1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
11 *
12 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Alan Cox, <A.Cox@swansea.ac.uk>
16 *
17 * Fixes:
18 * Alan Cox : Numerous verify_area() problems
19 * Alan Cox : Connecting on a connecting socket
20 * now returns an error for tcp.
21 * Alan Cox : sock->protocol is set correctly.
22 * and is not sometimes left as 0.
23 * Alan Cox : connect handles icmp errors on a
24 * connect properly. Unfortunately there
25 * is a restart syscall nasty there. I
26 * can't match BSD without hacking the C
27 * library. Ideas urgently sought!
28 * Alan Cox : Disallow bind() to addresses that are
29 * not ours - especially broadcast ones!!
30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
32 * instead they leave that for the DESTROY timer.
33 * Alan Cox : Clean up error flag in accept
34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
35 * was buggy. Put a remove_sock() in the handler
36 * for memory when we hit 0. Also altered the timer
37 * code. The ACK stuff can wait and needs major
38 * TCP layer surgery.
39 * Alan Cox : Fixed TCP ack bug, removed remove sock
40 * and fixed timer/inet_bh race.
41 * Alan Cox : Added zapped flag for TCP
42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
47 * Rick Sladkey : Relaxed UDP rules for matching packets.
48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
49 * Pauline Middelink : identd support
50 * Alan Cox : Fixed connect() taking signals I think.
51 * Alan Cox : SO_LINGER supported
52 * Alan Cox : Error reporting fixes
53 * Anonymous : inet_create tidied up (sk->reuse setting)
54 * Alan Cox : inet sockets don't set sk->type!
55 * Alan Cox : Split socket option code
56 * Alan Cox : Callbacks
57 * Alan Cox : Nagle flag for Charles & Johannes stuff
58 * Alex : Removed restriction on inet fioctl
59 * Alan Cox : Splitting INET from NET core
60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
62 * Alan Cox : Split IP from generic code
63 * Alan Cox : New kfree_skbmem()
64 * Alan Cox : Make SO_DEBUG superuser only.
65 * Alan Cox : Allow anyone to clear SO_DEBUG
66 * (compatibility fix)
67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
68 * Alan Cox : Allocator for a socket is settable.
69 * Alan Cox : SO_ERROR includes soft errors.
70 * Alan Cox : Allow NULL arguments on some SO_ opts
71 * Alan Cox : Generic socket allocation to make hooks
72 * easier (suggested by Craig Metz).
73 * Michael Pall : SO_ERROR returns positive errno again
74 * Steve Whitehouse: Added default destructor to free
75 * protocol private data.
76 * Steve Whitehouse: Added various other default routines
77 * common to several socket families.
78 * Chris Evans : Call suser() check last on F_SETOWN
79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
81 * Andi Kleen : Fix write_space callback
82 * Chris Evans : Security fixes - signedness again
83 * Arnaldo C. Melo : cleanups, use skb_queue_purge
84 *
85 * To Fix:
86 *
87 *
88 * This program is free software; you can redistribute it and/or
89 * modify it under the terms of the GNU General Public License
90 * as published by the Free Software Foundation; either version
91 * 2 of the License, or (at your option) any later version.
92 */
93
94 #include <linux/config.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
98 #include <linux/in.h>
99 #include <linux/kernel.h>
100 #include <linux/major.h>
101 #include <linux/module.h>
102 #include <linux/sched.h>
103 #include <linux/timer.h>
104 #include <linux/string.h>
105 #include <linux/sockios.h>
106 #include <linux/net.h>
107 #include <linux/mm.h>
108 #include <linux/slab.h>
109 #include <linux/interrupt.h>
110 #include <linux/poll.h>
111 #include <linux/tcp.h>
112 #include <linux/init.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/sock.h>
121 #include <net/xfrm.h>
122 #include <linux/ipsec.h>
123
124 #include <linux/filter.h>
125
126 #ifdef CONFIG_INET
127 #include <net/tcp.h>
128 #endif
129
130 /* Take into consideration the size of the struct sk_buff overhead in the
131 * determination of these values, since that is non-constant across
132 * platforms. This makes socket queueing behavior and performance
133 * not depend upon such differences.
134 */
135 #define _SK_MEM_PACKETS 256
136 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
137 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
138 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
139
140 /* Run time adjustable parameters. */
141 __u32 sysctl_wmem_max = SK_WMEM_MAX;
142 __u32 sysctl_rmem_max = SK_RMEM_MAX;
143 __u32 sysctl_wmem_default = SK_WMEM_MAX;
144 __u32 sysctl_rmem_default = SK_RMEM_MAX;
145
146 /* Maximal space eaten by iovec or ancilliary data plus some space */
147 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
148
149 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
150 {
151 struct timeval tv;
152
153 if (optlen < sizeof(tv))
154 return -EINVAL;
155 if (copy_from_user(&tv, optval, sizeof(tv)))
156 return -EFAULT;
157
158 *timeo_p = MAX_SCHEDULE_TIMEOUT;
159 if (tv.tv_sec == 0 && tv.tv_usec == 0)
160 return 0;
161 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
162 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
163 return 0;
164 }
165
166 static void sock_warn_obsolete_bsdism(const char *name)
167 {
168 static int warned;
169 static char warncomm[TASK_COMM_LEN];
170 if (strcmp(warncomm, current->comm) && warned < 5) {
171 strcpy(warncomm, current->comm);
172 printk(KERN_WARNING "process `%s' is using obsolete "
173 "%s SO_BSDCOMPAT\n", warncomm, name);
174 warned++;
175 }
176 }
177
178 static void sock_disable_timestamp(struct sock *sk)
179 {
180 if (sock_flag(sk, SOCK_TIMESTAMP)) {
181 sock_reset_flag(sk, SOCK_TIMESTAMP);
182 net_disable_timestamp();
183 }
184 }
185
186
187 /*
188 * This is meant for all protocols to use and covers goings on
189 * at the socket level. Everything here is generic.
190 */
191
192 int sock_setsockopt(struct socket *sock, int level, int optname,
193 char __user *optval, int optlen)
194 {
195 struct sock *sk=sock->sk;
196 struct sk_filter *filter;
197 int val;
198 int valbool;
199 struct linger ling;
200 int ret = 0;
201
202 /*
203 * Options without arguments
204 */
205
206 #ifdef SO_DONTLINGER /* Compatibility item... */
207 switch (optname) {
208 case SO_DONTLINGER:
209 sock_reset_flag(sk, SOCK_LINGER);
210 return 0;
211 }
212 #endif
213
214 if(optlen<sizeof(int))
215 return(-EINVAL);
216
217 if (get_user(val, (int __user *)optval))
218 return -EFAULT;
219
220 valbool = val?1:0;
221
222 lock_sock(sk);
223
224 switch(optname)
225 {
226 case SO_DEBUG:
227 if(val && !capable(CAP_NET_ADMIN))
228 {
229 ret = -EACCES;
230 }
231 else
232 sk->sk_debug = valbool;
233 break;
234 case SO_REUSEADDR:
235 sk->sk_reuse = valbool;
236 break;
237 case SO_TYPE:
238 case SO_ERROR:
239 ret = -ENOPROTOOPT;
240 break;
241 case SO_DONTROUTE:
242 sk->sk_localroute = valbool;
243 break;
244 case SO_BROADCAST:
245 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
246 break;
247 case SO_SNDBUF:
248 /* Don't error on this BSD doesn't and if you think
249 about it this is right. Otherwise apps have to
250 play 'guess the biggest size' games. RCVBUF/SNDBUF
251 are treated in BSD as hints */
252
253 if (val > sysctl_wmem_max)
254 val = sysctl_wmem_max;
255
256 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
257 if ((val * 2) < SOCK_MIN_SNDBUF)
258 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
259 else
260 sk->sk_sndbuf = val * 2;
261
262 /*
263 * Wake up sending tasks if we
264 * upped the value.
265 */
266 sk->sk_write_space(sk);
267 break;
268
269 case SO_RCVBUF:
270 /* Don't error on this BSD doesn't and if you think
271 about it this is right. Otherwise apps have to
272 play 'guess the biggest size' games. RCVBUF/SNDBUF
273 are treated in BSD as hints */
274
275 if (val > sysctl_rmem_max)
276 val = sysctl_rmem_max;
277
278 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
279 /* FIXME: is this lower bound the right one? */
280 if ((val * 2) < SOCK_MIN_RCVBUF)
281 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
282 else
283 sk->sk_rcvbuf = val * 2;
284 break;
285
286 case SO_KEEPALIVE:
287 #ifdef CONFIG_INET
288 if (sk->sk_protocol == IPPROTO_TCP)
289 tcp_set_keepalive(sk, valbool);
290 #endif
291 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
292 break;
293
294 case SO_OOBINLINE:
295 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
296 break;
297
298 case SO_NO_CHECK:
299 sk->sk_no_check = valbool;
300 break;
301
302 case SO_PRIORITY:
303 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
304 sk->sk_priority = val;
305 else
306 ret = -EPERM;
307 break;
308
309 case SO_LINGER:
310 if(optlen<sizeof(ling)) {
311 ret = -EINVAL; /* 1003.1g */
312 break;
313 }
314 if (copy_from_user(&ling,optval,sizeof(ling))) {
315 ret = -EFAULT;
316 break;
317 }
318 if (!ling.l_onoff)
319 sock_reset_flag(sk, SOCK_LINGER);
320 else {
321 #if (BITS_PER_LONG == 32)
322 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
323 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
324 else
325 #endif
326 sk->sk_lingertime = ling.l_linger * HZ;
327 sock_set_flag(sk, SOCK_LINGER);
328 }
329 break;
330
331 case SO_BSDCOMPAT:
332 sock_warn_obsolete_bsdism("setsockopt");
333 break;
334
335 case SO_PASSCRED:
336 sock->passcred = valbool;
337 break;
338
339 case SO_TIMESTAMP:
340 sk->sk_rcvtstamp = valbool;
341 if (valbool)
342 sock_enable_timestamp(sk);
343 break;
344
345 case SO_RCVLOWAT:
346 if (val < 0)
347 val = INT_MAX;
348 sk->sk_rcvlowat = val ? : 1;
349 break;
350
351 case SO_RCVTIMEO:
352 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
353 break;
354
355 case SO_SNDTIMEO:
356 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
357 break;
358
359 #ifdef CONFIG_NETDEVICES
360 case SO_BINDTODEVICE:
361 {
362 char devname[IFNAMSIZ];
363
364 /* Sorry... */
365 if (!capable(CAP_NET_RAW)) {
366 ret = -EPERM;
367 break;
368 }
369
370 /* Bind this socket to a particular device like "eth0",
371 * as specified in the passed interface name. If the
372 * name is "" or the option length is zero the socket
373 * is not bound.
374 */
375
376 if (!valbool) {
377 sk->sk_bound_dev_if = 0;
378 } else {
379 if (optlen > IFNAMSIZ)
380 optlen = IFNAMSIZ;
381 if (copy_from_user(devname, optval, optlen)) {
382 ret = -EFAULT;
383 break;
384 }
385
386 /* Remove any cached route for this socket. */
387 sk_dst_reset(sk);
388
389 if (devname[0] == '\0') {
390 sk->sk_bound_dev_if = 0;
391 } else {
392 struct net_device *dev = dev_get_by_name(devname);
393 if (!dev) {
394 ret = -ENODEV;
395 break;
396 }
397 sk->sk_bound_dev_if = dev->ifindex;
398 dev_put(dev);
399 }
400 }
401 break;
402 }
403 #endif
404
405
406 case SO_ATTACH_FILTER:
407 ret = -EINVAL;
408 if (optlen == sizeof(struct sock_fprog)) {
409 struct sock_fprog fprog;
410
411 ret = -EFAULT;
412 if (copy_from_user(&fprog, optval, sizeof(fprog)))
413 break;
414
415 ret = sk_attach_filter(&fprog, sk);
416 }
417 break;
418
419 case SO_DETACH_FILTER:
420 spin_lock_bh(&sk->sk_lock.slock);
421 filter = sk->sk_filter;
422 if (filter) {
423 sk->sk_filter = NULL;
424 spin_unlock_bh(&sk->sk_lock.slock);
425 sk_filter_release(sk, filter);
426 break;
427 }
428 spin_unlock_bh(&sk->sk_lock.slock);
429 ret = -ENONET;
430 break;
431
432 /* We implement the SO_SNDLOWAT etc to
433 not be settable (1003.1g 5.3) */
434 default:
435 ret = -ENOPROTOOPT;
436 break;
437 }
438 release_sock(sk);
439 return ret;
440 }
441
442
443 int sock_getsockopt(struct socket *sock, int level, int optname,
444 char __user *optval, int __user *optlen)
445 {
446 struct sock *sk = sock->sk;
447
448 union
449 {
450 int val;
451 struct linger ling;
452 struct timeval tm;
453 } v;
454
455 unsigned int lv = sizeof(int);
456 int len;
457
458 if(get_user(len,optlen))
459 return -EFAULT;
460 if(len < 0)
461 return -EINVAL;
462
463 switch(optname)
464 {
465 case SO_DEBUG:
466 v.val = sk->sk_debug;
467 break;
468
469 case SO_DONTROUTE:
470 v.val = sk->sk_localroute;
471 break;
472
473 case SO_BROADCAST:
474 v.val = !!sock_flag(sk, SOCK_BROADCAST);
475 break;
476
477 case SO_SNDBUF:
478 v.val = sk->sk_sndbuf;
479 break;
480
481 case SO_RCVBUF:
482 v.val = sk->sk_rcvbuf;
483 break;
484
485 case SO_REUSEADDR:
486 v.val = sk->sk_reuse;
487 break;
488
489 case SO_KEEPALIVE:
490 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
491 break;
492
493 case SO_TYPE:
494 v.val = sk->sk_type;
495 break;
496
497 case SO_ERROR:
498 v.val = -sock_error(sk);
499 if(v.val==0)
500 v.val = xchg(&sk->sk_err_soft, 0);
501 break;
502
503 case SO_OOBINLINE:
504 v.val = !!sock_flag(sk, SOCK_URGINLINE);
505 break;
506
507 case SO_NO_CHECK:
508 v.val = sk->sk_no_check;
509 break;
510
511 case SO_PRIORITY:
512 v.val = sk->sk_priority;
513 break;
514
515 case SO_LINGER:
516 lv = sizeof(v.ling);
517 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
518 v.ling.l_linger = sk->sk_lingertime / HZ;
519 break;
520
521 case SO_BSDCOMPAT:
522 sock_warn_obsolete_bsdism("getsockopt");
523 break;
524
525 case SO_TIMESTAMP:
526 v.val = sk->sk_rcvtstamp;
527 break;
528
529 case SO_RCVTIMEO:
530 lv=sizeof(struct timeval);
531 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
532 v.tm.tv_sec = 0;
533 v.tm.tv_usec = 0;
534 } else {
535 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
536 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
537 }
538 break;
539
540 case SO_SNDTIMEO:
541 lv=sizeof(struct timeval);
542 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
543 v.tm.tv_sec = 0;
544 v.tm.tv_usec = 0;
545 } else {
546 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
547 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
548 }
549 break;
550
551 case SO_RCVLOWAT:
552 v.val = sk->sk_rcvlowat;
553 break;
554
555 case SO_SNDLOWAT:
556 v.val=1;
557 break;
558
559 case SO_PASSCRED:
560 v.val = sock->passcred;
561 break;
562
563 case SO_PEERCRED:
564 if (len > sizeof(sk->sk_peercred))
565 len = sizeof(sk->sk_peercred);
566 if (copy_to_user(optval, &sk->sk_peercred, len))
567 return -EFAULT;
568 goto lenout;
569
570 case SO_PEERNAME:
571 {
572 char address[128];
573
574 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
575 return -ENOTCONN;
576 if (lv < len)
577 return -EINVAL;
578 if (copy_to_user(optval, address, len))
579 return -EFAULT;
580 goto lenout;
581 }
582
583 /* Dubious BSD thing... Probably nobody even uses it, but
584 * the UNIX standard wants it for whatever reason... -DaveM
585 */
586 case SO_ACCEPTCONN:
587 v.val = sk->sk_state == TCP_LISTEN;
588 break;
589
590 case SO_PEERSEC:
591 return security_socket_getpeersec(sock, optval, optlen, len);
592
593 default:
594 return(-ENOPROTOOPT);
595 }
596 if (len > lv)
597 len = lv;
598 if (copy_to_user(optval, &v, len))
599 return -EFAULT;
600 lenout:
601 if (put_user(len, optlen))
602 return -EFAULT;
603 return 0;
604 }
605
606 static kmem_cache_t *sk_cachep;
607
608 /**
609 * sk_alloc - All socket objects are allocated here
610 * @family - protocol family
611 * @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
612 * @zero_it - zeroes the allocated sock
613 * @slab - alternate slab
614 *
615 * All socket objects are allocated here. If @zero_it is non-zero
616 * it should have the size of the are to be zeroed, because the
617 * private slabcaches have different sizes of the generic struct sock.
618 * 1 has been kept as a way to say sizeof(struct sock).
619 */
620 struct sock *sk_alloc(int family, int priority, int zero_it, kmem_cache_t *slab)
621 {
622 struct sock *sk = NULL;
623
624 if (!slab)
625 slab = sk_cachep;
626 sk = kmem_cache_alloc(slab, priority);
627 if (sk) {
628 if (zero_it) {
629 memset(sk, 0,
630 zero_it == 1 ? sizeof(struct sock) : zero_it);
631 sk->sk_family = family;
632 sock_lock_init(sk);
633 }
634 sk->sk_slab = slab;
635
636 if (security_sk_alloc(sk, family, priority)) {
637 kmem_cache_free(slab, sk);
638 sk = NULL;
639 }
640 }
641 return sk;
642 }
643
644 void sk_free(struct sock *sk)
645 {
646 struct sk_filter *filter;
647 struct module *owner = sk->sk_owner;
648
649 if (sk->sk_destruct)
650 sk->sk_destruct(sk);
651
652 filter = sk->sk_filter;
653 if (filter) {
654 sk_filter_release(sk, filter);
655 sk->sk_filter = NULL;
656 }
657
658 sock_disable_timestamp(sk);
659
660 if (atomic_read(&sk->sk_omem_alloc))
661 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
662 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
663
664 security_sk_free(sk);
665 kmem_cache_free(sk->sk_slab, sk);
666 module_put(owner);
667 }
668
669 void __init sk_init(void)
670 {
671 sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
672 SLAB_HWCACHE_ALIGN, NULL, NULL);
673 if (!sk_cachep)
674 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
675
676 if (num_physpages <= 4096) {
677 sysctl_wmem_max = 32767;
678 sysctl_rmem_max = 32767;
679 sysctl_wmem_default = 32767;
680 sysctl_rmem_default = 32767;
681 } else if (num_physpages >= 131072) {
682 sysctl_wmem_max = 131071;
683 sysctl_rmem_max = 131071;
684 }
685 }
686
687 /*
688 * Simple resource managers for sockets.
689 */
690
691
692 /*
693 * Write buffer destructor automatically called from kfree_skb.
694 */
695 void sock_wfree(struct sk_buff *skb)
696 {
697 struct sock *sk = skb->sk;
698
699 /* In case it might be waiting for more memory. */
700 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
701 if (!sk->sk_use_write_queue)
702 sk->sk_write_space(sk);
703 sock_put(sk);
704 }
705
706 /*
707 * Read buffer destructor automatically called from kfree_skb.
708 */
709 void sock_rfree(struct sk_buff *skb)
710 {
711 struct sock *sk = skb->sk;
712
713 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
714 }
715
716
717 int sock_i_uid(struct sock *sk)
718 {
719 int uid;
720
721 read_lock(&sk->sk_callback_lock);
722 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
723 read_unlock(&sk->sk_callback_lock);
724 return uid;
725 }
726
727 unsigned long sock_i_ino(struct sock *sk)
728 {
729 unsigned long ino;
730
731 read_lock(&sk->sk_callback_lock);
732 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
733 read_unlock(&sk->sk_callback_lock);
734 return ino;
735 }
736
737 /*
738 * Allocate a skb from the socket's send buffer.
739 */
740 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
741 {
742 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
743 struct sk_buff * skb = alloc_skb(size, priority);
744 if (skb) {
745 skb_set_owner_w(skb, sk);
746 return skb;
747 }
748 }
749 return NULL;
750 }
751
752 /*
753 * Allocate a skb from the socket's receive buffer.
754 */
755 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
756 {
757 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
758 struct sk_buff *skb = alloc_skb(size, priority);
759 if (skb) {
760 skb_set_owner_r(skb, sk);
761 return skb;
762 }
763 }
764 return NULL;
765 }
766
767 /*
768 * Allocate a memory block from the socket's option memory buffer.
769 */
770 void *sock_kmalloc(struct sock *sk, int size, int priority)
771 {
772 if ((unsigned)size <= sysctl_optmem_max &&
773 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
774 void *mem;
775 /* First do the add, to avoid the race if kmalloc
776 * might sleep.
777 */
778 atomic_add(size, &sk->sk_omem_alloc);
779 mem = kmalloc(size, priority);
780 if (mem)
781 return mem;
782 atomic_sub(size, &sk->sk_omem_alloc);
783 }
784 return NULL;
785 }
786
787 /*
788 * Free an option memory block.
789 */
790 void sock_kfree_s(struct sock *sk, void *mem, int size)
791 {
792 kfree(mem);
793 atomic_sub(size, &sk->sk_omem_alloc);
794 }
795
796 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
797 I think, these locks should be removed for datagram sockets.
798 */
799 static long sock_wait_for_wmem(struct sock * sk, long timeo)
800 {
801 DEFINE_WAIT(wait);
802
803 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
804 for (;;) {
805 if (!timeo)
806 break;
807 if (signal_pending(current))
808 break;
809 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
810 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
811 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
812 break;
813 if (sk->sk_shutdown & SEND_SHUTDOWN)
814 break;
815 if (sk->sk_err)
816 break;
817 timeo = schedule_timeout(timeo);
818 }
819 finish_wait(sk->sk_sleep, &wait);
820 return timeo;
821 }
822
823
824 /*
825 * Generic send/receive buffer handlers
826 */
827
828 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
829 unsigned long header_len,
830 unsigned long data_len,
831 int noblock, int *errcode)
832 {
833 struct sk_buff *skb;
834 unsigned int gfp_mask;
835 long timeo;
836 int err;
837
838 gfp_mask = sk->sk_allocation;
839 if (gfp_mask & __GFP_WAIT)
840 gfp_mask |= __GFP_REPEAT;
841
842 timeo = sock_sndtimeo(sk, noblock);
843 while (1) {
844 err = sock_error(sk);
845 if (err != 0)
846 goto failure;
847
848 err = -EPIPE;
849 if (sk->sk_shutdown & SEND_SHUTDOWN)
850 goto failure;
851
852 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
853 skb = alloc_skb(header_len, sk->sk_allocation);
854 if (skb) {
855 int npages;
856 int i;
857
858 /* No pages, we're done... */
859 if (!data_len)
860 break;
861
862 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
863 skb->truesize += data_len;
864 skb_shinfo(skb)->nr_frags = npages;
865 for (i = 0; i < npages; i++) {
866 struct page *page;
867 skb_frag_t *frag;
868
869 page = alloc_pages(sk->sk_allocation, 0);
870 if (!page) {
871 err = -ENOBUFS;
872 skb_shinfo(skb)->nr_frags = i;
873 kfree_skb(skb);
874 goto failure;
875 }
876
877 frag = &skb_shinfo(skb)->frags[i];
878 frag->page = page;
879 frag->page_offset = 0;
880 frag->size = (data_len >= PAGE_SIZE ?
881 PAGE_SIZE :
882 data_len);
883 data_len -= PAGE_SIZE;
884 }
885
886 /* Full success... */
887 break;
888 }
889 err = -ENOBUFS;
890 goto failure;
891 }
892 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
893 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
894 err = -EAGAIN;
895 if (!timeo)
896 goto failure;
897 if (signal_pending(current))
898 goto interrupted;
899 timeo = sock_wait_for_wmem(sk, timeo);
900 }
901
902 skb_set_owner_w(skb, sk);
903 return skb;
904
905 interrupted:
906 err = sock_intr_errno(timeo);
907 failure:
908 *errcode = err;
909 return NULL;
910 }
911
912 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
913 int noblock, int *errcode)
914 {
915 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
916 }
917
918 static void __lock_sock(struct sock *sk)
919 {
920 DEFINE_WAIT(wait);
921
922 for(;;) {
923 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
924 TASK_UNINTERRUPTIBLE);
925 spin_unlock_bh(&sk->sk_lock.slock);
926 schedule();
927 spin_lock_bh(&sk->sk_lock.slock);
928 if(!sock_owned_by_user(sk))
929 break;
930 }
931 finish_wait(&sk->sk_lock.wq, &wait);
932 }
933
934 static void __release_sock(struct sock *sk)
935 {
936 struct sk_buff *skb = sk->sk_backlog.head;
937
938 do {
939 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
940 bh_unlock_sock(sk);
941
942 do {
943 struct sk_buff *next = skb->next;
944
945 skb->next = NULL;
946 sk->sk_backlog_rcv(sk, skb);
947
948 /*
949 * We are in process context here with softirqs
950 * disabled, use cond_resched_softirq() to preempt.
951 * This is safe to do because we've taken the backlog
952 * queue private:
953 */
954 cond_resched_softirq();
955
956 skb = next;
957 } while (skb != NULL);
958
959 bh_lock_sock(sk);
960 } while((skb = sk->sk_backlog.head) != NULL);
961 }
962
963 /**
964 * sk_wait_data - wait for data to arrive at sk_receive_queue
965 * sk - sock to wait on
966 * timeo - for how long
967 *
968 * Now socket state including sk->sk_err is changed only under lock,
969 * hence we may omit checks after joining wait queue.
970 * We check receive queue before schedule() only as optimization;
971 * it is very likely that release_sock() added new data.
972 */
973 int sk_wait_data(struct sock *sk, long *timeo)
974 {
975 int rc;
976 DEFINE_WAIT(wait);
977
978 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
979 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
980 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
981 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
982 finish_wait(sk->sk_sleep, &wait);
983 return rc;
984 }
985
986 EXPORT_SYMBOL(sk_wait_data);
987
988 /*
989 * Set of default routines for initialising struct proto_ops when
990 * the protocol does not support a particular function. In certain
991 * cases where it makes no sense for a protocol to have a "do nothing"
992 * function, some default processing is provided.
993 */
994
995 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
996 {
997 return -EOPNOTSUPP;
998 }
999
1000 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1001 int len, int flags)
1002 {
1003 return -EOPNOTSUPP;
1004 }
1005
1006 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1007 {
1008 return -EOPNOTSUPP;
1009 }
1010
1011 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1012 {
1013 return -EOPNOTSUPP;
1014 }
1015
1016 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1017 int *len, int peer)
1018 {
1019 return -EOPNOTSUPP;
1020 }
1021
1022 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1023 {
1024 return 0;
1025 }
1026
1027 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1028 {
1029 return -EOPNOTSUPP;
1030 }
1031
1032 int sock_no_listen(struct socket *sock, int backlog)
1033 {
1034 return -EOPNOTSUPP;
1035 }
1036
1037 int sock_no_shutdown(struct socket *sock, int how)
1038 {
1039 return -EOPNOTSUPP;
1040 }
1041
1042 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1043 char __user *optval, int optlen)
1044 {
1045 return -EOPNOTSUPP;
1046 }
1047
1048 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1049 char __user *optval, int __user *optlen)
1050 {
1051 return -EOPNOTSUPP;
1052 }
1053
1054 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1055 size_t len)
1056 {
1057 return -EOPNOTSUPP;
1058 }
1059
1060 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1061 size_t len, int flags)
1062 {
1063 return -EOPNOTSUPP;
1064 }
1065
1066 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1067 {
1068 /* Mirror missing mmap method error code */
1069 return -ENODEV;
1070 }
1071
1072 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1073 {
1074 ssize_t res;
1075 struct msghdr msg = {.msg_flags = flags};
1076 struct kvec iov;
1077 char *kaddr = kmap(page);
1078 iov.iov_base = kaddr + offset;
1079 iov.iov_len = size;
1080 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1081 kunmap(page);
1082 return res;
1083 }
1084
1085 /*
1086 * Default Socket Callbacks
1087 */
1088
1089 static void sock_def_wakeup(struct sock *sk)
1090 {
1091 read_lock(&sk->sk_callback_lock);
1092 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1093 wake_up_interruptible_all(sk->sk_sleep);
1094 read_unlock(&sk->sk_callback_lock);
1095 }
1096
1097 static void sock_def_error_report(struct sock *sk)
1098 {
1099 read_lock(&sk->sk_callback_lock);
1100 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1101 wake_up_interruptible(sk->sk_sleep);
1102 sk_wake_async(sk,0,POLL_ERR);
1103 read_unlock(&sk->sk_callback_lock);
1104 }
1105
1106 static void sock_def_readable(struct sock *sk, int len)
1107 {
1108 read_lock(&sk->sk_callback_lock);
1109 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1110 wake_up_interruptible(sk->sk_sleep);
1111 sk_wake_async(sk,1,POLL_IN);
1112 read_unlock(&sk->sk_callback_lock);
1113 }
1114
1115 static void sock_def_write_space(struct sock *sk)
1116 {
1117 read_lock(&sk->sk_callback_lock);
1118
1119 /* Do not wake up a writer until he can make "significant"
1120 * progress. --DaveM
1121 */
1122 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1123 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1124 wake_up_interruptible(sk->sk_sleep);
1125
1126 /* Should agree with poll, otherwise some programs break */
1127 if (sock_writeable(sk))
1128 sk_wake_async(sk, 2, POLL_OUT);
1129 }
1130
1131 read_unlock(&sk->sk_callback_lock);
1132 }
1133
1134 static void sock_def_destruct(struct sock *sk)
1135 {
1136 if (sk->sk_protinfo)
1137 kfree(sk->sk_protinfo);
1138 }
1139
1140 void sk_send_sigurg(struct sock *sk)
1141 {
1142 if (sk->sk_socket && sk->sk_socket->file)
1143 if (send_sigurg(&sk->sk_socket->file->f_owner))
1144 sk_wake_async(sk, 3, POLL_PRI);
1145 }
1146
1147 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1148 unsigned long expires)
1149 {
1150 if (!mod_timer(timer, expires))
1151 sock_hold(sk);
1152 }
1153
1154 EXPORT_SYMBOL(sk_reset_timer);
1155
1156 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1157 {
1158 if (timer_pending(timer) && del_timer(timer))
1159 __sock_put(sk);
1160 }
1161
1162 EXPORT_SYMBOL(sk_stop_timer);
1163
1164 void sock_init_data(struct socket *sock, struct sock *sk)
1165 {
1166 skb_queue_head_init(&sk->sk_receive_queue);
1167 skb_queue_head_init(&sk->sk_write_queue);
1168 skb_queue_head_init(&sk->sk_error_queue);
1169
1170 sk->sk_send_head = NULL;
1171
1172 init_timer(&sk->sk_timer);
1173
1174 sk->sk_allocation = GFP_KERNEL;
1175 sk->sk_rcvbuf = sysctl_rmem_default;
1176 sk->sk_sndbuf = sysctl_wmem_default;
1177 sk->sk_state = TCP_CLOSE;
1178 sk->sk_zapped = 1;
1179 sk->sk_socket = sock;
1180
1181 if(sock)
1182 {
1183 sk->sk_type = sock->type;
1184 sk->sk_sleep = &sock->wait;
1185 sock->sk = sk;
1186 } else
1187 sk->sk_sleep = NULL;
1188
1189 rwlock_init(&sk->sk_dst_lock);
1190 rwlock_init(&sk->sk_callback_lock);
1191
1192 sk->sk_state_change = sock_def_wakeup;
1193 sk->sk_data_ready = sock_def_readable;
1194 sk->sk_write_space = sock_def_write_space;
1195 sk->sk_error_report = sock_def_error_report;
1196 sk->sk_destruct = sock_def_destruct;
1197
1198 sk->sk_sndmsg_page = NULL;
1199 sk->sk_sndmsg_off = 0;
1200
1201 sk->sk_peercred.pid = 0;
1202 sk->sk_peercred.uid = -1;
1203 sk->sk_peercred.gid = -1;
1204 sk->sk_write_pending = 0;
1205 sk->sk_rcvlowat = 1;
1206 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1207 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1208 sk->sk_owner = NULL;
1209
1210 sk->sk_stamp.tv_sec = -1L;
1211 sk->sk_stamp.tv_usec = -1L;
1212
1213 atomic_set(&sk->sk_refcnt, 1);
1214 }
1215
1216 void fastcall lock_sock(struct sock *sk)
1217 {
1218 might_sleep();
1219 spin_lock_bh(&(sk->sk_lock.slock));
1220 if (sk->sk_lock.owner)
1221 __lock_sock(sk);
1222 sk->sk_lock.owner = (void *)1;
1223 spin_unlock_bh(&(sk->sk_lock.slock));
1224 }
1225
1226 EXPORT_SYMBOL(lock_sock);
1227
1228 void fastcall release_sock(struct sock *sk)
1229 {
1230 spin_lock_bh(&(sk->sk_lock.slock));
1231 if (sk->sk_backlog.tail)
1232 __release_sock(sk);
1233 sk->sk_lock.owner = NULL;
1234 if (waitqueue_active(&(sk->sk_lock.wq)))
1235 wake_up(&(sk->sk_lock.wq));
1236 spin_unlock_bh(&(sk->sk_lock.slock));
1237 }
1238 EXPORT_SYMBOL(release_sock);
1239
1240 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1241 {
1242 if (!sock_flag(sk, SOCK_TIMESTAMP))
1243 sock_enable_timestamp(sk);
1244 if (sk->sk_stamp.tv_sec == -1)
1245 return -ENOENT;
1246 if (sk->sk_stamp.tv_sec == 0)
1247 do_gettimeofday(&sk->sk_stamp);
1248 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1249 -EFAULT : 0;
1250 }
1251 EXPORT_SYMBOL(sock_get_timestamp);
1252
1253 void sock_enable_timestamp(struct sock *sk)
1254 {
1255 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1256 sock_set_flag(sk, SOCK_TIMESTAMP);
1257 net_enable_timestamp();
1258 }
1259 }
1260 EXPORT_SYMBOL(sock_enable_timestamp);
1261
1262 /*
1263 * Get a socket option on an socket.
1264 *
1265 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1266 * asynchronous errors should be reported by getsockopt. We assume
1267 * this means if you specify SO_ERROR (otherwise whats the point of it).
1268 */
1269 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1270 char __user *optval, int __user *optlen)
1271 {
1272 struct sock *sk = sock->sk;
1273
1274 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1275 }
1276
1277 EXPORT_SYMBOL(sock_common_getsockopt);
1278
1279 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1280 struct msghdr *msg, size_t size, int flags)
1281 {
1282 struct sock *sk = sock->sk;
1283 int addr_len = 0;
1284 int err;
1285
1286 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1287 flags & ~MSG_DONTWAIT, &addr_len);
1288 if (err >= 0)
1289 msg->msg_namelen = addr_len;
1290 return err;
1291 }
1292
1293 EXPORT_SYMBOL(sock_common_recvmsg);
1294
1295 /*
1296 * Set socket options on an inet socket.
1297 */
1298 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1299 char __user *optval, int optlen)
1300 {
1301 struct sock *sk = sock->sk;
1302
1303 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1304 }
1305
1306 EXPORT_SYMBOL(sock_common_setsockopt);
1307
1308 void sk_common_release(struct sock *sk)
1309 {
1310 if (sk->sk_prot->destroy)
1311 sk->sk_prot->destroy(sk);
1312
1313 /*
1314 * Observation: when sock_common_release is called, processes have
1315 * no access to socket. But net still has.
1316 * Step one, detach it from networking:
1317 *
1318 * A. Remove from hash tables.
1319 */
1320
1321 sk->sk_prot->unhash(sk);
1322
1323 /*
1324 * In this point socket cannot receive new packets, but it is possible
1325 * that some packets are in flight because some CPU runs receiver and
1326 * did hash table lookup before we unhashed socket. They will achieve
1327 * receive queue and will be purged by socket destructor.
1328 *
1329 * Also we still have packets pending on receive queue and probably,
1330 * our own packets waiting in device queues. sock_destroy will drain
1331 * receive queue, but transmitted packets will delay socket destruction
1332 * until the last reference will be released.
1333 */
1334
1335 sock_orphan(sk);
1336
1337 xfrm_sk_free_policy(sk);
1338
1339 #ifdef INET_REFCNT_DEBUG
1340 if (atomic_read(&sk->sk_refcnt) != 1)
1341 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1342 sk, atomic_read(&sk->sk_refcnt));
1343 #endif
1344 sock_put(sk);
1345 }
1346
1347 EXPORT_SYMBOL(sk_common_release);
1348
1349 int sk_alloc_slab(struct proto *prot, char *name)
1350 {
1351 prot->slab = kmem_cache_create(name,
1352 prot->slab_obj_size, 0,
1353 SLAB_HWCACHE_ALIGN, NULL, NULL);
1354
1355 return prot->slab != NULL ? 0 : -ENOBUFS;
1356 }
1357
1358 EXPORT_SYMBOL(sk_alloc_slab);
1359
1360 void sk_free_slab(struct proto *prot)
1361 {
1362 if (prot->slab != NULL) {
1363 kmem_cache_destroy(prot->slab);
1364 prot->slab = NULL;
1365 }
1366 }
1367
1368 EXPORT_SYMBOL(sk_free_slab);
1369
1370 EXPORT_SYMBOL(sk_alloc);
1371 EXPORT_SYMBOL(sk_free);
1372 EXPORT_SYMBOL(sk_send_sigurg);
1373 EXPORT_SYMBOL(sock_alloc_send_skb);
1374 EXPORT_SYMBOL(sock_init_data);
1375 EXPORT_SYMBOL(sock_kfree_s);
1376 EXPORT_SYMBOL(sock_kmalloc);
1377 EXPORT_SYMBOL(sock_no_accept);
1378 EXPORT_SYMBOL(sock_no_bind);
1379 EXPORT_SYMBOL(sock_no_connect);
1380 EXPORT_SYMBOL(sock_no_getname);
1381 EXPORT_SYMBOL(sock_no_getsockopt);
1382 EXPORT_SYMBOL(sock_no_ioctl);
1383 EXPORT_SYMBOL(sock_no_listen);
1384 EXPORT_SYMBOL(sock_no_mmap);
1385 EXPORT_SYMBOL(sock_no_poll);
1386 EXPORT_SYMBOL(sock_no_recvmsg);
1387 EXPORT_SYMBOL(sock_no_sendmsg);
1388 EXPORT_SYMBOL(sock_no_sendpage);
1389 EXPORT_SYMBOL(sock_no_setsockopt);
1390 EXPORT_SYMBOL(sock_no_shutdown);
1391 EXPORT_SYMBOL(sock_no_socketpair);
1392 EXPORT_SYMBOL(sock_rfree);
1393 EXPORT_SYMBOL(sock_setsockopt);
1394 EXPORT_SYMBOL(sock_wfree);
1395 EXPORT_SYMBOL(sock_wmalloc);
1396 EXPORT_SYMBOL(sock_i_uid);
1397 EXPORT_SYMBOL(sock_i_ino);
1398 #ifdef CONFIG_SYSCTL
1399 EXPORT_SYMBOL(sysctl_optmem_max);
1400 EXPORT_SYMBOL(sysctl_rmem_max);
1401 EXPORT_SYMBOL(sysctl_wmem_max);
1402 #endif
1403
|
This page was automatically generated by the
LXR engine.
|