projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
sock: accept SO_TIMESTAMPING flags in socket cmsg
[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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
99 #include <linux/in.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
120
121 #include <asm/uaccess.h>
122
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
135
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
138
139 #include <trace/events/sock.h>
140
141 #ifdef CONFIG_INET
142 #include <net/tcp.h>
143 #endif
144
145 #include <net/busy_poll.h>
146
147 static DEFINE_MUTEX(proto_list_mutex);
148 static LIST_HEAD(proto_list);
149
150 /**
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
155 *
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
159 */
160 bool sk_ns_capable(const struct sock *sk,
161 struct user_namespace *user_ns, int cap)
162 {
163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
164 ns_capable(user_ns, cap);
165 }
166 EXPORT_SYMBOL(sk_ns_capable);
167
168 /**
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
172 *
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
175 * namespaces.
176 */
177 bool sk_capable(const struct sock *sk, int cap)
178 {
179 return sk_ns_capable(sk, &init_user_ns, cap);
180 }
181 EXPORT_SYMBOL(sk_capable);
182
183 /**
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
187 *
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
191 */
192 bool sk_net_capable(const struct sock *sk, int cap)
193 {
194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195 }
196 EXPORT_SYMBOL(sk_net_capable);
197
198 /*
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family:
201 */
202 static struct lock_class_key af_family_keys[AF_MAX];
203 static struct lock_class_key af_family_slock_keys[AF_MAX];
204
205 /*
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
208 * locks is fast):
209 */
210 static const char *const af_family_key_strings[AF_MAX+1] = {
211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
225 };
226 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
227 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
228 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
229 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
230 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
231 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
232 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
233 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
234 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
235 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
236 "slock-27" , "slock-28" , "slock-AF_CAN" ,
237 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
238 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
239 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
240 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
241 };
242 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
243 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
244 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
245 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
246 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
247 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
248 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
249 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
250 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
251 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
252 "clock-27" , "clock-28" , "clock-AF_CAN" ,
253 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
254 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
255 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
256 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
257 };
258
259 /*
260 * sk_callback_lock locking rules are per-address-family,
261 * so split the lock classes by using a per-AF key:
262 */
263 static struct lock_class_key af_callback_keys[AF_MAX];
264
265 /* Take into consideration the size of the struct sk_buff overhead in the
266 * determination of these values, since that is non-constant across
267 * platforms. This makes socket queueing behavior and performance
268 * not depend upon such differences.
269 */
270 #define _SK_MEM_PACKETS 256
271 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
272 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
273 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
274
275 /* Run time adjustable parameters. */
276 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
277 EXPORT_SYMBOL(sysctl_wmem_max);
278 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
279 EXPORT_SYMBOL(sysctl_rmem_max);
280 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
281 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
282
283 /* Maximal space eaten by iovec or ancillary data plus some space */
284 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
285 EXPORT_SYMBOL(sysctl_optmem_max);
286
287 int sysctl_tstamp_allow_data __read_mostly = 1;
288
289 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
290 EXPORT_SYMBOL_GPL(memalloc_socks);
291
292 /**
293 * sk_set_memalloc - sets %SOCK_MEMALLOC
294 * @sk: socket to set it on
295 *
296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
297 * It's the responsibility of the admin to adjust min_free_kbytes
298 * to meet the requirements
299 */
300 void sk_set_memalloc(struct sock *sk)
301 {
302 sock_set_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation |= __GFP_MEMALLOC;
304 static_key_slow_inc(&memalloc_socks);
305 }
306 EXPORT_SYMBOL_GPL(sk_set_memalloc);
307
308 void sk_clear_memalloc(struct sock *sk)
309 {
310 sock_reset_flag(sk, SOCK_MEMALLOC);
311 sk->sk_allocation &= ~__GFP_MEMALLOC;
312 static_key_slow_dec(&memalloc_socks);
313
314 /*
315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
316 * progress of swapping. SOCK_MEMALLOC may be cleared while
317 * it has rmem allocations due to the last swapfile being deactivated
318 * but there is a risk that the socket is unusable due to exceeding
319 * the rmem limits. Reclaim the reserves and obey rmem limits again.
320 */
321 sk_mem_reclaim(sk);
322 }
323 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
324
325 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
326 {
327 int ret;
328 unsigned long pflags = current->flags;
329
330 /* these should have been dropped before queueing */
331 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
332
333 current->flags |= PF_MEMALLOC;
334 ret = sk->sk_backlog_rcv(sk, skb);
335 tsk_restore_flags(current, pflags, PF_MEMALLOC);
336
337 return ret;
338 }
339 EXPORT_SYMBOL(__sk_backlog_rcv);
340
341 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
342 {
343 struct timeval tv;
344
345 if (optlen < sizeof(tv))
346 return -EINVAL;
347 if (copy_from_user(&tv, optval, sizeof(tv)))
348 return -EFAULT;
349 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
350 return -EDOM;
351
352 if (tv.tv_sec < 0) {
353 static int warned __read_mostly;
354
355 *timeo_p = 0;
356 if (warned < 10 && net_ratelimit()) {
357 warned++;
358 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
359 __func__, current->comm, task_pid_nr(current));
360 }
361 return 0;
362 }
363 *timeo_p = MAX_SCHEDULE_TIMEOUT;
364 if (tv.tv_sec == 0 && tv.tv_usec == 0)
365 return 0;
366 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
367 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
368 return 0;
369 }
370
371 static void sock_warn_obsolete_bsdism(const char *name)
372 {
373 static int warned;
374 static char warncomm[TASK_COMM_LEN];
375 if (strcmp(warncomm, current->comm) && warned < 5) {
376 strcpy(warncomm, current->comm);
377 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
378 warncomm, name);
379 warned++;
380 }
381 }
382
383 static bool sock_needs_netstamp(const struct sock *sk)
384 {
385 switch (sk->sk_family) {
386 case AF_UNSPEC:
387 case AF_UNIX:
388 return false;
389 default:
390 return true;
391 }
392 }
393
394 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
395 {
396 if (sk->sk_flags & flags) {
397 sk->sk_flags &= ~flags;
398 if (sock_needs_netstamp(sk) &&
399 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
400 net_disable_timestamp();
401 }
402 }
403
404
405 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
406 {
407 int err;
408 unsigned long flags;
409 struct sk_buff_head *list = &sk->sk_receive_queue;
410
411 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
412 atomic_inc(&sk->sk_drops);
413 trace_sock_rcvqueue_full(sk, skb);
414 return -ENOMEM;
415 }
416
417 err = sk_filter(sk, skb);
418 if (err)
419 return err;
420
421 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
422 atomic_inc(&sk->sk_drops);
423 return -ENOBUFS;
424 }
425
426 skb->dev = NULL;
427 skb_set_owner_r(skb, sk);
428
429 /* we escape from rcu protected region, make sure we dont leak
430 * a norefcounted dst
431 */
432 skb_dst_force(skb);
433
434 spin_lock_irqsave(&list->lock, flags);
435 sock_skb_set_dropcount(sk, skb);
436 __skb_queue_tail(list, skb);
437 spin_unlock_irqrestore(&list->lock, flags);
438
439 if (!sock_flag(sk, SOCK_DEAD))
440 sk->sk_data_ready(sk);
441 return 0;
442 }
443 EXPORT_SYMBOL(sock_queue_rcv_skb);
444
445 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
446 {
447 int rc = NET_RX_SUCCESS;
448
449 if (sk_filter(sk, skb))
450 goto discard_and_relse;
451
452 skb->dev = NULL;
453
454 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
455 atomic_inc(&sk->sk_drops);
456 goto discard_and_relse;
457 }
458 if (nested)
459 bh_lock_sock_nested(sk);
460 else
461 bh_lock_sock(sk);
462 if (!sock_owned_by_user(sk)) {
463 /*
464 * trylock + unlock semantics:
465 */
466 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
467
468 rc = sk_backlog_rcv(sk, skb);
469
470 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
471 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
472 bh_unlock_sock(sk);
473 atomic_inc(&sk->sk_drops);
474 goto discard_and_relse;
475 }
476
477 bh_unlock_sock(sk);
478 out:
479 sock_put(sk);
480 return rc;
481 discard_and_relse:
482 kfree_skb(skb);
483 goto out;
484 }
485 EXPORT_SYMBOL(sk_receive_skb);
486
487 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
488 {
489 struct dst_entry *dst = __sk_dst_get(sk);
490
491 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
492 sk_tx_queue_clear(sk);
493 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
494 dst_release(dst);
495 return NULL;
496 }
497
498 return dst;
499 }
500 EXPORT_SYMBOL(__sk_dst_check);
501
502 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
503 {
504 struct dst_entry *dst = sk_dst_get(sk);
505
506 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
507 sk_dst_reset(sk);
508 dst_release(dst);
509 return NULL;
510 }
511
512 return dst;
513 }
514 EXPORT_SYMBOL(sk_dst_check);
515
516 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
517 int optlen)
518 {
519 int ret = -ENOPROTOOPT;
520 #ifdef CONFIG_NETDEVICES
521 struct net *net = sock_net(sk);
522 char devname[IFNAMSIZ];
523 int index;
524
525 /* Sorry... */
526 ret = -EPERM;
527 if (!ns_capable(net->user_ns, CAP_NET_RAW))
528 goto out;
529
530 ret = -EINVAL;
531 if (optlen < 0)
532 goto out;
533
534 /* Bind this socket to a particular device like "eth0",
535 * as specified in the passed interface name. If the
536 * name is "" or the option length is zero the socket
537 * is not bound.
538 */
539 if (optlen > IFNAMSIZ - 1)
540 optlen = IFNAMSIZ - 1;
541 memset(devname, 0, sizeof(devname));
542
543 ret = -EFAULT;
544 if (copy_from_user(devname, optval, optlen))
545 goto out;
546
547 index = 0;
548 if (devname[0] != '\0') {
549 struct net_device *dev;
550
551 rcu_read_lock();
552 dev = dev_get_by_name_rcu(net, devname);
553 if (dev)
554 index = dev->ifindex;
555 rcu_read_unlock();
556 ret = -ENODEV;
557 if (!dev)
558 goto out;
559 }
560
561 lock_sock(sk);
562 sk->sk_bound_dev_if = index;
563 sk_dst_reset(sk);
564 release_sock(sk);
565
566 ret = 0;
567
568 out:
569 #endif
570
571 return ret;
572 }
573
574 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
575 int __user *optlen, int len)
576 {
577 int ret = -ENOPROTOOPT;
578 #ifdef CONFIG_NETDEVICES
579 struct net *net = sock_net(sk);
580 char devname[IFNAMSIZ];
581
582 if (sk->sk_bound_dev_if == 0) {
583 len = 0;
584 goto zero;
585 }
586
587 ret = -EINVAL;
588 if (len < IFNAMSIZ)
589 goto out;
590
591 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
592 if (ret)
593 goto out;
594
595 len = strlen(devname) + 1;
596
597 ret = -EFAULT;
598 if (copy_to_user(optval, devname, len))
599 goto out;
600
601 zero:
602 ret = -EFAULT;
603 if (put_user(len, optlen))
604 goto out;
605
606 ret = 0;
607
608 out:
609 #endif
610
611 return ret;
612 }
613
614 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
615 {
616 if (valbool)
617 sock_set_flag(sk, bit);
618 else
619 sock_reset_flag(sk, bit);
620 }
621
622 bool sk_mc_loop(struct sock *sk)
623 {
624 if (dev_recursion_level())
625 return false;
626 if (!sk)
627 return true;
628 switch (sk->sk_family) {
629 case AF_INET:
630 return inet_sk(sk)->mc_loop;
631 #if IS_ENABLED(CONFIG_IPV6)
632 case AF_INET6:
633 return inet6_sk(sk)->mc_loop;
634 #endif
635 }
636 WARN_ON(1);
637 return true;
638 }
639 EXPORT_SYMBOL(sk_mc_loop);
640
641 /*
642 * This is meant for all protocols to use and covers goings on
643 * at the socket level. Everything here is generic.
644 */
645
646 int sock_setsockopt(struct socket *sock, int level, int optname,
647 char __user *optval, unsigned int optlen)
648 {
649 struct sock *sk = sock->sk;
650 int val;
651 int valbool;
652 struct linger ling;
653 int ret = 0;
654
655 /*
656 * Options without arguments
657 */
658
659 if (optname == SO_BINDTODEVICE)
660 return sock_setbindtodevice(sk, optval, optlen);
661
662 if (optlen < sizeof(int))
663 return -EINVAL;
664
665 if (get_user(val, (int __user *)optval))
666 return -EFAULT;
667
668 valbool = val ? 1 : 0;
669
670 lock_sock(sk);
671
672 switch (optname) {
673 case SO_DEBUG:
674 if (val && !capable(CAP_NET_ADMIN))
675 ret = -EACCES;
676 else
677 sock_valbool_flag(sk, SOCK_DBG, valbool);
678 break;
679 case SO_REUSEADDR:
680 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
681 break;
682 case SO_REUSEPORT:
683 sk->sk_reuseport = valbool;
684 break;
685 case SO_TYPE:
686 case SO_PROTOCOL:
687 case SO_DOMAIN:
688 case SO_ERROR:
689 ret = -ENOPROTOOPT;
690 break;
691 case SO_DONTROUTE:
692 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
693 break;
694 case SO_BROADCAST:
695 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
696 break;
697 case SO_SNDBUF:
698 /* Don't error on this BSD doesn't and if you think
699 * about it this is right. Otherwise apps have to
700 * play 'guess the biggest size' games. RCVBUF/SNDBUF
701 * are treated in BSD as hints
702 */
703 val = min_t(u32, val, sysctl_wmem_max);
704 set_sndbuf:
705 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
706 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
707 /* Wake up sending tasks if we upped the value. */
708 sk->sk_write_space(sk);
709 break;
710
711 case SO_SNDBUFFORCE:
712 if (!capable(CAP_NET_ADMIN)) {
713 ret = -EPERM;
714 break;
715 }
716 goto set_sndbuf;
717
718 case SO_RCVBUF:
719 /* Don't error on this BSD doesn't and if you think
720 * about it this is right. Otherwise apps have to
721 * play 'guess the biggest size' games. RCVBUF/SNDBUF
722 * are treated in BSD as hints
723 */
724 val = min_t(u32, val, sysctl_rmem_max);
725 set_rcvbuf:
726 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
727 /*
728 * We double it on the way in to account for
729 * "struct sk_buff" etc. overhead. Applications
730 * assume that the SO_RCVBUF setting they make will
731 * allow that much actual data to be received on that
732 * socket.
733 *
734 * Applications are unaware that "struct sk_buff" and
735 * other overheads allocate from the receive buffer
736 * during socket buffer allocation.
737 *
738 * And after considering the possible alternatives,
739 * returning the value we actually used in getsockopt
740 * is the most desirable behavior.
741 */
742 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
743 break;
744
745 case SO_RCVBUFFORCE:
746 if (!capable(CAP_NET_ADMIN)) {
747 ret = -EPERM;
748 break;
749 }
750 goto set_rcvbuf;
751
752 case SO_KEEPALIVE:
753 #ifdef CONFIG_INET
754 if (sk->sk_protocol == IPPROTO_TCP &&
755 sk->sk_type == SOCK_STREAM)
756 tcp_set_keepalive(sk, valbool);
757 #endif
758 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
759 break;
760
761 case SO_OOBINLINE:
762 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
763 break;
764
765 case SO_NO_CHECK:
766 sk->sk_no_check_tx = valbool;
767 break;
768
769 case SO_PRIORITY:
770 if ((val >= 0 && val <= 6) ||
771 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
772 sk->sk_priority = val;
773 else
774 ret = -EPERM;
775 break;
776
777 case SO_LINGER:
778 if (optlen < sizeof(ling)) {
779 ret = -EINVAL; /* 1003.1g */
780 break;
781 }
782 if (copy_from_user(&ling, optval, sizeof(ling))) {
783 ret = -EFAULT;
784 break;
785 }
786 if (!ling.l_onoff)
787 sock_reset_flag(sk, SOCK_LINGER);
788 else {
789 #if (BITS_PER_LONG == 32)
790 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
791 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
792 else
793 #endif
794 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
795 sock_set_flag(sk, SOCK_LINGER);
796 }
797 break;
798
799 case SO_BSDCOMPAT:
800 sock_warn_obsolete_bsdism("setsockopt");
801 break;
802
803 case SO_PASSCRED:
804 if (valbool)
805 set_bit(SOCK_PASSCRED, &sock->flags);
806 else
807 clear_bit(SOCK_PASSCRED, &sock->flags);
808 break;
809
810 case SO_TIMESTAMP:
811 case SO_TIMESTAMPNS:
812 if (valbool) {
813 if (optname == SO_TIMESTAMP)
814 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
815 else
816 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
817 sock_set_flag(sk, SOCK_RCVTSTAMP);
818 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
819 } else {
820 sock_reset_flag(sk, SOCK_RCVTSTAMP);
821 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
822 }
823 break;
824
825 case SO_TIMESTAMPING:
826 if (val & ~SOF_TIMESTAMPING_MASK) {
827 ret = -EINVAL;
828 break;
829 }
830
831 if (val & SOF_TIMESTAMPING_OPT_ID &&
832 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
833 if (sk->sk_protocol == IPPROTO_TCP &&
834 sk->sk_type == SOCK_STREAM) {
835 if ((1 << sk->sk_state) &
836 (TCPF_CLOSE | TCPF_LISTEN)) {
837 ret = -EINVAL;
838 break;
839 }
840 sk->sk_tskey = tcp_sk(sk)->snd_una;
841 } else {
842 sk->sk_tskey = 0;
843 }
844 }
845 sk->sk_tsflags = val;
846 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
847 sock_enable_timestamp(sk,
848 SOCK_TIMESTAMPING_RX_SOFTWARE);
849 else
850 sock_disable_timestamp(sk,
851 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
852 break;
853
854 case SO_RCVLOWAT:
855 if (val < 0)
856 val = INT_MAX;
857 sk->sk_rcvlowat = val ? : 1;
858 break;
859
860 case SO_RCVTIMEO:
861 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
862 break;
863
864 case SO_SNDTIMEO:
865 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
866 break;
867
868 case SO_ATTACH_FILTER:
869 ret = -EINVAL;
870 if (optlen == sizeof(struct sock_fprog)) {
871 struct sock_fprog fprog;
872
873 ret = -EFAULT;
874 if (copy_from_user(&fprog, optval, sizeof(fprog)))
875 break;
876
877 ret = sk_attach_filter(&fprog, sk);
878 }
879 break;
880
881 case SO_ATTACH_BPF:
882 ret = -EINVAL;
883 if (optlen == sizeof(u32)) {
884 u32 ufd;
885
886 ret = -EFAULT;
887 if (copy_from_user(&ufd, optval, sizeof(ufd)))
888 break;
889
890 ret = sk_attach_bpf(ufd, sk);
891 }
892 break;
893
894 case SO_ATTACH_REUSEPORT_CBPF:
895 ret = -EINVAL;
896 if (optlen == sizeof(struct sock_fprog)) {
897 struct sock_fprog fprog;
898
899 ret = -EFAULT;
900 if (copy_from_user(&fprog, optval, sizeof(fprog)))
901 break;
902
903 ret = sk_reuseport_attach_filter(&fprog, sk);
904 }
905 break;
906
907 case SO_ATTACH_REUSEPORT_EBPF:
908 ret = -EINVAL;
909 if (optlen == sizeof(u32)) {
910 u32 ufd;
911
912 ret = -EFAULT;
913 if (copy_from_user(&ufd, optval, sizeof(ufd)))
914 break;
915
916 ret = sk_reuseport_attach_bpf(ufd, sk);
917 }
918 break;
919
920 case SO_DETACH_FILTER:
921 ret = sk_detach_filter(sk);
922 break;
923
924 case SO_LOCK_FILTER:
925 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
926 ret = -EPERM;
927 else
928 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
929 break;
930
931 case SO_PASSSEC:
932 if (valbool)
933 set_bit(SOCK_PASSSEC, &sock->flags);
934 else
935 clear_bit(SOCK_PASSSEC, &sock->flags);
936 break;
937 case SO_MARK:
938 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
939 ret = -EPERM;
940 else
941 sk->sk_mark = val;
942 break;
943
944 case SO_RXQ_OVFL:
945 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
946 break;
947
948 case SO_WIFI_STATUS:
949 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
950 break;
951
952 case SO_PEEK_OFF:
953 if (sock->ops->set_peek_off)
954 ret = sock->ops->set_peek_off(sk, val);
955 else
956 ret = -EOPNOTSUPP;
957 break;
958
959 case SO_NOFCS:
960 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
961 break;
962
963 case SO_SELECT_ERR_QUEUE:
964 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
965 break;
966
967 #ifdef CONFIG_NET_RX_BUSY_POLL
968 case SO_BUSY_POLL:
969 /* allow unprivileged users to decrease the value */
970 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
971 ret = -EPERM;
972 else {
973 if (val < 0)
974 ret = -EINVAL;
975 else
976 sk->sk_ll_usec = val;
977 }
978 break;
979 #endif
980
981 case SO_MAX_PACING_RATE:
982 sk->sk_max_pacing_rate = val;
983 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
984 sk->sk_max_pacing_rate);
985 break;
986
987 case SO_INCOMING_CPU:
988 sk->sk_incoming_cpu = val;
989 break;
990
991 case SO_CNX_ADVICE:
992 if (val == 1)
993 dst_negative_advice(sk);
994 break;
995 default:
996 ret = -ENOPROTOOPT;
997 break;
998 }
999 release_sock(sk);
1000 return ret;
1001 }
1002 EXPORT_SYMBOL(sock_setsockopt);
1003
1004
1005 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1006 struct ucred *ucred)
1007 {
1008 ucred->pid = pid_vnr(pid);
1009 ucred->uid = ucred->gid = -1;
1010 if (cred) {
1011 struct user_namespace *current_ns = current_user_ns();
1012
1013 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1014 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1015 }
1016 }
1017
1018 int sock_getsockopt(struct socket *sock, int level, int optname,
1019 char __user *optval, int __user *optlen)
1020 {
1021 struct sock *sk = sock->sk;
1022
1023 union {
1024 int val;
1025 struct linger ling;
1026 struct timeval tm;
1027 } v;
1028
1029 int lv = sizeof(int);
1030 int len;
1031
1032 if (get_user(len, optlen))
1033 return -EFAULT;
1034 if (len < 0)
1035 return -EINVAL;
1036
1037 memset(&v, 0, sizeof(v));
1038
1039 switch (optname) {
1040 case SO_DEBUG:
1041 v.val = sock_flag(sk, SOCK_DBG);
1042 break;
1043
1044 case SO_DONTROUTE:
1045 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1046 break;
1047
1048 case SO_BROADCAST:
1049 v.val = sock_flag(sk, SOCK_BROADCAST);
1050 break;
1051
1052 case SO_SNDBUF:
1053 v.val = sk->sk_sndbuf;
1054 break;
1055
1056 case SO_RCVBUF:
1057 v.val = sk->sk_rcvbuf;
1058 break;
1059
1060 case SO_REUSEADDR:
1061 v.val = sk->sk_reuse;
1062 break;
1063
1064 case SO_REUSEPORT:
1065 v.val = sk->sk_reuseport;
1066 break;
1067
1068 case SO_KEEPALIVE:
1069 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1070 break;
1071
1072 case SO_TYPE:
1073 v.val = sk->sk_type;
1074 break;
1075
1076 case SO_PROTOCOL:
1077 v.val = sk->sk_protocol;
1078 break;
1079
1080 case SO_DOMAIN:
1081 v.val = sk->sk_family;
1082 break;
1083
1084 case SO_ERROR:
1085 v.val = -sock_error(sk);
1086 if (v.val == 0)
1087 v.val = xchg(&sk->sk_err_soft, 0);
1088 break;
1089
1090 case SO_OOBINLINE:
1091 v.val = sock_flag(sk, SOCK_URGINLINE);
1092 break;
1093
1094 case SO_NO_CHECK:
1095 v.val = sk->sk_no_check_tx;
1096 break;
1097
1098 case SO_PRIORITY:
1099 v.val = sk->sk_priority;
1100 break;
1101
1102 case SO_LINGER:
1103 lv = sizeof(v.ling);
1104 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1105 v.ling.l_linger = sk->sk_lingertime / HZ;
1106 break;
1107
1108 case SO_BSDCOMPAT:
1109 sock_warn_obsolete_bsdism("getsockopt");
1110 break;
1111
1112 case SO_TIMESTAMP:
1113 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1114 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1115 break;
1116
1117 case SO_TIMESTAMPNS:
1118 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1119 break;
1120
1121 case SO_TIMESTAMPING:
1122 v.val = sk->sk_tsflags;
1123 break;
1124
1125 case SO_RCVTIMEO:
1126 lv = sizeof(struct timeval);
1127 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1128 v.tm.tv_sec = 0;
1129 v.tm.tv_usec = 0;
1130 } else {
1131 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1132 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1133 }
1134 break;
1135
1136 case SO_SNDTIMEO:
1137 lv = sizeof(struct timeval);
1138 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1139 v.tm.tv_sec = 0;
1140 v.tm.tv_usec = 0;
1141 } else {
1142 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1143 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1144 }
1145 break;
1146
1147 case SO_RCVLOWAT:
1148 v.val = sk->sk_rcvlowat;
1149 break;
1150
1151 case SO_SNDLOWAT:
1152 v.val = 1;
1153 break;
1154
1155 case SO_PASSCRED:
1156 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1157 break;
1158
1159 case SO_PEERCRED:
1160 {
1161 struct ucred peercred;
1162 if (len > sizeof(peercred))
1163 len = sizeof(peercred);
1164 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1165 if (copy_to_user(optval, &peercred, len))
1166 return -EFAULT;
1167 goto lenout;
1168 }
1169
1170 case SO_PEERNAME:
1171 {
1172 char address[128];
1173
1174 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1175 return -ENOTCONN;
1176 if (lv < len)
1177 return -EINVAL;
1178 if (copy_to_user(optval, address, len))
1179 return -EFAULT;
1180 goto lenout;
1181 }
1182
1183 /* Dubious BSD thing... Probably nobody even uses it, but
1184 * the UNIX standard wants it for whatever reason... -DaveM
1185 */
1186 case SO_ACCEPTCONN:
1187 v.val = sk->sk_state == TCP_LISTEN;
1188 break;
1189
1190 case SO_PASSSEC:
1191 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1192 break;
1193
1194 case SO_PEERSEC:
1195 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1196
1197 case SO_MARK:
1198 v.val = sk->sk_mark;
1199 break;
1200
1201 case SO_RXQ_OVFL:
1202 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1203 break;
1204
1205 case SO_WIFI_STATUS:
1206 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1207 break;
1208
1209 case SO_PEEK_OFF:
1210 if (!sock->ops->set_peek_off)
1211 return -EOPNOTSUPP;
1212
1213 v.val = sk->sk_peek_off;
1214 break;
1215 case SO_NOFCS:
1216 v.val = sock_flag(sk, SOCK_NOFCS);
1217 break;
1218
1219 case SO_BINDTODEVICE:
1220 return sock_getbindtodevice(sk, optval, optlen, len);
1221
1222 case SO_GET_FILTER:
1223 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1224 if (len < 0)
1225 return len;
1226
1227 goto lenout;
1228
1229 case SO_LOCK_FILTER:
1230 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1231 break;
1232
1233 case SO_BPF_EXTENSIONS:
1234 v.val = bpf_tell_extensions();
1235 break;
1236
1237 case SO_SELECT_ERR_QUEUE:
1238 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1239 break;
1240
1241 #ifdef CONFIG_NET_RX_BUSY_POLL
1242 case SO_BUSY_POLL:
1243 v.val = sk->sk_ll_usec;
1244 break;
1245 #endif
1246
1247 case SO_MAX_PACING_RATE:
1248 v.val = sk->sk_max_pacing_rate;
1249 break;
1250
1251 case SO_INCOMING_CPU:
1252 v.val = sk->sk_incoming_cpu;
1253 break;
1254
1255 default:
1256 /* We implement the SO_SNDLOWAT etc to not be settable
1257 * (1003.1g 7).
1258 */
1259 return -ENOPROTOOPT;
1260 }
1261
1262 if (len > lv)
1263 len = lv;
1264 if (copy_to_user(optval, &v, len))
1265 return -EFAULT;
1266 lenout:
1267 if (put_user(len, optlen))
1268 return -EFAULT;
1269 return 0;
1270 }
1271
1272 /*
1273 * Initialize an sk_lock.
1274 *
1275 * (We also register the sk_lock with the lock validator.)
1276 */
1277 static inline void sock_lock_init(struct sock *sk)
1278 {
1279 sock_lock_init_class_and_name(sk,
1280 af_family_slock_key_strings[sk->sk_family],
1281 af_family_slock_keys + sk->sk_family,
1282 af_family_key_strings[sk->sk_family],
1283 af_family_keys + sk->sk_family);
1284 }
1285
1286 /*
1287 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1288 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1289 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1290 */
1291 static void sock_copy(struct sock *nsk, const struct sock *osk)
1292 {
1293 #ifdef CONFIG_SECURITY_NETWORK
1294 void *sptr = nsk->sk_security;
1295 #endif
1296 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1297
1298 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1299 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1300
1301 #ifdef CONFIG_SECURITY_NETWORK
1302 nsk->sk_security = sptr;
1303 security_sk_clone(osk, nsk);
1304 #endif
1305 }
1306
1307 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1308 {
1309 unsigned long nulls1, nulls2;
1310
1311 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1312 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1313 if (nulls1 > nulls2)
1314 swap(nulls1, nulls2);
1315
1316 if (nulls1 != 0)
1317 memset((char *)sk, 0, nulls1);
1318 memset((char *)sk + nulls1 + sizeof(void *), 0,
1319 nulls2 - nulls1 - sizeof(void *));
1320 memset((char *)sk + nulls2 + sizeof(void *), 0,
1321 size - nulls2 - sizeof(void *));
1322 }
1323 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1324
1325 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1326 int family)
1327 {
1328 struct sock *sk;
1329 struct kmem_cache *slab;
1330
1331 slab = prot->slab;
1332 if (slab != NULL) {
1333 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1334 if (!sk)
1335 return sk;
1336 if (priority & __GFP_ZERO) {
1337 if (prot->clear_sk)
1338 prot->clear_sk(sk, prot->obj_size);
1339 else
1340 sk_prot_clear_nulls(sk, prot->obj_size);
1341 }
1342 } else
1343 sk = kmalloc(prot->obj_size, priority);
1344
1345 if (sk != NULL) {
1346 kmemcheck_annotate_bitfield(sk, flags);
1347
1348 if (security_sk_alloc(sk, family, priority))
1349 goto out_free;
1350
1351 if (!try_module_get(prot->owner))
1352 goto out_free_sec;
1353 sk_tx_queue_clear(sk);
1354 cgroup_sk_alloc(&sk->sk_cgrp_data);
1355 }
1356
1357 return sk;
1358
1359 out_free_sec:
1360 security_sk_free(sk);
1361 out_free:
1362 if (slab != NULL)
1363 kmem_cache_free(slab, sk);
1364 else
1365 kfree(sk);
1366 return NULL;
1367 }
1368
1369 static void sk_prot_free(struct proto *prot, struct sock *sk)
1370 {
1371 struct kmem_cache *slab;
1372 struct module *owner;
1373
1374 owner = prot->owner;
1375 slab = prot->slab;
1376
1377 cgroup_sk_free(&sk->sk_cgrp_data);
1378 security_sk_free(sk);
1379 if (slab != NULL)
1380 kmem_cache_free(slab, sk);
1381 else
1382 kfree(sk);
1383 module_put(owner);
1384 }
1385
1386 /**
1387 * sk_alloc - All socket objects are allocated here
1388 * @net: the applicable net namespace
1389 * @family: protocol family
1390 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1391 * @prot: struct proto associated with this new sock instance
1392 * @kern: is this to be a kernel socket?
1393 */
1394 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1395 struct proto *prot, int kern)
1396 {
1397 struct sock *sk;
1398
1399 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1400 if (sk) {
1401 sk->sk_family = family;
1402 /*
1403 * See comment in struct sock definition to understand
1404 * why we need sk_prot_creator -acme
1405 */
1406 sk->sk_prot = sk->sk_prot_creator = prot;
1407 sock_lock_init(sk);
1408 sk->sk_net_refcnt = kern ? 0 : 1;
1409 if (likely(sk->sk_net_refcnt))
1410 get_net(net);
1411 sock_net_set(sk, net);
1412 atomic_set(&sk->sk_wmem_alloc, 1);
1413
1414 sock_update_classid(&sk->sk_cgrp_data);
1415 sock_update_netprioidx(&sk->sk_cgrp_data);
1416 }
1417
1418 return sk;
1419 }
1420 EXPORT_SYMBOL(sk_alloc);
1421
1422 void sk_destruct(struct sock *sk)
1423 {
1424 struct sk_filter *filter;
1425
1426 if (sk->sk_destruct)
1427 sk->sk_destruct(sk);
1428
1429 filter = rcu_dereference_check(sk->sk_filter,
1430 atomic_read(&sk->sk_wmem_alloc) == 0);
1431 if (filter) {
1432 sk_filter_uncharge(sk, filter);
1433 RCU_INIT_POINTER(sk->sk_filter, NULL);
1434 }
1435 if (rcu_access_pointer(sk->sk_reuseport_cb))
1436 reuseport_detach_sock(sk);
1437
1438 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1439
1440 if (atomic_read(&sk->sk_omem_alloc))
1441 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1442 __func__, atomic_read(&sk->sk_omem_alloc));
1443
1444 if (sk->sk_peer_cred)
1445 put_cred(sk->sk_peer_cred);
1446 put_pid(sk->sk_peer_pid);
1447 if (likely(sk->sk_net_refcnt))
1448 put_net(sock_net(sk));
1449 sk_prot_free(sk->sk_prot_creator, sk);
1450 }
1451
1452 static void __sk_free(struct sock *sk)
1453 {
1454 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1455 sock_diag_broadcast_destroy(sk);
1456 else
1457 sk_destruct(sk);
1458 }
1459
1460 void sk_free(struct sock *sk)
1461 {
1462 /*
1463 * We subtract one from sk_wmem_alloc and can know if
1464 * some packets are still in some tx queue.
1465 * If not null, sock_wfree() will call __sk_free(sk) later
1466 */
1467 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1468 __sk_free(sk);
1469 }
1470 EXPORT_SYMBOL(sk_free);
1471
1472 /**
1473 * sk_clone_lock - clone a socket, and lock its clone
1474 * @sk: the socket to clone
1475 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1476 *
1477 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1478 */
1479 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1480 {
1481 struct sock *newsk;
1482 bool is_charged = true;
1483
1484 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1485 if (newsk != NULL) {
1486 struct sk_filter *filter;
1487
1488 sock_copy(newsk, sk);
1489
1490 /* SANITY */
1491 if (likely(newsk->sk_net_refcnt))
1492 get_net(sock_net(newsk));
1493 sk_node_init(&newsk->sk_node);
1494 sock_lock_init(newsk);
1495 bh_lock_sock(newsk);
1496 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1497 newsk->sk_backlog.len = 0;
1498
1499 atomic_set(&newsk->sk_rmem_alloc, 0);
1500 /*
1501 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1502 */
1503 atomic_set(&newsk->sk_wmem_alloc, 1);
1504 atomic_set(&newsk->sk_omem_alloc, 0);
1505 skb_queue_head_init(&newsk->sk_receive_queue);
1506 skb_queue_head_init(&newsk->sk_write_queue);
1507
1508 rwlock_init(&newsk->sk_callback_lock);
1509 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1510 af_callback_keys + newsk->sk_family,
1511 af_family_clock_key_strings[newsk->sk_family]);
1512
1513 newsk->sk_dst_cache = NULL;
1514 newsk->sk_wmem_queued = 0;
1515 newsk->sk_forward_alloc = 0;
1516 newsk->sk_send_head = NULL;
1517 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1518
1519 sock_reset_flag(newsk, SOCK_DONE);
1520 skb_queue_head_init(&newsk->sk_error_queue);
1521
1522 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1523 if (filter != NULL)
1524 /* though it's an empty new sock, the charging may fail
1525 * if sysctl_optmem_max was changed between creation of
1526 * original socket and cloning
1527 */
1528 is_charged = sk_filter_charge(newsk, filter);
1529
1530 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1531 /* It is still raw copy of parent, so invalidate
1532 * destructor and make plain sk_free() */
1533 newsk->sk_destruct = NULL;
1534 bh_unlock_sock(newsk);
1535 sk_free(newsk);
1536 newsk = NULL;
1537 goto out;
1538 }
1539 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1540
1541 newsk->sk_err = 0;
1542 newsk->sk_priority = 0;
1543 newsk->sk_incoming_cpu = raw_smp_processor_id();
1544 atomic64_set(&newsk->sk_cookie, 0);
1545 /*
1546 * Before updating sk_refcnt, we must commit prior changes to memory
1547 * (Documentation/RCU/rculist_nulls.txt for details)
1548 */
1549 smp_wmb();
1550 atomic_set(&newsk->sk_refcnt, 2);
1551
1552 /*
1553 * Increment the counter in the same struct proto as the master
1554 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1555 * is the same as sk->sk_prot->socks, as this field was copied
1556 * with memcpy).
1557 *
1558 * This _changes_ the previous behaviour, where
1559 * tcp_create_openreq_child always was incrementing the
1560 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1561 * to be taken into account in all callers. -acme
1562 */
1563 sk_refcnt_debug_inc(newsk);
1564 sk_set_socket(newsk, NULL);
1565 newsk->sk_wq = NULL;
1566
1567 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1568 sock_update_memcg(newsk);
1569
1570 if (newsk->sk_prot->sockets_allocated)
1571 sk_sockets_allocated_inc(newsk);
1572
1573 if (sock_needs_netstamp(sk) &&
1574 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1575 net_enable_timestamp();
1576 }
1577 out:
1578 return newsk;
1579 }
1580 EXPORT_SYMBOL_GPL(sk_clone_lock);
1581
1582 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1583 {
1584 u32 max_segs = 1;
1585
1586 sk_dst_set(sk, dst);
1587 sk->sk_route_caps = dst->dev->features;
1588 if (sk->sk_route_caps & NETIF_F_GSO)
1589 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1590 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1591 if (sk_can_gso(sk)) {
1592 if (dst->header_len) {
1593 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1594 } else {
1595 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1596 sk->sk_gso_max_size = dst->dev->gso_max_size;
1597 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1598 }
1599 }
1600 sk->sk_gso_max_segs = max_segs;
1601 }
1602 EXPORT_SYMBOL_GPL(sk_setup_caps);
1603
1604 /*
1605 * Simple resource managers for sockets.
1606 */
1607
1608
1609 /*
1610 * Write buffer destructor automatically called from kfree_skb.
1611 */
1612 void sock_wfree(struct sk_buff *skb)
1613 {
1614 struct sock *sk = skb->sk;
1615 unsigned int len = skb->truesize;
1616
1617 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1618 /*
1619 * Keep a reference on sk_wmem_alloc, this will be released
1620 * after sk_write_space() call
1621 */
1622 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1623 sk->sk_write_space(sk);
1624 len = 1;
1625 }
1626 /*
1627 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1628 * could not do because of in-flight packets
1629 */
1630 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1631 __sk_free(sk);
1632 }
1633 EXPORT_SYMBOL(sock_wfree);
1634
1635 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1636 {
1637 skb_orphan(skb);
1638 skb->sk = sk;
1639 #ifdef CONFIG_INET
1640 if (unlikely(!sk_fullsock(sk))) {
1641 skb->destructor = sock_edemux;
1642 sock_hold(sk);
1643 return;
1644 }
1645 #endif
1646 skb->destructor = sock_wfree;
1647 skb_set_hash_from_sk(skb, sk);
1648 /*
1649 * We used to take a refcount on sk, but following operation
1650 * is enough to guarantee sk_free() wont free this sock until
1651 * all in-flight packets are completed
1652 */
1653 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1654 }
1655 EXPORT_SYMBOL(skb_set_owner_w);
1656
1657 void skb_orphan_partial(struct sk_buff *skb)
1658 {
1659 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1660 * so we do not completely orphan skb, but transfert all
1661 * accounted bytes but one, to avoid unexpected reorders.
1662 */
1663 if (skb->destructor == sock_wfree
1664 #ifdef CONFIG_INET
1665 || skb->destructor == tcp_wfree
1666 #endif
1667 ) {
1668 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1669 skb->truesize = 1;
1670 } else {
1671 skb_orphan(skb);
1672 }
1673 }
1674 EXPORT_SYMBOL(skb_orphan_partial);
1675
1676 /*
1677 * Read buffer destructor automatically called from kfree_skb.
1678 */
1679 void sock_rfree(struct sk_buff *skb)
1680 {
1681 struct sock *sk = skb->sk;
1682 unsigned int len = skb->truesize;
1683
1684 atomic_sub(len, &sk->sk_rmem_alloc);
1685 sk_mem_uncharge(sk, len);
1686 }
1687 EXPORT_SYMBOL(sock_rfree);
1688
1689 /*
1690 * Buffer destructor for skbs that are not used directly in read or write
1691 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1692 */
1693 void sock_efree(struct sk_buff *skb)
1694 {
1695 sock_put(skb->sk);
1696 }
1697 EXPORT_SYMBOL(sock_efree);
1698
1699 kuid_t sock_i_uid(struct sock *sk)
1700 {
1701 kuid_t uid;
1702
1703 read_lock_bh(&sk->sk_callback_lock);
1704 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1705 read_unlock_bh(&sk->sk_callback_lock);
1706 return uid;
1707 }
1708 EXPORT_SYMBOL(sock_i_uid);
1709
1710 unsigned long sock_i_ino(struct sock *sk)
1711 {
1712 unsigned long ino;
1713
1714 read_lock_bh(&sk->sk_callback_lock);
1715 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1716 read_unlock_bh(&sk->sk_callback_lock);
1717 return ino;
1718 }
1719 EXPORT_SYMBOL(sock_i_ino);
1720
1721 /*
1722 * Allocate a skb from the socket's send buffer.
1723 */
1724 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1725 gfp_t priority)
1726 {
1727 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1728 struct sk_buff *skb = alloc_skb(size, priority);
1729 if (skb) {
1730 skb_set_owner_w(skb, sk);
1731 return skb;
1732 }
1733 }
1734 return NULL;
1735 }
1736 EXPORT_SYMBOL(sock_wmalloc);
1737
1738 /*
1739 * Allocate a memory block from the socket's option memory buffer.
1740 */
1741 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1742 {
1743 if ((unsigned int)size <= sysctl_optmem_max &&
1744 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1745 void *mem;
1746 /* First do the add, to avoid the race if kmalloc
1747 * might sleep.
1748 */
1749 atomic_add(size, &sk->sk_omem_alloc);
1750 mem = kmalloc(size, priority);
1751 if (mem)
1752 return mem;
1753 atomic_sub(size, &sk->sk_omem_alloc);
1754 }
1755 return NULL;
1756 }
1757 EXPORT_SYMBOL(sock_kmalloc);
1758
1759 /* Free an option memory block. Note, we actually want the inline
1760 * here as this allows gcc to detect the nullify and fold away the
1761 * condition entirely.
1762 */
1763 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1764 const bool nullify)
1765 {
1766 if (WARN_ON_ONCE(!mem))
1767 return;
1768 if (nullify)
1769 kzfree(mem);
1770 else
1771 kfree(mem);
1772 atomic_sub(size, &sk->sk_omem_alloc);
1773 }
1774
1775 void sock_kfree_s(struct sock *sk, void *mem, int size)
1776 {
1777 __sock_kfree_s(sk, mem, size, false);
1778 }
1779 EXPORT_SYMBOL(sock_kfree_s);
1780
1781 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1782 {
1783 __sock_kfree_s(sk, mem, size, true);
1784 }
1785 EXPORT_SYMBOL(sock_kzfree_s);
1786
1787 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1788 I think, these locks should be removed for datagram sockets.
1789 */
1790 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1791 {
1792 DEFINE_WAIT(wait);
1793
1794 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1795 for (;;) {
1796 if (!timeo)
1797 break;
1798 if (signal_pending(current))
1799 break;
1800 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1801 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1802 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1803 break;
1804 if (sk->sk_shutdown & SEND_SHUTDOWN)
1805 break;
1806 if (sk->sk_err)
1807 break;
1808 timeo = schedule_timeout(timeo);
1809 }
1810 finish_wait(sk_sleep(sk), &wait);
1811 return timeo;
1812 }
1813
1814
1815 /*
1816 * Generic send/receive buffer handlers
1817 */
1818
1819 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1820 unsigned long data_len, int noblock,
1821 int *errcode, int max_page_order)
1822 {
1823 struct sk_buff *skb;
1824 long timeo;
1825 int err;
1826
1827 timeo = sock_sndtimeo(sk, noblock);
1828 for (;;) {
1829 err = sock_error(sk);
1830 if (err != 0)
1831 goto failure;
1832
1833 err = -EPIPE;
1834 if (sk->sk_shutdown & SEND_SHUTDOWN)
1835 goto failure;
1836
1837 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1838 break;
1839
1840 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1841 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1842 err = -EAGAIN;
1843 if (!timeo)
1844 goto failure;
1845 if (signal_pending(current))
1846 goto interrupted;
1847 timeo = sock_wait_for_wmem(sk, timeo);
1848 }
1849 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1850 errcode, sk->sk_allocation);
1851 if (skb)
1852 skb_set_owner_w(skb, sk);
1853 return skb;
1854
1855 interrupted:
1856 err = sock_intr_errno(timeo);
1857 failure:
1858 *errcode = err;
1859 return NULL;
1860 }
1861 EXPORT_SYMBOL(sock_alloc_send_pskb);
1862
1863 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1864 int noblock, int *errcode)
1865 {
1866 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1867 }
1868 EXPORT_SYMBOL(sock_alloc_send_skb);
1869
1870 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1871 struct sockcm_cookie *sockc)
1872 {
1873 u32 tsflags;
1874
1875 switch (cmsg->cmsg_type) {
1876 case SO_MARK:
1877 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1878 return -EPERM;
1879 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1880 return -EINVAL;
1881 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1882 break;
1883 case SO_TIMESTAMPING:
1884 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1885 return -EINVAL;
1886
1887 tsflags = *(u32 *)CMSG_DATA(cmsg);
1888 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1889 return -EINVAL;
1890
1891 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1892 sockc->tsflags |= tsflags;
1893 break;
1894 default:
1895 return -EINVAL;
1896 }
1897 return 0;
1898 }
1899 EXPORT_SYMBOL(__sock_cmsg_send);
1900
1901 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1902 struct sockcm_cookie *sockc)
1903 {
1904 struct cmsghdr *cmsg;
1905 int ret;
1906
1907 for_each_cmsghdr(cmsg, msg) {
1908 if (!CMSG_OK(msg, cmsg))
1909 return -EINVAL;
1910 if (cmsg->cmsg_level != SOL_SOCKET)
1911 continue;
1912 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1913 if (ret)
1914 return ret;
1915 }
1916 return 0;
1917 }
1918 EXPORT_SYMBOL(sock_cmsg_send);
1919
1920 /* On 32bit arches, an skb frag is limited to 2^15 */
1921 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1922
1923 /**
1924 * skb_page_frag_refill - check that a page_frag contains enough room
1925 * @sz: minimum size of the fragment we want to get
1926 * @pfrag: pointer to page_frag
1927 * @gfp: priority for memory allocation
1928 *
1929 * Note: While this allocator tries to use high order pages, there is
1930 * no guarantee that allocations succeed. Therefore, @sz MUST be
1931 * less or equal than PAGE_SIZE.
1932 */
1933 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1934 {
1935 if (pfrag->page) {
1936 if (page_ref_count(pfrag->page) == 1) {
1937 pfrag->offset = 0;
1938 return true;
1939 }
1940 if (pfrag->offset + sz <= pfrag->size)
1941 return true;
1942 put_page(pfrag->page);
1943 }
1944
1945 pfrag->offset = 0;
1946 if (SKB_FRAG_PAGE_ORDER) {
1947 /* Avoid direct reclaim but allow kswapd to wake */
1948 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1949 __GFP_COMP | __GFP_NOWARN |
1950 __GFP_NORETRY,
1951 SKB_FRAG_PAGE_ORDER);
1952 if (likely(pfrag->page)) {
1953 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1954 return true;
1955 }
1956 }
1957 pfrag->page = alloc_page(gfp);
1958 if (likely(pfrag->page)) {
1959 pfrag->size = PAGE_SIZE;
1960 return true;
1961 }
1962 return false;
1963 }
1964 EXPORT_SYMBOL(skb_page_frag_refill);
1965
1966 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1967 {
1968 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1969 return true;
1970
1971 sk_enter_memory_pressure(sk);
1972 sk_stream_moderate_sndbuf(sk);
1973 return false;
1974 }
1975 EXPORT_SYMBOL(sk_page_frag_refill);
1976
1977 static void __lock_sock(struct sock *sk)
1978 __releases(&sk->sk_lock.slock)
1979 __acquires(&sk->sk_lock.slock)
1980 {
1981 DEFINE_WAIT(wait);
1982
1983 for (;;) {
1984 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1985 TASK_UNINTERRUPTIBLE);
1986 spin_unlock_bh(&sk->sk_lock.slock);
1987 schedule();
1988 spin_lock_bh(&sk->sk_lock.slock);
1989 if (!sock_owned_by_user(sk))
1990 break;
1991 }
1992 finish_wait(&sk->sk_lock.wq, &wait);
1993 }
1994
1995 static void __release_sock(struct sock *sk)
1996 __releases(&sk->sk_lock.slock)
1997 __acquires(&sk->sk_lock.slock)
1998 {
1999 struct sk_buff *skb = sk->sk_backlog.head;
2000
2001 do {
2002 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2003 bh_unlock_sock(sk);
2004
2005 do {
2006 struct sk_buff *next = skb->next;
2007
2008 prefetch(next);
2009 WARN_ON_ONCE(skb_dst_is_noref(skb));
2010 skb->next = NULL;
2011 sk_backlog_rcv(sk, skb);
2012
2013 /*
2014 * We are in process context here with softirqs
2015 * disabled, use cond_resched_softirq() to preempt.
2016 * This is safe to do because we've taken the backlog
2017 * queue private:
2018 */
2019 cond_resched_softirq();
2020
2021 skb = next;
2022 } while (skb != NULL);
2023
2024 bh_lock_sock(sk);
2025 } while ((skb = sk->sk_backlog.head) != NULL);
2026
2027 /*
2028 * Doing the zeroing here guarantee we can not loop forever
2029 * while a wild producer attempts to flood us.
2030 */
2031 sk->sk_backlog.len = 0;
2032 }
2033
2034 /**
2035 * sk_wait_data - wait for data to arrive at sk_receive_queue
2036 * @sk: sock to wait on
2037 * @timeo: for how long
2038 * @skb: last skb seen on sk_receive_queue
2039 *
2040 * Now socket state including sk->sk_err is changed only under lock,
2041 * hence we may omit checks after joining wait queue.
2042 * We check receive queue before schedule() only as optimization;
2043 * it is very likely that release_sock() added new data.
2044 */
2045 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2046 {
2047 int rc;
2048 DEFINE_WAIT(wait);
2049
2050 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2051 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2052 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2053 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2054 finish_wait(sk_sleep(sk), &wait);
2055 return rc;
2056 }
2057 EXPORT_SYMBOL(sk_wait_data);
2058
2059 /**
2060 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2061 * @sk: socket
2062 * @size: memory size to allocate
2063 * @kind: allocation type
2064 *
2065 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2066 * rmem allocation. This function assumes that protocols which have
2067 * memory_pressure use sk_wmem_queued as write buffer accounting.
2068 */
2069 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2070 {
2071 struct proto *prot = sk->sk_prot;
2072 int amt = sk_mem_pages(size);
2073 long allocated;
2074
2075 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2076
2077 allocated = sk_memory_allocated_add(sk, amt);
2078
2079 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2080 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2081 goto suppress_allocation;
2082
2083 /* Under limit. */
2084 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2085 sk_leave_memory_pressure(sk);
2086 return 1;
2087 }
2088
2089 /* Under pressure. */
2090 if (allocated > sk_prot_mem_limits(sk, 1))
2091 sk_enter_memory_pressure(sk);
2092
2093 /* Over hard limit. */
2094 if (allocated > sk_prot_mem_limits(sk, 2))
2095 goto suppress_allocation;
2096
2097 /* guarantee minimum buffer size under pressure */
2098 if (kind == SK_MEM_RECV) {
2099 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2100 return 1;
2101
2102 } else { /* SK_MEM_SEND */
2103 if (sk->sk_type == SOCK_STREAM) {
2104 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2105 return 1;
2106 } else if (atomic_read(&sk->sk_wmem_alloc) <
2107 prot->sysctl_wmem[0])
2108 return 1;
2109 }
2110
2111 if (sk_has_memory_pressure(sk)) {
2112 int alloc;
2113
2114 if (!sk_under_memory_pressure(sk))
2115 return 1;
2116 alloc = sk_sockets_allocated_read_positive(sk);
2117 if (sk_prot_mem_limits(sk, 2) > alloc *
2118 sk_mem_pages(sk->sk_wmem_queued +
2119 atomic_read(&sk->sk_rmem_alloc) +
2120 sk->sk_forward_alloc))
2121 return 1;
2122 }
2123
2124 suppress_allocation:
2125
2126 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2127 sk_stream_moderate_sndbuf(sk);
2128
2129 /* Fail only if socket is _under_ its sndbuf.
2130 * In this case we cannot block, so that we have to fail.
2131 */
2132 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2133 return 1;
2134 }
2135
2136 trace_sock_exceed_buf_limit(sk, prot, allocated);
2137
2138 /* Alas. Undo changes. */
2139 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2140
2141 sk_memory_allocated_sub(sk, amt);
2142
2143 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2144 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2145
2146 return 0;
2147 }
2148 EXPORT_SYMBOL(__sk_mem_schedule);
2149
2150 /**
2151 * __sk_mem_reclaim - reclaim memory_allocated
2152 * @sk: socket
2153 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2154 */
2155 void __sk_mem_reclaim(struct sock *sk, int amount)
2156 {
2157 amount >>= SK_MEM_QUANTUM_SHIFT;
2158 sk_memory_allocated_sub(sk, amount);
2159 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2160
2161 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2162 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2163
2164 if (sk_under_memory_pressure(sk) &&
2165 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2166 sk_leave_memory_pressure(sk);
2167 }
2168 EXPORT_SYMBOL(__sk_mem_reclaim);
2169
2170
2171 /*
2172 * Set of default routines for initialising struct proto_ops when
2173 * the protocol does not support a particular function. In certain
2174 * cases where it makes no sense for a protocol to have a "do nothing"
2175 * function, some default processing is provided.
2176 */
2177
2178 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2179 {
2180 return -EOPNOTSUPP;
2181 }
2182 EXPORT_SYMBOL(sock_no_bind);
2183
2184 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2185 int len, int flags)
2186 {
2187 return -EOPNOTSUPP;
2188 }
2189 EXPORT_SYMBOL(sock_no_connect);
2190
2191 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2192 {
2193 return -EOPNOTSUPP;
2194 }
2195 EXPORT_SYMBOL(sock_no_socketpair);
2196
2197 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2198 {
2199 return -EOPNOTSUPP;
2200 }
2201 EXPORT_SYMBOL(sock_no_accept);
2202
2203 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2204 int *len, int peer)
2205 {
2206 return -EOPNOTSUPP;
2207 }
2208 EXPORT_SYMBOL(sock_no_getname);
2209
2210 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2211 {
2212 return 0;
2213 }
2214 EXPORT_SYMBOL(sock_no_poll);
2215
2216 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2217 {
2218 return -EOPNOTSUPP;
2219 }
2220 EXPORT_SYMBOL(sock_no_ioctl);
2221
2222 int sock_no_listen(struct socket *sock, int backlog)
2223 {
2224 return -EOPNOTSUPP;
2225 }
2226 EXPORT_SYMBOL(sock_no_listen);
2227
2228 int sock_no_shutdown(struct socket *sock, int how)
2229 {
2230 return -EOPNOTSUPP;
2231 }
2232 EXPORT_SYMBOL(sock_no_shutdown);
2233
2234 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2235 char __user *optval, unsigned int optlen)
2236 {
2237 return -EOPNOTSUPP;
2238 }
2239 EXPORT_SYMBOL(sock_no_setsockopt);
2240
2241 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2242 char __user *optval, int __user *optlen)
2243 {
2244 return -EOPNOTSUPP;
2245 }
2246 EXPORT_SYMBOL(sock_no_getsockopt);
2247
2248 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2249 {
2250 return -EOPNOTSUPP;
2251 }
2252 EXPORT_SYMBOL(sock_no_sendmsg);
2253
2254 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2255 int flags)
2256 {
2257 return -EOPNOTSUPP;
2258 }
2259 EXPORT_SYMBOL(sock_no_recvmsg);
2260
2261 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2262 {
2263 /* Mirror missing mmap method error code */
2264 return -ENODEV;
2265 }
2266 EXPORT_SYMBOL(sock_no_mmap);
2267
2268 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2269 {
2270 ssize_t res;
2271 struct msghdr msg = {.msg_flags = flags};
2272 struct kvec iov;
2273 char *kaddr = kmap(page);
2274 iov.iov_base = kaddr + offset;
2275 iov.iov_len = size;
2276 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2277 kunmap(page);
2278 return res;
2279 }
2280 EXPORT_SYMBOL(sock_no_sendpage);
2281
2282 /*
2283 * Default Socket Callbacks
2284 */
2285
2286 static void sock_def_wakeup(struct sock *sk)
2287 {
2288 struct socket_wq *wq;
2289
2290 rcu_read_lock();
2291 wq = rcu_dereference(sk->sk_wq);
2292 if (skwq_has_sleeper(wq))
2293 wake_up_interruptible_all(&wq->wait);
2294 rcu_read_unlock();
2295 }
2296
2297 static void sock_def_error_report(struct sock *sk)
2298 {
2299 struct socket_wq *wq;
2300
2301 rcu_read_lock();
2302 wq = rcu_dereference(sk->sk_wq);
2303 if (skwq_has_sleeper(wq))
2304 wake_up_interruptible_poll(&wq->wait, POLLERR);
2305 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2306 rcu_read_unlock();
2307 }
2308
2309 static void sock_def_readable(struct sock *sk)
2310 {
2311 struct socket_wq *wq;
2312
2313 rcu_read_lock();
2314 wq = rcu_dereference(sk->sk_wq);
2315 if (skwq_has_sleeper(wq))
2316 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2317 POLLRDNORM | POLLRDBAND);
2318 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2319 rcu_read_unlock();
2320 }
2321
2322 static void sock_def_write_space(struct sock *sk)
2323 {
2324 struct socket_wq *wq;
2325
2326 rcu_read_lock();
2327
2328 /* Do not wake up a writer until he can make "significant"
2329 * progress. --DaveM
2330 */
2331 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2332 wq = rcu_dereference(sk->sk_wq);
2333 if (skwq_has_sleeper(wq))
2334 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2335 POLLWRNORM | POLLWRBAND);
2336
2337 /* Should agree with poll, otherwise some programs break */
2338 if (sock_writeable(sk))
2339 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2340 }
2341
2342 rcu_read_unlock();
2343 }
2344
2345 static void sock_def_destruct(struct sock *sk)
2346 {
2347 }
2348
2349 void sk_send_sigurg(struct sock *sk)
2350 {
2351 if (sk->sk_socket && sk->sk_socket->file)
2352 if (send_sigurg(&sk->sk_socket->file->f_owner))
2353 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2354 }
2355 EXPORT_SYMBOL(sk_send_sigurg);
2356
2357 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2358 unsigned long expires)
2359 {
2360 if (!mod_timer(timer, expires))
2361 sock_hold(sk);
2362 }
2363 EXPORT_SYMBOL(sk_reset_timer);
2364
2365 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2366 {
2367 if (del_timer(timer))
2368 __sock_put(sk);
2369 }
2370 EXPORT_SYMBOL(sk_stop_timer);
2371
2372 void sock_init_data(struct socket *sock, struct sock *sk)
2373 {
2374 skb_queue_head_init(&sk->sk_receive_queue);
2375 skb_queue_head_init(&sk->sk_write_queue);
2376 skb_queue_head_init(&sk->sk_error_queue);
2377
2378 sk->sk_send_head = NULL;
2379
2380 init_timer(&sk->sk_timer);
2381
2382 sk->sk_allocation = GFP_KERNEL;
2383 sk->sk_rcvbuf = sysctl_rmem_default;
2384 sk->sk_sndbuf = sysctl_wmem_default;
2385 sk->sk_state = TCP_CLOSE;
2386 sk_set_socket(sk, sock);
2387
2388 sock_set_flag(sk, SOCK_ZAPPED);
2389
2390 if (sock) {
2391 sk->sk_type = sock->type;
2392 sk->sk_wq = sock->wq;
2393 sock->sk = sk;
2394 } else
2395 sk->sk_wq = NULL;
2396
2397 rwlock_init(&sk->sk_callback_lock);
2398 lockdep_set_class_and_name(&sk->sk_callback_lock,
2399 af_callback_keys + sk->sk_family,
2400 af_family_clock_key_strings[sk->sk_family]);
2401
2402 sk->sk_state_change = sock_def_wakeup;
2403 sk->sk_data_ready = sock_def_readable;
2404 sk->sk_write_space = sock_def_write_space;
2405 sk->sk_error_report = sock_def_error_report;
2406 sk->sk_destruct = sock_def_destruct;
2407
2408 sk->sk_frag.page = NULL;
2409 sk->sk_frag.offset = 0;
2410 sk->sk_peek_off = -1;
2411
2412 sk->sk_peer_pid = NULL;
2413 sk->sk_peer_cred = NULL;
2414 sk->sk_write_pending = 0;
2415 sk->sk_rcvlowat = 1;
2416 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2417 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2418
2419 sk->sk_stamp = ktime_set(-1L, 0);
2420
2421 #ifdef CONFIG_NET_RX_BUSY_POLL
2422 sk->sk_napi_id = 0;
2423 sk->sk_ll_usec = sysctl_net_busy_read;
2424 #endif
2425
2426 sk->sk_max_pacing_rate = ~0U;
2427 sk->sk_pacing_rate = ~0U;
2428 sk->sk_incoming_cpu = -1;
2429 /*
2430 * Before updating sk_refcnt, we must commit prior changes to memory
2431 * (Documentation/RCU/rculist_nulls.txt for details)
2432 */
2433 smp_wmb();
2434 atomic_set(&sk->sk_refcnt, 1);
2435 atomic_set(&sk->sk_drops, 0);
2436 }
2437 EXPORT_SYMBOL(sock_init_data);
2438
2439 void lock_sock_nested(struct sock *sk, int subclass)
2440 {
2441 might_sleep();
2442 spin_lock_bh(&sk->sk_lock.slock);
2443 if (sk->sk_lock.owned)
2444 __lock_sock(sk);
2445 sk->sk_lock.owned = 1;
2446 spin_unlock(&sk->sk_lock.slock);
2447 /*
2448 * The sk_lock has mutex_lock() semantics here:
2449 */
2450 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2451 local_bh_enable();
2452 }
2453 EXPORT_SYMBOL(lock_sock_nested);
2454
2455 void release_sock(struct sock *sk)
2456 {
2457 /*
2458 * The sk_lock has mutex_unlock() semantics:
2459 */
2460 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2461
2462 spin_lock_bh(&sk->sk_lock.slock);
2463 if (sk->sk_backlog.tail)
2464 __release_sock(sk);
2465
2466 /* Warning : release_cb() might need to release sk ownership,
2467 * ie call sock_release_ownership(sk) before us.
2468 */
2469 if (sk->sk_prot->release_cb)
2470 sk->sk_prot->release_cb(sk);
2471
2472 sock_release_ownership(sk);
2473 if (waitqueue_active(&sk->sk_lock.wq))
2474 wake_up(&sk->sk_lock.wq);
2475 spin_unlock_bh(&sk->sk_lock.slock);
2476 }
2477 EXPORT_SYMBOL(release_sock);
2478
2479 /**
2480 * lock_sock_fast - fast version of lock_sock
2481 * @sk: socket
2482 *
2483 * This version should be used for very small section, where process wont block
2484 * return false if fast path is taken
2485 * sk_lock.slock locked, owned = 0, BH disabled
2486 * return true if slow path is taken
2487 * sk_lock.slock unlocked, owned = 1, BH enabled
2488 */
2489 bool lock_sock_fast(struct sock *sk)
2490 {
2491 might_sleep();
2492 spin_lock_bh(&sk->sk_lock.slock);
2493
2494 if (!sk->sk_lock.owned)
2495 /*
2496 * Note : We must disable BH
2497 */
2498 return false;
2499
2500 __lock_sock(sk);
2501 sk->sk_lock.owned = 1;
2502 spin_unlock(&sk->sk_lock.slock);
2503 /*
2504 * The sk_lock has mutex_lock() semantics here:
2505 */
2506 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2507 local_bh_enable();
2508 return true;
2509 }
2510 EXPORT_SYMBOL(lock_sock_fast);
2511
2512 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2513 {
2514 struct timeval tv;
2515 if (!sock_flag(sk, SOCK_TIMESTAMP))
2516 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2517 tv = ktime_to_timeval(sk->sk_stamp);
2518 if (tv.tv_sec == -1)
2519 return -ENOENT;
2520 if (tv.tv_sec == 0) {
2521 sk->sk_stamp = ktime_get_real();
2522 tv = ktime_to_timeval(sk->sk_stamp);
2523 }
2524 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2525 }
2526 EXPORT_SYMBOL(sock_get_timestamp);
2527
2528 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2529 {
2530 struct timespec ts;
2531 if (!sock_flag(sk, SOCK_TIMESTAMP))
2532 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2533 ts = ktime_to_timespec(sk->sk_stamp);
2534 if (ts.tv_sec == -1)
2535 return -ENOENT;
2536 if (ts.tv_sec == 0) {
2537 sk->sk_stamp = ktime_get_real();
2538 ts = ktime_to_timespec(sk->sk_stamp);
2539 }
2540 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2541 }
2542 EXPORT_SYMBOL(sock_get_timestampns);
2543
2544 void sock_enable_timestamp(struct sock *sk, int flag)
2545 {
2546 if (!sock_flag(sk, flag)) {
2547 unsigned long previous_flags = sk->sk_flags;
2548
2549 sock_set_flag(sk, flag);
2550 /*
2551 * we just set one of the two flags which require net
2552 * time stamping, but time stamping might have been on
2553 * already because of the other one
2554 */
2555 if (sock_needs_netstamp(sk) &&
2556 !(previous_flags & SK_FLAGS_TIMESTAMP))
2557 net_enable_timestamp();
2558 }
2559 }
2560
2561 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2562 int level, int type)
2563 {
2564 struct sock_exterr_skb *serr;
2565 struct sk_buff *skb;
2566 int copied, err;
2567
2568 err = -EAGAIN;
2569 skb = sock_dequeue_err_skb(sk);
2570 if (skb == NULL)
2571 goto out;
2572
2573 copied = skb->len;
2574 if (copied > len) {
2575 msg->msg_flags |= MSG_TRUNC;
2576 copied = len;
2577 }
2578 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2579 if (err)
2580 goto out_free_skb;
2581
2582 sock_recv_timestamp(msg, sk, skb);
2583
2584 serr = SKB_EXT_ERR(skb);
2585 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2586
2587 msg->msg_flags |= MSG_ERRQUEUE;
2588 err = copied;
2589
2590 out_free_skb:
2591 kfree_skb(skb);
2592 out:
2593 return err;
2594 }
2595 EXPORT_SYMBOL(sock_recv_errqueue);
2596
2597 /*
2598 * Get a socket option on an socket.
2599 *
2600 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2601 * asynchronous errors should be reported by getsockopt. We assume
2602 * this means if you specify SO_ERROR (otherwise whats the point of it).
2603 */
2604 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2605 char __user *optval, int __user *optlen)
2606 {
2607 struct sock *sk = sock->sk;
2608
2609 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2610 }
2611 EXPORT_SYMBOL(sock_common_getsockopt);
2612
2613 #ifdef CONFIG_COMPAT
2614 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2615 char __user *optval, int __user *optlen)
2616 {
2617 struct sock *sk = sock->sk;
2618
2619 if (sk->sk_prot->compat_getsockopt != NULL)
2620 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2621 optval, optlen);
2622 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2623 }
2624 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2625 #endif
2626
2627 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2628 int flags)
2629 {
2630 struct sock *sk = sock->sk;
2631 int addr_len = 0;
2632 int err;
2633
2634 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2635 flags & ~MSG_DONTWAIT, &addr_len);
2636 if (err >= 0)
2637 msg->msg_namelen = addr_len;
2638 return err;
2639 }
2640 EXPORT_SYMBOL(sock_common_recvmsg);
2641
2642 /*
2643 * Set socket options on an inet socket.
2644 */
2645 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2646 char __user *optval, unsigned int optlen)
2647 {
2648 struct sock *sk = sock->sk;
2649
2650 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2651 }
2652 EXPORT_SYMBOL(sock_common_setsockopt);
2653
2654 #ifdef CONFIG_COMPAT
2655 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2656 char __user *optval, unsigned int optlen)
2657 {
2658 struct sock *sk = sock->sk;
2659
2660 if (sk->sk_prot->compat_setsockopt != NULL)
2661 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2662 optval, optlen);
2663 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2664 }
2665 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2666 #endif
2667
2668 void sk_common_release(struct sock *sk)
2669 {
2670 if (sk->sk_prot->destroy)
2671 sk->sk_prot->destroy(sk);
2672
2673 /*
2674 * Observation: when sock_common_release is called, processes have
2675 * no access to socket. But net still has.
2676 * Step one, detach it from networking:
2677 *
2678 * A. Remove from hash tables.
2679 */
2680
2681 sk->sk_prot->unhash(sk);
2682
2683 /*
2684 * In this point socket cannot receive new packets, but it is possible
2685 * that some packets are in flight because some CPU runs receiver and
2686 * did hash table lookup before we unhashed socket. They will achieve
2687 * receive queue and will be purged by socket destructor.
2688 *
2689 * Also we still have packets pending on receive queue and probably,
2690 * our own packets waiting in device queues. sock_destroy will drain
2691 * receive queue, but transmitted packets will delay socket destruction
2692 * until the last reference will be released.
2693 */
2694
2695 sock_orphan(sk);
2696
2697 xfrm_sk_free_policy(sk);
2698
2699 sk_refcnt_debug_release(sk);
2700
2701 if (sk->sk_frag.page) {
2702 put_page(sk->sk_frag.page);
2703 sk->sk_frag.page = NULL;
2704 }
2705
2706 sock_put(sk);
2707 }
2708 EXPORT_SYMBOL(sk_common_release);
2709
2710 #ifdef CONFIG_PROC_FS
2711 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2712 struct prot_inuse {
2713 int val[PROTO_INUSE_NR];
2714 };
2715
2716 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2717
2718 #ifdef CONFIG_NET_NS
2719 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2720 {
2721 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2722 }
2723 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2724
2725 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2726 {
2727 int cpu, idx = prot->inuse_idx;
2728 int res = 0;
2729
2730 for_each_possible_cpu(cpu)
2731 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2732
2733 return res >= 0 ? res : 0;
2734 }
2735 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2736
2737 static int __net_init sock_inuse_init_net(struct net *net)
2738 {
2739 net->core.inuse = alloc_percpu(struct prot_inuse);
2740 return net->core.inuse ? 0 : -ENOMEM;
2741 }
2742
2743 static void __net_exit sock_inuse_exit_net(struct net *net)
2744 {
2745 free_percpu(net->core.inuse);
2746 }
2747
2748 static struct pernet_operations net_inuse_ops = {
2749 .init = sock_inuse_init_net,
2750 .exit = sock_inuse_exit_net,
2751 };
2752
2753 static __init int net_inuse_init(void)
2754 {
2755 if (register_pernet_subsys(&net_inuse_ops))
2756 panic("Cannot initialize net inuse counters");
2757
2758 return 0;
2759 }
2760
2761 core_initcall(net_inuse_init);
2762 #else
2763 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2764
2765 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2766 {
2767 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2768 }
2769 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2770
2771 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2772 {
2773 int cpu, idx = prot->inuse_idx;
2774 int res = 0;
2775
2776 for_each_possible_cpu(cpu)
2777 res += per_cpu(prot_inuse, cpu).val[idx];
2778
2779 return res >= 0 ? res : 0;
2780 }
2781 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2782 #endif
2783
2784 static void assign_proto_idx(struct proto *prot)
2785 {
2786 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2787
2788 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2789 pr_err("PROTO_INUSE_NR exhausted\n");
2790 return;
2791 }
2792
2793 set_bit(prot->inuse_idx, proto_inuse_idx);
2794 }
2795
2796 static void release_proto_idx(struct proto *prot)
2797 {
2798 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2799 clear_bit(prot->inuse_idx, proto_inuse_idx);
2800 }
2801 #else
2802 static inline void assign_proto_idx(struct proto *prot)
2803 {
2804 }
2805
2806 static inline void release_proto_idx(struct proto *prot)
2807 {
2808 }
2809 #endif
2810
2811 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2812 {
2813 if (!rsk_prot)
2814 return;
2815 kfree(rsk_prot->slab_name);
2816 rsk_prot->slab_name = NULL;
2817 kmem_cache_destroy(rsk_prot->slab);
2818 rsk_prot->slab = NULL;
2819 }
2820
2821 static int req_prot_init(const struct proto *prot)
2822 {
2823 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2824
2825 if (!rsk_prot)
2826 return 0;
2827
2828 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2829 prot->name);
2830 if (!rsk_prot->slab_name)
2831 return -ENOMEM;
2832
2833 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2834 rsk_prot->obj_size, 0,
2835 prot->slab_flags, NULL);
2836
2837 if (!rsk_prot->slab) {
2838 pr_crit("%s: Can't create request sock SLAB cache!\n",
2839 prot->name);
2840 return -ENOMEM;
2841 }
2842 return 0;
2843 }
2844
2845 int proto_register(struct proto *prot, int alloc_slab)
2846 {
2847 if (alloc_slab) {
2848 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2849 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2850 NULL);
2851
2852 if (prot->slab == NULL) {
2853 pr_crit("%s: Can't create sock SLAB cache!\n",
2854 prot->name);
2855 goto out;
2856 }
2857
2858 if (req_prot_init(prot))
2859 goto out_free_request_sock_slab;
2860
2861 if (prot->twsk_prot != NULL) {
2862 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2863
2864 if (prot->twsk_prot->twsk_slab_name == NULL)
2865 goto out_free_request_sock_slab;
2866
2867 prot->twsk_prot->twsk_slab =
2868 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2869 prot->twsk_prot->twsk_obj_size,
2870 0,
2871 prot->slab_flags,
2872 NULL);
2873 if (prot->twsk_prot->twsk_slab == NULL)
2874 goto out_free_timewait_sock_slab_name;
2875 }
2876 }
2877
2878 mutex_lock(&proto_list_mutex);
2879 list_add(&prot->node, &proto_list);
2880 assign_proto_idx(prot);
2881 mutex_unlock(&proto_list_mutex);
2882 return 0;
2883
2884 out_free_timewait_sock_slab_name:
2885 kfree(prot->twsk_prot->twsk_slab_name);
2886 out_free_request_sock_slab:
2887 req_prot_cleanup(prot->rsk_prot);
2888
2889 kmem_cache_destroy(prot->slab);
2890 prot->slab = NULL;
2891 out:
2892 return -ENOBUFS;
2893 }
2894 EXPORT_SYMBOL(proto_register);
2895
2896 void proto_unregister(struct proto *prot)
2897 {
2898 mutex_lock(&proto_list_mutex);
2899 release_proto_idx(prot);
2900 list_del(&prot->node);
2901 mutex_unlock(&proto_list_mutex);
2902
2903 kmem_cache_destroy(prot->slab);
2904 prot->slab = NULL;
2905
2906 req_prot_cleanup(prot->rsk_prot);
2907
2908 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2909 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2910 kfree(prot->twsk_prot->twsk_slab_name);
2911 prot->twsk_prot->twsk_slab = NULL;
2912 }
2913 }
2914 EXPORT_SYMBOL(proto_unregister);
2915
2916 #ifdef CONFIG_PROC_FS
2917 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2918 __acquires(proto_list_mutex)
2919 {
2920 mutex_lock(&proto_list_mutex);
2921 return seq_list_start_head(&proto_list, *pos);
2922 }
2923
2924 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2925 {
2926 return seq_list_next(v, &proto_list, pos);
2927 }
2928
2929 static void proto_seq_stop(struct seq_file *seq, void *v)
2930 __releases(proto_list_mutex)
2931 {
2932 mutex_unlock(&proto_list_mutex);
2933 }
2934
2935 static char proto_method_implemented(const void *method)
2936 {
2937 return method == NULL ? 'n' : 'y';
2938 }
2939 static long sock_prot_memory_allocated(struct proto *proto)
2940 {
2941 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2942 }
2943
2944 static char *sock_prot_memory_pressure(struct proto *proto)
2945 {
2946 return proto->memory_pressure != NULL ?
2947 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2948 }
2949
2950 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2951 {
2952
2953 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2954 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2955 proto->name,
2956 proto->obj_size,
2957 sock_prot_inuse_get(seq_file_net(seq), proto),
2958 sock_prot_memory_allocated(proto),
2959 sock_prot_memory_pressure(proto),
2960 proto->max_header,
2961 proto->slab == NULL ? "no" : "yes",
2962 module_name(proto->owner),
2963 proto_method_implemented(proto->close),
2964 proto_method_implemented(proto->connect),
2965 proto_method_implemented(proto->disconnect),
2966 proto_method_implemented(proto->accept),
2967 proto_method_implemented(proto->ioctl),
2968 proto_method_implemented(proto->init),
2969 proto_method_implemented(proto->destroy),
2970 proto_method_implemented(proto->shutdown),
2971 proto_method_implemented(proto->setsockopt),
2972 proto_method_implemented(proto->getsockopt),
2973 proto_method_implemented(proto->sendmsg),
2974 proto_method_implemented(proto->recvmsg),
2975 proto_method_implemented(proto->sendpage),
2976 proto_method_implemented(proto->bind),
2977 proto_method_implemented(proto->backlog_rcv),
2978 proto_method_implemented(proto->hash),
2979 proto_method_implemented(proto->unhash),
2980 proto_method_implemented(proto->get_port),
2981 proto_method_implemented(proto->enter_memory_pressure));
2982 }
2983
2984 static int proto_seq_show(struct seq_file *seq, void *v)
2985 {
2986 if (v == &proto_list)
2987 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2988 "protocol",
2989 "size",
2990 "sockets",
2991 "memory",
2992 "press",
2993 "maxhdr",
2994 "slab",
2995 "module",
2996 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2997 else
2998 proto_seq_printf(seq, list_entry(v, struct proto, node));
2999 return 0;
3000 }
3001
3002 static const struct seq_operations proto_seq_ops = {
3003 .start = proto_seq_start,
3004 .next = proto_seq_next,
3005 .stop = proto_seq_stop,
3006 .show = proto_seq_show,
3007 };
3008
3009 static int proto_seq_open(struct inode *inode, struct file *file)
3010 {
3011 return seq_open_net(inode, file, &proto_seq_ops,
3012 sizeof(struct seq_net_private));
3013 }
3014
3015 static const struct file_operations proto_seq_fops = {
3016 .owner = THIS_MODULE,
3017 .open = proto_seq_open,
3018 .read = seq_read,
3019 .llseek = seq_lseek,
3020 .release = seq_release_net,
3021 };
3022
3023 static __net_init int proto_init_net(struct net *net)
3024 {
3025 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3026 return -ENOMEM;
3027
3028 return 0;
3029 }
3030
3031 static __net_exit void proto_exit_net(struct net *net)
3032 {
3033 remove_proc_entry("protocols", net->proc_net);
3034 }
3035
3036
3037 static __net_initdata struct pernet_operations proto_net_ops = {
3038 .init = proto_init_net,
3039 .exit = proto_exit_net,
3040 };
3041
3042 static int __init proto_init(void)
3043 {
3044 return register_pernet_subsys(&proto_net_ops);
3045 }
3046
3047 subsys_initcall(proto_init);
3048
3049 #endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.089794 seconds and 6 git commands to generate.