?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.31.6
[linux/fpc-iii.git] / net / core / sock.c
blobdd120d8cc8e62d2b63b0b91dcdddace13bff4c1a
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_IEEE802154",
159 "sk_lock-AF_MAX"
160 };
161 static const char *af_family_slock_key_strings[AF_MAX+1] = {
162 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
163 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
164 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
165 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
166 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
167 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
168 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
169 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
170 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
171 "slock-27" , "slock-28" , "slock-AF_CAN" ,
172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
173 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
174 "slock-AF_IEEE802154",
175 "slock-AF_MAX"
176 };
177 static const char *af_family_clock_key_strings[AF_MAX+1] = {
178 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
179 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
180 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
181 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
182 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
183 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
184 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
185 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
186 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
187 "clock-27" , "clock-28" , "clock-AF_CAN" ,
188 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
189 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
190 "clock-AF_IEEE802154",
191 "clock-AF_MAX"
192 };
193
194 /*
195 * sk_callback_lock locking rules are per-address-family,
196 * so split the lock classes by using a per-AF key:
197 */
198 static struct lock_class_key af_callback_keys[AF_MAX];
199
200 /* Take into consideration the size of the struct sk_buff overhead in the
201 * determination of these values, since that is non-constant across
202 * platforms. This makes socket queueing behavior and performance
203 * not depend upon such differences.
204 */
205 #define _SK_MEM_PACKETS 256
206 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
207 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
208 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
209
210 /* Run time adjustable parameters. */
211 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
212 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
213 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
214 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
215
216 /* Maximal space eaten by iovec or ancilliary data plus some space */
217 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
218 EXPORT_SYMBOL(sysctl_optmem_max);
219
220 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
221 {
222 struct timeval tv;
223
224 if (optlen < sizeof(tv))
225 return -EINVAL;
226 if (copy_from_user(&tv, optval, sizeof(tv)))
227 return -EFAULT;
228 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
229 return -EDOM;
230
231 if (tv.tv_sec < 0) {
232 static int warned __read_mostly;
233
234 *timeo_p = 0;
235 if (warned < 10 && net_ratelimit()) {
236 warned++;
237 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
238 "tries to set negative timeout\n",
239 current->comm, task_pid_nr(current));
240 }
241 return 0;
242 }
243 *timeo_p = MAX_SCHEDULE_TIMEOUT;
244 if (tv.tv_sec == 0 && tv.tv_usec == 0)
245 return 0;
246 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
247 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
248 return 0;
249 }
250
251 static void sock_warn_obsolete_bsdism(const char *name)
252 {
253 static int warned;
254 static char warncomm[TASK_COMM_LEN];
255 if (strcmp(warncomm, current->comm) && warned < 5) {
256 strcpy(warncomm, current->comm);
257 printk(KERN_WARNING "process `%s' is using obsolete "
258 "%s SO_BSDCOMPAT\n", warncomm, name);
259 warned++;
260 }
261 }
262
263 static void sock_disable_timestamp(struct sock *sk, int flag)
264 {
265 if (sock_flag(sk, flag)) {
266 sock_reset_flag(sk, flag);
267 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
268 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
269 net_disable_timestamp();
270 }
271 }
272 }
273
274
275 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
276 {
277 int err = 0;
278 int skb_len;
279
280 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
281 number of warnings when compiling with -W --ANK
282 */
283 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
284 (unsigned)sk->sk_rcvbuf) {
285 err = -ENOMEM;
286 goto out;
287 }
288
289 err = sk_filter(sk, skb);
290 if (err)
291 goto out;
292
293 if (!sk_rmem_schedule(sk, skb->truesize)) {
294 err = -ENOBUFS;
295 goto out;
296 }
297
298 skb->dev = NULL;
299 skb_set_owner_r(skb, sk);
300
301 /* Cache the SKB length before we tack it onto the receive
302 * queue. Once it is added it no longer belongs to us and
303 * may be freed by other threads of control pulling packets
304 * from the queue.
305 */
306 skb_len = skb->len;
307
308 skb_queue_tail(&sk->sk_receive_queue, skb);
309
310 if (!sock_flag(sk, SOCK_DEAD))
311 sk->sk_data_ready(sk, skb_len);
312 out:
313 return err;
314 }
315 EXPORT_SYMBOL(sock_queue_rcv_skb);
316
317 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
318 {
319 int rc = NET_RX_SUCCESS;
320
321 if (sk_filter(sk, skb))
322 goto discard_and_relse;
323
324 skb->dev = NULL;
325
326 if (nested)
327 bh_lock_sock_nested(sk);
328 else
329 bh_lock_sock(sk);
330 if (!sock_owned_by_user(sk)) {
331 /*
332 * trylock + unlock semantics:
333 */
334 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
335
336 rc = sk_backlog_rcv(sk, skb);
337
338 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
339 } else
340 sk_add_backlog(sk, skb);
341 bh_unlock_sock(sk);
342 out:
343 sock_put(sk);
344 return rc;
345 discard_and_relse:
346 kfree_skb(skb);
347 goto out;
348 }
349 EXPORT_SYMBOL(sk_receive_skb);
350
351 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
352 {
353 struct dst_entry *dst = sk->sk_dst_cache;
354
355 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
356 sk->sk_dst_cache = NULL;
357 dst_release(dst);
358 return NULL;
359 }
360
361 return dst;
362 }
363 EXPORT_SYMBOL(__sk_dst_check);
364
365 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
366 {
367 struct dst_entry *dst = sk_dst_get(sk);
368
369 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
370 sk_dst_reset(sk);
371 dst_release(dst);
372 return NULL;
373 }
374
375 return dst;
376 }
377 EXPORT_SYMBOL(sk_dst_check);
378
379 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
380 {
381 int ret = -ENOPROTOOPT;
382 #ifdef CONFIG_NETDEVICES
383 struct net *net = sock_net(sk);
384 char devname[IFNAMSIZ];
385 int index;
386
387 /* Sorry... */
388 ret = -EPERM;
389 if (!capable(CAP_NET_RAW))
390 goto out;
391
392 ret = -EINVAL;
393 if (optlen < 0)
394 goto out;
395
396 /* Bind this socket to a particular device like "eth0",
397 * as specified in the passed interface name. If the
398 * name is "" or the option length is zero the socket
399 * is not bound.
400 */
401 if (optlen > IFNAMSIZ - 1)
402 optlen = IFNAMSIZ - 1;
403 memset(devname, 0, sizeof(devname));
404
405 ret = -EFAULT;
406 if (copy_from_user(devname, optval, optlen))
407 goto out;
408
409 if (devname[0] == '\0') {
410 index = 0;
411 } else {
412 struct net_device *dev = dev_get_by_name(net, devname);
413
414 ret = -ENODEV;
415 if (!dev)
416 goto out;
417
418 index = dev->ifindex;
419 dev_put(dev);
420 }
421
422 lock_sock(sk);
423 sk->sk_bound_dev_if = index;
424 sk_dst_reset(sk);
425 release_sock(sk);
426
427 ret = 0;
428
429 out:
430 #endif
431
432 return ret;
433 }
434
435 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
436 {
437 if (valbool)
438 sock_set_flag(sk, bit);
439 else
440 sock_reset_flag(sk, bit);
441 }
442
443 /*
444 * This is meant for all protocols to use and covers goings on
445 * at the socket level. Everything here is generic.
446 */
447
448 int sock_setsockopt(struct socket *sock, int level, int optname,
449 char __user *optval, int optlen)
450 {
451 struct sock *sk = sock->sk;
452 int val;
453 int valbool;
454 struct linger ling;
455 int ret = 0;
456
457 /*
458 * Options without arguments
459 */
460
461 if (optname == SO_BINDTODEVICE)
462 return sock_bindtodevice(sk, optval, optlen);
463
464 if (optlen < sizeof(int))
465 return -EINVAL;
466
467 if (get_user(val, (int __user *)optval))
468 return -EFAULT;
469
470 valbool = val ? 1 : 0;
471
472 lock_sock(sk);
473
474 switch (optname) {
475 case SO_DEBUG:
476 if (val && !capable(CAP_NET_ADMIN))
477 ret = -EACCES;
478 else
479 sock_valbool_flag(sk, SOCK_DBG, valbool);
480 break;
481 case SO_REUSEADDR:
482 sk->sk_reuse = valbool;
483 break;
484 case SO_TYPE:
485 case SO_ERROR:
486 ret = -ENOPROTOOPT;
487 break;
488 case SO_DONTROUTE:
489 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
490 break;
491 case SO_BROADCAST:
492 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
493 break;
494 case SO_SNDBUF:
495 /* Don't error on this BSD doesn't and if you think
496 about it this is right. Otherwise apps have to
497 play 'guess the biggest size' games. RCVBUF/SNDBUF
498 are treated in BSD as hints */
499
500 if (val > sysctl_wmem_max)
501 val = sysctl_wmem_max;
502 set_sndbuf:
503 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
504 if ((val * 2) < SOCK_MIN_SNDBUF)
505 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
506 else
507 sk->sk_sndbuf = val * 2;
508
509 /*
510 * Wake up sending tasks if we
511 * upped the value.
512 */
513 sk->sk_write_space(sk);
514 break;
515
516 case SO_SNDBUFFORCE:
517 if (!capable(CAP_NET_ADMIN)) {
518 ret = -EPERM;
519 break;
520 }
521 goto set_sndbuf;
522
523 case SO_RCVBUF:
524 /* Don't error on this BSD doesn't and if you think
525 about it this is right. Otherwise apps have to
526 play 'guess the biggest size' games. RCVBUF/SNDBUF
527 are treated in BSD as hints */
528
529 if (val > sysctl_rmem_max)
530 val = sysctl_rmem_max;
531 set_rcvbuf:
532 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
533 /*
534 * We double it on the way in to account for
535 * "struct sk_buff" etc. overhead. Applications
536 * assume that the SO_RCVBUF setting they make will
537 * allow that much actual data to be received on that
538 * socket.
539 *
540 * Applications are unaware that "struct sk_buff" and
541 * other overheads allocate from the receive buffer
542 * during socket buffer allocation.
543 *
544 * And after considering the possible alternatives,
545 * returning the value we actually used in getsockopt
546 * is the most desirable behavior.
547 */
548 if ((val * 2) < SOCK_MIN_RCVBUF)
549 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
550 else
551 sk->sk_rcvbuf = val * 2;
552 break;
553
554 case SO_RCVBUFFORCE:
555 if (!capable(CAP_NET_ADMIN)) {
556 ret = -EPERM;
557 break;
558 }
559 goto set_rcvbuf;
560
561 case SO_KEEPALIVE:
562 #ifdef CONFIG_INET
563 if (sk->sk_protocol == IPPROTO_TCP)
564 tcp_set_keepalive(sk, valbool);
565 #endif
566 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
567 break;
568
569 case SO_OOBINLINE:
570 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
571 break;
572
573 case SO_NO_CHECK:
574 sk->sk_no_check = valbool;
575 break;
576
577 case SO_PRIORITY:
578 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
579 sk->sk_priority = val;
580 else
581 ret = -EPERM;
582 break;
583
584 case SO_LINGER:
585 if (optlen < sizeof(ling)) {
586 ret = -EINVAL; /* 1003.1g */
587 break;
588 }
589 if (copy_from_user(&ling, optval, sizeof(ling))) {
590 ret = -EFAULT;
591 break;
592 }
593 if (!ling.l_onoff)
594 sock_reset_flag(sk, SOCK_LINGER);
595 else {
596 #if (BITS_PER_LONG == 32)
597 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
598 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
599 else
600 #endif
601 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
602 sock_set_flag(sk, SOCK_LINGER);
603 }
604 break;
605
606 case SO_BSDCOMPAT:
607 sock_warn_obsolete_bsdism("setsockopt");
608 break;
609
610 case SO_PASSCRED:
611 if (valbool)
612 set_bit(SOCK_PASSCRED, &sock->flags);
613 else
614 clear_bit(SOCK_PASSCRED, &sock->flags);
615 break;
616
617 case SO_TIMESTAMP:
618 case SO_TIMESTAMPNS:
619 if (valbool) {
620 if (optname == SO_TIMESTAMP)
621 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
622 else
623 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
624 sock_set_flag(sk, SOCK_RCVTSTAMP);
625 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
626 } else {
627 sock_reset_flag(sk, SOCK_RCVTSTAMP);
628 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
629 }
630 break;
631
632 case SO_TIMESTAMPING:
633 if (val & ~SOF_TIMESTAMPING_MASK) {
634 ret = -EINVAL;
635 break;
636 }
637 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
638 val & SOF_TIMESTAMPING_TX_HARDWARE);
639 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
640 val & SOF_TIMESTAMPING_TX_SOFTWARE);
641 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
642 val & SOF_TIMESTAMPING_RX_HARDWARE);
643 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
644 sock_enable_timestamp(sk,
645 SOCK_TIMESTAMPING_RX_SOFTWARE);
646 else
647 sock_disable_timestamp(sk,
648 SOCK_TIMESTAMPING_RX_SOFTWARE);
649 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
650 val & SOF_TIMESTAMPING_SOFTWARE);
651 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
652 val & SOF_TIMESTAMPING_SYS_HARDWARE);
653 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
654 val & SOF_TIMESTAMPING_RAW_HARDWARE);
655 break;
656
657 case SO_RCVLOWAT:
658 if (val < 0)
659 val = INT_MAX;
660 sk->sk_rcvlowat = val ? : 1;
661 break;
662
663 case SO_RCVTIMEO:
664 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
665 break;
666
667 case SO_SNDTIMEO:
668 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
669 break;
670
671 case SO_ATTACH_FILTER:
672 ret = -EINVAL;
673 if (optlen == sizeof(struct sock_fprog)) {
674 struct sock_fprog fprog;
675
676 ret = -EFAULT;
677 if (copy_from_user(&fprog, optval, sizeof(fprog)))
678 break;
679
680 ret = sk_attach_filter(&fprog, sk);
681 }
682 break;
683
684 case SO_DETACH_FILTER:
685 ret = sk_detach_filter(sk);
686 break;
687
688 case SO_PASSSEC:
689 if (valbool)
690 set_bit(SOCK_PASSSEC, &sock->flags);
691 else
692 clear_bit(SOCK_PASSSEC, &sock->flags);
693 break;
694 case SO_MARK:
695 if (!capable(CAP_NET_ADMIN))
696 ret = -EPERM;
697 else
698 sk->sk_mark = val;
699 break;
700
701 /* We implement the SO_SNDLOWAT etc to
702 not be settable (1003.1g 5.3) */
703 default:
704 ret = -ENOPROTOOPT;
705 break;
706 }
707 release_sock(sk);
708 return ret;
709 }
710 EXPORT_SYMBOL(sock_setsockopt);
711
712
713 int sock_getsockopt(struct socket *sock, int level, int optname,
714 char __user *optval, int __user *optlen)
715 {
716 struct sock *sk = sock->sk;
717
718 union {
719 int val;
720 struct linger ling;
721 struct timeval tm;
722 } v;
723
724 unsigned int lv = sizeof(int);
725 int len;
726
727 if (get_user(len, optlen))
728 return -EFAULT;
729 if (len < 0)
730 return -EINVAL;
731
732 memset(&v, 0, sizeof(v));
733
734 switch (optname) {
735 case SO_DEBUG:
736 v.val = sock_flag(sk, SOCK_DBG);
737 break;
738
739 case SO_DONTROUTE:
740 v.val = sock_flag(sk, SOCK_LOCALROUTE);
741 break;
742
743 case SO_BROADCAST:
744 v.val = !!sock_flag(sk, SOCK_BROADCAST);
745 break;
746
747 case SO_SNDBUF:
748 v.val = sk->sk_sndbuf;
749 break;
750
751 case SO_RCVBUF:
752 v.val = sk->sk_rcvbuf;
753 break;
754
755 case SO_REUSEADDR:
756 v.val = sk->sk_reuse;
757 break;
758
759 case SO_KEEPALIVE:
760 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
761 break;
762
763 case SO_TYPE:
764 v.val = sk->sk_type;
765 break;
766
767 case SO_ERROR:
768 v.val = -sock_error(sk);
769 if (v.val == 0)
770 v.val = xchg(&sk->sk_err_soft, 0);
771 break;
772
773 case SO_OOBINLINE:
774 v.val = !!sock_flag(sk, SOCK_URGINLINE);
775 break;
776
777 case SO_NO_CHECK:
778 v.val = sk->sk_no_check;
779 break;
780
781 case SO_PRIORITY:
782 v.val = sk->sk_priority;
783 break;
784
785 case SO_LINGER:
786 lv = sizeof(v.ling);
787 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
788 v.ling.l_linger = sk->sk_lingertime / HZ;
789 break;
790
791 case SO_BSDCOMPAT:
792 sock_warn_obsolete_bsdism("getsockopt");
793 break;
794
795 case SO_TIMESTAMP:
796 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
797 !sock_flag(sk, SOCK_RCVTSTAMPNS);
798 break;
799
800 case SO_TIMESTAMPNS:
801 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
802 break;
803
804 case SO_TIMESTAMPING:
805 v.val = 0;
806 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
807 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
808 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
809 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
810 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
811 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
812 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
813 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
814 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
815 v.val |= SOF_TIMESTAMPING_SOFTWARE;
816 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
817 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
818 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
819 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
820 break;
821
822 case SO_RCVTIMEO:
823 lv = sizeof(struct timeval);
824 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
825 v.tm.tv_sec = 0;
826 v.tm.tv_usec = 0;
827 } else {
828 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
829 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
830 }
831 break;
832
833 case SO_SNDTIMEO:
834 lv = sizeof(struct timeval);
835 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
836 v.tm.tv_sec = 0;
837 v.tm.tv_usec = 0;
838 } else {
839 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
840 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
841 }
842 break;
843
844 case SO_RCVLOWAT:
845 v.val = sk->sk_rcvlowat;
846 break;
847
848 case SO_SNDLOWAT:
849 v.val = 1;
850 break;
851
852 case SO_PASSCRED:
853 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
854 break;
855
856 case SO_PEERCRED:
857 if (len > sizeof(sk->sk_peercred))
858 len = sizeof(sk->sk_peercred);
859 if (copy_to_user(optval, &sk->sk_peercred, len))
860 return -EFAULT;
861 goto lenout;
862
863 case SO_PEERNAME:
864 {
865 char address[128];
866
867 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
868 return -ENOTCONN;
869 if (lv < len)
870 return -EINVAL;
871 if (copy_to_user(optval, address, len))
872 return -EFAULT;
873 goto lenout;
874 }
875
876 /* Dubious BSD thing... Probably nobody even uses it, but
877 * the UNIX standard wants it for whatever reason... -DaveM
878 */
879 case SO_ACCEPTCONN:
880 v.val = sk->sk_state == TCP_LISTEN;
881 break;
882
883 case SO_PASSSEC:
884 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
885 break;
886
887 case SO_PEERSEC:
888 return security_socket_getpeersec_stream(sock, optval, optlen, len);
889
890 case SO_MARK:
891 v.val = sk->sk_mark;
892 break;
893
894 default:
895 return -ENOPROTOOPT;
896 }
897
898 if (len > lv)
899 len = lv;
900 if (copy_to_user(optval, &v, len))
901 return -EFAULT;
902 lenout:
903 if (put_user(len, optlen))
904 return -EFAULT;
905 return 0;
906 }
907
908 /*
909 * Initialize an sk_lock.
910 *
911 * (We also register the sk_lock with the lock validator.)
912 */
913 static inline void sock_lock_init(struct sock *sk)
914 {
915 sock_lock_init_class_and_name(sk,
916 af_family_slock_key_strings[sk->sk_family],
917 af_family_slock_keys + sk->sk_family,
918 af_family_key_strings[sk->sk_family],
919 af_family_keys + sk->sk_family);
920 }
921
922 /*
923 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
924 * even temporarly, because of RCU lookups. sk_node should also be left as is.
925 */
926 static void sock_copy(struct sock *nsk, const struct sock *osk)
927 {
928 #ifdef CONFIG_SECURITY_NETWORK
929 void *sptr = nsk->sk_security;
930 #endif
931 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
932 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt));
933 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
934 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
935 #ifdef CONFIG_SECURITY_NETWORK
936 nsk->sk_security = sptr;
937 security_sk_clone(osk, nsk);
938 #endif
939 }
940
941 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
942 int family)
943 {
944 struct sock *sk;
945 struct kmem_cache *slab;
946
947 slab = prot->slab;
948 if (slab != NULL) {
949 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
950 if (!sk)
951 return sk;
952 if (priority & __GFP_ZERO) {
953 /*
954 * caches using SLAB_DESTROY_BY_RCU should let
955 * sk_node.next un-modified. Special care is taken
956 * when initializing object to zero.
957 */
958 if (offsetof(struct sock, sk_node.next) != 0)
959 memset(sk, 0, offsetof(struct sock, sk_node.next));
960 memset(&sk->sk_node.pprev, 0,
961 prot->obj_size - offsetof(struct sock,
962 sk_node.pprev));
963 }
964 }
965 else
966 sk = kmalloc(prot->obj_size, priority);
967
968 if (sk != NULL) {
969 kmemcheck_annotate_bitfield(sk, flags);
970
971 if (security_sk_alloc(sk, family, priority))
972 goto out_free;
973
974 if (!try_module_get(prot->owner))
975 goto out_free_sec;
976 }
977
978 return sk;
979
980 out_free_sec:
981 security_sk_free(sk);
982 out_free:
983 if (slab != NULL)
984 kmem_cache_free(slab, sk);
985 else
986 kfree(sk);
987 return NULL;
988 }
989
990 static void sk_prot_free(struct proto *prot, struct sock *sk)
991 {
992 struct kmem_cache *slab;
993 struct module *owner;
994
995 owner = prot->owner;
996 slab = prot->slab;
997
998 security_sk_free(sk);
999 if (slab != NULL)
1000 kmem_cache_free(slab, sk);
1001 else
1002 kfree(sk);
1003 module_put(owner);
1004 }
1005
1006 /**
1007 * sk_alloc - All socket objects are allocated here
1008 * @net: the applicable net namespace
1009 * @family: protocol family
1010 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1011 * @prot: struct proto associated with this new sock instance
1012 */
1013 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1014 struct proto *prot)
1015 {
1016 struct sock *sk;
1017
1018 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1019 if (sk) {
1020 sk->sk_family = family;
1021 /*
1022 * See comment in struct sock definition to understand
1023 * why we need sk_prot_creator -acme
1024 */
1025 sk->sk_prot = sk->sk_prot_creator = prot;
1026 sock_lock_init(sk);
1027 sock_net_set(sk, get_net(net));
1028 atomic_set(&sk->sk_wmem_alloc, 1);
1029 }
1030
1031 return sk;
1032 }
1033 EXPORT_SYMBOL(sk_alloc);
1034
1035 static void __sk_free(struct sock *sk)
1036 {
1037 struct sk_filter *filter;
1038
1039 if (sk->sk_destruct)
1040 sk->sk_destruct(sk);
1041
1042 filter = rcu_dereference(sk->sk_filter);
1043 if (filter) {
1044 sk_filter_uncharge(sk, filter);
1045 rcu_assign_pointer(sk->sk_filter, NULL);
1046 }
1047
1048 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1049 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1050
1051 if (atomic_read(&sk->sk_omem_alloc))
1052 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1053 __func__, atomic_read(&sk->sk_omem_alloc));
1054
1055 put_net(sock_net(sk));
1056 sk_prot_free(sk->sk_prot_creator, sk);
1057 }
1058
1059 void sk_free(struct sock *sk)
1060 {
1061 /*
1062 * We substract one from sk_wmem_alloc and can know if
1063 * some packets are still in some tx queue.
1064 * If not null, sock_wfree() will call __sk_free(sk) later
1065 */
1066 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1067 __sk_free(sk);
1068 }
1069 EXPORT_SYMBOL(sk_free);
1070
1071 /*
1072 * Last sock_put should drop referrence to sk->sk_net. It has already
1073 * been dropped in sk_change_net. Taking referrence to stopping namespace
1074 * is not an option.
1075 * Take referrence to a socket to remove it from hash _alive_ and after that
1076 * destroy it in the context of init_net.
1077 */
1078 void sk_release_kernel(struct sock *sk)
1079 {
1080 if (sk == NULL || sk->sk_socket == NULL)
1081 return;
1082
1083 sock_hold(sk);
1084 sock_release(sk->sk_socket);
1085 release_net(sock_net(sk));
1086 sock_net_set(sk, get_net(&init_net));
1087 sock_put(sk);
1088 }
1089 EXPORT_SYMBOL(sk_release_kernel);
1090
1091 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1092 {
1093 struct sock *newsk;
1094
1095 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1096 if (newsk != NULL) {
1097 struct sk_filter *filter;
1098
1099 sock_copy(newsk, sk);
1100
1101 /* SANITY */
1102 get_net(sock_net(newsk));
1103 sk_node_init(&newsk->sk_node);
1104 sock_lock_init(newsk);
1105 bh_lock_sock(newsk);
1106 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1107
1108 atomic_set(&newsk->sk_rmem_alloc, 0);
1109 /*
1110 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1111 */
1112 atomic_set(&newsk->sk_wmem_alloc, 1);
1113 atomic_set(&newsk->sk_omem_alloc, 0);
1114 skb_queue_head_init(&newsk->sk_receive_queue);
1115 skb_queue_head_init(&newsk->sk_write_queue);
1116 #ifdef CONFIG_NET_DMA
1117 skb_queue_head_init(&newsk->sk_async_wait_queue);
1118 #endif
1119
1120 rwlock_init(&newsk->sk_dst_lock);
1121 rwlock_init(&newsk->sk_callback_lock);
1122 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1123 af_callback_keys + newsk->sk_family,
1124 af_family_clock_key_strings[newsk->sk_family]);
1125
1126 newsk->sk_dst_cache = NULL;
1127 newsk->sk_wmem_queued = 0;
1128 newsk->sk_forward_alloc = 0;
1129 newsk->sk_send_head = NULL;
1130 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1131
1132 sock_reset_flag(newsk, SOCK_DONE);
1133 skb_queue_head_init(&newsk->sk_error_queue);
1134
1135 filter = newsk->sk_filter;
1136 if (filter != NULL)
1137 sk_filter_charge(newsk, filter);
1138
1139 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1140 /* It is still raw copy of parent, so invalidate
1141 * destructor and make plain sk_free() */
1142 newsk->sk_destruct = NULL;
1143 sk_free(newsk);
1144 newsk = NULL;
1145 goto out;
1146 }
1147
1148 newsk->sk_err = 0;
1149 newsk->sk_priority = 0;
1150 /*
1151 * Before updating sk_refcnt, we must commit prior changes to memory
1152 * (Documentation/RCU/rculist_nulls.txt for details)
1153 */
1154 smp_wmb();
1155 atomic_set(&newsk->sk_refcnt, 2);
1156
1157 /*
1158 * Increment the counter in the same struct proto as the master
1159 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1160 * is the same as sk->sk_prot->socks, as this field was copied
1161 * with memcpy).
1162 *
1163 * This _changes_ the previous behaviour, where
1164 * tcp_create_openreq_child always was incrementing the
1165 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1166 * to be taken into account in all callers. -acme
1167 */
1168 sk_refcnt_debug_inc(newsk);
1169 sk_set_socket(newsk, NULL);
1170 newsk->sk_sleep = NULL;
1171
1172 if (newsk->sk_prot->sockets_allocated)
1173 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1174 }
1175 out:
1176 return newsk;
1177 }
1178 EXPORT_SYMBOL_GPL(sk_clone);
1179
1180 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1181 {
1182 __sk_dst_set(sk, dst);
1183 sk->sk_route_caps = dst->dev->features;
1184 if (sk->sk_route_caps & NETIF_F_GSO)
1185 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1186 if (sk_can_gso(sk)) {
1187 if (dst->header_len) {
1188 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1189 } else {
1190 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1191 sk->sk_gso_max_size = dst->dev->gso_max_size;
1192 }
1193 }
1194 }
1195 EXPORT_SYMBOL_GPL(sk_setup_caps);
1196
1197 void __init sk_init(void)
1198 {
1199 if (num_physpages <= 4096) {
1200 sysctl_wmem_max = 32767;
1201 sysctl_rmem_max = 32767;
1202 sysctl_wmem_default = 32767;
1203 sysctl_rmem_default = 32767;
1204 } else if (num_physpages >= 131072) {
1205 sysctl_wmem_max = 131071;
1206 sysctl_rmem_max = 131071;
1207 }
1208 }
1209
1210 /*
1211 * Simple resource managers for sockets.
1212 */
1213
1214
1215 /*
1216 * Write buffer destructor automatically called from kfree_skb.
1217 */
1218 void sock_wfree(struct sk_buff *skb)
1219 {
1220 struct sock *sk = skb->sk;
1221 unsigned int len = skb->truesize;
1222
1223 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1224 /*
1225 * Keep a reference on sk_wmem_alloc, this will be released
1226 * after sk_write_space() call
1227 */
1228 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1229 sk->sk_write_space(sk);
1230 len = 1;
1231 }
1232 /*
1233 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1234 * could not do because of in-flight packets
1235 */
1236 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1237 __sk_free(sk);
1238 }
1239 EXPORT_SYMBOL(sock_wfree);
1240
1241 /*
1242 * Read buffer destructor automatically called from kfree_skb.
1243 */
1244 void sock_rfree(struct sk_buff *skb)
1245 {
1246 struct sock *sk = skb->sk;
1247
1248 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1249 sk_mem_uncharge(skb->sk, skb->truesize);
1250 }
1251 EXPORT_SYMBOL(sock_rfree);
1252
1253
1254 int sock_i_uid(struct sock *sk)
1255 {
1256 int uid;
1257
1258 read_lock(&sk->sk_callback_lock);
1259 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1260 read_unlock(&sk->sk_callback_lock);
1261 return uid;
1262 }
1263 EXPORT_SYMBOL(sock_i_uid);
1264
1265 unsigned long sock_i_ino(struct sock *sk)
1266 {
1267 unsigned long ino;
1268
1269 read_lock(&sk->sk_callback_lock);
1270 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1271 read_unlock(&sk->sk_callback_lock);
1272 return ino;
1273 }
1274 EXPORT_SYMBOL(sock_i_ino);
1275
1276 /*
1277 * Allocate a skb from the socket's send buffer.
1278 */
1279 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1280 gfp_t priority)
1281 {
1282 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1283 struct sk_buff *skb = alloc_skb(size, priority);
1284 if (skb) {
1285 skb_set_owner_w(skb, sk);
1286 return skb;
1287 }
1288 }
1289 return NULL;
1290 }
1291 EXPORT_SYMBOL(sock_wmalloc);
1292
1293 /*
1294 * Allocate a skb from the socket's receive buffer.
1295 */
1296 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1297 gfp_t priority)
1298 {
1299 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1300 struct sk_buff *skb = alloc_skb(size, priority);
1301 if (skb) {
1302 skb_set_owner_r(skb, sk);
1303 return skb;
1304 }
1305 }
1306 return NULL;
1307 }
1308
1309 /*
1310 * Allocate a memory block from the socket's option memory buffer.
1311 */
1312 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1313 {
1314 if ((unsigned)size <= sysctl_optmem_max &&
1315 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1316 void *mem;
1317 /* First do the add, to avoid the race if kmalloc
1318 * might sleep.
1319 */
1320 atomic_add(size, &sk->sk_omem_alloc);
1321 mem = kmalloc(size, priority);
1322 if (mem)
1323 return mem;
1324 atomic_sub(size, &sk->sk_omem_alloc);
1325 }
1326 return NULL;
1327 }
1328 EXPORT_SYMBOL(sock_kmalloc);
1329
1330 /*
1331 * Free an option memory block.
1332 */
1333 void sock_kfree_s(struct sock *sk, void *mem, int size)
1334 {
1335 kfree(mem);
1336 atomic_sub(size, &sk->sk_omem_alloc);
1337 }
1338 EXPORT_SYMBOL(sock_kfree_s);
1339
1340 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1341 I think, these locks should be removed for datagram sockets.
1342 */
1343 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1344 {
1345 DEFINE_WAIT(wait);
1346
1347 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1348 for (;;) {
1349 if (!timeo)
1350 break;
1351 if (signal_pending(current))
1352 break;
1353 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1354 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1355 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1356 break;
1357 if (sk->sk_shutdown & SEND_SHUTDOWN)
1358 break;
1359 if (sk->sk_err)
1360 break;
1361 timeo = schedule_timeout(timeo);
1362 }
1363 finish_wait(sk->sk_sleep, &wait);
1364 return timeo;
1365 }
1366
1367
1368 /*
1369 * Generic send/receive buffer handlers
1370 */
1371
1372 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1373 unsigned long data_len, int noblock,
1374 int *errcode)
1375 {
1376 struct sk_buff *skb;
1377 gfp_t gfp_mask;
1378 long timeo;
1379 int err;
1380
1381 gfp_mask = sk->sk_allocation;
1382 if (gfp_mask & __GFP_WAIT)
1383 gfp_mask |= __GFP_REPEAT;
1384
1385 timeo = sock_sndtimeo(sk, noblock);
1386 while (1) {
1387 err = sock_error(sk);
1388 if (err != 0)
1389 goto failure;
1390
1391 err = -EPIPE;
1392 if (sk->sk_shutdown & SEND_SHUTDOWN)
1393 goto failure;
1394
1395 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1396 skb = alloc_skb(header_len, gfp_mask);
1397 if (skb) {
1398 int npages;
1399 int i;
1400
1401 /* No pages, we're done... */
1402 if (!data_len)
1403 break;
1404
1405 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1406 skb->truesize += data_len;
1407 skb_shinfo(skb)->nr_frags = npages;
1408 for (i = 0; i < npages; i++) {
1409 struct page *page;
1410 skb_frag_t *frag;
1411
1412 page = alloc_pages(sk->sk_allocation, 0);
1413 if (!page) {
1414 err = -ENOBUFS;
1415 skb_shinfo(skb)->nr_frags = i;
1416 kfree_skb(skb);
1417 goto failure;
1418 }
1419
1420 frag = &skb_shinfo(skb)->frags[i];
1421 frag->page = page;
1422 frag->page_offset = 0;
1423 frag->size = (data_len >= PAGE_SIZE ?
1424 PAGE_SIZE :
1425 data_len);
1426 data_len -= PAGE_SIZE;
1427 }
1428
1429 /* Full success... */
1430 break;
1431 }
1432 err = -ENOBUFS;
1433 goto failure;
1434 }
1435 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1436 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1437 err = -EAGAIN;
1438 if (!timeo)
1439 goto failure;
1440 if (signal_pending(current))
1441 goto interrupted;
1442 timeo = sock_wait_for_wmem(sk, timeo);
1443 }
1444
1445 skb_set_owner_w(skb, sk);
1446 return skb;
1447
1448 interrupted:
1449 err = sock_intr_errno(timeo);
1450 failure:
1451 *errcode = err;
1452 return NULL;
1453 }
1454 EXPORT_SYMBOL(sock_alloc_send_pskb);
1455
1456 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1457 int noblock, int *errcode)
1458 {
1459 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1460 }
1461 EXPORT_SYMBOL(sock_alloc_send_skb);
1462
1463 static void __lock_sock(struct sock *sk)
1464 {
1465 DEFINE_WAIT(wait);
1466
1467 for (;;) {
1468 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1469 TASK_UNINTERRUPTIBLE);
1470 spin_unlock_bh(&sk->sk_lock.slock);
1471 schedule();
1472 spin_lock_bh(&sk->sk_lock.slock);
1473 if (!sock_owned_by_user(sk))
1474 break;
1475 }
1476 finish_wait(&sk->sk_lock.wq, &wait);
1477 }
1478
1479 static void __release_sock(struct sock *sk)
1480 {
1481 struct sk_buff *skb = sk->sk_backlog.head;
1482
1483 do {
1484 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1485 bh_unlock_sock(sk);
1486
1487 do {
1488 struct sk_buff *next = skb->next;
1489
1490 skb->next = NULL;
1491 sk_backlog_rcv(sk, skb);
1492
1493 /*
1494 * We are in process context here with softirqs
1495 * disabled, use cond_resched_softirq() to preempt.
1496 * This is safe to do because we've taken the backlog
1497 * queue private:
1498 */
1499 cond_resched_softirq();
1500
1501 skb = next;
1502 } while (skb != NULL);
1503
1504 bh_lock_sock(sk);
1505 } while ((skb = sk->sk_backlog.head) != NULL);
1506 }
1507
1508 /**
1509 * sk_wait_data - wait for data to arrive at sk_receive_queue
1510 * @sk: sock to wait on
1511 * @timeo: for how long
1512 *
1513 * Now socket state including sk->sk_err is changed only under lock,
1514 * hence we may omit checks after joining wait queue.
1515 * We check receive queue before schedule() only as optimization;
1516 * it is very likely that release_sock() added new data.
1517 */
1518 int sk_wait_data(struct sock *sk, long *timeo)
1519 {
1520 int rc;
1521 DEFINE_WAIT(wait);
1522
1523 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1524 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1525 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1526 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1527 finish_wait(sk->sk_sleep, &wait);
1528 return rc;
1529 }
1530 EXPORT_SYMBOL(sk_wait_data);
1531
1532 /**
1533 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1534 * @sk: socket
1535 * @size: memory size to allocate
1536 * @kind: allocation type
1537 *
1538 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1539 * rmem allocation. This function assumes that protocols which have
1540 * memory_pressure use sk_wmem_queued as write buffer accounting.
1541 */
1542 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1543 {
1544 struct proto *prot = sk->sk_prot;
1545 int amt = sk_mem_pages(size);
1546 int allocated;
1547
1548 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1549 allocated = atomic_add_return(amt, prot->memory_allocated);
1550
1551 /* Under limit. */
1552 if (allocated <= prot->sysctl_mem[0]) {
1553 if (prot->memory_pressure && *prot->memory_pressure)
1554 *prot->memory_pressure = 0;
1555 return 1;
1556 }
1557
1558 /* Under pressure. */
1559 if (allocated > prot->sysctl_mem[1])
1560 if (prot->enter_memory_pressure)
1561 prot->enter_memory_pressure(sk);
1562
1563 /* Over hard limit. */
1564 if (allocated > prot->sysctl_mem[2])
1565 goto suppress_allocation;
1566
1567 /* guarantee minimum buffer size under pressure */
1568 if (kind == SK_MEM_RECV) {
1569 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1570 return 1;
1571 } else { /* SK_MEM_SEND */
1572 if (sk->sk_type == SOCK_STREAM) {
1573 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1574 return 1;
1575 } else if (atomic_read(&sk->sk_wmem_alloc) <
1576 prot->sysctl_wmem[0])
1577 return 1;
1578 }
1579
1580 if (prot->memory_pressure) {
1581 int alloc;
1582
1583 if (!*prot->memory_pressure)
1584 return 1;
1585 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1586 if (prot->sysctl_mem[2] > alloc *
1587 sk_mem_pages(sk->sk_wmem_queued +
1588 atomic_read(&sk->sk_rmem_alloc) +
1589 sk->sk_forward_alloc))
1590 return 1;
1591 }
1592
1593 suppress_allocation:
1594
1595 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1596 sk_stream_moderate_sndbuf(sk);
1597
1598 /* Fail only if socket is _under_ its sndbuf.
1599 * In this case we cannot block, so that we have to fail.
1600 */
1601 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1602 return 1;
1603 }
1604
1605 /* Alas. Undo changes. */
1606 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1607 atomic_sub(amt, prot->memory_allocated);
1608 return 0;
1609 }
1610 EXPORT_SYMBOL(__sk_mem_schedule);
1611
1612 /**
1613 * __sk_reclaim - reclaim memory_allocated
1614 * @sk: socket
1615 */
1616 void __sk_mem_reclaim(struct sock *sk)
1617 {
1618 struct proto *prot = sk->sk_prot;
1619
1620 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1621 prot->memory_allocated);
1622 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1623
1624 if (prot->memory_pressure && *prot->memory_pressure &&
1625 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1626 *prot->memory_pressure = 0;
1627 }
1628 EXPORT_SYMBOL(__sk_mem_reclaim);
1629
1630
1631 /*
1632 * Set of default routines for initialising struct proto_ops when
1633 * the protocol does not support a particular function. In certain
1634 * cases where it makes no sense for a protocol to have a "do nothing"
1635 * function, some default processing is provided.
1636 */
1637
1638 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1639 {
1640 return -EOPNOTSUPP;
1641 }
1642 EXPORT_SYMBOL(sock_no_bind);
1643
1644 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1645 int len, int flags)
1646 {
1647 return -EOPNOTSUPP;
1648 }
1649 EXPORT_SYMBOL(sock_no_connect);
1650
1651 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1652 {
1653 return -EOPNOTSUPP;
1654 }
1655 EXPORT_SYMBOL(sock_no_socketpair);
1656
1657 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1658 {
1659 return -EOPNOTSUPP;
1660 }
1661 EXPORT_SYMBOL(sock_no_accept);
1662
1663 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1664 int *len, int peer)
1665 {
1666 return -EOPNOTSUPP;
1667 }
1668 EXPORT_SYMBOL(sock_no_getname);
1669
1670 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1671 {
1672 return 0;
1673 }
1674 EXPORT_SYMBOL(sock_no_poll);
1675
1676 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1677 {
1678 return -EOPNOTSUPP;
1679 }
1680 EXPORT_SYMBOL(sock_no_ioctl);
1681
1682 int sock_no_listen(struct socket *sock, int backlog)
1683 {
1684 return -EOPNOTSUPP;
1685 }
1686 EXPORT_SYMBOL(sock_no_listen);
1687
1688 int sock_no_shutdown(struct socket *sock, int how)
1689 {
1690 return -EOPNOTSUPP;
1691 }
1692 EXPORT_SYMBOL(sock_no_shutdown);
1693
1694 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1695 char __user *optval, int optlen)
1696 {
1697 return -EOPNOTSUPP;
1698 }
1699 EXPORT_SYMBOL(sock_no_setsockopt);
1700
1701 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1702 char __user *optval, int __user *optlen)
1703 {
1704 return -EOPNOTSUPP;
1705 }
1706 EXPORT_SYMBOL(sock_no_getsockopt);
1707
1708 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1709 size_t len)
1710 {
1711 return -EOPNOTSUPP;
1712 }
1713 EXPORT_SYMBOL(sock_no_sendmsg);
1714
1715 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1716 size_t len, int flags)
1717 {
1718 return -EOPNOTSUPP;
1719 }
1720 EXPORT_SYMBOL(sock_no_recvmsg);
1721
1722 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1723 {
1724 /* Mirror missing mmap method error code */
1725 return -ENODEV;
1726 }
1727 EXPORT_SYMBOL(sock_no_mmap);
1728
1729 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1730 {
1731 ssize_t res;
1732 struct msghdr msg = {.msg_flags = flags};
1733 struct kvec iov;
1734 char *kaddr = kmap(page);
1735 iov.iov_base = kaddr + offset;
1736 iov.iov_len = size;
1737 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1738 kunmap(page);
1739 return res;
1740 }
1741 EXPORT_SYMBOL(sock_no_sendpage);
1742
1743 /*
1744 * Default Socket Callbacks
1745 */
1746
1747 static void sock_def_wakeup(struct sock *sk)
1748 {
1749 read_lock(&sk->sk_callback_lock);
1750 if (sk_has_sleeper(sk))
1751 wake_up_interruptible_all(sk->sk_sleep);
1752 read_unlock(&sk->sk_callback_lock);
1753 }
1754
1755 static void sock_def_error_report(struct sock *sk)
1756 {
1757 read_lock(&sk->sk_callback_lock);
1758 if (sk_has_sleeper(sk))
1759 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1760 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1761 read_unlock(&sk->sk_callback_lock);
1762 }
1763
1764 static void sock_def_readable(struct sock *sk, int len)
1765 {
1766 read_lock(&sk->sk_callback_lock);
1767 if (sk_has_sleeper(sk))
1768 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1769 POLLRDNORM | POLLRDBAND);
1770 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1771 read_unlock(&sk->sk_callback_lock);
1772 }
1773
1774 static void sock_def_write_space(struct sock *sk)
1775 {
1776 read_lock(&sk->sk_callback_lock);
1777
1778 /* Do not wake up a writer until he can make "significant"
1779 * progress. --DaveM
1780 */
1781 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1782 if (sk_has_sleeper(sk))
1783 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1784 POLLWRNORM | POLLWRBAND);
1785
1786 /* Should agree with poll, otherwise some programs break */
1787 if (sock_writeable(sk))
1788 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1789 }
1790
1791 read_unlock(&sk->sk_callback_lock);
1792 }
1793
1794 static void sock_def_destruct(struct sock *sk)
1795 {
1796 kfree(sk->sk_protinfo);
1797 }
1798
1799 void sk_send_sigurg(struct sock *sk)
1800 {
1801 if (sk->sk_socket && sk->sk_socket->file)
1802 if (send_sigurg(&sk->sk_socket->file->f_owner))
1803 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1804 }
1805 EXPORT_SYMBOL(sk_send_sigurg);
1806
1807 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1808 unsigned long expires)
1809 {
1810 if (!mod_timer(timer, expires))
1811 sock_hold(sk);
1812 }
1813 EXPORT_SYMBOL(sk_reset_timer);
1814
1815 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1816 {
1817 if (timer_pending(timer) && del_timer(timer))
1818 __sock_put(sk);
1819 }
1820 EXPORT_SYMBOL(sk_stop_timer);
1821
1822 void sock_init_data(struct socket *sock, struct sock *sk)
1823 {
1824 skb_queue_head_init(&sk->sk_receive_queue);
1825 skb_queue_head_init(&sk->sk_write_queue);
1826 skb_queue_head_init(&sk->sk_error_queue);
1827 #ifdef CONFIG_NET_DMA
1828 skb_queue_head_init(&sk->sk_async_wait_queue);
1829 #endif
1830
1831 sk->sk_send_head = NULL;
1832
1833 init_timer(&sk->sk_timer);
1834
1835 sk->sk_allocation = GFP_KERNEL;
1836 sk->sk_rcvbuf = sysctl_rmem_default;
1837 sk->sk_sndbuf = sysctl_wmem_default;
1838 sk->sk_state = TCP_CLOSE;
1839 sk_set_socket(sk, sock);
1840
1841 sock_set_flag(sk, SOCK_ZAPPED);
1842
1843 if (sock) {
1844 sk->sk_type = sock->type;
1845 sk->sk_sleep = &sock->wait;
1846 sock->sk = sk;
1847 } else
1848 sk->sk_sleep = NULL;
1849
1850 rwlock_init(&sk->sk_dst_lock);
1851 rwlock_init(&sk->sk_callback_lock);
1852 lockdep_set_class_and_name(&sk->sk_callback_lock,
1853 af_callback_keys + sk->sk_family,
1854 af_family_clock_key_strings[sk->sk_family]);
1855
1856 sk->sk_state_change = sock_def_wakeup;
1857 sk->sk_data_ready = sock_def_readable;
1858 sk->sk_write_space = sock_def_write_space;
1859 sk->sk_error_report = sock_def_error_report;
1860 sk->sk_destruct = sock_def_destruct;
1861
1862 sk->sk_sndmsg_page = NULL;
1863 sk->sk_sndmsg_off = 0;
1864
1865 sk->sk_peercred.pid = 0;
1866 sk->sk_peercred.uid = -1;
1867 sk->sk_peercred.gid = -1;
1868 sk->sk_write_pending = 0;
1869 sk->sk_rcvlowat = 1;
1870 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1871 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1872
1873 sk->sk_stamp = ktime_set(-1L, 0);
1874
1875 /*
1876 * Before updating sk_refcnt, we must commit prior changes to memory
1877 * (Documentation/RCU/rculist_nulls.txt for details)
1878 */
1879 smp_wmb();
1880 atomic_set(&sk->sk_refcnt, 1);
1881 atomic_set(&sk->sk_drops, 0);
1882 }
1883 EXPORT_SYMBOL(sock_init_data);
1884
1885 void lock_sock_nested(struct sock *sk, int subclass)
1886 {
1887 might_sleep();
1888 spin_lock_bh(&sk->sk_lock.slock);
1889 if (sk->sk_lock.owned)
1890 __lock_sock(sk);
1891 sk->sk_lock.owned = 1;
1892 spin_unlock(&sk->sk_lock.slock);
1893 /*
1894 * The sk_lock has mutex_lock() semantics here:
1895 */
1896 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1897 local_bh_enable();
1898 }
1899 EXPORT_SYMBOL(lock_sock_nested);
1900
1901 void release_sock(struct sock *sk)
1902 {
1903 /*
1904 * The sk_lock has mutex_unlock() semantics:
1905 */
1906 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1907
1908 spin_lock_bh(&sk->sk_lock.slock);
1909 if (sk->sk_backlog.tail)
1910 __release_sock(sk);
1911 sk->sk_lock.owned = 0;
1912 if (waitqueue_active(&sk->sk_lock.wq))
1913 wake_up(&sk->sk_lock.wq);
1914 spin_unlock_bh(&sk->sk_lock.slock);
1915 }
1916 EXPORT_SYMBOL(release_sock);
1917
1918 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1919 {
1920 struct timeval tv;
1921 if (!sock_flag(sk, SOCK_TIMESTAMP))
1922 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1923 tv = ktime_to_timeval(sk->sk_stamp);
1924 if (tv.tv_sec == -1)
1925 return -ENOENT;
1926 if (tv.tv_sec == 0) {
1927 sk->sk_stamp = ktime_get_real();
1928 tv = ktime_to_timeval(sk->sk_stamp);
1929 }
1930 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1931 }
1932 EXPORT_SYMBOL(sock_get_timestamp);
1933
1934 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1935 {
1936 struct timespec ts;
1937 if (!sock_flag(sk, SOCK_TIMESTAMP))
1938 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1939 ts = ktime_to_timespec(sk->sk_stamp);
1940 if (ts.tv_sec == -1)
1941 return -ENOENT;
1942 if (ts.tv_sec == 0) {
1943 sk->sk_stamp = ktime_get_real();
1944 ts = ktime_to_timespec(sk->sk_stamp);
1945 }
1946 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1947 }
1948 EXPORT_SYMBOL(sock_get_timestampns);
1949
1950 void sock_enable_timestamp(struct sock *sk, int flag)
1951 {
1952 if (!sock_flag(sk, flag)) {
1953 sock_set_flag(sk, flag);
1954 /*
1955 * we just set one of the two flags which require net
1956 * time stamping, but time stamping might have been on
1957 * already because of the other one
1958 */
1959 if (!sock_flag(sk,
1960 flag == SOCK_TIMESTAMP ?
1961 SOCK_TIMESTAMPING_RX_SOFTWARE :
1962 SOCK_TIMESTAMP))
1963 net_enable_timestamp();
1964 }
1965 }
1966
1967 /*
1968 * Get a socket option on an socket.
1969 *
1970 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1971 * asynchronous errors should be reported by getsockopt. We assume
1972 * this means if you specify SO_ERROR (otherwise whats the point of it).
1973 */
1974 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1975 char __user *optval, int __user *optlen)
1976 {
1977 struct sock *sk = sock->sk;
1978
1979 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1980 }
1981 EXPORT_SYMBOL(sock_common_getsockopt);
1982
1983 #ifdef CONFIG_COMPAT
1984 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1985 char __user *optval, int __user *optlen)
1986 {
1987 struct sock *sk = sock->sk;
1988
1989 if (sk->sk_prot->compat_getsockopt != NULL)
1990 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1991 optval, optlen);
1992 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1993 }
1994 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1995 #endif
1996
1997 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1998 struct msghdr *msg, size_t size, int flags)
1999 {
2000 struct sock *sk = sock->sk;
2001 int addr_len = 0;
2002 int err;
2003
2004 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2005 flags & ~MSG_DONTWAIT, &addr_len);
2006 if (err >= 0)
2007 msg->msg_namelen = addr_len;
2008 return err;
2009 }
2010 EXPORT_SYMBOL(sock_common_recvmsg);
2011
2012 /*
2013 * Set socket options on an inet socket.
2014 */
2015 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2016 char __user *optval, int optlen)
2017 {
2018 struct sock *sk = sock->sk;
2019
2020 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2021 }
2022 EXPORT_SYMBOL(sock_common_setsockopt);
2023
2024 #ifdef CONFIG_COMPAT
2025 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2026 char __user *optval, int optlen)
2027 {
2028 struct sock *sk = sock->sk;
2029
2030 if (sk->sk_prot->compat_setsockopt != NULL)
2031 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2032 optval, optlen);
2033 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2034 }
2035 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2036 #endif
2037
2038 void sk_common_release(struct sock *sk)
2039 {
2040 if (sk->sk_prot->destroy)
2041 sk->sk_prot->destroy(sk);
2042
2043 /*
2044 * Observation: when sock_common_release is called, processes have
2045 * no access to socket. But net still has.
2046 * Step one, detach it from networking:
2047 *
2048 * A. Remove from hash tables.
2049 */
2050
2051 sk->sk_prot->unhash(sk);
2052
2053 /*
2054 * In this point socket cannot receive new packets, but it is possible
2055 * that some packets are in flight because some CPU runs receiver and
2056 * did hash table lookup before we unhashed socket. They will achieve
2057 * receive queue and will be purged by socket destructor.
2058 *
2059 * Also we still have packets pending on receive queue and probably,
2060 * our own packets waiting in device queues. sock_destroy will drain
2061 * receive queue, but transmitted packets will delay socket destruction
2062 * until the last reference will be released.
2063 */
2064
2065 sock_orphan(sk);
2066
2067 xfrm_sk_free_policy(sk);
2068
2069 sk_refcnt_debug_release(sk);
2070 sock_put(sk);
2071 }
2072 EXPORT_SYMBOL(sk_common_release);
2073
2074 static DEFINE_RWLOCK(proto_list_lock);
2075 static LIST_HEAD(proto_list);
2076
2077 #ifdef CONFIG_PROC_FS
2078 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2079 struct prot_inuse {
2080 int val[PROTO_INUSE_NR];
2081 };
2082
2083 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2084
2085 #ifdef CONFIG_NET_NS
2086 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2087 {
2088 int cpu = smp_processor_id();
2089 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2090 }
2091 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2092
2093 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2094 {
2095 int cpu, idx = prot->inuse_idx;
2096 int res = 0;
2097
2098 for_each_possible_cpu(cpu)
2099 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2100
2101 return res >= 0 ? res : 0;
2102 }
2103 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2104
2105 static int sock_inuse_init_net(struct net *net)
2106 {
2107 net->core.inuse = alloc_percpu(struct prot_inuse);
2108 return net->core.inuse ? 0 : -ENOMEM;
2109 }
2110
2111 static void sock_inuse_exit_net(struct net *net)
2112 {
2113 free_percpu(net->core.inuse);
2114 }
2115
2116 static struct pernet_operations net_inuse_ops = {
2117 .init = sock_inuse_init_net,
2118 .exit = sock_inuse_exit_net,
2119 };
2120
2121 static __init int net_inuse_init(void)
2122 {
2123 if (register_pernet_subsys(&net_inuse_ops))
2124 panic("Cannot initialize net inuse counters");
2125
2126 return 0;
2127 }
2128
2129 core_initcall(net_inuse_init);
2130 #else
2131 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2132
2133 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2134 {
2135 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2136 }
2137 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2138
2139 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2140 {
2141 int cpu, idx = prot->inuse_idx;
2142 int res = 0;
2143
2144 for_each_possible_cpu(cpu)
2145 res += per_cpu(prot_inuse, cpu).val[idx];
2146
2147 return res >= 0 ? res : 0;
2148 }
2149 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2150 #endif
2151
2152 static void assign_proto_idx(struct proto *prot)
2153 {
2154 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2155
2156 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2157 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2158 return;
2159 }
2160
2161 set_bit(prot->inuse_idx, proto_inuse_idx);
2162 }
2163
2164 static void release_proto_idx(struct proto *prot)
2165 {
2166 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2167 clear_bit(prot->inuse_idx, proto_inuse_idx);
2168 }
2169 #else
2170 static inline void assign_proto_idx(struct proto *prot)
2171 {
2172 }
2173
2174 static inline void release_proto_idx(struct proto *prot)
2175 {
2176 }
2177 #endif
2178
2179 int proto_register(struct proto *prot, int alloc_slab)
2180 {
2181 if (alloc_slab) {
2182 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2183 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2184 NULL);
2185
2186 if (prot->slab == NULL) {
2187 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2188 prot->name);
2189 goto out;
2190 }
2191
2192 if (prot->rsk_prot != NULL) {
2193 static const char mask[] = "request_sock_%s";
2194
2195 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2196 if (prot->rsk_prot->slab_name == NULL)
2197 goto out_free_sock_slab;
2198
2199 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2200 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2201 prot->rsk_prot->obj_size, 0,
2202 SLAB_HWCACHE_ALIGN, NULL);
2203
2204 if (prot->rsk_prot->slab == NULL) {
2205 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2206 prot->name);
2207 goto out_free_request_sock_slab_name;
2208 }
2209 }
2210
2211 if (prot->twsk_prot != NULL) {
2212 static const char mask[] = "tw_sock_%s";
2213
2214 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2215
2216 if (prot->twsk_prot->twsk_slab_name == NULL)
2217 goto out_free_request_sock_slab;
2218
2219 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2220 prot->twsk_prot->twsk_slab =
2221 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2222 prot->twsk_prot->twsk_obj_size,
2223 0,
2224 SLAB_HWCACHE_ALIGN |
2225 prot->slab_flags,
2226 NULL);
2227 if (prot->twsk_prot->twsk_slab == NULL)
2228 goto out_free_timewait_sock_slab_name;
2229 }
2230 }
2231
2232 write_lock(&proto_list_lock);
2233 list_add(&prot->node, &proto_list);
2234 assign_proto_idx(prot);
2235 write_unlock(&proto_list_lock);
2236 return 0;
2237
2238 out_free_timewait_sock_slab_name:
2239 kfree(prot->twsk_prot->twsk_slab_name);
2240 out_free_request_sock_slab:
2241 if (prot->rsk_prot && prot->rsk_prot->slab) {
2242 kmem_cache_destroy(prot->rsk_prot->slab);
2243 prot->rsk_prot->slab = NULL;
2244 }
2245 out_free_request_sock_slab_name:
2246 kfree(prot->rsk_prot->slab_name);
2247 out_free_sock_slab:
2248 kmem_cache_destroy(prot->slab);
2249 prot->slab = NULL;
2250 out:
2251 return -ENOBUFS;
2252 }
2253 EXPORT_SYMBOL(proto_register);
2254
2255 void proto_unregister(struct proto *prot)
2256 {
2257 write_lock(&proto_list_lock);
2258 release_proto_idx(prot);
2259 list_del(&prot->node);
2260 write_unlock(&proto_list_lock);
2261
2262 if (prot->slab != NULL) {
2263 kmem_cache_destroy(prot->slab);
2264 prot->slab = NULL;
2265 }
2266
2267 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2268 kmem_cache_destroy(prot->rsk_prot->slab);
2269 kfree(prot->rsk_prot->slab_name);
2270 prot->rsk_prot->slab = NULL;
2271 }
2272
2273 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2274 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2275 kfree(prot->twsk_prot->twsk_slab_name);
2276 prot->twsk_prot->twsk_slab = NULL;
2277 }
2278 }
2279 EXPORT_SYMBOL(proto_unregister);
2280
2281 #ifdef CONFIG_PROC_FS
2282 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2283 __acquires(proto_list_lock)
2284 {
2285 read_lock(&proto_list_lock);
2286 return seq_list_start_head(&proto_list, *pos);
2287 }
2288
2289 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2290 {
2291 return seq_list_next(v, &proto_list, pos);
2292 }
2293
2294 static void proto_seq_stop(struct seq_file *seq, void *v)
2295 __releases(proto_list_lock)
2296 {
2297 read_unlock(&proto_list_lock);
2298 }
2299
2300 static char proto_method_implemented(const void *method)
2301 {
2302 return method == NULL ? 'n' : 'y';
2303 }
2304
2305 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2306 {
2307 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2308 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2309 proto->name,
2310 proto->obj_size,
2311 sock_prot_inuse_get(seq_file_net(seq), proto),
2312 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2313 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2314 proto->max_header,
2315 proto->slab == NULL ? "no" : "yes",
2316 module_name(proto->owner),
2317 proto_method_implemented(proto->close),
2318 proto_method_implemented(proto->connect),
2319 proto_method_implemented(proto->disconnect),
2320 proto_method_implemented(proto->accept),
2321 proto_method_implemented(proto->ioctl),
2322 proto_method_implemented(proto->init),
2323 proto_method_implemented(proto->destroy),
2324 proto_method_implemented(proto->shutdown),
2325 proto_method_implemented(proto->setsockopt),
2326 proto_method_implemented(proto->getsockopt),
2327 proto_method_implemented(proto->sendmsg),
2328 proto_method_implemented(proto->recvmsg),
2329 proto_method_implemented(proto->sendpage),
2330 proto_method_implemented(proto->bind),
2331 proto_method_implemented(proto->backlog_rcv),
2332 proto_method_implemented(proto->hash),
2333 proto_method_implemented(proto->unhash),
2334 proto_method_implemented(proto->get_port),
2335 proto_method_implemented(proto->enter_memory_pressure));
2336 }
2337
2338 static int proto_seq_show(struct seq_file *seq, void *v)
2339 {
2340 if (v == &proto_list)
2341 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2342 "protocol",
2343 "size",
2344 "sockets",
2345 "memory",
2346 "press",
2347 "maxhdr",
2348 "slab",
2349 "module",
2350 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2351 else
2352 proto_seq_printf(seq, list_entry(v, struct proto, node));
2353 return 0;
2354 }
2355
2356 static const struct seq_operations proto_seq_ops = {
2357 .start = proto_seq_start,
2358 .next = proto_seq_next,
2359 .stop = proto_seq_stop,
2360 .show = proto_seq_show,
2361 };
2362
2363 static int proto_seq_open(struct inode *inode, struct file *file)
2364 {
2365 return seq_open_net(inode, file, &proto_seq_ops,
2366 sizeof(struct seq_net_private));
2367 }
2368
2369 static const struct file_operations proto_seq_fops = {
2370 .owner = THIS_MODULE,
2371 .open = proto_seq_open,
2372 .read = seq_read,
2373 .llseek = seq_lseek,
2374 .release = seq_release_net,
2375 };
2376
2377 static __net_init int proto_init_net(struct net *net)
2378 {
2379 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2380 return -ENOMEM;
2381
2382 return 0;
2383 }
2384
2385 static __net_exit void proto_exit_net(struct net *net)
2386 {
2387 proc_net_remove(net, "protocols");
2388 }
2389
2390
2391 static __net_initdata struct pernet_operations proto_net_ops = {
2392 .init = proto_init_net,
2393 .exit = proto_exit_net,
2394 };
2395
2396 static int __init proto_init(void)
2397 {
2398 return register_pernet_subsys(&proto_net_ops);
2399 }
2400
2401 subsys_initcall(proto_init);
2402
2403 #endif /* PROC_FS */
Linux with added FPC-III support