projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: sk_add_backlog() take rmem_alloc into account
[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 *const 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 *const 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 *const 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;
278 int skb_len;
279 unsigned long flags;
280 struct sk_buff_head *list = &sk->sk_receive_queue;
281
282 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
283 number of warnings when compiling with -W --ANK
284 */
285 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
286 (unsigned)sk->sk_rcvbuf) {
287 atomic_inc(&sk->sk_drops);
288 return -ENOMEM;
289 }
290
291 err = sk_filter(sk, skb);
292 if (err)
293 return err;
294
295 if (!sk_rmem_schedule(sk, skb->truesize)) {
296 atomic_inc(&sk->sk_drops);
297 return -ENOBUFS;
298 }
299
300 skb->dev = NULL;
301 skb_set_owner_r(skb, sk);
302
303 /* Cache the SKB length before we tack it onto the receive
304 * queue. Once it is added it no longer belongs to us and
305 * may be freed by other threads of control pulling packets
306 * from the queue.
307 */
308 skb_len = skb->len;
309
310 spin_lock_irqsave(&list->lock, flags);
311 skb->dropcount = atomic_read(&sk->sk_drops);
312 __skb_queue_tail(list, skb);
313 spin_unlock_irqrestore(&list->lock, flags);
314
315 if (!sock_flag(sk, SOCK_DEAD))
316 sk->sk_data_ready(sk, skb_len);
317 return 0;
318 }
319 EXPORT_SYMBOL(sock_queue_rcv_skb);
320
321 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
322 {
323 int rc = NET_RX_SUCCESS;
324
325 if (sk_filter(sk, skb))
326 goto discard_and_relse;
327
328 skb->dev = NULL;
329
330 if (sk_rcvqueues_full(sk, skb)) {
331 atomic_inc(&sk->sk_drops);
332 goto discard_and_relse;
333 }
334 if (nested)
335 bh_lock_sock_nested(sk);
336 else
337 bh_lock_sock(sk);
338 if (!sock_owned_by_user(sk)) {
339 /*
340 * trylock + unlock semantics:
341 */
342 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
343
344 rc = sk_backlog_rcv(sk, skb);
345
346 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
347 } else if (sk_add_backlog(sk, skb)) {
348 bh_unlock_sock(sk);
349 atomic_inc(&sk->sk_drops);
350 goto discard_and_relse;
351 }
352
353 bh_unlock_sock(sk);
354 out:
355 sock_put(sk);
356 return rc;
357 discard_and_relse:
358 kfree_skb(skb);
359 goto out;
360 }
361 EXPORT_SYMBOL(sk_receive_skb);
362
363 void sk_reset_txq(struct sock *sk)
364 {
365 sk_tx_queue_clear(sk);
366 }
367 EXPORT_SYMBOL(sk_reset_txq);
368
369 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
370 {
371 struct dst_entry *dst = __sk_dst_get(sk);
372
373 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
374 sk_tx_queue_clear(sk);
375 rcu_assign_pointer(sk->sk_dst_cache, NULL);
376 dst_release(dst);
377 return NULL;
378 }
379
380 return dst;
381 }
382 EXPORT_SYMBOL(__sk_dst_check);
383
384 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
385 {
386 struct dst_entry *dst = sk_dst_get(sk);
387
388 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
389 sk_dst_reset(sk);
390 dst_release(dst);
391 return NULL;
392 }
393
394 return dst;
395 }
396 EXPORT_SYMBOL(sk_dst_check);
397
398 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
399 {
400 int ret = -ENOPROTOOPT;
401 #ifdef CONFIG_NETDEVICES
402 struct net *net = sock_net(sk);
403 char devname[IFNAMSIZ];
404 int index;
405
406 /* Sorry... */
407 ret = -EPERM;
408 if (!capable(CAP_NET_RAW))
409 goto out;
410
411 ret = -EINVAL;
412 if (optlen < 0)
413 goto out;
414
415 /* Bind this socket to a particular device like "eth0",
416 * as specified in the passed interface name. If the
417 * name is "" or the option length is zero the socket
418 * is not bound.
419 */
420 if (optlen > IFNAMSIZ - 1)
421 optlen = IFNAMSIZ - 1;
422 memset(devname, 0, sizeof(devname));
423
424 ret = -EFAULT;
425 if (copy_from_user(devname, optval, optlen))
426 goto out;
427
428 index = 0;
429 if (devname[0] != '\0') {
430 struct net_device *dev;
431
432 rcu_read_lock();
433 dev = dev_get_by_name_rcu(net, devname);
434 if (dev)
435 index = dev->ifindex;
436 rcu_read_unlock();
437 ret = -ENODEV;
438 if (!dev)
439 goto out;
440 }
441
442 lock_sock(sk);
443 sk->sk_bound_dev_if = index;
444 sk_dst_reset(sk);
445 release_sock(sk);
446
447 ret = 0;
448
449 out:
450 #endif
451
452 return ret;
453 }
454
455 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
456 {
457 if (valbool)
458 sock_set_flag(sk, bit);
459 else
460 sock_reset_flag(sk, bit);
461 }
462
463 /*
464 * This is meant for all protocols to use and covers goings on
465 * at the socket level. Everything here is generic.
466 */
467
468 int sock_setsockopt(struct socket *sock, int level, int optname,
469 char __user *optval, unsigned int optlen)
470 {
471 struct sock *sk = sock->sk;
472 int val;
473 int valbool;
474 struct linger ling;
475 int ret = 0;
476
477 /*
478 * Options without arguments
479 */
480
481 if (optname == SO_BINDTODEVICE)
482 return sock_bindtodevice(sk, optval, optlen);
483
484 if (optlen < sizeof(int))
485 return -EINVAL;
486
487 if (get_user(val, (int __user *)optval))
488 return -EFAULT;
489
490 valbool = val ? 1 : 0;
491
492 lock_sock(sk);
493
494 switch (optname) {
495 case SO_DEBUG:
496 if (val && !capable(CAP_NET_ADMIN))
497 ret = -EACCES;
498 else
499 sock_valbool_flag(sk, SOCK_DBG, valbool);
500 break;
501 case SO_REUSEADDR:
502 sk->sk_reuse = valbool;
503 break;
504 case SO_TYPE:
505 case SO_PROTOCOL:
506 case SO_DOMAIN:
507 case SO_ERROR:
508 ret = -ENOPROTOOPT;
509 break;
510 case SO_DONTROUTE:
511 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
512 break;
513 case SO_BROADCAST:
514 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
515 break;
516 case SO_SNDBUF:
517 /* Don't error on this BSD doesn't and if you think
518 about it this is right. Otherwise apps have to
519 play 'guess the biggest size' games. RCVBUF/SNDBUF
520 are treated in BSD as hints */
521
522 if (val > sysctl_wmem_max)
523 val = sysctl_wmem_max;
524 set_sndbuf:
525 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
526 if ((val * 2) < SOCK_MIN_SNDBUF)
527 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
528 else
529 sk->sk_sndbuf = val * 2;
530
531 /*
532 * Wake up sending tasks if we
533 * upped the value.
534 */
535 sk->sk_write_space(sk);
536 break;
537
538 case SO_SNDBUFFORCE:
539 if (!capable(CAP_NET_ADMIN)) {
540 ret = -EPERM;
541 break;
542 }
543 goto set_sndbuf;
544
545 case SO_RCVBUF:
546 /* Don't error on this BSD doesn't and if you think
547 about it this is right. Otherwise apps have to
548 play 'guess the biggest size' games. RCVBUF/SNDBUF
549 are treated in BSD as hints */
550
551 if (val > sysctl_rmem_max)
552 val = sysctl_rmem_max;
553 set_rcvbuf:
554 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
555 /*
556 * We double it on the way in to account for
557 * "struct sk_buff" etc. overhead. Applications
558 * assume that the SO_RCVBUF setting they make will
559 * allow that much actual data to be received on that
560 * socket.
561 *
562 * Applications are unaware that "struct sk_buff" and
563 * other overheads allocate from the receive buffer
564 * during socket buffer allocation.
565 *
566 * And after considering the possible alternatives,
567 * returning the value we actually used in getsockopt
568 * is the most desirable behavior.
569 */
570 if ((val * 2) < SOCK_MIN_RCVBUF)
571 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
572 else
573 sk->sk_rcvbuf = val * 2;
574 break;
575
576 case SO_RCVBUFFORCE:
577 if (!capable(CAP_NET_ADMIN)) {
578 ret = -EPERM;
579 break;
580 }
581 goto set_rcvbuf;
582
583 case SO_KEEPALIVE:
584 #ifdef CONFIG_INET
585 if (sk->sk_protocol == IPPROTO_TCP)
586 tcp_set_keepalive(sk, valbool);
587 #endif
588 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
589 break;
590
591 case SO_OOBINLINE:
592 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
593 break;
594
595 case SO_NO_CHECK:
596 sk->sk_no_check = valbool;
597 break;
598
599 case SO_PRIORITY:
600 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
601 sk->sk_priority = val;
602 else
603 ret = -EPERM;
604 break;
605
606 case SO_LINGER:
607 if (optlen < sizeof(ling)) {
608 ret = -EINVAL; /* 1003.1g */
609 break;
610 }
611 if (copy_from_user(&ling, optval, sizeof(ling))) {
612 ret = -EFAULT;
613 break;
614 }
615 if (!ling.l_onoff)
616 sock_reset_flag(sk, SOCK_LINGER);
617 else {
618 #if (BITS_PER_LONG == 32)
619 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
620 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
621 else
622 #endif
623 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
624 sock_set_flag(sk, SOCK_LINGER);
625 }
626 break;
627
628 case SO_BSDCOMPAT:
629 sock_warn_obsolete_bsdism("setsockopt");
630 break;
631
632 case SO_PASSCRED:
633 if (valbool)
634 set_bit(SOCK_PASSCRED, &sock->flags);
635 else
636 clear_bit(SOCK_PASSCRED, &sock->flags);
637 break;
638
639 case SO_TIMESTAMP:
640 case SO_TIMESTAMPNS:
641 if (valbool) {
642 if (optname == SO_TIMESTAMP)
643 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
644 else
645 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
646 sock_set_flag(sk, SOCK_RCVTSTAMP);
647 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
648 } else {
649 sock_reset_flag(sk, SOCK_RCVTSTAMP);
650 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
651 }
652 break;
653
654 case SO_TIMESTAMPING:
655 if (val & ~SOF_TIMESTAMPING_MASK) {
656 ret = -EINVAL;
657 break;
658 }
659 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
660 val & SOF_TIMESTAMPING_TX_HARDWARE);
661 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
662 val & SOF_TIMESTAMPING_TX_SOFTWARE);
663 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
664 val & SOF_TIMESTAMPING_RX_HARDWARE);
665 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
666 sock_enable_timestamp(sk,
667 SOCK_TIMESTAMPING_RX_SOFTWARE);
668 else
669 sock_disable_timestamp(sk,
670 SOCK_TIMESTAMPING_RX_SOFTWARE);
671 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
672 val & SOF_TIMESTAMPING_SOFTWARE);
673 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
674 val & SOF_TIMESTAMPING_SYS_HARDWARE);
675 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
676 val & SOF_TIMESTAMPING_RAW_HARDWARE);
677 break;
678
679 case SO_RCVLOWAT:
680 if (val < 0)
681 val = INT_MAX;
682 sk->sk_rcvlowat = val ? : 1;
683 break;
684
685 case SO_RCVTIMEO:
686 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
687 break;
688
689 case SO_SNDTIMEO:
690 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
691 break;
692
693 case SO_ATTACH_FILTER:
694 ret = -EINVAL;
695 if (optlen == sizeof(struct sock_fprog)) {
696 struct sock_fprog fprog;
697
698 ret = -EFAULT;
699 if (copy_from_user(&fprog, optval, sizeof(fprog)))
700 break;
701
702 ret = sk_attach_filter(&fprog, sk);
703 }
704 break;
705
706 case SO_DETACH_FILTER:
707 ret = sk_detach_filter(sk);
708 break;
709
710 case SO_PASSSEC:
711 if (valbool)
712 set_bit(SOCK_PASSSEC, &sock->flags);
713 else
714 clear_bit(SOCK_PASSSEC, &sock->flags);
715 break;
716 case SO_MARK:
717 if (!capable(CAP_NET_ADMIN))
718 ret = -EPERM;
719 else
720 sk->sk_mark = val;
721 break;
722
723 /* We implement the SO_SNDLOWAT etc to
724 not be settable (1003.1g 5.3) */
725 case SO_RXQ_OVFL:
726 if (valbool)
727 sock_set_flag(sk, SOCK_RXQ_OVFL);
728 else
729 sock_reset_flag(sk, SOCK_RXQ_OVFL);
730 break;
731 default:
732 ret = -ENOPROTOOPT;
733 break;
734 }
735 release_sock(sk);
736 return ret;
737 }
738 EXPORT_SYMBOL(sock_setsockopt);
739
740
741 int sock_getsockopt(struct socket *sock, int level, int optname,
742 char __user *optval, int __user *optlen)
743 {
744 struct sock *sk = sock->sk;
745
746 union {
747 int val;
748 struct linger ling;
749 struct timeval tm;
750 } v;
751
752 int lv = sizeof(int);
753 int len;
754
755 if (get_user(len, optlen))
756 return -EFAULT;
757 if (len < 0)
758 return -EINVAL;
759
760 memset(&v, 0, sizeof(v));
761
762 switch (optname) {
763 case SO_DEBUG:
764 v.val = sock_flag(sk, SOCK_DBG);
765 break;
766
767 case SO_DONTROUTE:
768 v.val = sock_flag(sk, SOCK_LOCALROUTE);
769 break;
770
771 case SO_BROADCAST:
772 v.val = !!sock_flag(sk, SOCK_BROADCAST);
773 break;
774
775 case SO_SNDBUF:
776 v.val = sk->sk_sndbuf;
777 break;
778
779 case SO_RCVBUF:
780 v.val = sk->sk_rcvbuf;
781 break;
782
783 case SO_REUSEADDR:
784 v.val = sk->sk_reuse;
785 break;
786
787 case SO_KEEPALIVE:
788 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
789 break;
790
791 case SO_TYPE:
792 v.val = sk->sk_type;
793 break;
794
795 case SO_PROTOCOL:
796 v.val = sk->sk_protocol;
797 break;
798
799 case SO_DOMAIN:
800 v.val = sk->sk_family;
801 break;
802
803 case SO_ERROR:
804 v.val = -sock_error(sk);
805 if (v.val == 0)
806 v.val = xchg(&sk->sk_err_soft, 0);
807 break;
808
809 case SO_OOBINLINE:
810 v.val = !!sock_flag(sk, SOCK_URGINLINE);
811 break;
812
813 case SO_NO_CHECK:
814 v.val = sk->sk_no_check;
815 break;
816
817 case SO_PRIORITY:
818 v.val = sk->sk_priority;
819 break;
820
821 case SO_LINGER:
822 lv = sizeof(v.ling);
823 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
824 v.ling.l_linger = sk->sk_lingertime / HZ;
825 break;
826
827 case SO_BSDCOMPAT:
828 sock_warn_obsolete_bsdism("getsockopt");
829 break;
830
831 case SO_TIMESTAMP:
832 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
833 !sock_flag(sk, SOCK_RCVTSTAMPNS);
834 break;
835
836 case SO_TIMESTAMPNS:
837 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
838 break;
839
840 case SO_TIMESTAMPING:
841 v.val = 0;
842 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
843 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
844 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
845 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
846 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
847 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
848 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
849 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
850 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
851 v.val |= SOF_TIMESTAMPING_SOFTWARE;
852 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
853 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
854 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
855 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
856 break;
857
858 case SO_RCVTIMEO:
859 lv = sizeof(struct timeval);
860 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
861 v.tm.tv_sec = 0;
862 v.tm.tv_usec = 0;
863 } else {
864 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
865 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
866 }
867 break;
868
869 case SO_SNDTIMEO:
870 lv = sizeof(struct timeval);
871 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
872 v.tm.tv_sec = 0;
873 v.tm.tv_usec = 0;
874 } else {
875 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
876 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
877 }
878 break;
879
880 case SO_RCVLOWAT:
881 v.val = sk->sk_rcvlowat;
882 break;
883
884 case SO_SNDLOWAT:
885 v.val = 1;
886 break;
887
888 case SO_PASSCRED:
889 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
890 break;
891
892 case SO_PEERCRED:
893 if (len > sizeof(sk->sk_peercred))
894 len = sizeof(sk->sk_peercred);
895 if (copy_to_user(optval, &sk->sk_peercred, len))
896 return -EFAULT;
897 goto lenout;
898
899 case SO_PEERNAME:
900 {
901 char address[128];
902
903 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
904 return -ENOTCONN;
905 if (lv < len)
906 return -EINVAL;
907 if (copy_to_user(optval, address, len))
908 return -EFAULT;
909 goto lenout;
910 }
911
912 /* Dubious BSD thing... Probably nobody even uses it, but
913 * the UNIX standard wants it for whatever reason... -DaveM
914 */
915 case SO_ACCEPTCONN:
916 v.val = sk->sk_state == TCP_LISTEN;
917 break;
918
919 case SO_PASSSEC:
920 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
921 break;
922
923 case SO_PEERSEC:
924 return security_socket_getpeersec_stream(sock, optval, optlen, len);
925
926 case SO_MARK:
927 v.val = sk->sk_mark;
928 break;
929
930 case SO_RXQ_OVFL:
931 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
932 break;
933
934 default:
935 return -ENOPROTOOPT;
936 }
937
938 if (len > lv)
939 len = lv;
940 if (copy_to_user(optval, &v, len))
941 return -EFAULT;
942 lenout:
943 if (put_user(len, optlen))
944 return -EFAULT;
945 return 0;
946 }
947
948 /*
949 * Initialize an sk_lock.
950 *
951 * (We also register the sk_lock with the lock validator.)
952 */
953 static inline void sock_lock_init(struct sock *sk)
954 {
955 sock_lock_init_class_and_name(sk,
956 af_family_slock_key_strings[sk->sk_family],
957 af_family_slock_keys + sk->sk_family,
958 af_family_key_strings[sk->sk_family],
959 af_family_keys + sk->sk_family);
960 }
961
962 /*
963 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
964 * even temporarly, because of RCU lookups. sk_node should also be left as is.
965 */
966 static void sock_copy(struct sock *nsk, const struct sock *osk)
967 {
968 #ifdef CONFIG_SECURITY_NETWORK
969 void *sptr = nsk->sk_security;
970 #endif
971 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
972 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
973 sizeof(osk->sk_tx_queue_mapping));
974 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
975 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
976 #ifdef CONFIG_SECURITY_NETWORK
977 nsk->sk_security = sptr;
978 security_sk_clone(osk, nsk);
979 #endif
980 }
981
982 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
983 int family)
984 {
985 struct sock *sk;
986 struct kmem_cache *slab;
987
988 slab = prot->slab;
989 if (slab != NULL) {
990 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
991 if (!sk)
992 return sk;
993 if (priority & __GFP_ZERO) {
994 /*
995 * caches using SLAB_DESTROY_BY_RCU should let
996 * sk_node.next un-modified. Special care is taken
997 * when initializing object to zero.
998 */
999 if (offsetof(struct sock, sk_node.next) != 0)
1000 memset(sk, 0, offsetof(struct sock, sk_node.next));
1001 memset(&sk->sk_node.pprev, 0,
1002 prot->obj_size - offsetof(struct sock,
1003 sk_node.pprev));
1004 }
1005 }
1006 else
1007 sk = kmalloc(prot->obj_size, priority);
1008
1009 if (sk != NULL) {
1010 kmemcheck_annotate_bitfield(sk, flags);
1011
1012 if (security_sk_alloc(sk, family, priority))
1013 goto out_free;
1014
1015 if (!try_module_get(prot->owner))
1016 goto out_free_sec;
1017 sk_tx_queue_clear(sk);
1018 }
1019
1020 return sk;
1021
1022 out_free_sec:
1023 security_sk_free(sk);
1024 out_free:
1025 if (slab != NULL)
1026 kmem_cache_free(slab, sk);
1027 else
1028 kfree(sk);
1029 return NULL;
1030 }
1031
1032 static void sk_prot_free(struct proto *prot, struct sock *sk)
1033 {
1034 struct kmem_cache *slab;
1035 struct module *owner;
1036
1037 owner = prot->owner;
1038 slab = prot->slab;
1039
1040 security_sk_free(sk);
1041 if (slab != NULL)
1042 kmem_cache_free(slab, sk);
1043 else
1044 kfree(sk);
1045 module_put(owner);
1046 }
1047
1048 /**
1049 * sk_alloc - All socket objects are allocated here
1050 * @net: the applicable net namespace
1051 * @family: protocol family
1052 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1053 * @prot: struct proto associated with this new sock instance
1054 */
1055 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1056 struct proto *prot)
1057 {
1058 struct sock *sk;
1059
1060 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1061 if (sk) {
1062 sk->sk_family = family;
1063 /*
1064 * See comment in struct sock definition to understand
1065 * why we need sk_prot_creator -acme
1066 */
1067 sk->sk_prot = sk->sk_prot_creator = prot;
1068 sock_lock_init(sk);
1069 sock_net_set(sk, get_net(net));
1070 atomic_set(&sk->sk_wmem_alloc, 1);
1071 }
1072
1073 return sk;
1074 }
1075 EXPORT_SYMBOL(sk_alloc);
1076
1077 static void __sk_free(struct sock *sk)
1078 {
1079 struct sk_filter *filter;
1080
1081 if (sk->sk_destruct)
1082 sk->sk_destruct(sk);
1083
1084 filter = rcu_dereference_check(sk->sk_filter,
1085 atomic_read(&sk->sk_wmem_alloc) == 0);
1086 if (filter) {
1087 sk_filter_uncharge(sk, filter);
1088 rcu_assign_pointer(sk->sk_filter, NULL);
1089 }
1090
1091 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1092 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1093
1094 if (atomic_read(&sk->sk_omem_alloc))
1095 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1096 __func__, atomic_read(&sk->sk_omem_alloc));
1097
1098 put_net(sock_net(sk));
1099 sk_prot_free(sk->sk_prot_creator, sk);
1100 }
1101
1102 void sk_free(struct sock *sk)
1103 {
1104 /*
1105 * We substract one from sk_wmem_alloc and can know if
1106 * some packets are still in some tx queue.
1107 * If not null, sock_wfree() will call __sk_free(sk) later
1108 */
1109 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1110 __sk_free(sk);
1111 }
1112 EXPORT_SYMBOL(sk_free);
1113
1114 /*
1115 * Last sock_put should drop referrence to sk->sk_net. It has already
1116 * been dropped in sk_change_net. Taking referrence to stopping namespace
1117 * is not an option.
1118 * Take referrence to a socket to remove it from hash _alive_ and after that
1119 * destroy it in the context of init_net.
1120 */
1121 void sk_release_kernel(struct sock *sk)
1122 {
1123 if (sk == NULL || sk->sk_socket == NULL)
1124 return;
1125
1126 sock_hold(sk);
1127 sock_release(sk->sk_socket);
1128 release_net(sock_net(sk));
1129 sock_net_set(sk, get_net(&init_net));
1130 sock_put(sk);
1131 }
1132 EXPORT_SYMBOL(sk_release_kernel);
1133
1134 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1135 {
1136 struct sock *newsk;
1137
1138 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1139 if (newsk != NULL) {
1140 struct sk_filter *filter;
1141
1142 sock_copy(newsk, sk);
1143
1144 /* SANITY */
1145 get_net(sock_net(newsk));
1146 sk_node_init(&newsk->sk_node);
1147 sock_lock_init(newsk);
1148 bh_lock_sock(newsk);
1149 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1150 newsk->sk_backlog.len = 0;
1151
1152 atomic_set(&newsk->sk_rmem_alloc, 0);
1153 /*
1154 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1155 */
1156 atomic_set(&newsk->sk_wmem_alloc, 1);
1157 atomic_set(&newsk->sk_omem_alloc, 0);
1158 skb_queue_head_init(&newsk->sk_receive_queue);
1159 skb_queue_head_init(&newsk->sk_write_queue);
1160 #ifdef CONFIG_NET_DMA
1161 skb_queue_head_init(&newsk->sk_async_wait_queue);
1162 #endif
1163
1164 spin_lock_init(&newsk->sk_dst_lock);
1165 rwlock_init(&newsk->sk_callback_lock);
1166 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1167 af_callback_keys + newsk->sk_family,
1168 af_family_clock_key_strings[newsk->sk_family]);
1169
1170 newsk->sk_dst_cache = NULL;
1171 newsk->sk_wmem_queued = 0;
1172 newsk->sk_forward_alloc = 0;
1173 newsk->sk_send_head = NULL;
1174 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1175
1176 sock_reset_flag(newsk, SOCK_DONE);
1177 skb_queue_head_init(&newsk->sk_error_queue);
1178
1179 filter = newsk->sk_filter;
1180 if (filter != NULL)
1181 sk_filter_charge(newsk, filter);
1182
1183 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1184 /* It is still raw copy of parent, so invalidate
1185 * destructor and make plain sk_free() */
1186 newsk->sk_destruct = NULL;
1187 sk_free(newsk);
1188 newsk = NULL;
1189 goto out;
1190 }
1191
1192 newsk->sk_err = 0;
1193 newsk->sk_priority = 0;
1194 /*
1195 * Before updating sk_refcnt, we must commit prior changes to memory
1196 * (Documentation/RCU/rculist_nulls.txt for details)
1197 */
1198 smp_wmb();
1199 atomic_set(&newsk->sk_refcnt, 2);
1200
1201 /*
1202 * Increment the counter in the same struct proto as the master
1203 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1204 * is the same as sk->sk_prot->socks, as this field was copied
1205 * with memcpy).
1206 *
1207 * This _changes_ the previous behaviour, where
1208 * tcp_create_openreq_child always was incrementing the
1209 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1210 * to be taken into account in all callers. -acme
1211 */
1212 sk_refcnt_debug_inc(newsk);
1213 sk_set_socket(newsk, NULL);
1214 newsk->sk_sleep = NULL;
1215
1216 if (newsk->sk_prot->sockets_allocated)
1217 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1218
1219 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1220 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1221 net_enable_timestamp();
1222 }
1223 out:
1224 return newsk;
1225 }
1226 EXPORT_SYMBOL_GPL(sk_clone);
1227
1228 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1229 {
1230 __sk_dst_set(sk, dst);
1231 sk->sk_route_caps = dst->dev->features;
1232 if (sk->sk_route_caps & NETIF_F_GSO)
1233 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1234 if (sk_can_gso(sk)) {
1235 if (dst->header_len) {
1236 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1237 } else {
1238 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1239 sk->sk_gso_max_size = dst->dev->gso_max_size;
1240 }
1241 }
1242 }
1243 EXPORT_SYMBOL_GPL(sk_setup_caps);
1244
1245 void __init sk_init(void)
1246 {
1247 if (totalram_pages <= 4096) {
1248 sysctl_wmem_max = 32767;
1249 sysctl_rmem_max = 32767;
1250 sysctl_wmem_default = 32767;
1251 sysctl_rmem_default = 32767;
1252 } else if (totalram_pages >= 131072) {
1253 sysctl_wmem_max = 131071;
1254 sysctl_rmem_max = 131071;
1255 }
1256 }
1257
1258 /*
1259 * Simple resource managers for sockets.
1260 */
1261
1262
1263 /*
1264 * Write buffer destructor automatically called from kfree_skb.
1265 */
1266 void sock_wfree(struct sk_buff *skb)
1267 {
1268 struct sock *sk = skb->sk;
1269 unsigned int len = skb->truesize;
1270
1271 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1272 /*
1273 * Keep a reference on sk_wmem_alloc, this will be released
1274 * after sk_write_space() call
1275 */
1276 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1277 sk->sk_write_space(sk);
1278 len = 1;
1279 }
1280 /*
1281 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1282 * could not do because of in-flight packets
1283 */
1284 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1285 __sk_free(sk);
1286 }
1287 EXPORT_SYMBOL(sock_wfree);
1288
1289 /*
1290 * Read buffer destructor automatically called from kfree_skb.
1291 */
1292 void sock_rfree(struct sk_buff *skb)
1293 {
1294 struct sock *sk = skb->sk;
1295
1296 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1297 sk_mem_uncharge(skb->sk, skb->truesize);
1298 }
1299 EXPORT_SYMBOL(sock_rfree);
1300
1301
1302 int sock_i_uid(struct sock *sk)
1303 {
1304 int uid;
1305
1306 read_lock(&sk->sk_callback_lock);
1307 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1308 read_unlock(&sk->sk_callback_lock);
1309 return uid;
1310 }
1311 EXPORT_SYMBOL(sock_i_uid);
1312
1313 unsigned long sock_i_ino(struct sock *sk)
1314 {
1315 unsigned long ino;
1316
1317 read_lock(&sk->sk_callback_lock);
1318 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1319 read_unlock(&sk->sk_callback_lock);
1320 return ino;
1321 }
1322 EXPORT_SYMBOL(sock_i_ino);
1323
1324 /*
1325 * Allocate a skb from the socket's send buffer.
1326 */
1327 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1328 gfp_t priority)
1329 {
1330 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1331 struct sk_buff *skb = alloc_skb(size, priority);
1332 if (skb) {
1333 skb_set_owner_w(skb, sk);
1334 return skb;
1335 }
1336 }
1337 return NULL;
1338 }
1339 EXPORT_SYMBOL(sock_wmalloc);
1340
1341 /*
1342 * Allocate a skb from the socket's receive buffer.
1343 */
1344 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1345 gfp_t priority)
1346 {
1347 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1348 struct sk_buff *skb = alloc_skb(size, priority);
1349 if (skb) {
1350 skb_set_owner_r(skb, sk);
1351 return skb;
1352 }
1353 }
1354 return NULL;
1355 }
1356
1357 /*
1358 * Allocate a memory block from the socket's option memory buffer.
1359 */
1360 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1361 {
1362 if ((unsigned)size <= sysctl_optmem_max &&
1363 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1364 void *mem;
1365 /* First do the add, to avoid the race if kmalloc
1366 * might sleep.
1367 */
1368 atomic_add(size, &sk->sk_omem_alloc);
1369 mem = kmalloc(size, priority);
1370 if (mem)
1371 return mem;
1372 atomic_sub(size, &sk->sk_omem_alloc);
1373 }
1374 return NULL;
1375 }
1376 EXPORT_SYMBOL(sock_kmalloc);
1377
1378 /*
1379 * Free an option memory block.
1380 */
1381 void sock_kfree_s(struct sock *sk, void *mem, int size)
1382 {
1383 kfree(mem);
1384 atomic_sub(size, &sk->sk_omem_alloc);
1385 }
1386 EXPORT_SYMBOL(sock_kfree_s);
1387
1388 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1389 I think, these locks should be removed for datagram sockets.
1390 */
1391 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1392 {
1393 DEFINE_WAIT(wait);
1394
1395 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1396 for (;;) {
1397 if (!timeo)
1398 break;
1399 if (signal_pending(current))
1400 break;
1401 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1402 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1403 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1404 break;
1405 if (sk->sk_shutdown & SEND_SHUTDOWN)
1406 break;
1407 if (sk->sk_err)
1408 break;
1409 timeo = schedule_timeout(timeo);
1410 }
1411 finish_wait(sk_sleep(sk), &wait);
1412 return timeo;
1413 }
1414
1415
1416 /*
1417 * Generic send/receive buffer handlers
1418 */
1419
1420 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1421 unsigned long data_len, int noblock,
1422 int *errcode)
1423 {
1424 struct sk_buff *skb;
1425 gfp_t gfp_mask;
1426 long timeo;
1427 int err;
1428
1429 gfp_mask = sk->sk_allocation;
1430 if (gfp_mask & __GFP_WAIT)
1431 gfp_mask |= __GFP_REPEAT;
1432
1433 timeo = sock_sndtimeo(sk, noblock);
1434 while (1) {
1435 err = sock_error(sk);
1436 if (err != 0)
1437 goto failure;
1438
1439 err = -EPIPE;
1440 if (sk->sk_shutdown & SEND_SHUTDOWN)
1441 goto failure;
1442
1443 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1444 skb = alloc_skb(header_len, gfp_mask);
1445 if (skb) {
1446 int npages;
1447 int i;
1448
1449 /* No pages, we're done... */
1450 if (!data_len)
1451 break;
1452
1453 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1454 skb->truesize += data_len;
1455 skb_shinfo(skb)->nr_frags = npages;
1456 for (i = 0; i < npages; i++) {
1457 struct page *page;
1458 skb_frag_t *frag;
1459
1460 page = alloc_pages(sk->sk_allocation, 0);
1461 if (!page) {
1462 err = -ENOBUFS;
1463 skb_shinfo(skb)->nr_frags = i;
1464 kfree_skb(skb);
1465 goto failure;
1466 }
1467
1468 frag = &skb_shinfo(skb)->frags[i];
1469 frag->page = page;
1470 frag->page_offset = 0;
1471 frag->size = (data_len >= PAGE_SIZE ?
1472 PAGE_SIZE :
1473 data_len);
1474 data_len -= PAGE_SIZE;
1475 }
1476
1477 /* Full success... */
1478 break;
1479 }
1480 err = -ENOBUFS;
1481 goto failure;
1482 }
1483 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1484 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1485 err = -EAGAIN;
1486 if (!timeo)
1487 goto failure;
1488 if (signal_pending(current))
1489 goto interrupted;
1490 timeo = sock_wait_for_wmem(sk, timeo);
1491 }
1492
1493 skb_set_owner_w(skb, sk);
1494 return skb;
1495
1496 interrupted:
1497 err = sock_intr_errno(timeo);
1498 failure:
1499 *errcode = err;
1500 return NULL;
1501 }
1502 EXPORT_SYMBOL(sock_alloc_send_pskb);
1503
1504 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1505 int noblock, int *errcode)
1506 {
1507 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1508 }
1509 EXPORT_SYMBOL(sock_alloc_send_skb);
1510
1511 static void __lock_sock(struct sock *sk)
1512 {
1513 DEFINE_WAIT(wait);
1514
1515 for (;;) {
1516 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1517 TASK_UNINTERRUPTIBLE);
1518 spin_unlock_bh(&sk->sk_lock.slock);
1519 schedule();
1520 spin_lock_bh(&sk->sk_lock.slock);
1521 if (!sock_owned_by_user(sk))
1522 break;
1523 }
1524 finish_wait(&sk->sk_lock.wq, &wait);
1525 }
1526
1527 static void __release_sock(struct sock *sk)
1528 {
1529 struct sk_buff *skb = sk->sk_backlog.head;
1530
1531 do {
1532 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1533 bh_unlock_sock(sk);
1534
1535 do {
1536 struct sk_buff *next = skb->next;
1537
1538 skb->next = NULL;
1539 sk_backlog_rcv(sk, skb);
1540
1541 /*
1542 * We are in process context here with softirqs
1543 * disabled, use cond_resched_softirq() to preempt.
1544 * This is safe to do because we've taken the backlog
1545 * queue private:
1546 */
1547 cond_resched_softirq();
1548
1549 skb = next;
1550 } while (skb != NULL);
1551
1552 bh_lock_sock(sk);
1553 } while ((skb = sk->sk_backlog.head) != NULL);
1554
1555 /*
1556 * Doing the zeroing here guarantee we can not loop forever
1557 * while a wild producer attempts to flood us.
1558 */
1559 sk->sk_backlog.len = 0;
1560 }
1561
1562 /**
1563 * sk_wait_data - wait for data to arrive at sk_receive_queue
1564 * @sk: sock to wait on
1565 * @timeo: for how long
1566 *
1567 * Now socket state including sk->sk_err is changed only under lock,
1568 * hence we may omit checks after joining wait queue.
1569 * We check receive queue before schedule() only as optimization;
1570 * it is very likely that release_sock() added new data.
1571 */
1572 int sk_wait_data(struct sock *sk, long *timeo)
1573 {
1574 int rc;
1575 DEFINE_WAIT(wait);
1576
1577 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1578 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1579 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1580 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1581 finish_wait(sk_sleep(sk), &wait);
1582 return rc;
1583 }
1584 EXPORT_SYMBOL(sk_wait_data);
1585
1586 /**
1587 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1588 * @sk: socket
1589 * @size: memory size to allocate
1590 * @kind: allocation type
1591 *
1592 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1593 * rmem allocation. This function assumes that protocols which have
1594 * memory_pressure use sk_wmem_queued as write buffer accounting.
1595 */
1596 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1597 {
1598 struct proto *prot = sk->sk_prot;
1599 int amt = sk_mem_pages(size);
1600 int allocated;
1601
1602 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1603 allocated = atomic_add_return(amt, prot->memory_allocated);
1604
1605 /* Under limit. */
1606 if (allocated <= prot->sysctl_mem[0]) {
1607 if (prot->memory_pressure && *prot->memory_pressure)
1608 *prot->memory_pressure = 0;
1609 return 1;
1610 }
1611
1612 /* Under pressure. */
1613 if (allocated > prot->sysctl_mem[1])
1614 if (prot->enter_memory_pressure)
1615 prot->enter_memory_pressure(sk);
1616
1617 /* Over hard limit. */
1618 if (allocated > prot->sysctl_mem[2])
1619 goto suppress_allocation;
1620
1621 /* guarantee minimum buffer size under pressure */
1622 if (kind == SK_MEM_RECV) {
1623 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1624 return 1;
1625 } else { /* SK_MEM_SEND */
1626 if (sk->sk_type == SOCK_STREAM) {
1627 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1628 return 1;
1629 } else if (atomic_read(&sk->sk_wmem_alloc) <
1630 prot->sysctl_wmem[0])
1631 return 1;
1632 }
1633
1634 if (prot->memory_pressure) {
1635 int alloc;
1636
1637 if (!*prot->memory_pressure)
1638 return 1;
1639 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1640 if (prot->sysctl_mem[2] > alloc *
1641 sk_mem_pages(sk->sk_wmem_queued +
1642 atomic_read(&sk->sk_rmem_alloc) +
1643 sk->sk_forward_alloc))
1644 return 1;
1645 }
1646
1647 suppress_allocation:
1648
1649 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1650 sk_stream_moderate_sndbuf(sk);
1651
1652 /* Fail only if socket is _under_ its sndbuf.
1653 * In this case we cannot block, so that we have to fail.
1654 */
1655 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1656 return 1;
1657 }
1658
1659 /* Alas. Undo changes. */
1660 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1661 atomic_sub(amt, prot->memory_allocated);
1662 return 0;
1663 }
1664 EXPORT_SYMBOL(__sk_mem_schedule);
1665
1666 /**
1667 * __sk_reclaim - reclaim memory_allocated
1668 * @sk: socket
1669 */
1670 void __sk_mem_reclaim(struct sock *sk)
1671 {
1672 struct proto *prot = sk->sk_prot;
1673
1674 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1675 prot->memory_allocated);
1676 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1677
1678 if (prot->memory_pressure && *prot->memory_pressure &&
1679 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1680 *prot->memory_pressure = 0;
1681 }
1682 EXPORT_SYMBOL(__sk_mem_reclaim);
1683
1684
1685 /*
1686 * Set of default routines for initialising struct proto_ops when
1687 * the protocol does not support a particular function. In certain
1688 * cases where it makes no sense for a protocol to have a "do nothing"
1689 * function, some default processing is provided.
1690 */
1691
1692 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1693 {
1694 return -EOPNOTSUPP;
1695 }
1696 EXPORT_SYMBOL(sock_no_bind);
1697
1698 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1699 int len, int flags)
1700 {
1701 return -EOPNOTSUPP;
1702 }
1703 EXPORT_SYMBOL(sock_no_connect);
1704
1705 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1706 {
1707 return -EOPNOTSUPP;
1708 }
1709 EXPORT_SYMBOL(sock_no_socketpair);
1710
1711 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1712 {
1713 return -EOPNOTSUPP;
1714 }
1715 EXPORT_SYMBOL(sock_no_accept);
1716
1717 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1718 int *len, int peer)
1719 {
1720 return -EOPNOTSUPP;
1721 }
1722 EXPORT_SYMBOL(sock_no_getname);
1723
1724 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1725 {
1726 return 0;
1727 }
1728 EXPORT_SYMBOL(sock_no_poll);
1729
1730 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1731 {
1732 return -EOPNOTSUPP;
1733 }
1734 EXPORT_SYMBOL(sock_no_ioctl);
1735
1736 int sock_no_listen(struct socket *sock, int backlog)
1737 {
1738 return -EOPNOTSUPP;
1739 }
1740 EXPORT_SYMBOL(sock_no_listen);
1741
1742 int sock_no_shutdown(struct socket *sock, int how)
1743 {
1744 return -EOPNOTSUPP;
1745 }
1746 EXPORT_SYMBOL(sock_no_shutdown);
1747
1748 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1749 char __user *optval, unsigned int optlen)
1750 {
1751 return -EOPNOTSUPP;
1752 }
1753 EXPORT_SYMBOL(sock_no_setsockopt);
1754
1755 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1756 char __user *optval, int __user *optlen)
1757 {
1758 return -EOPNOTSUPP;
1759 }
1760 EXPORT_SYMBOL(sock_no_getsockopt);
1761
1762 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1763 size_t len)
1764 {
1765 return -EOPNOTSUPP;
1766 }
1767 EXPORT_SYMBOL(sock_no_sendmsg);
1768
1769 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1770 size_t len, int flags)
1771 {
1772 return -EOPNOTSUPP;
1773 }
1774 EXPORT_SYMBOL(sock_no_recvmsg);
1775
1776 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1777 {
1778 /* Mirror missing mmap method error code */
1779 return -ENODEV;
1780 }
1781 EXPORT_SYMBOL(sock_no_mmap);
1782
1783 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1784 {
1785 ssize_t res;
1786 struct msghdr msg = {.msg_flags = flags};
1787 struct kvec iov;
1788 char *kaddr = kmap(page);
1789 iov.iov_base = kaddr + offset;
1790 iov.iov_len = size;
1791 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1792 kunmap(page);
1793 return res;
1794 }
1795 EXPORT_SYMBOL(sock_no_sendpage);
1796
1797 /*
1798 * Default Socket Callbacks
1799 */
1800
1801 static void sock_def_wakeup(struct sock *sk)
1802 {
1803 read_lock(&sk->sk_callback_lock);
1804 if (sk_has_sleeper(sk))
1805 wake_up_interruptible_all(sk_sleep(sk));
1806 read_unlock(&sk->sk_callback_lock);
1807 }
1808
1809 static void sock_def_error_report(struct sock *sk)
1810 {
1811 read_lock(&sk->sk_callback_lock);
1812 if (sk_has_sleeper(sk))
1813 wake_up_interruptible_poll(sk_sleep(sk), POLLERR);
1814 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1815 read_unlock(&sk->sk_callback_lock);
1816 }
1817
1818 static void sock_def_readable(struct sock *sk, int len)
1819 {
1820 read_lock(&sk->sk_callback_lock);
1821 if (sk_has_sleeper(sk))
1822 wake_up_interruptible_sync_poll(sk_sleep(sk), POLLIN |
1823 POLLRDNORM | POLLRDBAND);
1824 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1825 read_unlock(&sk->sk_callback_lock);
1826 }
1827
1828 static void sock_def_write_space(struct sock *sk)
1829 {
1830 read_lock(&sk->sk_callback_lock);
1831
1832 /* Do not wake up a writer until he can make "significant"
1833 * progress. --DaveM
1834 */
1835 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1836 if (sk_has_sleeper(sk))
1837 wake_up_interruptible_sync_poll(sk_sleep(sk), POLLOUT |
1838 POLLWRNORM | POLLWRBAND);
1839
1840 /* Should agree with poll, otherwise some programs break */
1841 if (sock_writeable(sk))
1842 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1843 }
1844
1845 read_unlock(&sk->sk_callback_lock);
1846 }
1847
1848 static void sock_def_destruct(struct sock *sk)
1849 {
1850 kfree(sk->sk_protinfo);
1851 }
1852
1853 void sk_send_sigurg(struct sock *sk)
1854 {
1855 if (sk->sk_socket && sk->sk_socket->file)
1856 if (send_sigurg(&sk->sk_socket->file->f_owner))
1857 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1858 }
1859 EXPORT_SYMBOL(sk_send_sigurg);
1860
1861 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1862 unsigned long expires)
1863 {
1864 if (!mod_timer(timer, expires))
1865 sock_hold(sk);
1866 }
1867 EXPORT_SYMBOL(sk_reset_timer);
1868
1869 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1870 {
1871 if (timer_pending(timer) && del_timer(timer))
1872 __sock_put(sk);
1873 }
1874 EXPORT_SYMBOL(sk_stop_timer);
1875
1876 void sock_init_data(struct socket *sock, struct sock *sk)
1877 {
1878 skb_queue_head_init(&sk->sk_receive_queue);
1879 skb_queue_head_init(&sk->sk_write_queue);
1880 skb_queue_head_init(&sk->sk_error_queue);
1881 #ifdef CONFIG_NET_DMA
1882 skb_queue_head_init(&sk->sk_async_wait_queue);
1883 #endif
1884
1885 sk->sk_send_head = NULL;
1886
1887 init_timer(&sk->sk_timer);
1888
1889 sk->sk_allocation = GFP_KERNEL;
1890 sk->sk_rcvbuf = sysctl_rmem_default;
1891 sk->sk_sndbuf = sysctl_wmem_default;
1892 sk->sk_state = TCP_CLOSE;
1893 sk_set_socket(sk, sock);
1894
1895 sock_set_flag(sk, SOCK_ZAPPED);
1896
1897 if (sock) {
1898 sk->sk_type = sock->type;
1899 sk->sk_sleep = &sock->wait;
1900 sock->sk = sk;
1901 } else
1902 sk->sk_sleep = NULL;
1903
1904 spin_lock_init(&sk->sk_dst_lock);
1905 rwlock_init(&sk->sk_callback_lock);
1906 lockdep_set_class_and_name(&sk->sk_callback_lock,
1907 af_callback_keys + sk->sk_family,
1908 af_family_clock_key_strings[sk->sk_family]);
1909
1910 sk->sk_state_change = sock_def_wakeup;
1911 sk->sk_data_ready = sock_def_readable;
1912 sk->sk_write_space = sock_def_write_space;
1913 sk->sk_error_report = sock_def_error_report;
1914 sk->sk_destruct = sock_def_destruct;
1915
1916 sk->sk_sndmsg_page = NULL;
1917 sk->sk_sndmsg_off = 0;
1918
1919 sk->sk_peercred.pid = 0;
1920 sk->sk_peercred.uid = -1;
1921 sk->sk_peercred.gid = -1;
1922 sk->sk_write_pending = 0;
1923 sk->sk_rcvlowat = 1;
1924 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1925 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1926
1927 sk->sk_stamp = ktime_set(-1L, 0);
1928
1929 /*
1930 * Before updating sk_refcnt, we must commit prior changes to memory
1931 * (Documentation/RCU/rculist_nulls.txt for details)
1932 */
1933 smp_wmb();
1934 atomic_set(&sk->sk_refcnt, 1);
1935 atomic_set(&sk->sk_drops, 0);
1936 }
1937 EXPORT_SYMBOL(sock_init_data);
1938
1939 void lock_sock_nested(struct sock *sk, int subclass)
1940 {
1941 might_sleep();
1942 spin_lock_bh(&sk->sk_lock.slock);
1943 if (sk->sk_lock.owned)
1944 __lock_sock(sk);
1945 sk->sk_lock.owned = 1;
1946 spin_unlock(&sk->sk_lock.slock);
1947 /*
1948 * The sk_lock has mutex_lock() semantics here:
1949 */
1950 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1951 local_bh_enable();
1952 }
1953 EXPORT_SYMBOL(lock_sock_nested);
1954
1955 void release_sock(struct sock *sk)
1956 {
1957 /*
1958 * The sk_lock has mutex_unlock() semantics:
1959 */
1960 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1961
1962 spin_lock_bh(&sk->sk_lock.slock);
1963 if (sk->sk_backlog.tail)
1964 __release_sock(sk);
1965 sk->sk_lock.owned = 0;
1966 if (waitqueue_active(&sk->sk_lock.wq))
1967 wake_up(&sk->sk_lock.wq);
1968 spin_unlock_bh(&sk->sk_lock.slock);
1969 }
1970 EXPORT_SYMBOL(release_sock);
1971
1972 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1973 {
1974 struct timeval tv;
1975 if (!sock_flag(sk, SOCK_TIMESTAMP))
1976 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1977 tv = ktime_to_timeval(sk->sk_stamp);
1978 if (tv.tv_sec == -1)
1979 return -ENOENT;
1980 if (tv.tv_sec == 0) {
1981 sk->sk_stamp = ktime_get_real();
1982 tv = ktime_to_timeval(sk->sk_stamp);
1983 }
1984 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1985 }
1986 EXPORT_SYMBOL(sock_get_timestamp);
1987
1988 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1989 {
1990 struct timespec ts;
1991 if (!sock_flag(sk, SOCK_TIMESTAMP))
1992 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1993 ts = ktime_to_timespec(sk->sk_stamp);
1994 if (ts.tv_sec == -1)
1995 return -ENOENT;
1996 if (ts.tv_sec == 0) {
1997 sk->sk_stamp = ktime_get_real();
1998 ts = ktime_to_timespec(sk->sk_stamp);
1999 }
2000 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2001 }
2002 EXPORT_SYMBOL(sock_get_timestampns);
2003
2004 void sock_enable_timestamp(struct sock *sk, int flag)
2005 {
2006 if (!sock_flag(sk, flag)) {
2007 sock_set_flag(sk, flag);
2008 /*
2009 * we just set one of the two flags which require net
2010 * time stamping, but time stamping might have been on
2011 * already because of the other one
2012 */
2013 if (!sock_flag(sk,
2014 flag == SOCK_TIMESTAMP ?
2015 SOCK_TIMESTAMPING_RX_SOFTWARE :
2016 SOCK_TIMESTAMP))
2017 net_enable_timestamp();
2018 }
2019 }
2020
2021 /*
2022 * Get a socket option on an socket.
2023 *
2024 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2025 * asynchronous errors should be reported by getsockopt. We assume
2026 * this means if you specify SO_ERROR (otherwise whats the point of it).
2027 */
2028 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2029 char __user *optval, int __user *optlen)
2030 {
2031 struct sock *sk = sock->sk;
2032
2033 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2034 }
2035 EXPORT_SYMBOL(sock_common_getsockopt);
2036
2037 #ifdef CONFIG_COMPAT
2038 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2039 char __user *optval, int __user *optlen)
2040 {
2041 struct sock *sk = sock->sk;
2042
2043 if (sk->sk_prot->compat_getsockopt != NULL)
2044 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2045 optval, optlen);
2046 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2047 }
2048 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2049 #endif
2050
2051 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2052 struct msghdr *msg, size_t size, int flags)
2053 {
2054 struct sock *sk = sock->sk;
2055 int addr_len = 0;
2056 int err;
2057
2058 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2059 flags & ~MSG_DONTWAIT, &addr_len);
2060 if (err >= 0)
2061 msg->msg_namelen = addr_len;
2062 return err;
2063 }
2064 EXPORT_SYMBOL(sock_common_recvmsg);
2065
2066 /*
2067 * Set socket options on an inet socket.
2068 */
2069 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2070 char __user *optval, unsigned int optlen)
2071 {
2072 struct sock *sk = sock->sk;
2073
2074 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2075 }
2076 EXPORT_SYMBOL(sock_common_setsockopt);
2077
2078 #ifdef CONFIG_COMPAT
2079 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2080 char __user *optval, unsigned int optlen)
2081 {
2082 struct sock *sk = sock->sk;
2083
2084 if (sk->sk_prot->compat_setsockopt != NULL)
2085 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2086 optval, optlen);
2087 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2088 }
2089 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2090 #endif
2091
2092 void sk_common_release(struct sock *sk)
2093 {
2094 if (sk->sk_prot->destroy)
2095 sk->sk_prot->destroy(sk);
2096
2097 /*
2098 * Observation: when sock_common_release is called, processes have
2099 * no access to socket. But net still has.
2100 * Step one, detach it from networking:
2101 *
2102 * A. Remove from hash tables.
2103 */
2104
2105 sk->sk_prot->unhash(sk);
2106
2107 /*
2108 * In this point socket cannot receive new packets, but it is possible
2109 * that some packets are in flight because some CPU runs receiver and
2110 * did hash table lookup before we unhashed socket. They will achieve
2111 * receive queue and will be purged by socket destructor.
2112 *
2113 * Also we still have packets pending on receive queue and probably,
2114 * our own packets waiting in device queues. sock_destroy will drain
2115 * receive queue, but transmitted packets will delay socket destruction
2116 * until the last reference will be released.
2117 */
2118
2119 sock_orphan(sk);
2120
2121 xfrm_sk_free_policy(sk);
2122
2123 sk_refcnt_debug_release(sk);
2124 sock_put(sk);
2125 }
2126 EXPORT_SYMBOL(sk_common_release);
2127
2128 static DEFINE_RWLOCK(proto_list_lock);
2129 static LIST_HEAD(proto_list);
2130
2131 #ifdef CONFIG_PROC_FS
2132 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2133 struct prot_inuse {
2134 int val[PROTO_INUSE_NR];
2135 };
2136
2137 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2138
2139 #ifdef CONFIG_NET_NS
2140 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2141 {
2142 int cpu = smp_processor_id();
2143 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2144 }
2145 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2146
2147 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2148 {
2149 int cpu, idx = prot->inuse_idx;
2150 int res = 0;
2151
2152 for_each_possible_cpu(cpu)
2153 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2154
2155 return res >= 0 ? res : 0;
2156 }
2157 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2158
2159 static int __net_init sock_inuse_init_net(struct net *net)
2160 {
2161 net->core.inuse = alloc_percpu(struct prot_inuse);
2162 return net->core.inuse ? 0 : -ENOMEM;
2163 }
2164
2165 static void __net_exit sock_inuse_exit_net(struct net *net)
2166 {
2167 free_percpu(net->core.inuse);
2168 }
2169
2170 static struct pernet_operations net_inuse_ops = {
2171 .init = sock_inuse_init_net,
2172 .exit = sock_inuse_exit_net,
2173 };
2174
2175 static __init int net_inuse_init(void)
2176 {
2177 if (register_pernet_subsys(&net_inuse_ops))
2178 panic("Cannot initialize net inuse counters");
2179
2180 return 0;
2181 }
2182
2183 core_initcall(net_inuse_init);
2184 #else
2185 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2186
2187 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2188 {
2189 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2190 }
2191 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2192
2193 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2194 {
2195 int cpu, idx = prot->inuse_idx;
2196 int res = 0;
2197
2198 for_each_possible_cpu(cpu)
2199 res += per_cpu(prot_inuse, cpu).val[idx];
2200
2201 return res >= 0 ? res : 0;
2202 }
2203 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2204 #endif
2205
2206 static void assign_proto_idx(struct proto *prot)
2207 {
2208 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2209
2210 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2211 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2212 return;
2213 }
2214
2215 set_bit(prot->inuse_idx, proto_inuse_idx);
2216 }
2217
2218 static void release_proto_idx(struct proto *prot)
2219 {
2220 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2221 clear_bit(prot->inuse_idx, proto_inuse_idx);
2222 }
2223 #else
2224 static inline void assign_proto_idx(struct proto *prot)
2225 {
2226 }
2227
2228 static inline void release_proto_idx(struct proto *prot)
2229 {
2230 }
2231 #endif
2232
2233 int proto_register(struct proto *prot, int alloc_slab)
2234 {
2235 if (alloc_slab) {
2236 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2237 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2238 NULL);
2239
2240 if (prot->slab == NULL) {
2241 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2242 prot->name);
2243 goto out;
2244 }
2245
2246 if (prot->rsk_prot != NULL) {
2247 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2248 if (prot->rsk_prot->slab_name == NULL)
2249 goto out_free_sock_slab;
2250
2251 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2252 prot->rsk_prot->obj_size, 0,
2253 SLAB_HWCACHE_ALIGN, NULL);
2254
2255 if (prot->rsk_prot->slab == NULL) {
2256 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2257 prot->name);
2258 goto out_free_request_sock_slab_name;
2259 }
2260 }
2261
2262 if (prot->twsk_prot != NULL) {
2263 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2264
2265 if (prot->twsk_prot->twsk_slab_name == NULL)
2266 goto out_free_request_sock_slab;
2267
2268 prot->twsk_prot->twsk_slab =
2269 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2270 prot->twsk_prot->twsk_obj_size,
2271 0,
2272 SLAB_HWCACHE_ALIGN |
2273 prot->slab_flags,
2274 NULL);
2275 if (prot->twsk_prot->twsk_slab == NULL)
2276 goto out_free_timewait_sock_slab_name;
2277 }
2278 }
2279
2280 write_lock(&proto_list_lock);
2281 list_add(&prot->node, &proto_list);
2282 assign_proto_idx(prot);
2283 write_unlock(&proto_list_lock);
2284 return 0;
2285
2286 out_free_timewait_sock_slab_name:
2287 kfree(prot->twsk_prot->twsk_slab_name);
2288 out_free_request_sock_slab:
2289 if (prot->rsk_prot && prot->rsk_prot->slab) {
2290 kmem_cache_destroy(prot->rsk_prot->slab);
2291 prot->rsk_prot->slab = NULL;
2292 }
2293 out_free_request_sock_slab_name:
2294 if (prot->rsk_prot)
2295 kfree(prot->rsk_prot->slab_name);
2296 out_free_sock_slab:
2297 kmem_cache_destroy(prot->slab);
2298 prot->slab = NULL;
2299 out:
2300 return -ENOBUFS;
2301 }
2302 EXPORT_SYMBOL(proto_register);
2303
2304 void proto_unregister(struct proto *prot)
2305 {
2306 write_lock(&proto_list_lock);
2307 release_proto_idx(prot);
2308 list_del(&prot->node);
2309 write_unlock(&proto_list_lock);
2310
2311 if (prot->slab != NULL) {
2312 kmem_cache_destroy(prot->slab);
2313 prot->slab = NULL;
2314 }
2315
2316 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2317 kmem_cache_destroy(prot->rsk_prot->slab);
2318 kfree(prot->rsk_prot->slab_name);
2319 prot->rsk_prot->slab = NULL;
2320 }
2321
2322 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2323 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2324 kfree(prot->twsk_prot->twsk_slab_name);
2325 prot->twsk_prot->twsk_slab = NULL;
2326 }
2327 }
2328 EXPORT_SYMBOL(proto_unregister);
2329
2330 #ifdef CONFIG_PROC_FS
2331 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2332 __acquires(proto_list_lock)
2333 {
2334 read_lock(&proto_list_lock);
2335 return seq_list_start_head(&proto_list, *pos);
2336 }
2337
2338 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2339 {
2340 return seq_list_next(v, &proto_list, pos);
2341 }
2342
2343 static void proto_seq_stop(struct seq_file *seq, void *v)
2344 __releases(proto_list_lock)
2345 {
2346 read_unlock(&proto_list_lock);
2347 }
2348
2349 static char proto_method_implemented(const void *method)
2350 {
2351 return method == NULL ? 'n' : 'y';
2352 }
2353
2354 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2355 {
2356 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2357 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2358 proto->name,
2359 proto->obj_size,
2360 sock_prot_inuse_get(seq_file_net(seq), proto),
2361 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2362 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2363 proto->max_header,
2364 proto->slab == NULL ? "no" : "yes",
2365 module_name(proto->owner),
2366 proto_method_implemented(proto->close),
2367 proto_method_implemented(proto->connect),
2368 proto_method_implemented(proto->disconnect),
2369 proto_method_implemented(proto->accept),
2370 proto_method_implemented(proto->ioctl),
2371 proto_method_implemented(proto->init),
2372 proto_method_implemented(proto->destroy),
2373 proto_method_implemented(proto->shutdown),
2374 proto_method_implemented(proto->setsockopt),
2375 proto_method_implemented(proto->getsockopt),
2376 proto_method_implemented(proto->sendmsg),
2377 proto_method_implemented(proto->recvmsg),
2378 proto_method_implemented(proto->sendpage),
2379 proto_method_implemented(proto->bind),
2380 proto_method_implemented(proto->backlog_rcv),
2381 proto_method_implemented(proto->hash),
2382 proto_method_implemented(proto->unhash),
2383 proto_method_implemented(proto->get_port),
2384 proto_method_implemented(proto->enter_memory_pressure));
2385 }
2386
2387 static int proto_seq_show(struct seq_file *seq, void *v)
2388 {
2389 if (v == &proto_list)
2390 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2391 "protocol",
2392 "size",
2393 "sockets",
2394 "memory",
2395 "press",
2396 "maxhdr",
2397 "slab",
2398 "module",
2399 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2400 else
2401 proto_seq_printf(seq, list_entry(v, struct proto, node));
2402 return 0;
2403 }
2404
2405 static const struct seq_operations proto_seq_ops = {
2406 .start = proto_seq_start,
2407 .next = proto_seq_next,
2408 .stop = proto_seq_stop,
2409 .show = proto_seq_show,
2410 };
2411
2412 static int proto_seq_open(struct inode *inode, struct file *file)
2413 {
2414 return seq_open_net(inode, file, &proto_seq_ops,
2415 sizeof(struct seq_net_private));
2416 }
2417
2418 static const struct file_operations proto_seq_fops = {
2419 .owner = THIS_MODULE,
2420 .open = proto_seq_open,
2421 .read = seq_read,
2422 .llseek = seq_lseek,
2423 .release = seq_release_net,
2424 };
2425
2426 static __net_init int proto_init_net(struct net *net)
2427 {
2428 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2429 return -ENOMEM;
2430
2431 return 0;
2432 }
2433
2434 static __net_exit void proto_exit_net(struct net *net)
2435 {
2436 proc_net_remove(net, "protocols");
2437 }
2438
2439
2440 static __net_initdata struct pernet_operations proto_net_ops = {
2441 .init = proto_init_net,
2442 .exit = proto_exit_net,
2443 };
2444
2445 static int __init proto_init(void)
2446 {
2447 return register_pernet_subsys(&proto_net_ops);
2448 }
2449
2450 subsys_initcall(proto_init);
2451
2452 #endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.082105 seconds and 5 git commands to generate.