projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: socket infrastructure for SO_TIMESTAMPING
[deliverable/linux.git] / net / core / sock.c
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-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
154 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
155 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
156 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
157 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
158 "sk_lock-AF_MAX"
159 };
160 static const char *af_family_slock_key_strings[AF_MAX+1] = {
161 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
162 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
163 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
164 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
165 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
166 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
167 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
168 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
169 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
170 "slock-27" , "slock-28" , "slock-AF_CAN" ,
171 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
172 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
173 "slock-AF_MAX"
174 };
175 static const char *af_family_clock_key_strings[AF_MAX+1] = {
176 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
177 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
178 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
179 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
180 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
181 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
182 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
183 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
184 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
185 "clock-27" , "clock-28" , "clock-AF_CAN" ,
186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
187 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
188 "clock-AF_MAX"
189 };
190
191 /*
192 * sk_callback_lock locking rules are per-address-family,
193 * so split the lock classes by using a per-AF key:
194 */
195 static struct lock_class_key af_callback_keys[AF_MAX];
196
197 /* Take into consideration the size of the struct sk_buff overhead in the
198 * determination of these values, since that is non-constant across
199 * platforms. This makes socket queueing behavior and performance
200 * not depend upon such differences.
201 */
202 #define _SK_MEM_PACKETS 256
203 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
204 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
206
207 /* Run time adjustable parameters. */
208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
212
213 /* Maximal space eaten by iovec or ancilliary data plus some space */
214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
215
216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
217 {
218 struct timeval tv;
219
220 if (optlen < sizeof(tv))
221 return -EINVAL;
222 if (copy_from_user(&tv, optval, sizeof(tv)))
223 return -EFAULT;
224 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
225 return -EDOM;
226
227 if (tv.tv_sec < 0) {
228 static int warned __read_mostly;
229
230 *timeo_p = 0;
231 if (warned < 10 && net_ratelimit()) {
232 warned++;
233 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
234 "tries to set negative timeout\n",
235 current->comm, task_pid_nr(current));
236 }
237 return 0;
238 }
239 *timeo_p = MAX_SCHEDULE_TIMEOUT;
240 if (tv.tv_sec == 0 && tv.tv_usec == 0)
241 return 0;
242 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
243 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
244 return 0;
245 }
246
247 static void sock_warn_obsolete_bsdism(const char *name)
248 {
249 static int warned;
250 static char warncomm[TASK_COMM_LEN];
251 if (strcmp(warncomm, current->comm) && warned < 5) {
252 strcpy(warncomm, current->comm);
253 printk(KERN_WARNING "process `%s' is using obsolete "
254 "%s SO_BSDCOMPAT\n", warncomm, name);
255 warned++;
256 }
257 }
258
259 static void sock_disable_timestamp(struct sock *sk, int flag)
260 {
261 if (sock_flag(sk, flag)) {
262 sock_reset_flag(sk, flag);
263 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
264 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
265 net_disable_timestamp();
266 }
267 }
268 }
269
270
271 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
272 {
273 int err = 0;
274 int skb_len;
275
276 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
277 number of warnings when compiling with -W --ANK
278 */
279 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
280 (unsigned)sk->sk_rcvbuf) {
281 err = -ENOMEM;
282 goto out;
283 }
284
285 err = sk_filter(sk, skb);
286 if (err)
287 goto out;
288
289 if (!sk_rmem_schedule(sk, skb->truesize)) {
290 err = -ENOBUFS;
291 goto out;
292 }
293
294 skb->dev = NULL;
295 skb_set_owner_r(skb, sk);
296
297 /* Cache the SKB length before we tack it onto the receive
298 * queue. Once it is added it no longer belongs to us and
299 * may be freed by other threads of control pulling packets
300 * from the queue.
301 */
302 skb_len = skb->len;
303
304 skb_queue_tail(&sk->sk_receive_queue, skb);
305
306 if (!sock_flag(sk, SOCK_DEAD))
307 sk->sk_data_ready(sk, skb_len);
308 out:
309 return err;
310 }
311 EXPORT_SYMBOL(sock_queue_rcv_skb);
312
313 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
314 {
315 int rc = NET_RX_SUCCESS;
316
317 if (sk_filter(sk, skb))
318 goto discard_and_relse;
319
320 skb->dev = NULL;
321
322 if (nested)
323 bh_lock_sock_nested(sk);
324 else
325 bh_lock_sock(sk);
326 if (!sock_owned_by_user(sk)) {
327 /*
328 * trylock + unlock semantics:
329 */
330 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
331
332 rc = sk_backlog_rcv(sk, skb);
333
334 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
335 } else
336 sk_add_backlog(sk, skb);
337 bh_unlock_sock(sk);
338 out:
339 sock_put(sk);
340 return rc;
341 discard_and_relse:
342 kfree_skb(skb);
343 goto out;
344 }
345 EXPORT_SYMBOL(sk_receive_skb);
346
347 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
348 {
349 struct dst_entry *dst = sk->sk_dst_cache;
350
351 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
352 sk->sk_dst_cache = NULL;
353 dst_release(dst);
354 return NULL;
355 }
356
357 return dst;
358 }
359 EXPORT_SYMBOL(__sk_dst_check);
360
361 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
362 {
363 struct dst_entry *dst = sk_dst_get(sk);
364
365 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
366 sk_dst_reset(sk);
367 dst_release(dst);
368 return NULL;
369 }
370
371 return dst;
372 }
373 EXPORT_SYMBOL(sk_dst_check);
374
375 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
376 {
377 int ret = -ENOPROTOOPT;
378 #ifdef CONFIG_NETDEVICES
379 struct net *net = sock_net(sk);
380 char devname[IFNAMSIZ];
381 int index;
382
383 /* Sorry... */
384 ret = -EPERM;
385 if (!capable(CAP_NET_RAW))
386 goto out;
387
388 ret = -EINVAL;
389 if (optlen < 0)
390 goto out;
391
392 /* Bind this socket to a particular device like "eth0",
393 * as specified in the passed interface name. If the
394 * name is "" or the option length is zero the socket
395 * is not bound.
396 */
397 if (optlen > IFNAMSIZ - 1)
398 optlen = IFNAMSIZ - 1;
399 memset(devname, 0, sizeof(devname));
400
401 ret = -EFAULT;
402 if (copy_from_user(devname, optval, optlen))
403 goto out;
404
405 if (devname[0] == '\0') {
406 index = 0;
407 } else {
408 struct net_device *dev = dev_get_by_name(net, devname);
409
410 ret = -ENODEV;
411 if (!dev)
412 goto out;
413
414 index = dev->ifindex;
415 dev_put(dev);
416 }
417
418 lock_sock(sk);
419 sk->sk_bound_dev_if = index;
420 sk_dst_reset(sk);
421 release_sock(sk);
422
423 ret = 0;
424
425 out:
426 #endif
427
428 return ret;
429 }
430
431 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
432 {
433 if (valbool)
434 sock_set_flag(sk, bit);
435 else
436 sock_reset_flag(sk, bit);
437 }
438
439 /*
440 * This is meant for all protocols to use and covers goings on
441 * at the socket level. Everything here is generic.
442 */
443
444 int sock_setsockopt(struct socket *sock, int level, int optname,
445 char __user *optval, int optlen)
446 {
447 struct sock *sk=sock->sk;
448 int val;
449 int valbool;
450 struct linger ling;
451 int ret = 0;
452
453 /*
454 * Options without arguments
455 */
456
457 if (optname == SO_BINDTODEVICE)
458 return sock_bindtodevice(sk, optval, optlen);
459
460 if (optlen < sizeof(int))
461 return -EINVAL;
462
463 if (get_user(val, (int __user *)optval))
464 return -EFAULT;
465
466 valbool = val?1:0;
467
468 lock_sock(sk);
469
470 switch(optname) {
471 case SO_DEBUG:
472 if (val && !capable(CAP_NET_ADMIN)) {
473 ret = -EACCES;
474 } else
475 sock_valbool_flag(sk, SOCK_DBG, valbool);
476 break;
477 case SO_REUSEADDR:
478 sk->sk_reuse = valbool;
479 break;
480 case SO_TYPE:
481 case SO_ERROR:
482 ret = -ENOPROTOOPT;
483 break;
484 case SO_DONTROUTE:
485 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
486 break;
487 case SO_BROADCAST:
488 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
489 break;
490 case SO_SNDBUF:
491 /* Don't error on this BSD doesn't and if you think
492 about it this is right. Otherwise apps have to
493 play 'guess the biggest size' games. RCVBUF/SNDBUF
494 are treated in BSD as hints */
495
496 if (val > sysctl_wmem_max)
497 val = sysctl_wmem_max;
498 set_sndbuf:
499 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
500 if ((val * 2) < SOCK_MIN_SNDBUF)
501 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
502 else
503 sk->sk_sndbuf = val * 2;
504
505 /*
506 * Wake up sending tasks if we
507 * upped the value.
508 */
509 sk->sk_write_space(sk);
510 break;
511
512 case SO_SNDBUFFORCE:
513 if (!capable(CAP_NET_ADMIN)) {
514 ret = -EPERM;
515 break;
516 }
517 goto set_sndbuf;
518
519 case SO_RCVBUF:
520 /* Don't error on this BSD doesn't and if you think
521 about it this is right. Otherwise apps have to
522 play 'guess the biggest size' games. RCVBUF/SNDBUF
523 are treated in BSD as hints */
524
525 if (val > sysctl_rmem_max)
526 val = sysctl_rmem_max;
527 set_rcvbuf:
528 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
529 /*
530 * We double it on the way in to account for
531 * "struct sk_buff" etc. overhead. Applications
532 * assume that the SO_RCVBUF setting they make will
533 * allow that much actual data to be received on that
534 * socket.
535 *
536 * Applications are unaware that "struct sk_buff" and
537 * other overheads allocate from the receive buffer
538 * during socket buffer allocation.
539 *
540 * And after considering the possible alternatives,
541 * returning the value we actually used in getsockopt
542 * is the most desirable behavior.
543 */
544 if ((val * 2) < SOCK_MIN_RCVBUF)
545 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
546 else
547 sk->sk_rcvbuf = val * 2;
548 break;
549
550 case SO_RCVBUFFORCE:
551 if (!capable(CAP_NET_ADMIN)) {
552 ret = -EPERM;
553 break;
554 }
555 goto set_rcvbuf;
556
557 case SO_KEEPALIVE:
558 #ifdef CONFIG_INET
559 if (sk->sk_protocol == IPPROTO_TCP)
560 tcp_set_keepalive(sk, valbool);
561 #endif
562 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
563 break;
564
565 case SO_OOBINLINE:
566 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
567 break;
568
569 case SO_NO_CHECK:
570 sk->sk_no_check = valbool;
571 break;
572
573 case SO_PRIORITY:
574 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
575 sk->sk_priority = val;
576 else
577 ret = -EPERM;
578 break;
579
580 case SO_LINGER:
581 if (optlen < sizeof(ling)) {
582 ret = -EINVAL; /* 1003.1g */
583 break;
584 }
585 if (copy_from_user(&ling,optval,sizeof(ling))) {
586 ret = -EFAULT;
587 break;
588 }
589 if (!ling.l_onoff)
590 sock_reset_flag(sk, SOCK_LINGER);
591 else {
592 #if (BITS_PER_LONG == 32)
593 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
594 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
595 else
596 #endif
597 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
598 sock_set_flag(sk, SOCK_LINGER);
599 }
600 break;
601
602 case SO_BSDCOMPAT:
603 sock_warn_obsolete_bsdism("setsockopt");
604 break;
605
606 case SO_PASSCRED:
607 if (valbool)
608 set_bit(SOCK_PASSCRED, &sock->flags);
609 else
610 clear_bit(SOCK_PASSCRED, &sock->flags);
611 break;
612
613 case SO_TIMESTAMP:
614 case SO_TIMESTAMPNS:
615 if (valbool) {
616 if (optname == SO_TIMESTAMP)
617 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
618 else
619 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
620 sock_set_flag(sk, SOCK_RCVTSTAMP);
621 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
622 } else {
623 sock_reset_flag(sk, SOCK_RCVTSTAMP);
624 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
625 }
626 break;
627
628 case SO_TIMESTAMPING:
629 if (val & ~SOF_TIMESTAMPING_MASK) {
630 ret = EINVAL;
631 break;
632 }
633 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
634 val & SOF_TIMESTAMPING_TX_HARDWARE);
635 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
636 val & SOF_TIMESTAMPING_TX_SOFTWARE);
637 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
638 val & SOF_TIMESTAMPING_RX_HARDWARE);
639 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
640 sock_enable_timestamp(sk,
641 SOCK_TIMESTAMPING_RX_SOFTWARE);
642 else
643 sock_disable_timestamp(sk,
644 SOCK_TIMESTAMPING_RX_SOFTWARE);
645 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
646 val & SOF_TIMESTAMPING_SOFTWARE);
647 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
648 val & SOF_TIMESTAMPING_SYS_HARDWARE);
649 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
650 val & SOF_TIMESTAMPING_RAW_HARDWARE);
651 break;
652
653 case SO_RCVLOWAT:
654 if (val < 0)
655 val = INT_MAX;
656 sk->sk_rcvlowat = val ? : 1;
657 break;
658
659 case SO_RCVTIMEO:
660 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
661 break;
662
663 case SO_SNDTIMEO:
664 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
665 break;
666
667 case SO_ATTACH_FILTER:
668 ret = -EINVAL;
669 if (optlen == sizeof(struct sock_fprog)) {
670 struct sock_fprog fprog;
671
672 ret = -EFAULT;
673 if (copy_from_user(&fprog, optval, sizeof(fprog)))
674 break;
675
676 ret = sk_attach_filter(&fprog, sk);
677 }
678 break;
679
680 case SO_DETACH_FILTER:
681 ret = sk_detach_filter(sk);
682 break;
683
684 case SO_PASSSEC:
685 if (valbool)
686 set_bit(SOCK_PASSSEC, &sock->flags);
687 else
688 clear_bit(SOCK_PASSSEC, &sock->flags);
689 break;
690 case SO_MARK:
691 if (!capable(CAP_NET_ADMIN))
692 ret = -EPERM;
693 else {
694 sk->sk_mark = val;
695 }
696 break;
697
698 /* We implement the SO_SNDLOWAT etc to
699 not be settable (1003.1g 5.3) */
700 default:
701 ret = -ENOPROTOOPT;
702 break;
703 }
704 release_sock(sk);
705 return ret;
706 }
707
708
709 int sock_getsockopt(struct socket *sock, int level, int optname,
710 char __user *optval, int __user *optlen)
711 {
712 struct sock *sk = sock->sk;
713
714 union {
715 int val;
716 struct linger ling;
717 struct timeval tm;
718 } v;
719
720 unsigned int lv = sizeof(int);
721 int len;
722
723 if (get_user(len, optlen))
724 return -EFAULT;
725 if (len < 0)
726 return -EINVAL;
727
728 v.val = 0;
729
730 switch(optname) {
731 case SO_DEBUG:
732 v.val = sock_flag(sk, SOCK_DBG);
733 break;
734
735 case SO_DONTROUTE:
736 v.val = sock_flag(sk, SOCK_LOCALROUTE);
737 break;
738
739 case SO_BROADCAST:
740 v.val = !!sock_flag(sk, SOCK_BROADCAST);
741 break;
742
743 case SO_SNDBUF:
744 v.val = sk->sk_sndbuf;
745 break;
746
747 case SO_RCVBUF:
748 v.val = sk->sk_rcvbuf;
749 break;
750
751 case SO_REUSEADDR:
752 v.val = sk->sk_reuse;
753 break;
754
755 case SO_KEEPALIVE:
756 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
757 break;
758
759 case SO_TYPE:
760 v.val = sk->sk_type;
761 break;
762
763 case SO_ERROR:
764 v.val = -sock_error(sk);
765 if (v.val==0)
766 v.val = xchg(&sk->sk_err_soft, 0);
767 break;
768
769 case SO_OOBINLINE:
770 v.val = !!sock_flag(sk, SOCK_URGINLINE);
771 break;
772
773 case SO_NO_CHECK:
774 v.val = sk->sk_no_check;
775 break;
776
777 case SO_PRIORITY:
778 v.val = sk->sk_priority;
779 break;
780
781 case SO_LINGER:
782 lv = sizeof(v.ling);
783 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
784 v.ling.l_linger = sk->sk_lingertime / HZ;
785 break;
786
787 case SO_BSDCOMPAT:
788 sock_warn_obsolete_bsdism("getsockopt");
789 break;
790
791 case SO_TIMESTAMP:
792 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
793 !sock_flag(sk, SOCK_RCVTSTAMPNS);
794 break;
795
796 case SO_TIMESTAMPNS:
797 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
798 break;
799
800 case SO_TIMESTAMPING:
801 v.val = 0;
802 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
803 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
804 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
805 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
806 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
807 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
808 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
809 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
810 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
811 v.val |= SOF_TIMESTAMPING_SOFTWARE;
812 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
813 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
814 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
815 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
816 break;
817
818 case SO_RCVTIMEO:
819 lv=sizeof(struct timeval);
820 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
821 v.tm.tv_sec = 0;
822 v.tm.tv_usec = 0;
823 } else {
824 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
825 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
826 }
827 break;
828
829 case SO_SNDTIMEO:
830 lv=sizeof(struct timeval);
831 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
832 v.tm.tv_sec = 0;
833 v.tm.tv_usec = 0;
834 } else {
835 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
836 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
837 }
838 break;
839
840 case SO_RCVLOWAT:
841 v.val = sk->sk_rcvlowat;
842 break;
843
844 case SO_SNDLOWAT:
845 v.val=1;
846 break;
847
848 case SO_PASSCRED:
849 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
850 break;
851
852 case SO_PEERCRED:
853 if (len > sizeof(sk->sk_peercred))
854 len = sizeof(sk->sk_peercred);
855 if (copy_to_user(optval, &sk->sk_peercred, len))
856 return -EFAULT;
857 goto lenout;
858
859 case SO_PEERNAME:
860 {
861 char address[128];
862
863 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
864 return -ENOTCONN;
865 if (lv < len)
866 return -EINVAL;
867 if (copy_to_user(optval, address, len))
868 return -EFAULT;
869 goto lenout;
870 }
871
872 /* Dubious BSD thing... Probably nobody even uses it, but
873 * the UNIX standard wants it for whatever reason... -DaveM
874 */
875 case SO_ACCEPTCONN:
876 v.val = sk->sk_state == TCP_LISTEN;
877 break;
878
879 case SO_PASSSEC:
880 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
881 break;
882
883 case SO_PEERSEC:
884 return security_socket_getpeersec_stream(sock, optval, optlen, len);
885
886 case SO_MARK:
887 v.val = sk->sk_mark;
888 break;
889
890 default:
891 return -ENOPROTOOPT;
892 }
893
894 if (len > lv)
895 len = lv;
896 if (copy_to_user(optval, &v, len))
897 return -EFAULT;
898 lenout:
899 if (put_user(len, optlen))
900 return -EFAULT;
901 return 0;
902 }
903
904 /*
905 * Initialize an sk_lock.
906 *
907 * (We also register the sk_lock with the lock validator.)
908 */
909 static inline void sock_lock_init(struct sock *sk)
910 {
911 sock_lock_init_class_and_name(sk,
912 af_family_slock_key_strings[sk->sk_family],
913 af_family_slock_keys + sk->sk_family,
914 af_family_key_strings[sk->sk_family],
915 af_family_keys + sk->sk_family);
916 }
917
918 static void sock_copy(struct sock *nsk, const struct sock *osk)
919 {
920 #ifdef CONFIG_SECURITY_NETWORK
921 void *sptr = nsk->sk_security;
922 #endif
923
924 memcpy(nsk, osk, osk->sk_prot->obj_size);
925 #ifdef CONFIG_SECURITY_NETWORK
926 nsk->sk_security = sptr;
927 security_sk_clone(osk, nsk);
928 #endif
929 }
930
931 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
932 int family)
933 {
934 struct sock *sk;
935 struct kmem_cache *slab;
936
937 slab = prot->slab;
938 if (slab != NULL)
939 sk = kmem_cache_alloc(slab, priority);
940 else
941 sk = kmalloc(prot->obj_size, priority);
942
943 if (sk != NULL) {
944 if (security_sk_alloc(sk, family, priority))
945 goto out_free;
946
947 if (!try_module_get(prot->owner))
948 goto out_free_sec;
949 }
950
951 return sk;
952
953 out_free_sec:
954 security_sk_free(sk);
955 out_free:
956 if (slab != NULL)
957 kmem_cache_free(slab, sk);
958 else
959 kfree(sk);
960 return NULL;
961 }
962
963 static void sk_prot_free(struct proto *prot, struct sock *sk)
964 {
965 struct kmem_cache *slab;
966 struct module *owner;
967
968 owner = prot->owner;
969 slab = prot->slab;
970
971 security_sk_free(sk);
972 if (slab != NULL)
973 kmem_cache_free(slab, sk);
974 else
975 kfree(sk);
976 module_put(owner);
977 }
978
979 /**
980 * sk_alloc - All socket objects are allocated here
981 * @net: the applicable net namespace
982 * @family: protocol family
983 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
984 * @prot: struct proto associated with this new sock instance
985 */
986 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
987 struct proto *prot)
988 {
989 struct sock *sk;
990
991 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
992 if (sk) {
993 sk->sk_family = family;
994 /*
995 * See comment in struct sock definition to understand
996 * why we need sk_prot_creator -acme
997 */
998 sk->sk_prot = sk->sk_prot_creator = prot;
999 sock_lock_init(sk);
1000 sock_net_set(sk, get_net(net));
1001 }
1002
1003 return sk;
1004 }
1005
1006 void sk_free(struct sock *sk)
1007 {
1008 struct sk_filter *filter;
1009
1010 if (sk->sk_destruct)
1011 sk->sk_destruct(sk);
1012
1013 filter = rcu_dereference(sk->sk_filter);
1014 if (filter) {
1015 sk_filter_uncharge(sk, filter);
1016 rcu_assign_pointer(sk->sk_filter, NULL);
1017 }
1018
1019 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1020 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1021
1022 if (atomic_read(&sk->sk_omem_alloc))
1023 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1024 __func__, atomic_read(&sk->sk_omem_alloc));
1025
1026 put_net(sock_net(sk));
1027 sk_prot_free(sk->sk_prot_creator, sk);
1028 }
1029
1030 /*
1031 * Last sock_put should drop referrence to sk->sk_net. It has already
1032 * been dropped in sk_change_net. Taking referrence to stopping namespace
1033 * is not an option.
1034 * Take referrence to a socket to remove it from hash _alive_ and after that
1035 * destroy it in the context of init_net.
1036 */
1037 void sk_release_kernel(struct sock *sk)
1038 {
1039 if (sk == NULL || sk->sk_socket == NULL)
1040 return;
1041
1042 sock_hold(sk);
1043 sock_release(sk->sk_socket);
1044 release_net(sock_net(sk));
1045 sock_net_set(sk, get_net(&init_net));
1046 sock_put(sk);
1047 }
1048 EXPORT_SYMBOL(sk_release_kernel);
1049
1050 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1051 {
1052 struct sock *newsk;
1053
1054 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1055 if (newsk != NULL) {
1056 struct sk_filter *filter;
1057
1058 sock_copy(newsk, sk);
1059
1060 /* SANITY */
1061 get_net(sock_net(newsk));
1062 sk_node_init(&newsk->sk_node);
1063 sock_lock_init(newsk);
1064 bh_lock_sock(newsk);
1065 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1066
1067 atomic_set(&newsk->sk_rmem_alloc, 0);
1068 atomic_set(&newsk->sk_wmem_alloc, 0);
1069 atomic_set(&newsk->sk_omem_alloc, 0);
1070 skb_queue_head_init(&newsk->sk_receive_queue);
1071 skb_queue_head_init(&newsk->sk_write_queue);
1072 #ifdef CONFIG_NET_DMA
1073 skb_queue_head_init(&newsk->sk_async_wait_queue);
1074 #endif
1075
1076 rwlock_init(&newsk->sk_dst_lock);
1077 rwlock_init(&newsk->sk_callback_lock);
1078 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1079 af_callback_keys + newsk->sk_family,
1080 af_family_clock_key_strings[newsk->sk_family]);
1081
1082 newsk->sk_dst_cache = NULL;
1083 newsk->sk_wmem_queued = 0;
1084 newsk->sk_forward_alloc = 0;
1085 newsk->sk_send_head = NULL;
1086 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1087
1088 sock_reset_flag(newsk, SOCK_DONE);
1089 skb_queue_head_init(&newsk->sk_error_queue);
1090
1091 filter = newsk->sk_filter;
1092 if (filter != NULL)
1093 sk_filter_charge(newsk, filter);
1094
1095 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1096 /* It is still raw copy of parent, so invalidate
1097 * destructor and make plain sk_free() */
1098 newsk->sk_destruct = NULL;
1099 sk_free(newsk);
1100 newsk = NULL;
1101 goto out;
1102 }
1103
1104 newsk->sk_err = 0;
1105 newsk->sk_priority = 0;
1106 atomic_set(&newsk->sk_refcnt, 2);
1107
1108 /*
1109 * Increment the counter in the same struct proto as the master
1110 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1111 * is the same as sk->sk_prot->socks, as this field was copied
1112 * with memcpy).
1113 *
1114 * This _changes_ the previous behaviour, where
1115 * tcp_create_openreq_child always was incrementing the
1116 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1117 * to be taken into account in all callers. -acme
1118 */
1119 sk_refcnt_debug_inc(newsk);
1120 sk_set_socket(newsk, NULL);
1121 newsk->sk_sleep = NULL;
1122
1123 if (newsk->sk_prot->sockets_allocated)
1124 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1125 }
1126 out:
1127 return newsk;
1128 }
1129
1130 EXPORT_SYMBOL_GPL(sk_clone);
1131
1132 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1133 {
1134 __sk_dst_set(sk, dst);
1135 sk->sk_route_caps = dst->dev->features;
1136 if (sk->sk_route_caps & NETIF_F_GSO)
1137 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1138 if (sk_can_gso(sk)) {
1139 if (dst->header_len) {
1140 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1141 } else {
1142 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1143 sk->sk_gso_max_size = dst->dev->gso_max_size;
1144 }
1145 }
1146 }
1147 EXPORT_SYMBOL_GPL(sk_setup_caps);
1148
1149 void __init sk_init(void)
1150 {
1151 if (num_physpages <= 4096) {
1152 sysctl_wmem_max = 32767;
1153 sysctl_rmem_max = 32767;
1154 sysctl_wmem_default = 32767;
1155 sysctl_rmem_default = 32767;
1156 } else if (num_physpages >= 131072) {
1157 sysctl_wmem_max = 131071;
1158 sysctl_rmem_max = 131071;
1159 }
1160 }
1161
1162 /*
1163 * Simple resource managers for sockets.
1164 */
1165
1166
1167 /*
1168 * Write buffer destructor automatically called from kfree_skb.
1169 */
1170 void sock_wfree(struct sk_buff *skb)
1171 {
1172 struct sock *sk = skb->sk;
1173
1174 /* In case it might be waiting for more memory. */
1175 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1176 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1177 sk->sk_write_space(sk);
1178 sock_put(sk);
1179 }
1180
1181 /*
1182 * Read buffer destructor automatically called from kfree_skb.
1183 */
1184 void sock_rfree(struct sk_buff *skb)
1185 {
1186 struct sock *sk = skb->sk;
1187
1188 skb_truesize_check(skb);
1189 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1190 sk_mem_uncharge(skb->sk, skb->truesize);
1191 }
1192
1193
1194 int sock_i_uid(struct sock *sk)
1195 {
1196 int uid;
1197
1198 read_lock(&sk->sk_callback_lock);
1199 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1200 read_unlock(&sk->sk_callback_lock);
1201 return uid;
1202 }
1203
1204 unsigned long sock_i_ino(struct sock *sk)
1205 {
1206 unsigned long ino;
1207
1208 read_lock(&sk->sk_callback_lock);
1209 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1210 read_unlock(&sk->sk_callback_lock);
1211 return ino;
1212 }
1213
1214 /*
1215 * Allocate a skb from the socket's send buffer.
1216 */
1217 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1218 gfp_t priority)
1219 {
1220 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1221 struct sk_buff * skb = alloc_skb(size, priority);
1222 if (skb) {
1223 skb_set_owner_w(skb, sk);
1224 return skb;
1225 }
1226 }
1227 return NULL;
1228 }
1229
1230 /*
1231 * Allocate a skb from the socket's receive buffer.
1232 */
1233 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1234 gfp_t priority)
1235 {
1236 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1237 struct sk_buff *skb = alloc_skb(size, priority);
1238 if (skb) {
1239 skb_set_owner_r(skb, sk);
1240 return skb;
1241 }
1242 }
1243 return NULL;
1244 }
1245
1246 /*
1247 * Allocate a memory block from the socket's option memory buffer.
1248 */
1249 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1250 {
1251 if ((unsigned)size <= sysctl_optmem_max &&
1252 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1253 void *mem;
1254 /* First do the add, to avoid the race if kmalloc
1255 * might sleep.
1256 */
1257 atomic_add(size, &sk->sk_omem_alloc);
1258 mem = kmalloc(size, priority);
1259 if (mem)
1260 return mem;
1261 atomic_sub(size, &sk->sk_omem_alloc);
1262 }
1263 return NULL;
1264 }
1265
1266 /*
1267 * Free an option memory block.
1268 */
1269 void sock_kfree_s(struct sock *sk, void *mem, int size)
1270 {
1271 kfree(mem);
1272 atomic_sub(size, &sk->sk_omem_alloc);
1273 }
1274
1275 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1276 I think, these locks should be removed for datagram sockets.
1277 */
1278 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1279 {
1280 DEFINE_WAIT(wait);
1281
1282 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1283 for (;;) {
1284 if (!timeo)
1285 break;
1286 if (signal_pending(current))
1287 break;
1288 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1289 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1290 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1291 break;
1292 if (sk->sk_shutdown & SEND_SHUTDOWN)
1293 break;
1294 if (sk->sk_err)
1295 break;
1296 timeo = schedule_timeout(timeo);
1297 }
1298 finish_wait(sk->sk_sleep, &wait);
1299 return timeo;
1300 }
1301
1302
1303 /*
1304 * Generic send/receive buffer handlers
1305 */
1306
1307 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1308 unsigned long data_len, int noblock,
1309 int *errcode)
1310 {
1311 struct sk_buff *skb;
1312 gfp_t gfp_mask;
1313 long timeo;
1314 int err;
1315
1316 gfp_mask = sk->sk_allocation;
1317 if (gfp_mask & __GFP_WAIT)
1318 gfp_mask |= __GFP_REPEAT;
1319
1320 timeo = sock_sndtimeo(sk, noblock);
1321 while (1) {
1322 err = sock_error(sk);
1323 if (err != 0)
1324 goto failure;
1325
1326 err = -EPIPE;
1327 if (sk->sk_shutdown & SEND_SHUTDOWN)
1328 goto failure;
1329
1330 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1331 skb = alloc_skb(header_len, gfp_mask);
1332 if (skb) {
1333 int npages;
1334 int i;
1335
1336 /* No pages, we're done... */
1337 if (!data_len)
1338 break;
1339
1340 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1341 skb->truesize += data_len;
1342 skb_shinfo(skb)->nr_frags = npages;
1343 for (i = 0; i < npages; i++) {
1344 struct page *page;
1345 skb_frag_t *frag;
1346
1347 page = alloc_pages(sk->sk_allocation, 0);
1348 if (!page) {
1349 err = -ENOBUFS;
1350 skb_shinfo(skb)->nr_frags = i;
1351 kfree_skb(skb);
1352 goto failure;
1353 }
1354
1355 frag = &skb_shinfo(skb)->frags[i];
1356 frag->page = page;
1357 frag->page_offset = 0;
1358 frag->size = (data_len >= PAGE_SIZE ?
1359 PAGE_SIZE :
1360 data_len);
1361 data_len -= PAGE_SIZE;
1362 }
1363
1364 /* Full success... */
1365 break;
1366 }
1367 err = -ENOBUFS;
1368 goto failure;
1369 }
1370 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1371 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1372 err = -EAGAIN;
1373 if (!timeo)
1374 goto failure;
1375 if (signal_pending(current))
1376 goto interrupted;
1377 timeo = sock_wait_for_wmem(sk, timeo);
1378 }
1379
1380 skb_set_owner_w(skb, sk);
1381 return skb;
1382
1383 interrupted:
1384 err = sock_intr_errno(timeo);
1385 failure:
1386 *errcode = err;
1387 return NULL;
1388 }
1389 EXPORT_SYMBOL(sock_alloc_send_pskb);
1390
1391 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1392 int noblock, int *errcode)
1393 {
1394 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1395 }
1396
1397 static void __lock_sock(struct sock *sk)
1398 {
1399 DEFINE_WAIT(wait);
1400
1401 for (;;) {
1402 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1403 TASK_UNINTERRUPTIBLE);
1404 spin_unlock_bh(&sk->sk_lock.slock);
1405 schedule();
1406 spin_lock_bh(&sk->sk_lock.slock);
1407 if (!sock_owned_by_user(sk))
1408 break;
1409 }
1410 finish_wait(&sk->sk_lock.wq, &wait);
1411 }
1412
1413 static void __release_sock(struct sock *sk)
1414 {
1415 struct sk_buff *skb = sk->sk_backlog.head;
1416
1417 do {
1418 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1419 bh_unlock_sock(sk);
1420
1421 do {
1422 struct sk_buff *next = skb->next;
1423
1424 skb->next = NULL;
1425 sk_backlog_rcv(sk, skb);
1426
1427 /*
1428 * We are in process context here with softirqs
1429 * disabled, use cond_resched_softirq() to preempt.
1430 * This is safe to do because we've taken the backlog
1431 * queue private:
1432 */
1433 cond_resched_softirq();
1434
1435 skb = next;
1436 } while (skb != NULL);
1437
1438 bh_lock_sock(sk);
1439 } while ((skb = sk->sk_backlog.head) != NULL);
1440 }
1441
1442 /**
1443 * sk_wait_data - wait for data to arrive at sk_receive_queue
1444 * @sk: sock to wait on
1445 * @timeo: for how long
1446 *
1447 * Now socket state including sk->sk_err is changed only under lock,
1448 * hence we may omit checks after joining wait queue.
1449 * We check receive queue before schedule() only as optimization;
1450 * it is very likely that release_sock() added new data.
1451 */
1452 int sk_wait_data(struct sock *sk, long *timeo)
1453 {
1454 int rc;
1455 DEFINE_WAIT(wait);
1456
1457 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1458 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1459 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1460 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1461 finish_wait(sk->sk_sleep, &wait);
1462 return rc;
1463 }
1464
1465 EXPORT_SYMBOL(sk_wait_data);
1466
1467 /**
1468 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1469 * @sk: socket
1470 * @size: memory size to allocate
1471 * @kind: allocation type
1472 *
1473 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1474 * rmem allocation. This function assumes that protocols which have
1475 * memory_pressure use sk_wmem_queued as write buffer accounting.
1476 */
1477 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1478 {
1479 struct proto *prot = sk->sk_prot;
1480 int amt = sk_mem_pages(size);
1481 int allocated;
1482
1483 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1484 allocated = atomic_add_return(amt, prot->memory_allocated);
1485
1486 /* Under limit. */
1487 if (allocated <= prot->sysctl_mem[0]) {
1488 if (prot->memory_pressure && *prot->memory_pressure)
1489 *prot->memory_pressure = 0;
1490 return 1;
1491 }
1492
1493 /* Under pressure. */
1494 if (allocated > prot->sysctl_mem[1])
1495 if (prot->enter_memory_pressure)
1496 prot->enter_memory_pressure(sk);
1497
1498 /* Over hard limit. */
1499 if (allocated > prot->sysctl_mem[2])
1500 goto suppress_allocation;
1501
1502 /* guarantee minimum buffer size under pressure */
1503 if (kind == SK_MEM_RECV) {
1504 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1505 return 1;
1506 } else { /* SK_MEM_SEND */
1507 if (sk->sk_type == SOCK_STREAM) {
1508 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1509 return 1;
1510 } else if (atomic_read(&sk->sk_wmem_alloc) <
1511 prot->sysctl_wmem[0])
1512 return 1;
1513 }
1514
1515 if (prot->memory_pressure) {
1516 int alloc;
1517
1518 if (!*prot->memory_pressure)
1519 return 1;
1520 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1521 if (prot->sysctl_mem[2] > alloc *
1522 sk_mem_pages(sk->sk_wmem_queued +
1523 atomic_read(&sk->sk_rmem_alloc) +
1524 sk->sk_forward_alloc))
1525 return 1;
1526 }
1527
1528 suppress_allocation:
1529
1530 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1531 sk_stream_moderate_sndbuf(sk);
1532
1533 /* Fail only if socket is _under_ its sndbuf.
1534 * In this case we cannot block, so that we have to fail.
1535 */
1536 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1537 return 1;
1538 }
1539
1540 /* Alas. Undo changes. */
1541 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1542 atomic_sub(amt, prot->memory_allocated);
1543 return 0;
1544 }
1545
1546 EXPORT_SYMBOL(__sk_mem_schedule);
1547
1548 /**
1549 * __sk_reclaim - reclaim memory_allocated
1550 * @sk: socket
1551 */
1552 void __sk_mem_reclaim(struct sock *sk)
1553 {
1554 struct proto *prot = sk->sk_prot;
1555
1556 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1557 prot->memory_allocated);
1558 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1559
1560 if (prot->memory_pressure && *prot->memory_pressure &&
1561 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1562 *prot->memory_pressure = 0;
1563 }
1564
1565 EXPORT_SYMBOL(__sk_mem_reclaim);
1566
1567
1568 /*
1569 * Set of default routines for initialising struct proto_ops when
1570 * the protocol does not support a particular function. In certain
1571 * cases where it makes no sense for a protocol to have a "do nothing"
1572 * function, some default processing is provided.
1573 */
1574
1575 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1576 {
1577 return -EOPNOTSUPP;
1578 }
1579
1580 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1581 int len, int flags)
1582 {
1583 return -EOPNOTSUPP;
1584 }
1585
1586 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1587 {
1588 return -EOPNOTSUPP;
1589 }
1590
1591 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1592 {
1593 return -EOPNOTSUPP;
1594 }
1595
1596 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1597 int *len, int peer)
1598 {
1599 return -EOPNOTSUPP;
1600 }
1601
1602 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1603 {
1604 return 0;
1605 }
1606
1607 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1608 {
1609 return -EOPNOTSUPP;
1610 }
1611
1612 int sock_no_listen(struct socket *sock, int backlog)
1613 {
1614 return -EOPNOTSUPP;
1615 }
1616
1617 int sock_no_shutdown(struct socket *sock, int how)
1618 {
1619 return -EOPNOTSUPP;
1620 }
1621
1622 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1623 char __user *optval, int optlen)
1624 {
1625 return -EOPNOTSUPP;
1626 }
1627
1628 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1629 char __user *optval, int __user *optlen)
1630 {
1631 return -EOPNOTSUPP;
1632 }
1633
1634 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1635 size_t len)
1636 {
1637 return -EOPNOTSUPP;
1638 }
1639
1640 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1641 size_t len, int flags)
1642 {
1643 return -EOPNOTSUPP;
1644 }
1645
1646 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1647 {
1648 /* Mirror missing mmap method error code */
1649 return -ENODEV;
1650 }
1651
1652 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1653 {
1654 ssize_t res;
1655 struct msghdr msg = {.msg_flags = flags};
1656 struct kvec iov;
1657 char *kaddr = kmap(page);
1658 iov.iov_base = kaddr + offset;
1659 iov.iov_len = size;
1660 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1661 kunmap(page);
1662 return res;
1663 }
1664
1665 /*
1666 * Default Socket Callbacks
1667 */
1668
1669 static void sock_def_wakeup(struct sock *sk)
1670 {
1671 read_lock(&sk->sk_callback_lock);
1672 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1673 wake_up_interruptible_all(sk->sk_sleep);
1674 read_unlock(&sk->sk_callback_lock);
1675 }
1676
1677 static void sock_def_error_report(struct sock *sk)
1678 {
1679 read_lock(&sk->sk_callback_lock);
1680 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1681 wake_up_interruptible(sk->sk_sleep);
1682 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1683 read_unlock(&sk->sk_callback_lock);
1684 }
1685
1686 static void sock_def_readable(struct sock *sk, int len)
1687 {
1688 read_lock(&sk->sk_callback_lock);
1689 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1690 wake_up_interruptible_sync(sk->sk_sleep);
1691 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1692 read_unlock(&sk->sk_callback_lock);
1693 }
1694
1695 static void sock_def_write_space(struct sock *sk)
1696 {
1697 read_lock(&sk->sk_callback_lock);
1698
1699 /* Do not wake up a writer until he can make "significant"
1700 * progress. --DaveM
1701 */
1702 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1703 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1704 wake_up_interruptible_sync(sk->sk_sleep);
1705
1706 /* Should agree with poll, otherwise some programs break */
1707 if (sock_writeable(sk))
1708 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1709 }
1710
1711 read_unlock(&sk->sk_callback_lock);
1712 }
1713
1714 static void sock_def_destruct(struct sock *sk)
1715 {
1716 kfree(sk->sk_protinfo);
1717 }
1718
1719 void sk_send_sigurg(struct sock *sk)
1720 {
1721 if (sk->sk_socket && sk->sk_socket->file)
1722 if (send_sigurg(&sk->sk_socket->file->f_owner))
1723 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1724 }
1725
1726 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1727 unsigned long expires)
1728 {
1729 if (!mod_timer(timer, expires))
1730 sock_hold(sk);
1731 }
1732
1733 EXPORT_SYMBOL(sk_reset_timer);
1734
1735 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1736 {
1737 if (timer_pending(timer) && del_timer(timer))
1738 __sock_put(sk);
1739 }
1740
1741 EXPORT_SYMBOL(sk_stop_timer);
1742
1743 void sock_init_data(struct socket *sock, struct sock *sk)
1744 {
1745 skb_queue_head_init(&sk->sk_receive_queue);
1746 skb_queue_head_init(&sk->sk_write_queue);
1747 skb_queue_head_init(&sk->sk_error_queue);
1748 #ifdef CONFIG_NET_DMA
1749 skb_queue_head_init(&sk->sk_async_wait_queue);
1750 #endif
1751
1752 sk->sk_send_head = NULL;
1753
1754 init_timer(&sk->sk_timer);
1755
1756 sk->sk_allocation = GFP_KERNEL;
1757 sk->sk_rcvbuf = sysctl_rmem_default;
1758 sk->sk_sndbuf = sysctl_wmem_default;
1759 sk->sk_state = TCP_CLOSE;
1760 sk_set_socket(sk, sock);
1761
1762 sock_set_flag(sk, SOCK_ZAPPED);
1763
1764 if (sock) {
1765 sk->sk_type = sock->type;
1766 sk->sk_sleep = &sock->wait;
1767 sock->sk = sk;
1768 } else
1769 sk->sk_sleep = NULL;
1770
1771 rwlock_init(&sk->sk_dst_lock);
1772 rwlock_init(&sk->sk_callback_lock);
1773 lockdep_set_class_and_name(&sk->sk_callback_lock,
1774 af_callback_keys + sk->sk_family,
1775 af_family_clock_key_strings[sk->sk_family]);
1776
1777 sk->sk_state_change = sock_def_wakeup;
1778 sk->sk_data_ready = sock_def_readable;
1779 sk->sk_write_space = sock_def_write_space;
1780 sk->sk_error_report = sock_def_error_report;
1781 sk->sk_destruct = sock_def_destruct;
1782
1783 sk->sk_sndmsg_page = NULL;
1784 sk->sk_sndmsg_off = 0;
1785
1786 sk->sk_peercred.pid = 0;
1787 sk->sk_peercred.uid = -1;
1788 sk->sk_peercred.gid = -1;
1789 sk->sk_write_pending = 0;
1790 sk->sk_rcvlowat = 1;
1791 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1792 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1793
1794 sk->sk_stamp = ktime_set(-1L, 0);
1795
1796 atomic_set(&sk->sk_refcnt, 1);
1797 atomic_set(&sk->sk_drops, 0);
1798 }
1799
1800 void lock_sock_nested(struct sock *sk, int subclass)
1801 {
1802 might_sleep();
1803 spin_lock_bh(&sk->sk_lock.slock);
1804 if (sk->sk_lock.owned)
1805 __lock_sock(sk);
1806 sk->sk_lock.owned = 1;
1807 spin_unlock(&sk->sk_lock.slock);
1808 /*
1809 * The sk_lock has mutex_lock() semantics here:
1810 */
1811 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1812 local_bh_enable();
1813 }
1814
1815 EXPORT_SYMBOL(lock_sock_nested);
1816
1817 void release_sock(struct sock *sk)
1818 {
1819 /*
1820 * The sk_lock has mutex_unlock() semantics:
1821 */
1822 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1823
1824 spin_lock_bh(&sk->sk_lock.slock);
1825 if (sk->sk_backlog.tail)
1826 __release_sock(sk);
1827 sk->sk_lock.owned = 0;
1828 if (waitqueue_active(&sk->sk_lock.wq))
1829 wake_up(&sk->sk_lock.wq);
1830 spin_unlock_bh(&sk->sk_lock.slock);
1831 }
1832 EXPORT_SYMBOL(release_sock);
1833
1834 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1835 {
1836 struct timeval tv;
1837 if (!sock_flag(sk, SOCK_TIMESTAMP))
1838 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1839 tv = ktime_to_timeval(sk->sk_stamp);
1840 if (tv.tv_sec == -1)
1841 return -ENOENT;
1842 if (tv.tv_sec == 0) {
1843 sk->sk_stamp = ktime_get_real();
1844 tv = ktime_to_timeval(sk->sk_stamp);
1845 }
1846 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1847 }
1848 EXPORT_SYMBOL(sock_get_timestamp);
1849
1850 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1851 {
1852 struct timespec ts;
1853 if (!sock_flag(sk, SOCK_TIMESTAMP))
1854 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1855 ts = ktime_to_timespec(sk->sk_stamp);
1856 if (ts.tv_sec == -1)
1857 return -ENOENT;
1858 if (ts.tv_sec == 0) {
1859 sk->sk_stamp = ktime_get_real();
1860 ts = ktime_to_timespec(sk->sk_stamp);
1861 }
1862 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1863 }
1864 EXPORT_SYMBOL(sock_get_timestampns);
1865
1866 void sock_enable_timestamp(struct sock *sk, int flag)
1867 {
1868 if (!sock_flag(sk, flag)) {
1869 sock_set_flag(sk, flag);
1870 /*
1871 * we just set one of the two flags which require net
1872 * time stamping, but time stamping might have been on
1873 * already because of the other one
1874 */
1875 if (!sock_flag(sk,
1876 flag == SOCK_TIMESTAMP ?
1877 SOCK_TIMESTAMPING_RX_SOFTWARE :
1878 SOCK_TIMESTAMP))
1879 net_enable_timestamp();
1880 }
1881 }
1882
1883 /*
1884 * Get a socket option on an socket.
1885 *
1886 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1887 * asynchronous errors should be reported by getsockopt. We assume
1888 * this means if you specify SO_ERROR (otherwise whats the point of it).
1889 */
1890 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1891 char __user *optval, int __user *optlen)
1892 {
1893 struct sock *sk = sock->sk;
1894
1895 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1896 }
1897
1898 EXPORT_SYMBOL(sock_common_getsockopt);
1899
1900 #ifdef CONFIG_COMPAT
1901 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1902 char __user *optval, int __user *optlen)
1903 {
1904 struct sock *sk = sock->sk;
1905
1906 if (sk->sk_prot->compat_getsockopt != NULL)
1907 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1908 optval, optlen);
1909 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1910 }
1911 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1912 #endif
1913
1914 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1915 struct msghdr *msg, size_t size, int flags)
1916 {
1917 struct sock *sk = sock->sk;
1918 int addr_len = 0;
1919 int err;
1920
1921 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1922 flags & ~MSG_DONTWAIT, &addr_len);
1923 if (err >= 0)
1924 msg->msg_namelen = addr_len;
1925 return err;
1926 }
1927
1928 EXPORT_SYMBOL(sock_common_recvmsg);
1929
1930 /*
1931 * Set socket options on an inet socket.
1932 */
1933 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1934 char __user *optval, int optlen)
1935 {
1936 struct sock *sk = sock->sk;
1937
1938 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1939 }
1940
1941 EXPORT_SYMBOL(sock_common_setsockopt);
1942
1943 #ifdef CONFIG_COMPAT
1944 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1945 char __user *optval, int optlen)
1946 {
1947 struct sock *sk = sock->sk;
1948
1949 if (sk->sk_prot->compat_setsockopt != NULL)
1950 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1951 optval, optlen);
1952 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1953 }
1954 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1955 #endif
1956
1957 void sk_common_release(struct sock *sk)
1958 {
1959 if (sk->sk_prot->destroy)
1960 sk->sk_prot->destroy(sk);
1961
1962 /*
1963 * Observation: when sock_common_release is called, processes have
1964 * no access to socket. But net still has.
1965 * Step one, detach it from networking:
1966 *
1967 * A. Remove from hash tables.
1968 */
1969
1970 sk->sk_prot->unhash(sk);
1971
1972 /*
1973 * In this point socket cannot receive new packets, but it is possible
1974 * that some packets are in flight because some CPU runs receiver and
1975 * did hash table lookup before we unhashed socket. They will achieve
1976 * receive queue and will be purged by socket destructor.
1977 *
1978 * Also we still have packets pending on receive queue and probably,
1979 * our own packets waiting in device queues. sock_destroy will drain
1980 * receive queue, but transmitted packets will delay socket destruction
1981 * until the last reference will be released.
1982 */
1983
1984 sock_orphan(sk);
1985
1986 xfrm_sk_free_policy(sk);
1987
1988 sk_refcnt_debug_release(sk);
1989 sock_put(sk);
1990 }
1991
1992 EXPORT_SYMBOL(sk_common_release);
1993
1994 static DEFINE_RWLOCK(proto_list_lock);
1995 static LIST_HEAD(proto_list);
1996
1997 #ifdef CONFIG_PROC_FS
1998 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
1999 struct prot_inuse {
2000 int val[PROTO_INUSE_NR];
2001 };
2002
2003 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2004
2005 #ifdef CONFIG_NET_NS
2006 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2007 {
2008 int cpu = smp_processor_id();
2009 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2010 }
2011 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2012
2013 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2014 {
2015 int cpu, idx = prot->inuse_idx;
2016 int res = 0;
2017
2018 for_each_possible_cpu(cpu)
2019 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2020
2021 return res >= 0 ? res : 0;
2022 }
2023 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2024
2025 static int sock_inuse_init_net(struct net *net)
2026 {
2027 net->core.inuse = alloc_percpu(struct prot_inuse);
2028 return net->core.inuse ? 0 : -ENOMEM;
2029 }
2030
2031 static void sock_inuse_exit_net(struct net *net)
2032 {
2033 free_percpu(net->core.inuse);
2034 }
2035
2036 static struct pernet_operations net_inuse_ops = {
2037 .init = sock_inuse_init_net,
2038 .exit = sock_inuse_exit_net,
2039 };
2040
2041 static __init int net_inuse_init(void)
2042 {
2043 if (register_pernet_subsys(&net_inuse_ops))
2044 panic("Cannot initialize net inuse counters");
2045
2046 return 0;
2047 }
2048
2049 core_initcall(net_inuse_init);
2050 #else
2051 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2052
2053 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2054 {
2055 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2056 }
2057 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2058
2059 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2060 {
2061 int cpu, idx = prot->inuse_idx;
2062 int res = 0;
2063
2064 for_each_possible_cpu(cpu)
2065 res += per_cpu(prot_inuse, cpu).val[idx];
2066
2067 return res >= 0 ? res : 0;
2068 }
2069 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2070 #endif
2071
2072 static void assign_proto_idx(struct proto *prot)
2073 {
2074 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2075
2076 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2077 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2078 return;
2079 }
2080
2081 set_bit(prot->inuse_idx, proto_inuse_idx);
2082 }
2083
2084 static void release_proto_idx(struct proto *prot)
2085 {
2086 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2087 clear_bit(prot->inuse_idx, proto_inuse_idx);
2088 }
2089 #else
2090 static inline void assign_proto_idx(struct proto *prot)
2091 {
2092 }
2093
2094 static inline void release_proto_idx(struct proto *prot)
2095 {
2096 }
2097 #endif
2098
2099 int proto_register(struct proto *prot, int alloc_slab)
2100 {
2101 if (alloc_slab) {
2102 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2103 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2104 NULL);
2105
2106 if (prot->slab == NULL) {
2107 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2108 prot->name);
2109 goto out;
2110 }
2111
2112 if (prot->rsk_prot != NULL) {
2113 static const char mask[] = "request_sock_%s";
2114
2115 prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2116 if (prot->rsk_prot->slab_name == NULL)
2117 goto out_free_sock_slab;
2118
2119 sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2120 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2121 prot->rsk_prot->obj_size, 0,
2122 SLAB_HWCACHE_ALIGN, NULL);
2123
2124 if (prot->rsk_prot->slab == NULL) {
2125 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2126 prot->name);
2127 goto out_free_request_sock_slab_name;
2128 }
2129 }
2130
2131 if (prot->twsk_prot != NULL) {
2132 static const char mask[] = "tw_sock_%s";
2133
2134 prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2135
2136 if (prot->twsk_prot->twsk_slab_name == NULL)
2137 goto out_free_request_sock_slab;
2138
2139 sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2140 prot->twsk_prot->twsk_slab =
2141 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2142 prot->twsk_prot->twsk_obj_size,
2143 0,
2144 SLAB_HWCACHE_ALIGN |
2145 prot->slab_flags,
2146 NULL);
2147 if (prot->twsk_prot->twsk_slab == NULL)
2148 goto out_free_timewait_sock_slab_name;
2149 }
2150 }
2151
2152 write_lock(&proto_list_lock);
2153 list_add(&prot->node, &proto_list);
2154 assign_proto_idx(prot);
2155 write_unlock(&proto_list_lock);
2156 return 0;
2157
2158 out_free_timewait_sock_slab_name:
2159 kfree(prot->twsk_prot->twsk_slab_name);
2160 out_free_request_sock_slab:
2161 if (prot->rsk_prot && prot->rsk_prot->slab) {
2162 kmem_cache_destroy(prot->rsk_prot->slab);
2163 prot->rsk_prot->slab = NULL;
2164 }
2165 out_free_request_sock_slab_name:
2166 kfree(prot->rsk_prot->slab_name);
2167 out_free_sock_slab:
2168 kmem_cache_destroy(prot->slab);
2169 prot->slab = NULL;
2170 out:
2171 return -ENOBUFS;
2172 }
2173
2174 EXPORT_SYMBOL(proto_register);
2175
2176 void proto_unregister(struct proto *prot)
2177 {
2178 write_lock(&proto_list_lock);
2179 release_proto_idx(prot);
2180 list_del(&prot->node);
2181 write_unlock(&proto_list_lock);
2182
2183 if (prot->slab != NULL) {
2184 kmem_cache_destroy(prot->slab);
2185 prot->slab = NULL;
2186 }
2187
2188 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2189 kmem_cache_destroy(prot->rsk_prot->slab);
2190 kfree(prot->rsk_prot->slab_name);
2191 prot->rsk_prot->slab = NULL;
2192 }
2193
2194 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2195 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2196 kfree(prot->twsk_prot->twsk_slab_name);
2197 prot->twsk_prot->twsk_slab = NULL;
2198 }
2199 }
2200
2201 EXPORT_SYMBOL(proto_unregister);
2202
2203 #ifdef CONFIG_PROC_FS
2204 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2205 __acquires(proto_list_lock)
2206 {
2207 read_lock(&proto_list_lock);
2208 return seq_list_start_head(&proto_list, *pos);
2209 }
2210
2211 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2212 {
2213 return seq_list_next(v, &proto_list, pos);
2214 }
2215
2216 static void proto_seq_stop(struct seq_file *seq, void *v)
2217 __releases(proto_list_lock)
2218 {
2219 read_unlock(&proto_list_lock);
2220 }
2221
2222 static char proto_method_implemented(const void *method)
2223 {
2224 return method == NULL ? 'n' : 'y';
2225 }
2226
2227 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2228 {
2229 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2230 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2231 proto->name,
2232 proto->obj_size,
2233 sock_prot_inuse_get(seq_file_net(seq), proto),
2234 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2235 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2236 proto->max_header,
2237 proto->slab == NULL ? "no" : "yes",
2238 module_name(proto->owner),
2239 proto_method_implemented(proto->close),
2240 proto_method_implemented(proto->connect),
2241 proto_method_implemented(proto->disconnect),
2242 proto_method_implemented(proto->accept),
2243 proto_method_implemented(proto->ioctl),
2244 proto_method_implemented(proto->init),
2245 proto_method_implemented(proto->destroy),
2246 proto_method_implemented(proto->shutdown),
2247 proto_method_implemented(proto->setsockopt),
2248 proto_method_implemented(proto->getsockopt),
2249 proto_method_implemented(proto->sendmsg),
2250 proto_method_implemented(proto->recvmsg),
2251 proto_method_implemented(proto->sendpage),
2252 proto_method_implemented(proto->bind),
2253 proto_method_implemented(proto->backlog_rcv),
2254 proto_method_implemented(proto->hash),
2255 proto_method_implemented(proto->unhash),
2256 proto_method_implemented(proto->get_port),
2257 proto_method_implemented(proto->enter_memory_pressure));
2258 }
2259
2260 static int proto_seq_show(struct seq_file *seq, void *v)
2261 {
2262 if (v == &proto_list)
2263 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2264 "protocol",
2265 "size",
2266 "sockets",
2267 "memory",
2268 "press",
2269 "maxhdr",
2270 "slab",
2271 "module",
2272 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2273 else
2274 proto_seq_printf(seq, list_entry(v, struct proto, node));
2275 return 0;
2276 }
2277
2278 static const struct seq_operations proto_seq_ops = {
2279 .start = proto_seq_start,
2280 .next = proto_seq_next,
2281 .stop = proto_seq_stop,
2282 .show = proto_seq_show,
2283 };
2284
2285 static int proto_seq_open(struct inode *inode, struct file *file)
2286 {
2287 return seq_open_net(inode, file, &proto_seq_ops,
2288 sizeof(struct seq_net_private));
2289 }
2290
2291 static const struct file_operations proto_seq_fops = {
2292 .owner = THIS_MODULE,
2293 .open = proto_seq_open,
2294 .read = seq_read,
2295 .llseek = seq_lseek,
2296 .release = seq_release_net,
2297 };
2298
2299 static __net_init int proto_init_net(struct net *net)
2300 {
2301 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2302 return -ENOMEM;
2303
2304 return 0;
2305 }
2306
2307 static __net_exit void proto_exit_net(struct net *net)
2308 {
2309 proc_net_remove(net, "protocols");
2310 }
2311
2312
2313 static __net_initdata struct pernet_operations proto_net_ops = {
2314 .init = proto_init_net,
2315 .exit = proto_exit_net,
2316 };
2317
2318 static int __init proto_init(void)
2319 {
2320 return register_pernet_subsys(&proto_net_ops);
2321 }
2322
2323 subsys_initcall(proto_init);
2324
2325 #endif /* PROC_FS */
2326
2327 EXPORT_SYMBOL(sk_alloc);
2328 EXPORT_SYMBOL(sk_free);
2329 EXPORT_SYMBOL(sk_send_sigurg);
2330 EXPORT_SYMBOL(sock_alloc_send_skb);
2331 EXPORT_SYMBOL(sock_init_data);
2332 EXPORT_SYMBOL(sock_kfree_s);
2333 EXPORT_SYMBOL(sock_kmalloc);
2334 EXPORT_SYMBOL(sock_no_accept);
2335 EXPORT_SYMBOL(sock_no_bind);
2336 EXPORT_SYMBOL(sock_no_connect);
2337 EXPORT_SYMBOL(sock_no_getname);
2338 EXPORT_SYMBOL(sock_no_getsockopt);
2339 EXPORT_SYMBOL(sock_no_ioctl);
2340 EXPORT_SYMBOL(sock_no_listen);
2341 EXPORT_SYMBOL(sock_no_mmap);
2342 EXPORT_SYMBOL(sock_no_poll);
2343 EXPORT_SYMBOL(sock_no_recvmsg);
2344 EXPORT_SYMBOL(sock_no_sendmsg);
2345 EXPORT_SYMBOL(sock_no_sendpage);
2346 EXPORT_SYMBOL(sock_no_setsockopt);
2347 EXPORT_SYMBOL(sock_no_shutdown);
2348 EXPORT_SYMBOL(sock_no_socketpair);
2349 EXPORT_SYMBOL(sock_rfree);
2350 EXPORT_SYMBOL(sock_setsockopt);
2351 EXPORT_SYMBOL(sock_wfree);
2352 EXPORT_SYMBOL(sock_wmalloc);
2353 EXPORT_SYMBOL(sock_i_uid);
2354 EXPORT_SYMBOL(sock_i_ino);
2355 EXPORT_SYMBOL(sysctl_optmem_max);
Linux kernel - Deliverables
This page took 0.100606 seconds and 6 git commands to generate.