?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
powerpc/8xx: Fix regression introduced by cache coherency rewrite
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / core / sock.c
blob3f5e77e728e4c40b6bd4dd3ec0d01ff3047a3eb8
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 <linux/net_tstamp.h>
124 #include <net/xfrm.h>
125 #include <linux/ipsec.h>
126
127 #include <linux/filter.h>
128
129 #ifdef CONFIG_INET
130 #include <net/tcp.h>
131 #endif
132
133 /*
134 * Each address family might have different locking rules, so we have
135 * one slock key per address family:
136 */
137 static struct lock_class_key af_family_keys[AF_MAX];
138 static struct lock_class_key af_family_slock_keys[AF_MAX];
139
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-AF_RDS" , "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-AF_RDS" , "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-AF_RDS" , "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
191 /*
192 * sk_callback_lock locking rules are per-address-family,
193 * so split the lock classes by using a per-AF key:
194 */
195 static struct lock_class_key af_callback_keys[AF_MAX];
196
197 /* Take into consideration the size of the struct sk_buff overhead in the
198 * determination of these values, since that is non-constant across
199 * platforms. This makes socket queueing behavior and performance
200 * not depend upon such differences.
201 */
202 #define _SK_MEM_PACKETS 256
203 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
204 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
206
207 /* Run time adjustable parameters. */
208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
212
213 /* Maximal space eaten by iovec or ancilliary data plus some space */
214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
215
216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
217 {
218 struct timeval tv;
219
220 if (optlen < sizeof(tv))
221 return -EINVAL;
222 if (copy_from_user(&tv, optval, sizeof(tv)))
223 return -EFAULT;
224 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
225 return -EDOM;
226
227 if (tv.tv_sec < 0) {
228 static int warned __read_mostly;
229
230 *timeo_p = 0;
231 if (warned < 10 && net_ratelimit()) {
232 warned++;
233 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
234 "tries to set negative timeout\n",
235 current->comm, task_pid_nr(current));
236 }
237 return 0;
238 }
239 *timeo_p = MAX_SCHEDULE_TIMEOUT;
240 if (tv.tv_sec == 0 && tv.tv_usec == 0)
241 return 0;
242 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
243 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
244 return 0;
245 }
246
247 static void sock_warn_obsolete_bsdism(const char *name)
248 {
249 static int warned;
250 static char warncomm[TASK_COMM_LEN];
251 if (strcmp(warncomm, current->comm) && warned < 5) {
252 strcpy(warncomm, current->comm);
253 printk(KERN_WARNING "process `%s' is using obsolete "
254 "%s SO_BSDCOMPAT\n", warncomm, name);
255 warned++;
256 }
257 }
258
259 static void sock_disable_timestamp(struct sock *sk, int flag)
260 {
261 if (sock_flag(sk, flag)) {
262 sock_reset_flag(sk, flag);
263 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
264 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
265 net_disable_timestamp();
266 }
267 }
268 }
269
270
271 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
272 {
273 int err = 0;
274 int skb_len;
275
276 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
277 number of warnings when compiling with -W --ANK
278 */
279 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
280 (unsigned)sk->sk_rcvbuf) {
281 err = -ENOMEM;
282 goto out;
283 }
284
285 err = sk_filter(sk, skb);
286 if (err)
287 goto out;
288
289 if (!sk_rmem_schedule(sk, skb->truesize)) {
290 err = -ENOBUFS;
291 goto out;
292 }
293
294 skb->dev = NULL;
295 skb_set_owner_r(skb, sk);
296
297 /* Cache the SKB length before we tack it onto the receive
298 * queue. Once it is added it no longer belongs to us and
299 * may be freed by other threads of control pulling packets
300 * from the queue.
301 */
302 skb_len = skb->len;
303
304 skb_queue_tail(&sk->sk_receive_queue, skb);
305
306 if (!sock_flag(sk, SOCK_DEAD))
307 sk->sk_data_ready(sk, skb_len);
308 out:
309 return err;
310 }
311 EXPORT_SYMBOL(sock_queue_rcv_skb);
312
313 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
314 {
315 int rc = NET_RX_SUCCESS;
316
317 if (sk_filter(sk, skb))
318 goto discard_and_relse;
319
320 skb->dev = NULL;
321
322 if (nested)
323 bh_lock_sock_nested(sk);
324 else
325 bh_lock_sock(sk);
326 if (!sock_owned_by_user(sk)) {
327 /*
328 * trylock + unlock semantics:
329 */
330 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
331
332 rc = sk_backlog_rcv(sk, skb);
333
334 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
335 } else
336 sk_add_backlog(sk, skb);
337 bh_unlock_sock(sk);
338 out:
339 sock_put(sk);
340 return rc;
341 discard_and_relse:
342 kfree_skb(skb);
343 goto out;
344 }
345 EXPORT_SYMBOL(sk_receive_skb);
346
347 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
348 {
349 struct dst_entry *dst = sk->sk_dst_cache;
350
351 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
352 sk->sk_dst_cache = NULL;
353 dst_release(dst);
354 return NULL;
355 }
356
357 return dst;
358 }
359 EXPORT_SYMBOL(__sk_dst_check);
360
361 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
362 {
363 struct dst_entry *dst = sk_dst_get(sk);
364
365 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
366 sk_dst_reset(sk);
367 dst_release(dst);
368 return NULL;
369 }
370
371 return dst;
372 }
373 EXPORT_SYMBOL(sk_dst_check);
374
375 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
376 {
377 int ret = -ENOPROTOOPT;
378 #ifdef CONFIG_NETDEVICES
379 struct net *net = sock_net(sk);
380 char devname[IFNAMSIZ];
381 int index;
382
383 /* Sorry... */
384 ret = -EPERM;
385 if (!capable(CAP_NET_RAW))
386 goto out;
387
388 ret = -EINVAL;
389 if (optlen < 0)
390 goto out;
391
392 /* Bind this socket to a particular device like "eth0",
393 * as specified in the passed interface name. If the
394 * name is "" or the option length is zero the socket
395 * is not bound.
396 */
397 if (optlen > IFNAMSIZ - 1)
398 optlen = IFNAMSIZ - 1;
399 memset(devname, 0, sizeof(devname));
400
401 ret = -EFAULT;
402 if (copy_from_user(devname, optval, optlen))
403 goto out;
404
405 if (devname[0] == '\0') {
406 index = 0;
407 } else {
408 struct net_device *dev = dev_get_by_name(net, devname);
409
410 ret = -ENODEV;
411 if (!dev)
412 goto out;
413
414 index = dev->ifindex;
415 dev_put(dev);
416 }
417
418 lock_sock(sk);
419 sk->sk_bound_dev_if = index;
420 sk_dst_reset(sk);
421 release_sock(sk);
422
423 ret = 0;
424
425 out:
426 #endif
427
428 return ret;
429 }
430
431 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
432 {
433 if (valbool)
434 sock_set_flag(sk, bit);
435 else
436 sock_reset_flag(sk, bit);
437 }
438
439 /*
440 * This is meant for all protocols to use and covers goings on
441 * at the socket level. Everything here is generic.
442 */
443
444 int sock_setsockopt(struct socket *sock, int level, int optname,
445 char __user *optval, int optlen)
446 {
447 struct sock *sk=sock->sk;
448 int val;
449 int valbool;
450 struct linger ling;
451 int ret = 0;
452
453 /*
454 * Options without arguments
455 */
456
457 if (optname == SO_BINDTODEVICE)
458 return sock_bindtodevice(sk, optval, optlen);
459
460 if (optlen < sizeof(int))
461 return -EINVAL;
462
463 if (get_user(val, (int __user *)optval))
464 return -EFAULT;
465
466 valbool = val?1:0;
467
468 lock_sock(sk);
469
470 switch(optname) {
471 case SO_DEBUG:
472 if (val && !capable(CAP_NET_ADMIN)) {
473 ret = -EACCES;
474 } else
475 sock_valbool_flag(sk, SOCK_DBG, valbool);
476 break;
477 case SO_REUSEADDR:
478 sk->sk_reuse = valbool;
479 break;
480 case SO_TYPE:
481 case SO_ERROR:
482 ret = -ENOPROTOOPT;
483 break;
484 case SO_DONTROUTE:
485 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
486 break;
487 case SO_BROADCAST:
488 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
489 break;
490 case SO_SNDBUF:
491 /* Don't error on this BSD doesn't and if you think
492 about it this is right. Otherwise apps have to
493 play 'guess the biggest size' games. RCVBUF/SNDBUF
494 are treated in BSD as hints */
495
496 if (val > sysctl_wmem_max)
497 val = sysctl_wmem_max;
498 set_sndbuf:
499 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
500 if ((val * 2) < SOCK_MIN_SNDBUF)
501 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
502 else
503 sk->sk_sndbuf = val * 2;
504
505 /*
506 * Wake up sending tasks if we
507 * upped the value.
508 */
509 sk->sk_write_space(sk);
510 break;
511
512 case SO_SNDBUFFORCE:
513 if (!capable(CAP_NET_ADMIN)) {
514 ret = -EPERM;
515 break;
516 }
517 goto set_sndbuf;
518
519 case SO_RCVBUF:
520 /* Don't error on this BSD doesn't and if you think
521 about it this is right. Otherwise apps have to
522 play 'guess the biggest size' games. RCVBUF/SNDBUF
523 are treated in BSD as hints */
524
525 if (val > sysctl_rmem_max)
526 val = sysctl_rmem_max;
527 set_rcvbuf:
528 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
529 /*
530 * We double it on the way in to account for
531 * "struct sk_buff" etc. overhead. Applications
532 * assume that the SO_RCVBUF setting they make will
533 * allow that much actual data to be received on that
534 * socket.
535 *
536 * Applications are unaware that "struct sk_buff" and
537 * other overheads allocate from the receive buffer
538 * during socket buffer allocation.
539 *
540 * And after considering the possible alternatives,
541 * returning the value we actually used in getsockopt
542 * is the most desirable behavior.
543 */
544 if ((val * 2) < SOCK_MIN_RCVBUF)
545 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
546 else
547 sk->sk_rcvbuf = val * 2;
548 break;
549
550 case SO_RCVBUFFORCE:
551 if (!capable(CAP_NET_ADMIN)) {
552 ret = -EPERM;
553 break;
554 }
555 goto set_rcvbuf;
556
557 case SO_KEEPALIVE:
558 #ifdef CONFIG_INET
559 if (sk->sk_protocol == IPPROTO_TCP)
560 tcp_set_keepalive(sk, valbool);
561 #endif
562 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
563 break;
564
565 case SO_OOBINLINE:
566 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
567 break;
568
569 case SO_NO_CHECK:
570 sk->sk_no_check = valbool;
571 break;
572
573 case SO_PRIORITY:
574 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
575 sk->sk_priority = val;
576 else
577 ret = -EPERM;
578 break;
579
580 case SO_LINGER:
581 if (optlen < sizeof(ling)) {
582 ret = -EINVAL; /* 1003.1g */
583 break;
584 }
585 if (copy_from_user(&ling,optval,sizeof(ling))) {
586 ret = -EFAULT;
587 break;
588 }
589 if (!ling.l_onoff)
590 sock_reset_flag(sk, SOCK_LINGER);
591 else {
592 #if (BITS_PER_LONG == 32)
593 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
594 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
595 else
596 #endif
597 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
598 sock_set_flag(sk, SOCK_LINGER);
599 }
600 break;
601
602 case SO_BSDCOMPAT:
603 sock_warn_obsolete_bsdism("setsockopt");
604 break;
605
606 case SO_PASSCRED:
607 if (valbool)
608 set_bit(SOCK_PASSCRED, &sock->flags);
609 else
610 clear_bit(SOCK_PASSCRED, &sock->flags);
611 break;
612
613 case SO_TIMESTAMP:
614 case SO_TIMESTAMPNS:
615 if (valbool) {
616 if (optname == SO_TIMESTAMP)
617 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
618 else
619 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
620 sock_set_flag(sk, SOCK_RCVTSTAMP);
621 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
622 } else {
623 sock_reset_flag(sk, SOCK_RCVTSTAMP);
624 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
625 }
626 break;
627
628 case SO_TIMESTAMPING:
629 if (val & ~SOF_TIMESTAMPING_MASK) {
630 ret = EINVAL;
631 break;
632 }
633 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
634 val & SOF_TIMESTAMPING_TX_HARDWARE);
635 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
636 val & SOF_TIMESTAMPING_TX_SOFTWARE);
637 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
638 val & SOF_TIMESTAMPING_RX_HARDWARE);
639 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
640 sock_enable_timestamp(sk,
641 SOCK_TIMESTAMPING_RX_SOFTWARE);
642 else
643 sock_disable_timestamp(sk,
644 SOCK_TIMESTAMPING_RX_SOFTWARE);
645 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
646 val & SOF_TIMESTAMPING_SOFTWARE);
647 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
648 val & SOF_TIMESTAMPING_SYS_HARDWARE);
649 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
650 val & SOF_TIMESTAMPING_RAW_HARDWARE);
651 break;
652
653 case SO_RCVLOWAT:
654 if (val < 0)
655 val = INT_MAX;
656 sk->sk_rcvlowat = val ? : 1;
657 break;
658
659 case SO_RCVTIMEO:
660 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
661 break;
662
663 case SO_SNDTIMEO:
664 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
665 break;
666
667 case SO_ATTACH_FILTER:
668 ret = -EINVAL;
669 if (optlen == sizeof(struct sock_fprog)) {
670 struct sock_fprog fprog;
671
672 ret = -EFAULT;
673 if (copy_from_user(&fprog, optval, sizeof(fprog)))
674 break;
675
676 ret = sk_attach_filter(&fprog, sk);
677 }
678 break;
679
680 case SO_DETACH_FILTER:
681 ret = sk_detach_filter(sk);
682 break;
683
684 case SO_PASSSEC:
685 if (valbool)
686 set_bit(SOCK_PASSSEC, &sock->flags);
687 else
688 clear_bit(SOCK_PASSSEC, &sock->flags);
689 break;
690 case SO_MARK:
691 if (!capable(CAP_NET_ADMIN))
692 ret = -EPERM;
693 else {
694 sk->sk_mark = val;
695 }
696 break;
697
698 /* We implement the SO_SNDLOWAT etc to
699 not be settable (1003.1g 5.3) */
700 default:
701 ret = -ENOPROTOOPT;
702 break;
703 }
704 release_sock(sk);
705 return ret;
706 }
707
708
709 int sock_getsockopt(struct socket *sock, int level, int optname,
710 char __user *optval, int __user *optlen)
711 {
712 struct sock *sk = sock->sk;
713
714 union {
715 int val;
716 struct linger ling;
717 struct timeval tm;
718 } v;
719
720 unsigned int lv = sizeof(int);
721 int len;
722
723 if (get_user(len, optlen))
724 return -EFAULT;
725 if (len < 0)
726 return -EINVAL;
727
728 memset(&v, 0, sizeof(v));
729
730 switch(optname) {
731 case SO_DEBUG:
732 v.val = sock_flag(sk, SOCK_DBG);
733 break;
734
735 case SO_DONTROUTE:
736 v.val = sock_flag(sk, SOCK_LOCALROUTE);
737 break;
738
739 case SO_BROADCAST:
740 v.val = !!sock_flag(sk, SOCK_BROADCAST);
741 break;
742
743 case SO_SNDBUF:
744 v.val = sk->sk_sndbuf;
745 break;
746
747 case SO_RCVBUF:
748 v.val = sk->sk_rcvbuf;
749 break;
750
751 case SO_REUSEADDR:
752 v.val = sk->sk_reuse;
753 break;
754
755 case SO_KEEPALIVE:
756 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
757 break;
758
759 case SO_TYPE:
760 v.val = sk->sk_type;
761 break;
762
763 case SO_ERROR:
764 v.val = -sock_error(sk);
765 if (v.val==0)
766 v.val = xchg(&sk->sk_err_soft, 0);
767 break;
768
769 case SO_OOBINLINE:
770 v.val = !!sock_flag(sk, SOCK_URGINLINE);
771 break;
772
773 case SO_NO_CHECK:
774 v.val = sk->sk_no_check;
775 break;
776
777 case SO_PRIORITY:
778 v.val = sk->sk_priority;
779 break;
780
781 case SO_LINGER:
782 lv = sizeof(v.ling);
783 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
784 v.ling.l_linger = sk->sk_lingertime / HZ;
785 break;
786
787 case SO_BSDCOMPAT:
788 sock_warn_obsolete_bsdism("getsockopt");
789 break;
790
791 case SO_TIMESTAMP:
792 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
793 !sock_flag(sk, SOCK_RCVTSTAMPNS);
794 break;
795
796 case SO_TIMESTAMPNS:
797 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
798 break;
799
800 case SO_TIMESTAMPING:
801 v.val = 0;
802 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
803 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
804 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
805 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
806 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
807 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
808 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
809 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
810 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
811 v.val |= SOF_TIMESTAMPING_SOFTWARE;
812 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
813 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
814 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
815 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
816 break;
817
818 case SO_RCVTIMEO:
819 lv=sizeof(struct timeval);
820 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
821 v.tm.tv_sec = 0;
822 v.tm.tv_usec = 0;
823 } else {
824 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
825 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
826 }
827 break;
828
829 case SO_SNDTIMEO:
830 lv=sizeof(struct timeval);
831 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
832 v.tm.tv_sec = 0;
833 v.tm.tv_usec = 0;
834 } else {
835 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
836 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
837 }
838 break;
839
840 case SO_RCVLOWAT:
841 v.val = sk->sk_rcvlowat;
842 break;
843
844 case SO_SNDLOWAT:
845 v.val=1;
846 break;
847
848 case SO_PASSCRED:
849 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
850 break;
851
852 case SO_PEERCRED:
853 if (len > sizeof(sk->sk_peercred))
854 len = sizeof(sk->sk_peercred);
855 if (copy_to_user(optval, &sk->sk_peercred, len))
856 return -EFAULT;
857 goto lenout;
858
859 case SO_PEERNAME:
860 {
861 char address[128];
862
863 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
864 return -ENOTCONN;
865 if (lv < len)
866 return -EINVAL;
867 if (copy_to_user(optval, address, len))
868 return -EFAULT;
869 goto lenout;
870 }
871
872 /* Dubious BSD thing... Probably nobody even uses it, but
873 * the UNIX standard wants it for whatever reason... -DaveM
874 */
875 case SO_ACCEPTCONN:
876 v.val = sk->sk_state == TCP_LISTEN;
877 break;
878
879 case SO_PASSSEC:
880 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
881 break;
882
883 case SO_PEERSEC:
884 return security_socket_getpeersec_stream(sock, optval, optlen, len);
885
886 case SO_MARK:
887 v.val = sk->sk_mark;
888 break;
889
890 default:
891 return -ENOPROTOOPT;
892 }
893
894 if (len > lv)
895 len = lv;
896 if (copy_to_user(optval, &v, len))
897 return -EFAULT;
898 lenout:
899 if (put_user(len, optlen))
900 return -EFAULT;
901 return 0;
902 }
903
904 /*
905 * Initialize an sk_lock.
906 *
907 * (We also register the sk_lock with the lock validator.)
908 */
909 static inline void sock_lock_init(struct sock *sk)
910 {
911 sock_lock_init_class_and_name(sk,
912 af_family_slock_key_strings[sk->sk_family],
913 af_family_slock_keys + sk->sk_family,
914 af_family_key_strings[sk->sk_family],
915 af_family_keys + sk->sk_family);
916 }
917
918 /*
919 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
920 * even temporarly, because of RCU lookups. sk_node should also be left as is.
921 */
922 static void sock_copy(struct sock *nsk, const struct sock *osk)
923 {
924 #ifdef CONFIG_SECURITY_NETWORK
925 void *sptr = nsk->sk_security;
926 #endif
927 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
928 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt));
929 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
930 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
931 #ifdef CONFIG_SECURITY_NETWORK
932 nsk->sk_security = sptr;
933 security_sk_clone(osk, nsk);
934 #endif
935 }
936
937 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
938 int family)
939 {
940 struct sock *sk;
941 struct kmem_cache *slab;
942
943 slab = prot->slab;
944 if (slab != NULL) {
945 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
946 if (!sk)
947 return sk;
948 if (priority & __GFP_ZERO) {
949 /*
950 * caches using SLAB_DESTROY_BY_RCU should let
951 * sk_node.next un-modified. Special care is taken
952 * when initializing object to zero.
953 */
954 if (offsetof(struct sock, sk_node.next) != 0)
955 memset(sk, 0, offsetof(struct sock, sk_node.next));
956 memset(&sk->sk_node.pprev, 0,
957 prot->obj_size - offsetof(struct sock,
958 sk_node.pprev));
959 }
960 }
961 else
962 sk = kmalloc(prot->obj_size, priority);
963
964 if (sk != NULL) {
965 if (security_sk_alloc(sk, family, priority))
966 goto out_free;
967
968 if (!try_module_get(prot->owner))
969 goto out_free_sec;
970 }
971
972 return sk;
973
974 out_free_sec:
975 security_sk_free(sk);
976 out_free:
977 if (slab != NULL)
978 kmem_cache_free(slab, sk);
979 else
980 kfree(sk);
981 return NULL;
982 }
983
984 static void sk_prot_free(struct proto *prot, struct sock *sk)
985 {
986 struct kmem_cache *slab;
987 struct module *owner;
988
989 owner = prot->owner;
990 slab = prot->slab;
991
992 security_sk_free(sk);
993 if (slab != NULL)
994 kmem_cache_free(slab, sk);
995 else
996 kfree(sk);
997 module_put(owner);
998 }
999
1000 /**
1001 * sk_alloc - All socket objects are allocated here
1002 * @net: the applicable net namespace
1003 * @family: protocol family
1004 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1005 * @prot: struct proto associated with this new sock instance
1006 */
1007 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1008 struct proto *prot)
1009 {
1010 struct sock *sk;
1011
1012 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1013 if (sk) {
1014 sk->sk_family = family;
1015 /*
1016 * See comment in struct sock definition to understand
1017 * why we need sk_prot_creator -acme
1018 */
1019 sk->sk_prot = sk->sk_prot_creator = prot;
1020 sock_lock_init(sk);
1021 sock_net_set(sk, get_net(net));
1022 }
1023
1024 return sk;
1025 }
1026
1027 void sk_free(struct sock *sk)
1028 {
1029 struct sk_filter *filter;
1030
1031 if (sk->sk_destruct)
1032 sk->sk_destruct(sk);
1033
1034 filter = rcu_dereference(sk->sk_filter);
1035 if (filter) {
1036 sk_filter_uncharge(sk, filter);
1037 rcu_assign_pointer(sk->sk_filter, NULL);
1038 }
1039
1040 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1041 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1042
1043 if (atomic_read(&sk->sk_omem_alloc))
1044 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1045 __func__, atomic_read(&sk->sk_omem_alloc));
1046
1047 put_net(sock_net(sk));
1048 sk_prot_free(sk->sk_prot_creator, sk);
1049 }
1050
1051 /*
1052 * Last sock_put should drop referrence to sk->sk_net. It has already
1053 * been dropped in sk_change_net. Taking referrence to stopping namespace
1054 * is not an option.
1055 * Take referrence to a socket to remove it from hash _alive_ and after that
1056 * destroy it in the context of init_net.
1057 */
1058 void sk_release_kernel(struct sock *sk)
1059 {
1060 if (sk == NULL || sk->sk_socket == NULL)
1061 return;
1062
1063 sock_hold(sk);
1064 sock_release(sk->sk_socket);
1065 release_net(sock_net(sk));
1066 sock_net_set(sk, get_net(&init_net));
1067 sock_put(sk);
1068 }
1069 EXPORT_SYMBOL(sk_release_kernel);
1070
1071 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1072 {
1073 struct sock *newsk;
1074
1075 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1076 if (newsk != NULL) {
1077 struct sk_filter *filter;
1078
1079 sock_copy(newsk, sk);
1080
1081 /* SANITY */
1082 get_net(sock_net(newsk));
1083 sk_node_init(&newsk->sk_node);
1084 sock_lock_init(newsk);
1085 bh_lock_sock(newsk);
1086 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1087
1088 atomic_set(&newsk->sk_rmem_alloc, 0);
1089 atomic_set(&newsk->sk_wmem_alloc, 0);
1090 atomic_set(&newsk->sk_omem_alloc, 0);
1091 skb_queue_head_init(&newsk->sk_receive_queue);
1092 skb_queue_head_init(&newsk->sk_write_queue);
1093 #ifdef CONFIG_NET_DMA
1094 skb_queue_head_init(&newsk->sk_async_wait_queue);
1095 #endif
1096
1097 rwlock_init(&newsk->sk_dst_lock);
1098 rwlock_init(&newsk->sk_callback_lock);
1099 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1100 af_callback_keys + newsk->sk_family,
1101 af_family_clock_key_strings[newsk->sk_family]);
1102
1103 newsk->sk_dst_cache = NULL;
1104 newsk->sk_wmem_queued = 0;
1105 newsk->sk_forward_alloc = 0;
1106 newsk->sk_send_head = NULL;
1107 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1108
1109 sock_reset_flag(newsk, SOCK_DONE);
1110 skb_queue_head_init(&newsk->sk_error_queue);
1111
1112 filter = newsk->sk_filter;
1113 if (filter != NULL)
1114 sk_filter_charge(newsk, filter);
1115
1116 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1117 /* It is still raw copy of parent, so invalidate
1118 * destructor and make plain sk_free() */
1119 newsk->sk_destruct = NULL;
1120 sk_free(newsk);
1121 newsk = NULL;
1122 goto out;
1123 }
1124
1125 newsk->sk_err = 0;
1126 newsk->sk_priority = 0;
1127 /*
1128 * Before updating sk_refcnt, we must commit prior changes to memory
1129 * (Documentation/RCU/rculist_nulls.txt for details)
1130 */
1131 smp_wmb();
1132 atomic_set(&newsk->sk_refcnt, 2);
1133
1134 /*
1135 * Increment the counter in the same struct proto as the master
1136 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1137 * is the same as sk->sk_prot->socks, as this field was copied
1138 * with memcpy).
1139 *
1140 * This _changes_ the previous behaviour, where
1141 * tcp_create_openreq_child always was incrementing the
1142 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1143 * to be taken into account in all callers. -acme
1144 */
1145 sk_refcnt_debug_inc(newsk);
1146 sk_set_socket(newsk, NULL);
1147 newsk->sk_sleep = NULL;
1148
1149 if (newsk->sk_prot->sockets_allocated)
1150 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1151 }
1152 out:
1153 return newsk;
1154 }
1155
1156 EXPORT_SYMBOL_GPL(sk_clone);
1157
1158 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1159 {
1160 __sk_dst_set(sk, dst);
1161 sk->sk_route_caps = dst->dev->features;
1162 if (sk->sk_route_caps & NETIF_F_GSO)
1163 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1164 if (sk_can_gso(sk)) {
1165 if (dst->header_len) {
1166 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1167 } else {
1168 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1169 sk->sk_gso_max_size = dst->dev->gso_max_size;
1170 }
1171 }
1172 }
1173 EXPORT_SYMBOL_GPL(sk_setup_caps);
1174
1175 void __init sk_init(void)
1176 {
1177 if (num_physpages <= 4096) {
1178 sysctl_wmem_max = 32767;
1179 sysctl_rmem_max = 32767;
1180 sysctl_wmem_default = 32767;
1181 sysctl_rmem_default = 32767;
1182 } else if (num_physpages >= 131072) {
1183 sysctl_wmem_max = 131071;
1184 sysctl_rmem_max = 131071;
1185 }
1186 }
1187
1188 /*
1189 * Simple resource managers for sockets.
1190 */
1191
1192
1193 /*
1194 * Write buffer destructor automatically called from kfree_skb.
1195 */
1196 void sock_wfree(struct sk_buff *skb)
1197 {
1198 struct sock *sk = skb->sk;
1199
1200 /* In case it might be waiting for more memory. */
1201 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1202 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1203 sk->sk_write_space(sk);
1204 sock_put(sk);
1205 }
1206
1207 /*
1208 * Read buffer destructor automatically called from kfree_skb.
1209 */
1210 void sock_rfree(struct sk_buff *skb)
1211 {
1212 struct sock *sk = skb->sk;
1213
1214 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1215 sk_mem_uncharge(skb->sk, skb->truesize);
1216 }
1217
1218
1219 int sock_i_uid(struct sock *sk)
1220 {
1221 int uid;
1222
1223 read_lock(&sk->sk_callback_lock);
1224 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1225 read_unlock(&sk->sk_callback_lock);
1226 return uid;
1227 }
1228
1229 unsigned long sock_i_ino(struct sock *sk)
1230 {
1231 unsigned long ino;
1232
1233 read_lock(&sk->sk_callback_lock);
1234 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1235 read_unlock(&sk->sk_callback_lock);
1236 return ino;
1237 }
1238
1239 /*
1240 * Allocate a skb from the socket's send buffer.
1241 */
1242 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1243 gfp_t priority)
1244 {
1245 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1246 struct sk_buff * skb = alloc_skb(size, priority);
1247 if (skb) {
1248 skb_set_owner_w(skb, sk);
1249 return skb;
1250 }
1251 }
1252 return NULL;
1253 }
1254
1255 /*
1256 * Allocate a skb from the socket's receive buffer.
1257 */
1258 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1259 gfp_t priority)
1260 {
1261 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1262 struct sk_buff *skb = alloc_skb(size, priority);
1263 if (skb) {
1264 skb_set_owner_r(skb, sk);
1265 return skb;
1266 }
1267 }
1268 return NULL;
1269 }
1270
1271 /*
1272 * Allocate a memory block from the socket's option memory buffer.
1273 */
1274 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1275 {
1276 if ((unsigned)size <= sysctl_optmem_max &&
1277 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1278 void *mem;
1279 /* First do the add, to avoid the race if kmalloc
1280 * might sleep.
1281 */
1282 atomic_add(size, &sk->sk_omem_alloc);
1283 mem = kmalloc(size, priority);
1284 if (mem)
1285 return mem;
1286 atomic_sub(size, &sk->sk_omem_alloc);
1287 }
1288 return NULL;
1289 }
1290
1291 /*
1292 * Free an option memory block.
1293 */
1294 void sock_kfree_s(struct sock *sk, void *mem, int size)
1295 {
1296 kfree(mem);
1297 atomic_sub(size, &sk->sk_omem_alloc);
1298 }
1299
1300 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1301 I think, these locks should be removed for datagram sockets.
1302 */
1303 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1304 {
1305 DEFINE_WAIT(wait);
1306
1307 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1308 for (;;) {
1309 if (!timeo)
1310 break;
1311 if (signal_pending(current))
1312 break;
1313 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1314 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1315 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1316 break;
1317 if (sk->sk_shutdown & SEND_SHUTDOWN)
1318 break;
1319 if (sk->sk_err)
1320 break;
1321 timeo = schedule_timeout(timeo);
1322 }
1323 finish_wait(sk->sk_sleep, &wait);
1324 return timeo;
1325 }
1326
1327
1328 /*
1329 * Generic send/receive buffer handlers
1330 */
1331
1332 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1333 unsigned long data_len, int noblock,
1334 int *errcode)
1335 {
1336 struct sk_buff *skb;
1337 gfp_t gfp_mask;
1338 long timeo;
1339 int err;
1340
1341 gfp_mask = sk->sk_allocation;
1342 if (gfp_mask & __GFP_WAIT)
1343 gfp_mask |= __GFP_REPEAT;
1344
1345 timeo = sock_sndtimeo(sk, noblock);
1346 while (1) {
1347 err = sock_error(sk);
1348 if (err != 0)
1349 goto failure;
1350
1351 err = -EPIPE;
1352 if (sk->sk_shutdown & SEND_SHUTDOWN)
1353 goto failure;
1354
1355 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1356 skb = alloc_skb(header_len, gfp_mask);
1357 if (skb) {
1358 int npages;
1359 int i;
1360
1361 /* No pages, we're done... */
1362 if (!data_len)
1363 break;
1364
1365 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1366 skb->truesize += data_len;
1367 skb_shinfo(skb)->nr_frags = npages;
1368 for (i = 0; i < npages; i++) {
1369 struct page *page;
1370 skb_frag_t *frag;
1371
1372 page = alloc_pages(sk->sk_allocation, 0);
1373 if (!page) {
1374 err = -ENOBUFS;
1375 skb_shinfo(skb)->nr_frags = i;
1376 kfree_skb(skb);
1377 goto failure;
1378 }
1379
1380 frag = &skb_shinfo(skb)->frags[i];
1381 frag->page = page;
1382 frag->page_offset = 0;
1383 frag->size = (data_len >= PAGE_SIZE ?
1384 PAGE_SIZE :
1385 data_len);
1386 data_len -= PAGE_SIZE;
1387 }
1388
1389 /* Full success... */
1390 break;
1391 }
1392 err = -ENOBUFS;
1393 goto failure;
1394 }
1395 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1396 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1397 err = -EAGAIN;
1398 if (!timeo)
1399 goto failure;
1400 if (signal_pending(current))
1401 goto interrupted;
1402 timeo = sock_wait_for_wmem(sk, timeo);
1403 }
1404
1405 skb_set_owner_w(skb, sk);
1406 return skb;
1407
1408 interrupted:
1409 err = sock_intr_errno(timeo);
1410 failure:
1411 *errcode = err;
1412 return NULL;
1413 }
1414 EXPORT_SYMBOL(sock_alloc_send_pskb);
1415
1416 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1417 int noblock, int *errcode)
1418 {
1419 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1420 }
1421
1422 static void __lock_sock(struct sock *sk)
1423 {
1424 DEFINE_WAIT(wait);
1425
1426 for (;;) {
1427 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1428 TASK_UNINTERRUPTIBLE);
1429 spin_unlock_bh(&sk->sk_lock.slock);
1430 schedule();
1431 spin_lock_bh(&sk->sk_lock.slock);
1432 if (!sock_owned_by_user(sk))
1433 break;
1434 }
1435 finish_wait(&sk->sk_lock.wq, &wait);
1436 }
1437
1438 static void __release_sock(struct sock *sk)
1439 {
1440 struct sk_buff *skb = sk->sk_backlog.head;
1441
1442 do {
1443 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1444 bh_unlock_sock(sk);
1445
1446 do {
1447 struct sk_buff *next = skb->next;
1448
1449 skb->next = NULL;
1450 sk_backlog_rcv(sk, skb);
1451
1452 /*
1453 * We are in process context here with softirqs
1454 * disabled, use cond_resched_softirq() to preempt.
1455 * This is safe to do because we've taken the backlog
1456 * queue private:
1457 */
1458 cond_resched_softirq();
1459
1460 skb = next;
1461 } while (skb != NULL);
1462
1463 bh_lock_sock(sk);
1464 } while ((skb = sk->sk_backlog.head) != NULL);
1465 }
1466
1467 /**
1468 * sk_wait_data - wait for data to arrive at sk_receive_queue
1469 * @sk: sock to wait on
1470 * @timeo: for how long
1471 *
1472 * Now socket state including sk->sk_err is changed only under lock,
1473 * hence we may omit checks after joining wait queue.
1474 * We check receive queue before schedule() only as optimization;
1475 * it is very likely that release_sock() added new data.
1476 */
1477 int sk_wait_data(struct sock *sk, long *timeo)
1478 {
1479 int rc;
1480 DEFINE_WAIT(wait);
1481
1482 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1483 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1484 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1485 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1486 finish_wait(sk->sk_sleep, &wait);
1487 return rc;
1488 }
1489
1490 EXPORT_SYMBOL(sk_wait_data);
1491
1492 /**
1493 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1494 * @sk: socket
1495 * @size: memory size to allocate
1496 * @kind: allocation type
1497 *
1498 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1499 * rmem allocation. This function assumes that protocols which have
1500 * memory_pressure use sk_wmem_queued as write buffer accounting.
1501 */
1502 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1503 {
1504 struct proto *prot = sk->sk_prot;
1505 int amt = sk_mem_pages(size);
1506 int allocated;
1507
1508 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1509 allocated = atomic_add_return(amt, prot->memory_allocated);
1510
1511 /* Under limit. */
1512 if (allocated <= prot->sysctl_mem[0]) {
1513 if (prot->memory_pressure && *prot->memory_pressure)
1514 *prot->memory_pressure = 0;
1515 return 1;
1516 }
1517
1518 /* Under pressure. */
1519 if (allocated > prot->sysctl_mem[1])
1520 if (prot->enter_memory_pressure)
1521 prot->enter_memory_pressure(sk);
1522
1523 /* Over hard limit. */
1524 if (allocated > prot->sysctl_mem[2])
1525 goto suppress_allocation;
1526
1527 /* guarantee minimum buffer size under pressure */
1528 if (kind == SK_MEM_RECV) {
1529 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1530 return 1;
1531 } else { /* SK_MEM_SEND */
1532 if (sk->sk_type == SOCK_STREAM) {
1533 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1534 return 1;
1535 } else if (atomic_read(&sk->sk_wmem_alloc) <
1536 prot->sysctl_wmem[0])
1537 return 1;
1538 }
1539
1540 if (prot->memory_pressure) {
1541 int alloc;
1542
1543 if (!*prot->memory_pressure)
1544 return 1;
1545 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1546 if (prot->sysctl_mem[2] > alloc *
1547 sk_mem_pages(sk->sk_wmem_queued +
1548 atomic_read(&sk->sk_rmem_alloc) +
1549 sk->sk_forward_alloc))
1550 return 1;
1551 }
1552
1553 suppress_allocation:
1554
1555 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1556 sk_stream_moderate_sndbuf(sk);
1557
1558 /* Fail only if socket is _under_ its sndbuf.
1559 * In this case we cannot block, so that we have to fail.
1560 */
1561 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1562 return 1;
1563 }
1564
1565 /* Alas. Undo changes. */
1566 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1567 atomic_sub(amt, prot->memory_allocated);
1568 return 0;
1569 }
1570
1571 EXPORT_SYMBOL(__sk_mem_schedule);
1572
1573 /**
1574 * __sk_reclaim - reclaim memory_allocated
1575 * @sk: socket
1576 */
1577 void __sk_mem_reclaim(struct sock *sk)
1578 {
1579 struct proto *prot = sk->sk_prot;
1580
1581 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1582 prot->memory_allocated);
1583 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1584
1585 if (prot->memory_pressure && *prot->memory_pressure &&
1586 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1587 *prot->memory_pressure = 0;
1588 }
1589
1590 EXPORT_SYMBOL(__sk_mem_reclaim);
1591
1592
1593 /*
1594 * Set of default routines for initialising struct proto_ops when
1595 * the protocol does not support a particular function. In certain
1596 * cases where it makes no sense for a protocol to have a "do nothing"
1597 * function, some default processing is provided.
1598 */
1599
1600 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1601 {
1602 return -EOPNOTSUPP;
1603 }
1604
1605 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1606 int len, int flags)
1607 {
1608 return -EOPNOTSUPP;
1609 }
1610
1611 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1612 {
1613 return -EOPNOTSUPP;
1614 }
1615
1616 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1617 {
1618 return -EOPNOTSUPP;
1619 }
1620
1621 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1622 int *len, int peer)
1623 {
1624 return -EOPNOTSUPP;
1625 }
1626
1627 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1628 {
1629 return 0;
1630 }
1631
1632 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1633 {
1634 return -EOPNOTSUPP;
1635 }
1636
1637 int sock_no_listen(struct socket *sock, int backlog)
1638 {
1639 return -EOPNOTSUPP;
1640 }
1641
1642 int sock_no_shutdown(struct socket *sock, int how)
1643 {
1644 return -EOPNOTSUPP;
1645 }
1646
1647 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1648 char __user *optval, int optlen)
1649 {
1650 return -EOPNOTSUPP;
1651 }
1652
1653 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1654 char __user *optval, int __user *optlen)
1655 {
1656 return -EOPNOTSUPP;
1657 }
1658
1659 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1660 size_t len)
1661 {
1662 return -EOPNOTSUPP;
1663 }
1664
1665 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1666 size_t len, int flags)
1667 {
1668 return -EOPNOTSUPP;
1669 }
1670
1671 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1672 {
1673 /* Mirror missing mmap method error code */
1674 return -ENODEV;
1675 }
1676
1677 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1678 {
1679 ssize_t res;
1680 struct msghdr msg = {.msg_flags = flags};
1681 struct kvec iov;
1682 char *kaddr = kmap(page);
1683 iov.iov_base = kaddr + offset;
1684 iov.iov_len = size;
1685 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1686 kunmap(page);
1687 return res;
1688 }
1689
1690 /*
1691 * Default Socket Callbacks
1692 */
1693
1694 static void sock_def_wakeup(struct sock *sk)
1695 {
1696 read_lock(&sk->sk_callback_lock);
1697 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1698 wake_up_interruptible_all(sk->sk_sleep);
1699 read_unlock(&sk->sk_callback_lock);
1700 }
1701
1702 static void sock_def_error_report(struct sock *sk)
1703 {
1704 read_lock(&sk->sk_callback_lock);
1705 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1706 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1707 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1708 read_unlock(&sk->sk_callback_lock);
1709 }
1710
1711 static void sock_def_readable(struct sock *sk, int len)
1712 {
1713 read_lock(&sk->sk_callback_lock);
1714 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1715 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1716 POLLRDNORM | POLLRDBAND);
1717 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1718 read_unlock(&sk->sk_callback_lock);
1719 }
1720
1721 static void sock_def_write_space(struct sock *sk)
1722 {
1723 read_lock(&sk->sk_callback_lock);
1724
1725 /* Do not wake up a writer until he can make "significant"
1726 * progress. --DaveM
1727 */
1728 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1729 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1730 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1731 POLLWRNORM | POLLWRBAND);
1732
1733 /* Should agree with poll, otherwise some programs break */
1734 if (sock_writeable(sk))
1735 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1736 }
1737
1738 read_unlock(&sk->sk_callback_lock);
1739 }
1740
1741 static void sock_def_destruct(struct sock *sk)
1742 {
1743 kfree(sk->sk_protinfo);
1744 }
1745
1746 void sk_send_sigurg(struct sock *sk)
1747 {
1748 if (sk->sk_socket && sk->sk_socket->file)
1749 if (send_sigurg(&sk->sk_socket->file->f_owner))
1750 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1751 }
1752
1753 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1754 unsigned long expires)
1755 {
1756 if (!mod_timer(timer, expires))
1757 sock_hold(sk);
1758 }
1759
1760 EXPORT_SYMBOL(sk_reset_timer);
1761
1762 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1763 {
1764 if (timer_pending(timer) && del_timer(timer))
1765 __sock_put(sk);
1766 }
1767
1768 EXPORT_SYMBOL(sk_stop_timer);
1769
1770 void sock_init_data(struct socket *sock, struct sock *sk)
1771 {
1772 skb_queue_head_init(&sk->sk_receive_queue);
1773 skb_queue_head_init(&sk->sk_write_queue);
1774 skb_queue_head_init(&sk->sk_error_queue);
1775 #ifdef CONFIG_NET_DMA
1776 skb_queue_head_init(&sk->sk_async_wait_queue);
1777 #endif
1778
1779 sk->sk_send_head = NULL;
1780
1781 init_timer(&sk->sk_timer);
1782
1783 sk->sk_allocation = GFP_KERNEL;
1784 sk->sk_rcvbuf = sysctl_rmem_default;
1785 sk->sk_sndbuf = sysctl_wmem_default;
1786 sk->sk_state = TCP_CLOSE;
1787 sk_set_socket(sk, sock);
1788
1789 sock_set_flag(sk, SOCK_ZAPPED);
1790
1791 if (sock) {
1792 sk->sk_type = sock->type;
1793 sk->sk_sleep = &sock->wait;
1794 sock->sk = sk;
1795 } else
1796 sk->sk_sleep = NULL;
1797
1798 rwlock_init(&sk->sk_dst_lock);
1799 rwlock_init(&sk->sk_callback_lock);
1800 lockdep_set_class_and_name(&sk->sk_callback_lock,
1801 af_callback_keys + sk->sk_family,
1802 af_family_clock_key_strings[sk->sk_family]);
1803
1804 sk->sk_state_change = sock_def_wakeup;
1805 sk->sk_data_ready = sock_def_readable;
1806 sk->sk_write_space = sock_def_write_space;
1807 sk->sk_error_report = sock_def_error_report;
1808 sk->sk_destruct = sock_def_destruct;
1809
1810 sk->sk_sndmsg_page = NULL;
1811 sk->sk_sndmsg_off = 0;
1812
1813 sk->sk_peercred.pid = 0;
1814 sk->sk_peercred.uid = -1;
1815 sk->sk_peercred.gid = -1;
1816 sk->sk_write_pending = 0;
1817 sk->sk_rcvlowat = 1;
1818 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1819 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1820
1821 sk->sk_stamp = ktime_set(-1L, 0);
1822
1823 /*
1824 * Before updating sk_refcnt, we must commit prior changes to memory
1825 * (Documentation/RCU/rculist_nulls.txt for details)
1826 */
1827 smp_wmb();
1828 atomic_set(&sk->sk_refcnt, 1);
1829 atomic_set(&sk->sk_drops, 0);
1830 }
1831
1832 void lock_sock_nested(struct sock *sk, int subclass)
1833 {
1834 might_sleep();
1835 spin_lock_bh(&sk->sk_lock.slock);
1836 if (sk->sk_lock.owned)
1837 __lock_sock(sk);
1838 sk->sk_lock.owned = 1;
1839 spin_unlock(&sk->sk_lock.slock);
1840 /*
1841 * The sk_lock has mutex_lock() semantics here:
1842 */
1843 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1844 local_bh_enable();
1845 }
1846
1847 EXPORT_SYMBOL(lock_sock_nested);
1848
1849 void release_sock(struct sock *sk)
1850 {
1851 /*
1852 * The sk_lock has mutex_unlock() semantics:
1853 */
1854 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1855
1856 spin_lock_bh(&sk->sk_lock.slock);
1857 if (sk->sk_backlog.tail)
1858 __release_sock(sk);
1859 sk->sk_lock.owned = 0;
1860 if (waitqueue_active(&sk->sk_lock.wq))
1861 wake_up(&sk->sk_lock.wq);
1862 spin_unlock_bh(&sk->sk_lock.slock);
1863 }
1864 EXPORT_SYMBOL(release_sock);
1865
1866 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1867 {
1868 struct timeval tv;
1869 if (!sock_flag(sk, SOCK_TIMESTAMP))
1870 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1871 tv = ktime_to_timeval(sk->sk_stamp);
1872 if (tv.tv_sec == -1)
1873 return -ENOENT;
1874 if (tv.tv_sec == 0) {
1875 sk->sk_stamp = ktime_get_real();
1876 tv = ktime_to_timeval(sk->sk_stamp);
1877 }
1878 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1879 }
1880 EXPORT_SYMBOL(sock_get_timestamp);
1881
1882 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1883 {
1884 struct timespec ts;
1885 if (!sock_flag(sk, SOCK_TIMESTAMP))
1886 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1887 ts = ktime_to_timespec(sk->sk_stamp);
1888 if (ts.tv_sec == -1)
1889 return -ENOENT;
1890 if (ts.tv_sec == 0) {
1891 sk->sk_stamp = ktime_get_real();
1892 ts = ktime_to_timespec(sk->sk_stamp);
1893 }
1894 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1895 }
1896 EXPORT_SYMBOL(sock_get_timestampns);
1897
1898 void sock_enable_timestamp(struct sock *sk, int flag)
1899 {
1900 if (!sock_flag(sk, flag)) {
1901 sock_set_flag(sk, flag);
1902 /*
1903 * we just set one of the two flags which require net
1904 * time stamping, but time stamping might have been on
1905 * already because of the other one
1906 */
1907 if (!sock_flag(sk,
1908 flag == SOCK_TIMESTAMP ?
1909 SOCK_TIMESTAMPING_RX_SOFTWARE :
1910 SOCK_TIMESTAMP))
1911 net_enable_timestamp();
1912 }
1913 }
1914
1915 /*
1916 * Get a socket option on an socket.
1917 *
1918 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1919 * asynchronous errors should be reported by getsockopt. We assume
1920 * this means if you specify SO_ERROR (otherwise whats the point of it).
1921 */
1922 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1923 char __user *optval, int __user *optlen)
1924 {
1925 struct sock *sk = sock->sk;
1926
1927 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1928 }
1929
1930 EXPORT_SYMBOL(sock_common_getsockopt);
1931
1932 #ifdef CONFIG_COMPAT
1933 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1934 char __user *optval, int __user *optlen)
1935 {
1936 struct sock *sk = sock->sk;
1937
1938 if (sk->sk_prot->compat_getsockopt != NULL)
1939 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1940 optval, optlen);
1941 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1942 }
1943 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1944 #endif
1945
1946 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1947 struct msghdr *msg, size_t size, int flags)
1948 {
1949 struct sock *sk = sock->sk;
1950 int addr_len = 0;
1951 int err;
1952
1953 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1954 flags & ~MSG_DONTWAIT, &addr_len);
1955 if (err >= 0)
1956 msg->msg_namelen = addr_len;
1957 return err;
1958 }
1959
1960 EXPORT_SYMBOL(sock_common_recvmsg);
1961
1962 /*
1963 * Set socket options on an inet socket.
1964 */
1965 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1966 char __user *optval, int optlen)
1967 {
1968 struct sock *sk = sock->sk;
1969
1970 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1971 }
1972
1973 EXPORT_SYMBOL(sock_common_setsockopt);
1974
1975 #ifdef CONFIG_COMPAT
1976 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1977 char __user *optval, int optlen)
1978 {
1979 struct sock *sk = sock->sk;
1980
1981 if (sk->sk_prot->compat_setsockopt != NULL)
1982 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1983 optval, optlen);
1984 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1985 }
1986 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1987 #endif
1988
1989 void sk_common_release(struct sock *sk)
1990 {
1991 if (sk->sk_prot->destroy)
1992 sk->sk_prot->destroy(sk);
1993
1994 /*
1995 * Observation: when sock_common_release is called, processes have
1996 * no access to socket. But net still has.
1997 * Step one, detach it from networking:
1998 *
1999 * A. Remove from hash tables.
2000 */
2001
2002 sk->sk_prot->unhash(sk);
2003
2004 /*
2005 * In this point socket cannot receive new packets, but it is possible
2006 * that some packets are in flight because some CPU runs receiver and
2007 * did hash table lookup before we unhashed socket. They will achieve
2008 * receive queue and will be purged by socket destructor.
2009 *
2010 * Also we still have packets pending on receive queue and probably,
2011 * our own packets waiting in device queues. sock_destroy will drain
2012 * receive queue, but transmitted packets will delay socket destruction
2013 * until the last reference will be released.
2014 */
2015
2016 sock_orphan(sk);
2017
2018 xfrm_sk_free_policy(sk);
2019
2020 sk_refcnt_debug_release(sk);
2021 sock_put(sk);
2022 }
2023
2024 EXPORT_SYMBOL(sk_common_release);
2025
2026 static DEFINE_RWLOCK(proto_list_lock);
2027 static LIST_HEAD(proto_list);
2028
2029 #ifdef CONFIG_PROC_FS
2030 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2031 struct prot_inuse {
2032 int val[PROTO_INUSE_NR];
2033 };
2034
2035 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2036
2037 #ifdef CONFIG_NET_NS
2038 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2039 {
2040 int cpu = smp_processor_id();
2041 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2042 }
2043 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2044
2045 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2046 {
2047 int cpu, idx = prot->inuse_idx;
2048 int res = 0;
2049
2050 for_each_possible_cpu(cpu)
2051 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2052
2053 return res >= 0 ? res : 0;
2054 }
2055 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2056
2057 static int sock_inuse_init_net(struct net *net)
2058 {
2059 net->core.inuse = alloc_percpu(struct prot_inuse);
2060 return net->core.inuse ? 0 : -ENOMEM;
2061 }
2062
2063 static void sock_inuse_exit_net(struct net *net)
2064 {
2065 free_percpu(net->core.inuse);
2066 }
2067
2068 static struct pernet_operations net_inuse_ops = {
2069 .init = sock_inuse_init_net,
2070 .exit = sock_inuse_exit_net,
2071 };
2072
2073 static __init int net_inuse_init(void)
2074 {
2075 if (register_pernet_subsys(&net_inuse_ops))
2076 panic("Cannot initialize net inuse counters");
2077
2078 return 0;
2079 }
2080
2081 core_initcall(net_inuse_init);
2082 #else
2083 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2084
2085 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2086 {
2087 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2088 }
2089 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2090
2091 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2092 {
2093 int cpu, idx = prot->inuse_idx;
2094 int res = 0;
2095
2096 for_each_possible_cpu(cpu)
2097 res += per_cpu(prot_inuse, cpu).val[idx];
2098
2099 return res >= 0 ? res : 0;
2100 }
2101 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2102 #endif
2103
2104 static void assign_proto_idx(struct proto *prot)
2105 {
2106 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2107
2108 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2109 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2110 return;
2111 }
2112
2113 set_bit(prot->inuse_idx, proto_inuse_idx);
2114 }
2115
2116 static void release_proto_idx(struct proto *prot)
2117 {
2118 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2119 clear_bit(prot->inuse_idx, proto_inuse_idx);
2120 }
2121 #else
2122 static inline void assign_proto_idx(struct proto *prot)
2123 {
2124 }
2125
2126 static inline void release_proto_idx(struct proto *prot)
2127 {
2128 }
2129 #endif
2130
2131 int proto_register(struct proto *prot, int alloc_slab)
2132 {
2133 if (alloc_slab) {
2134 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2135 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2136 NULL);
2137
2138 if (prot->slab == NULL) {
2139 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2140 prot->name);
2141 goto out;
2142 }
2143
2144 if (prot->rsk_prot != NULL) {
2145 static const char mask[] = "request_sock_%s";
2146
2147 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2148 if (prot->rsk_prot->slab_name == NULL)
2149 goto out_free_sock_slab;
2150
2151 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2152 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2153 prot->rsk_prot->obj_size, 0,
2154 SLAB_HWCACHE_ALIGN, NULL);
2155
2156 if (prot->rsk_prot->slab == NULL) {
2157 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2158 prot->name);
2159 goto out_free_request_sock_slab_name;
2160 }
2161 }
2162
2163 if (prot->twsk_prot != NULL) {
2164 static const char mask[] = "tw_sock_%s";
2165
2166 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2167
2168 if (prot->twsk_prot->twsk_slab_name == NULL)
2169 goto out_free_request_sock_slab;
2170
2171 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2172 prot->twsk_prot->twsk_slab =
2173 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2174 prot->twsk_prot->twsk_obj_size,
2175 0,
2176 SLAB_HWCACHE_ALIGN |
2177 prot->slab_flags,
2178 NULL);
2179 if (prot->twsk_prot->twsk_slab == NULL)
2180 goto out_free_timewait_sock_slab_name;
2181 }
2182 }
2183
2184 write_lock(&proto_list_lock);
2185 list_add(&prot->node, &proto_list);
2186 assign_proto_idx(prot);
2187 write_unlock(&proto_list_lock);
2188 return 0;
2189
2190 out_free_timewait_sock_slab_name:
2191 kfree(prot->twsk_prot->twsk_slab_name);
2192 out_free_request_sock_slab:
2193 if (prot->rsk_prot && prot->rsk_prot->slab) {
2194 kmem_cache_destroy(prot->rsk_prot->slab);
2195 prot->rsk_prot->slab = NULL;
2196 }
2197 out_free_request_sock_slab_name:
2198 kfree(prot->rsk_prot->slab_name);
2199 out_free_sock_slab:
2200 kmem_cache_destroy(prot->slab);
2201 prot->slab = NULL;
2202 out:
2203 return -ENOBUFS;
2204 }
2205
2206 EXPORT_SYMBOL(proto_register);
2207
2208 void proto_unregister(struct proto *prot)
2209 {
2210 write_lock(&proto_list_lock);
2211 release_proto_idx(prot);
2212 list_del(&prot->node);
2213 write_unlock(&proto_list_lock);
2214
2215 if (prot->slab != NULL) {
2216 kmem_cache_destroy(prot->slab);
2217 prot->slab = NULL;
2218 }
2219
2220 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2221 kmem_cache_destroy(prot->rsk_prot->slab);
2222 kfree(prot->rsk_prot->slab_name);
2223 prot->rsk_prot->slab = NULL;
2224 }
2225
2226 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2227 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2228 kfree(prot->twsk_prot->twsk_slab_name);
2229 prot->twsk_prot->twsk_slab = NULL;
2230 }
2231 }
2232
2233 EXPORT_SYMBOL(proto_unregister);
2234
2235 #ifdef CONFIG_PROC_FS
2236 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2237 __acquires(proto_list_lock)
2238 {
2239 read_lock(&proto_list_lock);
2240 return seq_list_start_head(&proto_list, *pos);
2241 }
2242
2243 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2244 {
2245 return seq_list_next(v, &proto_list, pos);
2246 }
2247
2248 static void proto_seq_stop(struct seq_file *seq, void *v)
2249 __releases(proto_list_lock)
2250 {
2251 read_unlock(&proto_list_lock);
2252 }
2253
2254 static char proto_method_implemented(const void *method)
2255 {
2256 return method == NULL ? 'n' : 'y';
2257 }
2258
2259 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2260 {
2261 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2262 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2263 proto->name,
2264 proto->obj_size,
2265 sock_prot_inuse_get(seq_file_net(seq), proto),
2266 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2267 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2268 proto->max_header,
2269 proto->slab == NULL ? "no" : "yes",
2270 module_name(proto->owner),
2271 proto_method_implemented(proto->close),
2272 proto_method_implemented(proto->connect),
2273 proto_method_implemented(proto->disconnect),
2274 proto_method_implemented(proto->accept),
2275 proto_method_implemented(proto->ioctl),
2276 proto_method_implemented(proto->init),
2277 proto_method_implemented(proto->destroy),
2278 proto_method_implemented(proto->shutdown),
2279 proto_method_implemented(proto->setsockopt),
2280 proto_method_implemented(proto->getsockopt),
2281 proto_method_implemented(proto->sendmsg),
2282 proto_method_implemented(proto->recvmsg),
2283 proto_method_implemented(proto->sendpage),
2284 proto_method_implemented(proto->bind),
2285 proto_method_implemented(proto->backlog_rcv),
2286 proto_method_implemented(proto->hash),
2287 proto_method_implemented(proto->unhash),
2288 proto_method_implemented(proto->get_port),
2289 proto_method_implemented(proto->enter_memory_pressure));
2290 }
2291
2292 static int proto_seq_show(struct seq_file *seq, void *v)
2293 {
2294 if (v == &proto_list)
2295 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2296 "protocol",
2297 "size",
2298 "sockets",
2299 "memory",
2300 "press",
2301 "maxhdr",
2302 "slab",
2303 "module",
2304 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2305 else
2306 proto_seq_printf(seq, list_entry(v, struct proto, node));
2307 return 0;
2308 }
2309
2310 static const struct seq_operations proto_seq_ops = {
2311 .start = proto_seq_start,
2312 .next = proto_seq_next,
2313 .stop = proto_seq_stop,
2314 .show = proto_seq_show,
2315 };
2316
2317 static int proto_seq_open(struct inode *inode, struct file *file)
2318 {
2319 return seq_open_net(inode, file, &proto_seq_ops,
2320 sizeof(struct seq_net_private));
2321 }
2322
2323 static const struct file_operations proto_seq_fops = {
2324 .owner = THIS_MODULE,
2325 .open = proto_seq_open,
2326 .read = seq_read,
2327 .llseek = seq_lseek,
2328 .release = seq_release_net,
2329 };
2330
2331 static __net_init int proto_init_net(struct net *net)
2332 {
2333 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2334 return -ENOMEM;
2335
2336 return 0;
2337 }
2338
2339 static __net_exit void proto_exit_net(struct net *net)
2340 {
2341 proc_net_remove(net, "protocols");
2342 }
2343
2344
2345 static __net_initdata struct pernet_operations proto_net_ops = {
2346 .init = proto_init_net,
2347 .exit = proto_exit_net,
2348 };
2349
2350 static int __init proto_init(void)
2351 {
2352 return register_pernet_subsys(&proto_net_ops);
2353 }
2354
2355 subsys_initcall(proto_init);
2356
2357 #endif /* PROC_FS */
2358
2359 EXPORT_SYMBOL(sk_alloc);
2360 EXPORT_SYMBOL(sk_free);
2361 EXPORT_SYMBOL(sk_send_sigurg);
2362 EXPORT_SYMBOL(sock_alloc_send_skb);
2363 EXPORT_SYMBOL(sock_init_data);
2364 EXPORT_SYMBOL(sock_kfree_s);
2365 EXPORT_SYMBOL(sock_kmalloc);
2366 EXPORT_SYMBOL(sock_no_accept);
2367 EXPORT_SYMBOL(sock_no_bind);
2368 EXPORT_SYMBOL(sock_no_connect);
2369 EXPORT_SYMBOL(sock_no_getname);
2370 EXPORT_SYMBOL(sock_no_getsockopt);
2371 EXPORT_SYMBOL(sock_no_ioctl);
2372 EXPORT_SYMBOL(sock_no_listen);
2373 EXPORT_SYMBOL(sock_no_mmap);
2374 EXPORT_SYMBOL(sock_no_poll);
2375 EXPORT_SYMBOL(sock_no_recvmsg);
2376 EXPORT_SYMBOL(sock_no_sendmsg);
2377 EXPORT_SYMBOL(sock_no_sendpage);
2378 EXPORT_SYMBOL(sock_no_setsockopt);
2379 EXPORT_SYMBOL(sock_no_shutdown);
2380 EXPORT_SYMBOL(sock_no_socketpair);
2381 EXPORT_SYMBOL(sock_rfree);
2382 EXPORT_SYMBOL(sock_setsockopt);
2383 EXPORT_SYMBOL(sock_wfree);
2384 EXPORT_SYMBOL(sock_wmalloc);
2385 EXPORT_SYMBOL(sock_i_uid);
2386 EXPORT_SYMBOL(sock_i_ino);
2387 EXPORT_SYMBOL(sysctl_optmem_max);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree