 if (!sk_rmem_schedule(sk, skb->truesize)) {
286 err = -ENOBUFS;
287 goto out;
288 }
289
290 skb->dev = NULL;
291 skb_set_owner_r(skb, sk);
292
293 /* Cache the SKB length before we tack it onto the receive
294 * queue. Once it is added it no longer belongs to us and
295 * may be freed by other threads of control pulling packets
296 * from the queue.
297 */
298 skb_len = skb->len;
299
300 skb_queue_tail(&sk->sk_receive_queue, skb);
301
302 if (!sock_flag(sk, SOCK_DEAD))
303 sk->sk_data_ready(sk, skb_len);
304 out:
305 return err;
306 }
307 EXPORT_SYMBOL(sock_queue_rcv_skb);
308
309 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
310 {
311 int rc = NET_RX_SUCCESS;
312
313 if (sk_filter(sk, skb))
314 goto discard_and_relse;
315
316 skb->dev = NULL;
317
318 if (nested)
319 bh_lock_sock_nested(sk);
320 else
321 bh_lock_sock(sk);
322 if (!sock_owned_by_user(sk)) {
323 /*
324 * trylock + unlock semantics:
325 */
326 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
327
328 rc = sk_backlog_rcv(sk, skb);
329
330 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
331 } else
332 sk_add_backlog(sk, skb);
333 bh_unlock_sock(sk);
334 out:
335 sock_put(sk);
336 return rc;
337 discard_and_relse:
338 kfree_skb(skb);
339 goto out;
340 }
341 EXPORT_SYMBOL(sk_receive_skb);
342
343 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
344 {
345 struct dst_entry *dst = sk->sk_dst_cache;
346
347 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
348 sk->sk_dst_cache = NULL;
349 dst_release(dst);
350 return NULL;
351 }
352
353 return dst;
354 }
355 EXPORT_SYMBOL(__sk_dst_check);
356
357 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
358 {
359 struct dst_entry *dst = sk_dst_get(sk);
360
361 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
362 sk_dst_reset(sk);
363 dst_release(dst);
364 return NULL;
365 }
366
367 return dst;
368 }
369 EXPORT_SYMBOL(sk_dst_check);
370
371 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
372 {
373 int ret = -ENOPROTOOPT;
374 #ifdef CONFIG_NETDEVICES
375 struct net *net = sock_net(sk);
376 char devname[IFNAMSIZ];
377 int index;
378
379 /* Sorry... */
380 ret = -EPERM;
381 if (!capable(CAP_NET_RAW))
382 goto out;
383
384 ret = -EINVAL;
385 if (optlen < 0)
386 goto out;
387
388 /* Bind this socket to a particular device like "eth0",
389 * as specified in the passed interface name. If the
390 * name is "" or the option length is zero the socket
391 * is not bound.
392 */
393 if (optlen > IFNAMSIZ - 1)
394 optlen = IFNAMSIZ - 1;
395 memset(devname, 0, sizeof(devname));
396
397 ret = -EFAULT;
398 if (copy_from_user(devname, optval, optlen))
399 goto out;
400
401 if (devname[0] == '\0') {
402 index = 0;
403 } else {
404 struct net_device *dev = dev_get_by_name(net, devname);
405
406 ret = -ENODEV;
407 if (!dev)
408 goto out;
409
410 index = dev->ifindex;
411 dev_put(dev);
412 }
413
414 lock_sock(sk);
415 sk->sk_bound_dev_if = index;
416 sk_dst_reset(sk);
417 release_sock(sk);
418
419 ret = 0;
420
421 out:
422 #endif
423
424 return ret;
425 }
426
427 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
428 {
429 if (valbool)
430 sock_set_flag(sk, bit);
431 else
432 sock_reset_flag(sk, bit);
433 }
434
435 /*
436 * This is meant for all protocols to use and covers goings on
437 * at the socket level. Everything here is generic.
438 */
439
440 int sock_setsockopt(struct socket *sock, int level, int optname,
441 char __user *optval, int optlen)
442 {
443 struct sock *sk=sock->sk;
444 int val;
445 int valbool;
446 struct linger ling;
447 int ret = 0;
448
449 /*
450 * Options without arguments
451 */
452
453 if (optname == SO_BINDTODEVICE)
454 return sock_bindtodevice(sk, optval, optlen);
455
456 if (optlen < sizeof(int))
457 return -EINVAL;
458
459 if (get_user(val, (int __user *)optval))
460 return -EFAULT;
461
462 valbool = val?1:0;
463
464 lock_sock(sk);
465
466 switch(optname) {
467 case SO_DEBUG:
468 if (val && !capable(CAP_NET_ADMIN)) {
469 ret = -EACCES;
470 } else
471 sock_valbool_flag(sk, SOCK_DBG, valbool);
472 break;
473 case SO_REUSEADDR:
474 sk->sk_reuse = valbool;
475 break;
476 case SO_TYPE:
477 case SO_ERROR:
478 ret = -ENOPROTOOPT;
479 break;
480 case SO_DONTROUTE:
481 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
482 break;
483 case SO_BROADCAST:
484 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
485 break;
486 case SO_SNDBUF:
487 /* Don't error on this BSD doesn't and if you think
488 about it this is right. Otherwise apps have to
489 play 'guess the biggest size' games. RCVBUF/SNDBUF
490 are treated in BSD as hints */
491
492 if (val > sysctl_wmem_max)
493 val = sysctl_wmem_max;
494 set_sndbuf:
495 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
496 if ((val * 2) < SOCK_MIN_SNDBUF)
497 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
498 else
499 sk->sk_sndbuf = val * 2;
500
501 /*
502 * Wake up sending tasks if we
503 * upped the value.
504 */
505 sk->sk_write_space(sk);
506 break;
507
508 case SO_SNDBUFFORCE:
509 if (!capable(CAP_NET_ADMIN)) {
510 ret = -EPERM;
511 break;
512 }
513 goto set_sndbuf;
514
515 case SO_RCVBUF:
516 /* Don't error on this BSD doesn't and if you think
517 about it this is right. Otherwise apps have to
518 play 'guess the biggest size' games. RCVBUF/SNDBUF
519 are treated in BSD as hints */
520
521 if (val > sysctl_rmem_max)
522 val = sysctl_rmem_max;
523 set_rcvbuf:
524 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
525 /*
526 * We double it on the way in to account for
527 * "struct sk_buff" etc. overhead. Applications
528 * assume that the SO_RCVBUF setting they make will
529 * allow that much actual data to be received on that
530 * socket.
531 *
532 * Applications are unaware that "struct sk_buff" and
533 * other overheads allocate from the receive buffer
534 * during socket buffer allocation.
535 *
536 * And after considering the possible alternatives,
537 * returning the value we actually used in getsockopt
538 * is the most desirable behavior.
539 */
540 if ((val * 2) < SOCK_MIN_RCVBUF)
541 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
542 else
543 sk->sk_rcvbuf = val * 2;
544 break;
545
546 case SO_RCVBUFFORCE:
547 if (!capable(CAP_NET_ADMIN)) {
548 ret = -EPERM;
549 break;
550 }
551 goto set_rcvbuf;
552
553 case SO_KEEPALIVE:
554 #ifdef CONFIG_INET
555 if (sk->sk_protocol == IPPROTO_TCP)
556 tcp_set_keepalive(sk, valbool);
557 #endif
558 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
559 break;
560
561 case SO_OOBINLINE:
562 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
563 break;
564
565 case SO_NO_CHECK:
566 sk->sk_no_check = valbool;
567 break;
568
569 case SO_PRIORITY:
570 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
571 sk->sk_priority = val;
572 else
573 ret = -EPERM;
574 break;
575
576 case SO_LINGER:
577 if (optlen < sizeof(ling)) {
578 ret = -EINVAL; /* 1003.1g */
579 break;
580 }
581 if (copy_from_user(&ling,optval,sizeof(ling))) {
582 ret = -EFAULT;
583 break;
584 }
585 if (!ling.l_onoff)
586 sock_reset_flag(sk, SOCK_LINGER);
587 else {
588 #if (BITS_PER_LONG == 32)
589 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
590 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
591 else
592 #endif
593 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
594 sock_set_flag(sk, SOCK_LINGER);
595 }
596 break;
597
598 case SO_BSDCOMPAT:
599 sock_warn_obsolete_bsdism("setsockopt");
600 break;
601
602 case SO_PASSCRED:
603 if (valbool)
604 set_bit(SOCK_PASSCRED, &sock->flags);
605 else
606 clear_bit(SOCK_PASSCRED, &sock->flags);
607 break;
608
609 case SO_TIMESTAMP:
610 case SO_TIMESTAMPNS:
611 if (valbool) {
612 if (optname == SO_TIMESTAMP)
613 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
614 else
615 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
616 sock_set_flag(sk, SOCK_RCVTSTAMP);
617 sock_enable_timestamp(sk);
618 } else {
619 sock_reset_flag(sk, SOCK_RCVTSTAMP);
620 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
621 }
622 break;
623
624 case SO_RCVLOWAT:
625 if (val < 0)
626 val = INT_MAX;
627 sk->sk_rcvlowat = val ? : 1;
628 break;
629
630 case SO_RCVTIMEO:
631 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
632 break;
633
634 case SO_SNDTIMEO:
635 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
636 break;
637
638 case SO_ATTACH_FILTER:
639 ret = -EINVAL;
640 if (optlen == sizeof(struct sock_fprog)) {
641 struct sock_fprog fprog;
642
643 ret = -EFAULT;
644 if (copy_from_user(&fprog, optval, sizeof(fprog)))
645 break;
646
647 ret = sk_attach_filter(&fprog, sk);
648 }
649 break;
650
651 case SO_DETACH_FILTER:
652 ret = sk_detach_filter(sk);
653 break;
654
655 case SO_PASSSEC:
656 if (valbool)
657 set_bit(SOCK_PASSSEC, &sock->flags);
658 else
659 clear_bit(SOCK_PASSSEC, &sock->flags);
660 break;
661 case SO_MARK:
662 if (!capable(CAP_NET_ADMIN))
663 ret = -EPERM;
664 else {
665 sk->sk_mark = val;
666 }
667 break;
668
669 /* We implement the SO_SNDLOWAT etc to
670 not be settable (1003.1g 5.3) */
671 default:
672 ret = -ENOPROTOOPT;
673 break;
674 }
675 release_sock(sk);
676 return ret;
677 }
678
679
680 int sock_getsockopt(struct socket *sock, int level, int optname,
681 char __user *optval, int __user *optlen)
682 {
683 struct sock *sk = sock->sk;
684
685 union {
686 int val;
687 struct linger ling;
688 struct timeval tm;
689 } v;
690
691 unsigned int lv = sizeof(int);
692 int len;
693
694 if (get_user(len, optlen))
695 return -EFAULT;
696 if (len < 0)
697 return -EINVAL;
698
699 memset(&v, 0, sizeof(v));
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 kmemcheck_annotate_bitfield(sk->flags);
898
899 if (security_sk_alloc(sk, family, priority))
900 goto out_free;
901
902 if (!try_module_get(prot->owner))
903 goto out_free_sec;
904 }
905
906 return sk;
907
908 out_free_sec:
909 security_sk_free(sk);
910 out_free:
911 if (slab != NULL)
912 kmem_cache_free(slab, sk);
913 else
914 kfree(sk);
915 return NULL;
916 }
917
918 static void sk_prot_free(struct proto *prot, struct sock *sk)
919 {
920 struct kmem_cache *slab;
921 struct module *owner;
922
923 owner = prot->owner;
924 slab = prot->slab;
925
926 security_sk_free(sk);
927 if (slab != NULL)
928 kmem_cache_free(slab, sk);
929 else
930 kfree(sk);
931 module_put(owner);
932 }
933
934 /**
935 * sk_alloc - All socket objects are allocated here
936 * @net: the applicable net namespace
937 * @family: protocol family
938 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
939 * @prot: struct proto associated with this new sock instance
940 */
941 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
942 struct proto *prot)
943 {
944 struct sock *sk;
945
946 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
947 if (sk) {
948 sk->sk_family = family;
949 /*
950 * See comment in struct sock definition to understand
951 * why we need sk_prot_creator -acme
952 */
953 sk->sk_prot = sk->sk_prot_creator = prot;
954 sock_lock_init(sk);
955 sock_net_set(sk, get_net(net));
956 }
957
958 return sk;
959 }
960
961 void sk_free(struct sock *sk)
962 {
963 struct sk_filter *filter;
964
965 if (sk->sk_destruct)
966 sk->sk_destruct(sk);
967
968 filter = rcu_dereference(sk->sk_filter);
969 if (filter) {
970 sk_filter_uncharge(sk, filter);
971 rcu_assign_pointer(sk->sk_filter, NULL);
972 }
973
974 sock_disable_timestamp(sk);
975
976 if (atomic_read(&sk->sk_omem_alloc))
977 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
978 __func__, atomic_read(&sk->sk_omem_alloc));
979
980 put_net(sock_net(sk));
981 sk_prot_free(sk->sk_prot_creator, sk);
982 }
983
984 /*
985 * Last sock_put should drop referrence to sk->sk_net. It has already
986 * been dropped in sk_change_net. Taking referrence to stopping namespace
987 * is not an option.
988 * Take referrence to a socket to remove it from hash _alive_ and after that
989 * destroy it in the context of init_net.
990 */
991 void sk_release_kernel(struct sock *sk)
992 {
993 if (sk == NULL || sk->sk_socket == NULL)
994 return;
995
996 sock_hold(sk);
997 sock_release(sk->sk_socket);
998 release_net(sock_net(sk));
999 sock_net_set(sk, get_net(&init_net));
1000 sock_put(sk);
1001 }
1002 EXPORT_SYMBOL(sk_release_kernel);
1003
1004 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1005 {
1006 struct sock *newsk;
1007
1008 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1009 if (newsk != NULL) {
1010 struct sk_filter *filter;
1011
1012 sock_copy(newsk, sk);
1013
1014 /* SANITY */
1015 get_net(sock_net(newsk));
1016 sk_node_init(&newsk->sk_node);
1017 sock_lock_init(newsk);
1018 bh_lock_sock(newsk);
1019 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1020
1021 atomic_set(&newsk->sk_rmem_alloc, 0);
1022 atomic_set(&newsk->sk_wmem_alloc, 0);
1023 atomic_set(&newsk->sk_omem_alloc, 0);
1024 skb_queue_head_init(&newsk->sk_receive_queue);
1025 skb_queue_head_init(&newsk->sk_write_queue);
1026 #ifdef CONFIG_NET_DMA
1027 skb_queue_head_init(&newsk->sk_async_wait_queue);
1028 #endif
1029
1030 rwlock_init(&newsk->sk_dst_lock);
1031 rwlock_init(&newsk->sk_callback_lock);
1032 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1033 af_callback_keys + newsk->sk_family,
1034 af_family_clock_key_strings[newsk->sk_family]);
1035
1036 newsk->sk_dst_cache = NULL;
1037 newsk->sk_wmem_queued = 0;
1038 newsk->sk_forward_alloc = 0;
1039 newsk->sk_send_head = NULL;
1040 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1041
1042 sock_reset_flag(newsk, SOCK_DONE);
1043 skb_queue_head_init(&newsk->sk_error_queue);
1044
1045 filter = newsk->sk_filter;
1046 if (filter != NULL)
1047 sk_filter_charge(newsk, filter);
1048
1049 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1050 /* It is still raw copy of parent, so invalidate
1051 * destructor and make plain sk_free() */
1052 newsk->sk_destruct = NULL;
1053 sk_free(newsk);
1054 newsk = NULL;
1055 goto out;
1056 }
1057
1058 newsk->sk_err = 0;
1059 newsk->sk_priority = 0;
1060 atomic_set(&newsk->sk_refcnt, 2);
1061
1062 /*
1063 * Increment the counter in the same struct proto as the master
1064 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1065 * is the same as sk->sk_prot->socks, as this field was copied
1066 * with memcpy).
1067 *
1068 * This _changes_ the previous behaviour, where
1069 * tcp_create_openreq_child always was incrementing the
1070 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1071 * to be taken into account in all callers. -acme
1072 */
1073 sk_refcnt_debug_inc(newsk);
1074 sk_set_socket(newsk, NULL);
1075 newsk->sk_sleep = NULL;
1076
1077 if (newsk->sk_prot->sockets_allocated)
1078 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1079 }
1080 out:
1081 return newsk;
1082 }
1083
1084 EXPORT_SYMBOL_GPL(sk_clone);
1085
1086 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1087 {
1088 __sk_dst_set(sk, dst);
1089 sk->sk_route_caps = dst->dev->features;
1090 if (sk->sk_route_caps & NETIF_F_GSO)
1091 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1092 if (sk_can_gso(sk)) {
1093 if (dst->header_len) {
1094 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1095 } else {
1096 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1097 sk->sk_gso_max_size = dst->dev->gso_max_size;
1098 }
1099 }
1100 }
1101 EXPORT_SYMBOL_GPL(sk_setup_caps);
1102
1103 void __init sk_init(void)
1104 {
1105 if (num_physpages <= 4096) {
1106 sysctl_wmem_max = 32767;
1107 sysctl_rmem_max = 32767;
1108 sysctl_wmem_default = 32767;
1109 sysctl_rmem_default = 32767;
1110 } else if (num_physpages >= 131072) {
1111 sysctl_wmem_max = 131071;
1112 sysctl_rmem_max = 131071;
1113 }
1114 }
1115
1116 /*
1117 * Simple resource managers for sockets.
1118 */
1119
1120
1121 /*
1122 * Write buffer destructor automatically called from kfree_skb.
1123 */
1124 void sock_wfree(struct sk_buff *skb)
1125 {
1126 struct sock *sk = skb->sk;
1127
1128 /* In case it might be waiting for more memory. */
1129 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1130 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1131 sk->sk_write_space(sk);
1132 sock_put(sk);
1133 }
1134
1135 /*
1136 * Read buffer destructor automatically called from kfree_skb.
1137 */
1138 void sock_rfree(struct sk_buff *skb)
1139 {
1140 struct sock *sk = skb->sk;
1141
1142 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1143 sk_mem_uncharge(skb->sk, skb->truesize);
1144 }
1145
1146
1147 int sock_i_uid(struct sock *sk)
1148 {
1149 int uid;
1150
1151 read_lock(&sk->sk_callback_lock);
1152 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1153 read_unlock(&sk->sk_callback_lock);
1154 return uid;
1155 }
1156
1157 unsigned long sock_i_ino(struct sock *sk)
1158 {
1159 unsigned long ino;
1160
1161 read_lock(&sk->sk_callback_lock);
1162 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1163 read_unlock(&sk->sk_callback_lock);
1164 return ino;
1165 }
1166
1167 /*
1168 * Allocate a skb from the socket's send buffer.
1169 */
1170 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1171 gfp_t priority)
1172 {
1173 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1174 struct sk_buff * skb = alloc_skb(size, priority);
1175 if (skb) {
1176 skb_set_owner_w(skb, sk);
1177 return skb;
1178 }
1179 }
1180 return NULL;
1181 }
1182
1183 /*
1184 * Allocate a skb from the socket's receive buffer.
1185 */
1186 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1187 gfp_t priority)
1188 {
1189 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1190 struct sk_buff *skb = alloc_skb(size, priority);
1191 if (skb) {
1192 skb_set_owner_r(skb, sk);
1193 return skb;
1194 }
1195 }
1196 return NULL;
1197 }
1198
1199 /*
1200 * Allocate a memory block from the socket's option memory buffer.
1201 */
1202 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1203 {
1204 if ((unsigned)size <= sysctl_optmem_max &&
1205 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1206 void *mem;
1207 /* First do the add, to avoid the race if kmalloc
1208 * might sleep.
1209 */
1210 atomic_add(size, &sk->sk_omem_alloc);
1211 mem = kmalloc(size, priority);
1212 if (mem)
1213 return mem;
1214 atomic_sub(size, &sk->sk_omem_alloc);
1215 }
1216 return NULL;
1217 }
1218
1219 /*
1220 * Free an option memory block.
1221 */
1222 void sock_kfree_s(struct sock *sk, void *mem, int size)
1223 {
1224 kfree(mem);
1225 atomic_sub(size, &sk->sk_omem_alloc);
1226 }
1227
1228 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1229 I think, these locks should be removed for datagram sockets.
1230 */
1231 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1232 {
1233 DEFINE_WAIT(wait);
1234
1235 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1236 for (;;) {
1237 if (!timeo)
1238 break;
1239 if (signal_pending(current))
1240 break;
1241 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1242 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1243 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1244 break;
1245 if (sk->sk_shutdown & SEND_SHUTDOWN)
1246 break;
1247 if (sk->sk_err)
1248 break;
1249 timeo = schedule_timeout(timeo);
1250 }
1251 finish_wait(sk->sk_sleep, &wait);
1252 return timeo;
1253 }
1254
1255
1256 /*
1257 * Generic send/receive buffer handlers
1258 */
1259
1260 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1261 unsigned long header_len,
1262 unsigned long data_len,
1263 int noblock, int *errcode)
1264 {
1265 struct sk_buff *skb;
1266 gfp_t gfp_mask;
1267 long timeo;
1268 int err;
1269
1270 gfp_mask = sk->sk_allocation;
1271 if (gfp_mask & __GFP_WAIT)
1272 gfp_mask |= __GFP_REPEAT;
1273
1274 timeo = sock_sndtimeo(sk, noblock);
1275 while (1) {
1276 err = sock_error(sk);
1277 if (err != 0)
1278 goto failure;
1279
1280 err = -EPIPE;
1281 if (sk->sk_shutdown & SEND_SHUTDOWN)
1282 goto failure;
1283
1284 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1285 skb = alloc_skb(header_len, gfp_mask);
1286 if (skb) {
1287 int npages;
1288 int i;
1289
1290 /* No pages, we're done... */
1291 if (!data_len)
1292 break;
1293
1294 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1295 skb->truesize += data_len;
1296 skb_shinfo(skb)->nr_frags = npages;
1297 for (i = 0; i < npages; i++) {
1298 struct page *page;
1299 skb_frag_t *frag;
1300
1301 page = alloc_pages(sk->sk_allocation, 0);
1302 if (!page) {
1303 err = -ENOBUFS;
1304 skb_shinfo(skb)->nr_frags = i;
1305 kfree_skb(skb);
1306 goto failure;
1307 }
1308
1309 frag = &skb_shinfo(skb)->frags[i];
1310 frag->page = page;
1311 frag->page_offset = 0;
1312 frag->size = (data_len >= PAGE_SIZE ?
1313 PAGE_SIZE :
1314 data_len);
1315 data_len -= PAGE_SIZE;
1316 }
1317
1318 /* Full success... */
1319 break;
1320 }
1321 err = -ENOBUFS;
1322 goto failure;
1323 }
1324 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1325 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1326 err = -EAGAIN;
1327 if (!timeo)
1328 goto failure;
1329 if (signal_pending(current))
1330 goto interrupted;
1331 timeo = sock_wait_for_wmem(sk, timeo);
1332 }
1333
1334 skb_set_owner_w(skb, sk);
1335 return skb;
1336
1337 interrupted:
1338 err = sock_intr_errno(timeo);
1339 failure:
1340 *errcode = err;
1341 return NULL;
1342 }
1343
1344 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1345 int noblock, int *errcode)
1346 {
1347 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1348 }
1349
1350 static void __lock_sock(struct sock *sk)
1351 {
1352 DEFINE_WAIT(wait);
1353
1354 for (;;) {
1355 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1356 TASK_UNINTERRUPTIBLE);
1357 spin_unlock_bh(&sk->sk_lock.slock);
1358 schedule();
1359 spin_lock_bh(&sk->sk_lock.slock);
1360 if (!sock_owned_by_user(sk))
1361 break;
1362 }
1363 finish_wait(&sk->sk_lock.wq, &wait);
1364 }
1365
1366 static void __release_sock(struct sock *sk)
1367 {
1368 struct sk_buff *skb = sk->sk_backlog.head;
1369
1370 do {
1371 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1372 bh_unlock_sock(sk);
1373
1374 do {
1375 struct sk_buff *next = skb->next;
1376
1377 skb->next = NULL;
1378 sk_backlog_rcv(sk, skb);
1379
1380 /*
1381 * We are in process context here with softirqs
1382 * disabled, use cond_resched_softirq() to preempt.
1383 * This is safe to do because we've taken the backlog
1384 * queue private:
1385 */
1386 cond_resched_softirq();
1387
1388 skb = next;
1389 } while (skb != NULL);
1390
1391 bh_lock_sock(sk);
1392 } while ((skb = sk->sk_backlog.head) != NULL);
1393 }
1394
1395 /**
1396 * sk_wait_data - wait for data to arrive at sk_receive_queue
1397 * @sk: sock to wait on
1398 * @timeo: for how long
1399 *
1400 * Now socket state including sk->sk_err is changed only under lock,
1401 * hence we may omit checks after joining wait queue.
1402 * We check receive queue before schedule() only as optimization;
1403 * it is very likely that release_sock() added new data.
1404 */
1405 int sk_wait_data(struct sock *sk, long *timeo)
1406 {
1407 int rc;
1408 DEFINE_WAIT(wait);
1409
1410 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1411 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1412 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1413 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1414 finish_wait(sk->sk_sleep, &wait);
1415 return rc;
1416 }
1417
1418 EXPORT_SYMBOL(sk_wait_data);
1419
1420 /**
1421 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1422 * @sk: socket
1423 * @size: memory size to allocate
1424 * @kind: allocation type
1425 *
1426 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1427 * rmem allocation. This function assumes that protocols which have
1428 * memory_pressure use sk_wmem_queued as write buffer accounting.
1429 */
1430 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1431 {
1432 struct proto *prot = sk->sk_prot;
1433 int amt = sk_mem_pages(size);
1434 int allocated;
1435
1436 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1437 allocated = atomic_add_return(amt, prot->memory_allocated);
1438
1439 /* Under limit. */
1440 if (allocated <= prot->sysctl_mem[0]) {
1441 if (prot->memory_pressure && *prot->memory_pressure)
1442 *prot->memory_pressure = 0;
1443 return 1;
1444 }
1445
1446 /* Under pressure. */
1447 if (allocated > prot->sysctl_mem[1])
1448 if (prot->enter_memory_pressure)
1449 prot->enter_memory_pressure(sk);
1450
1451 /* Over hard limit. */
1452 if (allocated > prot->sysctl_mem[2])
1453 goto suppress_allocation;
1454
1455 /* guarantee minimum buffer size under pressure */
1456 if (kind == SK_MEM_RECV) {
1457 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1458 return 1;
1459 } else { /* SK_MEM_SEND */
1460 if (sk->sk_type == SOCK_STREAM) {
1461 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1462 return 1;
1463 } else if (atomic_read(&sk->sk_wmem_alloc) <
1464 prot->sysctl_wmem[0])
1465 return 1;
1466 }
1467
1468 if (prot->memory_pressure) {
1469 int alloc;
1470
1471 if (!*prot->memory_pressure)
1472 return 1;
1473 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1474 if (prot->sysctl_mem[2] > alloc *
1475 sk_mem_pages(sk->sk_wmem_queued +
1476 atomic_read(&sk->sk_rmem_alloc) +
1477 sk->sk_forward_alloc))
1478 return 1;
1479 }
1480
1481 suppress_allocation:
1482
1483 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1484 sk_stream_moderate_sndbuf(sk);
1485
1486 /* Fail only if socket is _under_ its sndbuf.
1487 * In this case we cannot block, so that we have to fail.
1488 */
1489 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1490 return 1;
1491 }
1492
1493 /* Alas. Undo changes. */
1494 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1495 atomic_sub(amt, prot->memory_allocated);
1496 return 0;
1497 }
1498
1499 EXPORT_SYMBOL(__sk_mem_schedule);
1500
1501 /**
1502 * __sk_reclaim - reclaim memory_allocated
1503 * @sk: socket
1504 */
1505 void __sk_mem_reclaim(struct sock *sk)
1506 {
1507 struct proto *prot = sk->sk_prot;
1508
1509 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1510 prot->memory_allocated);
1511 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1512
1513 if (prot->memory_pressure && *prot->memory_pressure &&
1514 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1515 *prot->memory_pressure = 0;
1516 }
1517
1518 EXPORT_SYMBOL(__sk_mem_reclaim);
1519
1520
1521 /*
1522 * Set of default routines for initialising struct proto_ops when
1523 * the protocol does not support a particular function. In certain
1524 * cases where it makes no sense for a protocol to have a "do nothing"
1525 * function, some default processing is provided.
1526 */
1527
1528 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1529 {
1530 return -EOPNOTSUPP;
1531 }
1532
1533 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1534 int len, int flags)
1535 {
1536 return -EOPNOTSUPP;
1537 }
1538
1539 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1540 {
1541 return -EOPNOTSUPP;
1542 }
1543
1544 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1545 {
1546 return -EOPNOTSUPP;
1547 }
1548
1549 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1550 int *len, int peer)
1551 {
1552 return -EOPNOTSUPP;
1553 }
1554
1555 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1556 {
1557 return 0;
1558 }
1559
1560 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1561 {
1562 return -EOPNOTSUPP;
1563 }
1564
1565 int sock_no_listen(struct socket *sock, int backlog)
1566 {
1567 return -EOPNOTSUPP;
1568 }
1569
1570 int sock_no_shutdown(struct socket *sock, int how)
1571 {
1572 return -EOPNOTSUPP;
1573 }
1574
1575 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1576 char __user *optval, int optlen)
1577 {
1578 return -EOPNOTSUPP;
1579 }
1580
1581 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1582 char __user *optval, int __user *optlen)
1583 {
1584 return -EOPNOTSUPP;
1585 }
1586
1587 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1588 size_t len)
1589 {
1590 return -EOPNOTSUPP;
1591 }
1592
1593 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1594 size_t len, int flags)
1595 {
1596 return -EOPNOTSUPP;
1597 }
1598
1599 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1600 {
1601 /* Mirror missing mmap method error code */
1602 return -ENODEV;
1603 }
1604
1605 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1606 {
1607 ssize_t res;
1608 struct msghdr msg = {.msg_flags = flags};
1609 struct kvec iov;
1610 char *kaddr = kmap(page);
1611 iov.iov_base = kaddr + offset;
1612 iov.iov_len = size;
1613 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1614 kunmap(page);
1615 return res;
1616 }
1617
1618 /*
1619 * Default Socket Callbacks
1620 */
1621
1622 static void sock_def_wakeup(struct sock *sk)
1623 {
1624 read_lock(&sk->sk_callback_lock);
1625 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1626 wake_up_interruptible_all(sk->sk_sleep);
1627 read_unlock(&sk->sk_callback_lock);
1628 }
1629
1630 static void sock_def_error_report(struct sock *sk)
1631 {
1632 read_lock(&sk->sk_callback_lock);
1633 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1634 wake_up_interruptible(sk->sk_sleep);
1635 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1636 read_unlock(&sk->sk_callback_lock);
1637 }
1638
1639 static void sock_def_readable(struct sock *sk, int len)
1640 {
1641 read_lock(&sk->sk_callback_lock);
1642 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1643 wake_up_interruptible_sync(sk->sk_sleep);
1644 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1645 read_unlock(&sk->sk_callback_lock);
1646 }
1647
1648 static void sock_def_write_space(struct sock *sk)
1649 {
1650 read_lock(&sk->sk_callback_lock);
1651
1652 /* Do not wake up a writer until he can make "significant"
1653 * progress. --DaveM
1654 */
1655 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1656 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1657 wake_up_interruptible_sync(sk->sk_sleep);
1658
1659 /* Should agree with poll, otherwise some programs break */
1660 if (sock_writeable(sk))
1661 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1662 }
1663
1664 read_unlock(&sk->sk_callback_lock);
1665 }
1666
1667 static void sock_def_destruct(struct sock *sk)
1668 {
1669 kfree(sk->sk_protinfo);
1670 }
1671
1672 void sk_send_sigurg(struct sock *sk)
1673 {
1674 if (sk->sk_socket && sk->sk_socket->file)
1675 if (send_sigurg(&sk->sk_socket->file->f_owner))
1676 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1677 }
1678
1679 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1680 unsigned long expires)
1681 {
1682 if (!mod_timer(timer, expires))
1683 sock_hold(sk);
1684 }
1685
1686 EXPORT_SYMBOL(sk_reset_timer);
1687
1688 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1689 {
1690 if (timer_pending(timer) && del_timer(timer))
1691 __sock_put(sk);
1692 }
1693
1694 EXPORT_SYMBOL(sk_stop_timer);
1695
1696 void sock_init_data(struct socket *sock, struct sock *sk)
1697 {
1698 skb_queue_head_init(&sk->sk_receive_queue);
1699 skb_queue_head_init(&sk->sk_write_queue);
1700 skb_queue_head_init(&sk->sk_error_queue);
1701 #ifdef CONFIG_NET_DMA
1702 skb_queue_head_init(&sk->sk_async_wait_queue);
1703 #endif
1704
1705 sk->sk_send_head = NULL;
1706
1707 init_timer(&sk->sk_timer);
1708
1709 sk->sk_allocation = GFP_KERNEL;
1710 sk->sk_rcvbuf = sysctl_rmem_default;
1711 sk->sk_sndbuf = sysctl_wmem_default;
1712 sk->sk_state = TCP_CLOSE;
1713 sk_set_socket(sk, sock);
1714
1715 sock_set_flag(sk, SOCK_ZAPPED);
1716
1717 if (sock) {
1718 sk->sk_type = sock->type;
1719 sk->sk_sleep = &sock->wait;
1720 sock->sk = sk;
1721 } else
1722 sk->sk_sleep = NULL;
1723
1724 rwlock_init(&sk->sk_dst_lock);
1725 rwlock_init(&sk->sk_callback_lock);
1726 lockdep_set_class_and_name(&sk->sk_callback_lock,
1727 af_callback_keys + sk->sk_family,
1728 af_family_clock_key_strings[sk->sk_family]);
1729
1730 sk->sk_state_change = sock_def_wakeup;
1731 sk->sk_data_ready = sock_def_readable;
1732 sk->sk_write_space = sock_def_write_space;
1733 sk->sk_error_report = sock_def_error_report;
1734 sk->sk_destruct = sock_def_destruct;
1735
1736 sk->sk_sndmsg_page = NULL;
1737 sk->sk_sndmsg_off = 0;
1738
1739 sk->sk_peercred.pid = 0;
1740 sk->sk_peercred.uid = -1;
1741 sk->sk_peercred.gid = -1;
1742 sk->sk_write_pending = 0;
1743 sk->sk_rcvlowat = 1;
1744 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1745 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1746
1747 sk->sk_stamp = ktime_set(-1L, 0);
1748
1749 atomic_set(&sk->sk_refcnt, 1);
1750 atomic_set(&sk->sk_drops, 0);
1751 }
1752
1753 void lock_sock_nested(struct sock *sk, int subclass)
1754 {
1755 might_sleep();
1756 spin_lock_bh(&sk->sk_lock.slock);
1757 if (sk->sk_lock.owned)
1758 __lock_sock(sk);
1759 sk->sk_lock.owned = 1;
1760 spin_unlock(&sk->sk_lock.slock);
1761 /*
1762 * The sk_lock has mutex_lock() semantics here:
1763 */
1764 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1765 local_bh_enable();
1766 }
1767
1768 EXPORT_SYMBOL(lock_sock_nested);
1769
1770 void release_sock(struct sock *sk)
1771 {
1772 /*
1773 * The sk_lock has mutex_unlock() semantics:
1774 */
1775 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1776
1777 spin_lock_bh(&sk->sk_lock.slock);
1778 if (sk->sk_backlog.tail)
1779 __release_sock(sk);
1780 sk->sk_lock.owned = 0;
1781 if (waitqueue_active(&sk->sk_lock.wq))
1782 wake_up(&sk->sk_lock.wq);
1783 spin_unlock_bh(&sk->sk_lock.slock);
1784 }
1785 EXPORT_SYMBOL(release_sock);
1786
1787 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1788 {
1789 struct timeval tv;
1790 if (!sock_flag(sk, SOCK_TIMESTAMP))
1791 sock_enable_timestamp(sk);
1792 tv = ktime_to_timeval(sk->sk_stamp);
1793 if (tv.tv_sec == -1)
1794 return -ENOENT;
1795 if (tv.tv_sec == 0) {
1796 sk->sk_stamp = ktime_get_real();
1797 tv = ktime_to_timeval(sk->sk_stamp);
1798 }
1799 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1800 }
1801 EXPORT_SYMBOL(sock_get_timestamp);
1802
1803 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1804 {
1805 struct timespec ts;
1806 if (!sock_flag(sk, SOCK_TIMESTAMP))
1807 sock_enable_timestamp(sk);
1808 ts = ktime_to_timespec(sk->sk_stamp);
1809 if (ts.tv_sec == -1)
1810 return -ENOENT;
1811 if (ts.tv_sec == 0) {
1812 sk->sk_stamp = ktime_get_real();
1813 ts = ktime_to_timespec(sk->sk_stamp);
1814 }
1815 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1816 }
1817 EXPORT_SYMBOL(sock_get_timestampns);
1818
1819 void sock_enable_timestamp(struct sock *sk)
1820 {
1821 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1822 sock_set_flag(sk, SOCK_TIMESTAMP);
1823 net_enable_timestamp();
1824 }
1825 }
1826
1827 /*
1828 * Get a socket option on an socket.
1829 *
1830 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1831 * asynchronous errors should be reported by getsockopt. We assume
1832 * this means if you specify SO_ERROR (otherwise whats the point of it).
1833 */
1834 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1835 char __user *optval, int __user *optlen)
1836 {
1837 struct sock *sk = sock->sk;
1838
1839 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1840 }
1841
1842 EXPORT_SYMBOL(sock_common_getsockopt);
1843
1844 #ifdef CONFIG_COMPAT
1845 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1846 char __user *optval, int __user *optlen)
1847 {
1848 struct sock *sk = sock->sk;
1849
1850 if (sk->sk_prot->compat_getsockopt != NULL)
1851 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1852 optval, optlen);
1853 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1854 }
1855 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1856 #endif
1857
1858 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1859 struct msghdr *msg, size_t size, int flags)
1860 {
1861 struct sock *sk = sock->sk;
1862 int addr_len = 0;
1863 int err;
1864
1865 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1866 flags & ~MSG_DONTWAIT, &addr_len);
1867 if (err >= 0)
1868 msg->msg_namelen = addr_len;
1869 return err;
1870 }
1871
1872 EXPORT_SYMBOL(sock_common_recvmsg);
1873
1874 /*
1875 * Set socket options on an inet socket.
1876 */
1877 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1878 char __user *optval, int optlen)
1879 {
1880 struct sock *sk = sock->sk;
1881
1882 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1883 }
1884
1885 EXPORT_SYMBOL(sock_common_setsockopt);
1886
1887 #ifdef CONFIG_COMPAT
1888 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1889 char __user *optval, int optlen)
1890 {
1891 struct sock *sk = sock->sk;
1892
1893 if (sk->sk_prot->compat_setsockopt != NULL)
1894 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1895 optval, optlen);
1896 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1897 }
1898 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1899 #endif
1900
1901 void sk_common_release(struct sock *sk)
1902 {
1903 if (sk->sk_prot->destroy)
1904 sk->sk_prot->destroy(sk);
1905
1906 /*
1907 * Observation: when sock_common_release is called, processes have
1908 * no access to socket. But net still has.
1909 * Step one, detach it from networking:
1910 *
1911 * A. Remove from hash tables.
1912 */
1913
1914 sk->sk_prot->unhash(sk);
1915
1916 /*
1917 * In this point socket cannot receive new packets, but it is possible
1918 * that some packets are in flight because some CPU runs receiver and
1919 * did hash table lookup before we unhashed socket. They will achieve
1920 * receive queue and will be purged by socket destructor.
1921 *
1922 * Also we still have packets pending on receive queue and probably,
1923 * our own packets waiting in device queues. sock_destroy will drain
1924 * receive queue, but transmitted packets will delay socket destruction
1925 * until the last reference will be released.
1926 */
1927
1928 sock_orphan(sk);
1929
1930 xfrm_sk_free_policy(sk);
1931
1932 sk_refcnt_debug_release(sk);
1933 sock_put(sk);
1934 }
1935
1936 EXPORT_SYMBOL(sk_common_release);
1937
1938 static DEFINE_RWLOCK(proto_list_lock);
1939 static LIST_HEAD(proto_list);
1940
1941 #ifdef CONFIG_PROC_FS
1942 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1943 struct prot_inuse {
1944 int val[PROTO_INUSE_NR];
1945 };
1946
1947 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
1948
1949 #ifdef CONFIG_NET_NS
1950 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1951 {
1952 int cpu = get_cpu();
1953 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1954 put_cpu();
1955 }
1956 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1957
1958 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1959 {
1960 int cpu, idx = prot->inuse_idx;
1961 int res = 0;
1962
1963 for_each_possible_cpu(cpu)
1964 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1965
1966 return res >= 0 ? res : 0;
1967 }
1968 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1969
1970 static int sock_inuse_init_net(struct net *net)
1971 {
1972 net->core.inuse = alloc_percpu(struct prot_inuse);
1973 return net->core.inuse ? 0 : -ENOMEM;
1974 }
1975
1976 static void sock_inuse_exit_net(struct net *net)
1977 {
1978 free_percpu(net->core.inuse);
1979 }
1980
1981 static struct pernet_operations net_inuse_ops = {
1982 .init = sock_inuse_init_net,
1983 .exit = sock_inuse_exit_net,
1984 };
1985
1986 static __init int net_inuse_init(void)
1987 {
1988 if (register_pernet_subsys(&net_inuse_ops))
1989 panic("Cannot initialize net inuse counters");
1990
1991 return 0;
1992 }
1993
1994 core_initcall(net_inuse_init);
1995 #else
1996 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1997
1998 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1999 {
2000 int cpu = get_cpu();
2001 per_cpu(prot_inuse, cpu).val[prot->inuse_idx] += val;
2002 put_cpu();
2003 }
2004 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2005
2006 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2007 {
2008 int cpu, idx = prot->inuse_idx;
2009 int res = 0;
2010
2011 for_each_possible_cpu(cpu)
2012 res += per_cpu(prot_inuse, cpu).val[idx];
2013
2014 return res >= 0 ? res : 0;
2015 }
2016 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2017 #endif
2018
2019 static void assign_proto_idx(struct proto *prot)
2020 {
2021 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2022
2023 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2024 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2025 return;
2026 }
2027
2028 set_bit(prot->inuse_idx, proto_inuse_idx);
2029 }
2030
2031 static void release_proto_idx(struct proto *prot)
2032 {
2033 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2034 clear_bit(prot->inuse_idx, proto_inuse_idx);
2035 }
2036 #else
2037 static inline void assign_proto_idx(struct proto *prot)
2038 {
2039 }
2040
2041 static inline void release_proto_idx(struct proto *prot)
2042 {
2043 }
2044 #endif
2045
2046 int proto_register(struct proto *prot, int alloc_slab)
2047 {
2048 if (alloc_slab) {
2049 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2050 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2051 NULL);
2052
2053 if (prot->slab == NULL) {
2054 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2055 prot->name);
2056 goto out;
2057 }
2058
2059 if (prot->rsk_prot != NULL) {
2060 static const char mask[] = "request_sock_%s";
2061
2062 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2063 if (prot->rsk_prot->slab_name == NULL)
2064 goto out_free_sock_slab;
2065
2066 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2067 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2068 prot->rsk_prot->obj_size, 0,
2069 SLAB_HWCACHE_ALIGN, NULL);
2070
2071 if (prot->rsk_prot->slab == NULL) {
2072 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2073 prot->name);
2074 goto out_free_request_sock_slab_name;
2075 }
2076 }
2077
2078 if (prot->twsk_prot != NULL) {
2079 static const char mask[] = "tw_sock_%s";
2080
2081 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2082
2083 if (prot->twsk_prot->twsk_slab_name == NULL)
2084 goto out_free_request_sock_slab;
2085
2086 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2087 prot->twsk_prot->twsk_slab =
2088 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2089 prot->twsk_prot->twsk_obj_size,
2090 0,
2091 SLAB_HWCACHE_ALIGN |
2092 prot->slab_flags,
2093 NULL);
2094 if (prot->twsk_prot->twsk_slab == NULL)
2095 goto out_free_timewait_sock_slab_name;
2096 }
2097 }
2098
2099 write_lock(&proto_list_lock);
2100 list_add(&prot->node, &proto_list);
2101 assign_proto_idx(prot);
2102 write_unlock(&proto_list_lock);
2103 return 0;
2104
2105 out_free_timewait_sock_slab_name:
2106 kfree(prot->twsk_prot->twsk_slab_name);
2107 out_free_request_sock_slab:
2108 if (prot->rsk_prot && prot->rsk_prot->slab) {
2109 kmem_cache_destroy(prot->rsk_prot->slab);
2110 prot->rsk_prot->slab = NULL;
2111 }
2112 out_free_request_sock_slab_name:
2113 kfree(prot->rsk_prot->slab_name);
2114 out_free_sock_slab:
2115 kmem_cache_destroy(prot->slab);
2116 prot->slab = NULL;
2117 out:
2118 return -ENOBUFS;
2119 }
2120
2121 EXPORT_SYMBOL(proto_register);
2122
2123 void proto_unregister(struct proto *prot)
2124 {
2125 write_lock(&proto_list_lock);
2126 release_proto_idx(prot);
2127 list_del(&prot->node);
2128 write_unlock(&proto_list_lock);
2129
2130 if (prot->slab != NULL) {
2131 kmem_cache_destroy(prot->slab);
2132 prot->slab = NULL;
2133 }
2134
2135 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2136 kmem_cache_destroy(prot->rsk_prot->slab);
2137 kfree(prot->rsk_prot->slab_name);
2138 prot->rsk_prot->slab = NULL;
2139 }
2140
2141 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2142 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2143 kfree(prot->twsk_prot->twsk_slab_name);
2144 prot->twsk_prot->twsk_slab = NULL;
2145 }
2146 }
2147
2148 EXPORT_SYMBOL(proto_unregister);
2149
2150 #ifdef CONFIG_PROC_FS
2151 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2152 __acquires(proto_list_lock)
2153 {
2154 read_lock(&proto_list_lock);
2155 return seq_list_start_head(&proto_list, *pos);
2156 }
2157
2158 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2159 {
2160 return seq_list_next(v, &proto_list, pos);
2161 }
2162
2163 static void proto_seq_stop(struct seq_file *seq, void *v)
2164 __releases(proto_list_lock)
2165 {
2166 read_unlock(&proto_list_lock);
2167 }
2168
2169 static char proto_method_implemented(const void *method)
2170 {
2171 return method == NULL ? 'n' : 'y';
2172 }
2173
2174 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2175 {
2176 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2177 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2178 proto->name,
2179 proto->obj_size,
2180 sock_prot_inuse_get(seq_file_net(seq), proto),
2181 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2182 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2183 proto->max_header,
2184 proto->slab == NULL ? "no" : "yes",
2185 module_name(proto->owner),
2186 proto_method_implemented(proto->close),
2187 proto_method_implemented(proto->connect),
2188 proto_method_implemented(proto->disconnect),
2189 proto_method_implemented(proto->accept),
2190 proto_method_implemented(proto->ioctl),
2191 proto_method_implemented(proto->init),
2192 proto_method_implemented(proto->destroy),
2193 proto_method_implemented(proto->shutdown),
2194 proto_method_implemented(proto->setsockopt),
2195 proto_method_implemented(proto->getsockopt),
2196 proto_method_implemented(proto->sendmsg),
2197 proto_method_implemented(proto->recvmsg),
2198 proto_method_implemented(proto->sendpage),
2199 proto_method_implemented(proto->bind),
2200 proto_method_implemented(proto->backlog_rcv),
2201 proto_method_implemented(proto->hash),
2202 proto_method_implemented(proto->unhash),
2203 proto_method_implemented(proto->get_port),
2204 proto_method_implemented(proto->enter_memory_pressure));
2205 }
2206
2207 static int proto_seq_show(struct seq_file *seq, void *v)
2208 {
2209 if (v == &proto_list)
2210 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2211 "protocol",
2212 "size",
2213 "sockets",
2214 "memory",
2215 "press",
2216 "maxhdr",
2217 "slab",
2218 "module",
2219 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2220 else
2221 proto_seq_printf(seq, list_entry(v, struct proto, node));
2222 return 0;
2223 }
2224
2225 static const struct seq_operations proto_seq_ops = {
2226 .start = proto_seq_start,
2227 .next = proto_seq_next,
2228 .stop = proto_seq_stop,
2229 .show = proto_seq_show,
2230 };
2231
2232 static int proto_seq_open(struct inode *inode, struct file *file)
2233 {
2234 return seq_open_net(inode, file, &proto_seq_ops,
2235 sizeof(struct seq_net_private));
2236 }
2237
2238 static const struct file_operations proto_seq_fops = {
2239 .owner = THIS_MODULE,
2240 .open = proto_seq_open,
2241 .read = seq_read,
2242 .llseek = seq_lseek,
2243 .release = seq_release_net,
2244 };
2245
2246 static __net_init int proto_init_net(struct net *net)
2247 {
2248 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2249 return -ENOMEM;
2250
2251 return 0;
2252 }
2253
2254 static __net_exit void proto_exit_net(struct net *net)
2255 {
2256 proc_net_remove(net, "protocols");
2257 }
2258
2259
2260 static __net_initdata struct pernet_operations proto_net_ops = {
2261 .init = proto_init_net,
2262 .exit = proto_exit_net,
2263 };
2264
2265 static int __init proto_init(void)
2266 {
2267 return register_pernet_subsys(&proto_net_ops);
2268 }
2269
2270 subsys_initcall(proto_init);
2271
2272 #endif /* PROC_FS */
2273
2274 EXPORT_SYMBOL(sk_alloc);
2275 EXPORT_SYMBOL(sk_free);
2276 EXPORT_SYMBOL(sk_send_sigurg);
2277 EXPORT_SYMBOL(sock_alloc_send_skb);
2278 EXPORT_SYMBOL(sock_init_data);
2279 EXPORT_SYMBOL(sock_kfree_s);
2280 EXPORT_SYMBOL(sock_kmalloc);
2281 EXPORT_SYMBOL(sock_no_accept);
2282 EXPORT_SYMBOL(sock_no_bind);
2283 EXPORT_SYMBOL(sock_no_connect);
2284 EXPORT_SYMBOL(sock_no_getname);
2285 EXPORT_SYMBOL(sock_no_getsockopt);
2286 EXPORT_SYMBOL(sock_no_ioctl);
2287 EXPORT_SYMBOL(sock_no_listen);
2288 EXPORT_SYMBOL(sock_no_mmap);
2289 EXPORT_SYMBOL(sock_no_poll);
2290 EXPORT_SYMBOL(sock_no_recvmsg);
2291 EXPORT_SYMBOL(sock_no_sendmsg);
2292 EXPORT_SYMBOL(sock_no_sendpage);
2293 EXPORT_SYMBOL(sock_no_setsockopt);
2294 EXPORT_SYMBOL(sock_no_shutdown);
2295 EXPORT_SYMBOL(sock_no_socketpair);
2296 EXPORT_SYMBOL(sock_rfree);
2297 EXPORT_SYMBOL(sock_setsockopt);
2298 EXPORT_SYMBOL(sock_wfree);
2299 EXPORT_SYMBOL(sock_wmalloc);
2300 EXPORT_SYMBOL(sock_i_uid);
2301 EXPORT_SYMBOL(sock_i_ino);
2302 EXPORT_SYMBOL(sysctl_optmem_max);
Nao low-level kernelspace/userspace library that runs on our robots, Nao-Team HTWK