projects / deliverable / linux.git / blob
? search:


2f517ea56786127ce2e5c12684a35d9833f68d9b
[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 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1423 * grace period. This is the case for UDP sockets and TCP listeners.
1424 */
1425 static void __sk_destruct(struct rcu_head *head)
1426 {
1427 struct sock *sk = container_of(head, struct sock, sk_rcu);
1428 struct sk_filter *filter;
1429
1430 if (sk->sk_destruct)
1431 sk->sk_destruct(sk);
1432
1433 filter = rcu_dereference_check(sk->sk_filter,
1434 atomic_read(&sk->sk_wmem_alloc) == 0);
1435 if (filter) {
1436 sk_filter_uncharge(sk, filter);
1437 RCU_INIT_POINTER(sk->sk_filter, NULL);
1438 }
1439 if (rcu_access_pointer(sk->sk_reuseport_cb))
1440 reuseport_detach_sock(sk);
1441
1442 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1443
1444 if (atomic_read(&sk->sk_omem_alloc))
1445 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1446 __func__, atomic_read(&sk->sk_omem_alloc));
1447
1448 if (sk->sk_peer_cred)
1449 put_cred(sk->sk_peer_cred);
1450 put_pid(sk->sk_peer_pid);
1451 if (likely(sk->sk_net_refcnt))
1452 put_net(sock_net(sk));
1453 sk_prot_free(sk->sk_prot_creator, sk);
1454 }
1455
1456 void sk_destruct(struct sock *sk)
1457 {
1458 if (sock_flag(sk, SOCK_RCU_FREE))
1459 call_rcu(&sk->sk_rcu, __sk_destruct);
1460 else
1461 __sk_destruct(&sk->sk_rcu);
1462 }
1463
1464 static void __sk_free(struct sock *sk)
1465 {
1466 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1467 sock_diag_broadcast_destroy(sk);
1468 else
1469 sk_destruct(sk);
1470 }
1471
1472 void sk_free(struct sock *sk)
1473 {
1474 /*
1475 * We subtract one from sk_wmem_alloc and can know if
1476 * some packets are still in some tx queue.
1477 * If not null, sock_wfree() will call __sk_free(sk) later
1478 */
1479 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1480 __sk_free(sk);
1481 }
1482 EXPORT_SYMBOL(sk_free);
1483
1484 /**
1485 * sk_clone_lock - clone a socket, and lock its clone
1486 * @sk: the socket to clone
1487 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1488 *
1489 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1490 */
1491 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1492 {
1493 struct sock *newsk;
1494 bool is_charged = true;
1495
1496 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1497 if (newsk != NULL) {
1498 struct sk_filter *filter;
1499
1500 sock_copy(newsk, sk);
1501
1502 /* SANITY */
1503 if (likely(newsk->sk_net_refcnt))
1504 get_net(sock_net(newsk));
1505 sk_node_init(&newsk->sk_node);
1506 sock_lock_init(newsk);
1507 bh_lock_sock(newsk);
1508 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1509 newsk->sk_backlog.len = 0;
1510
1511 atomic_set(&newsk->sk_rmem_alloc, 0);
1512 /*
1513 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1514 */
1515 atomic_set(&newsk->sk_wmem_alloc, 1);
1516 atomic_set(&newsk->sk_omem_alloc, 0);
1517 skb_queue_head_init(&newsk->sk_receive_queue);
1518 skb_queue_head_init(&newsk->sk_write_queue);
1519
1520 rwlock_init(&newsk->sk_callback_lock);
1521 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1522 af_callback_keys + newsk->sk_family,
1523 af_family_clock_key_strings[newsk->sk_family]);
1524
1525 newsk->sk_dst_cache = NULL;
1526 newsk->sk_wmem_queued = 0;
1527 newsk->sk_forward_alloc = 0;
1528 atomic_set(&newsk->sk_drops, 0);
1529 newsk->sk_send_head = NULL;
1530 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1531
1532 sock_reset_flag(newsk, SOCK_DONE);
1533 skb_queue_head_init(&newsk->sk_error_queue);
1534
1535 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1536 if (filter != NULL)
1537 /* though it's an empty new sock, the charging may fail
1538 * if sysctl_optmem_max was changed between creation of
1539 * original socket and cloning
1540 */
1541 is_charged = sk_filter_charge(newsk, filter);
1542
1543 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1544 /* It is still raw copy of parent, so invalidate
1545 * destructor and make plain sk_free() */
1546 newsk->sk_destruct = NULL;
1547 bh_unlock_sock(newsk);
1548 sk_free(newsk);
1549 newsk = NULL;
1550 goto out;
1551 }
1552 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1553
1554 newsk->sk_err = 0;
1555 newsk->sk_priority = 0;
1556 newsk->sk_incoming_cpu = raw_smp_processor_id();
1557 atomic64_set(&newsk->sk_cookie, 0);
1558 /*
1559 * Before updating sk_refcnt, we must commit prior changes to memory
1560 * (Documentation/RCU/rculist_nulls.txt for details)
1561 */
1562 smp_wmb();
1563 atomic_set(&newsk->sk_refcnt, 2);
1564
1565 /*
1566 * Increment the counter in the same struct proto as the master
1567 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1568 * is the same as sk->sk_prot->socks, as this field was copied
1569 * with memcpy).
1570 *
1571 * This _changes_ the previous behaviour, where
1572 * tcp_create_openreq_child always was incrementing the
1573 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1574 * to be taken into account in all callers. -acme
1575 */
1576 sk_refcnt_debug_inc(newsk);
1577 sk_set_socket(newsk, NULL);
1578 newsk->sk_wq = NULL;
1579
1580 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1581 sock_update_memcg(newsk);
1582
1583 if (newsk->sk_prot->sockets_allocated)
1584 sk_sockets_allocated_inc(newsk);
1585
1586 if (sock_needs_netstamp(sk) &&
1587 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1588 net_enable_timestamp();
1589 }
1590 out:
1591 return newsk;
1592 }
1593 EXPORT_SYMBOL_GPL(sk_clone_lock);
1594
1595 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1596 {
1597 u32 max_segs = 1;
1598
1599 sk_dst_set(sk, dst);
1600 sk->sk_route_caps = dst->dev->features;
1601 if (sk->sk_route_caps & NETIF_F_GSO)
1602 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1603 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1604 if (sk_can_gso(sk)) {
1605 if (dst->header_len) {
1606 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1607 } else {
1608 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1609 sk->sk_gso_max_size = dst->dev->gso_max_size;
1610 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1611 }
1612 }
1613 sk->sk_gso_max_segs = max_segs;
1614 }
1615 EXPORT_SYMBOL_GPL(sk_setup_caps);
1616
1617 /*
1618 * Simple resource managers for sockets.
1619 */
1620
1621
1622 /*
1623 * Write buffer destructor automatically called from kfree_skb.
1624 */
1625 void sock_wfree(struct sk_buff *skb)
1626 {
1627 struct sock *sk = skb->sk;
1628 unsigned int len = skb->truesize;
1629
1630 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1631 /*
1632 * Keep a reference on sk_wmem_alloc, this will be released
1633 * after sk_write_space() call
1634 */
1635 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1636 sk->sk_write_space(sk);
1637 len = 1;
1638 }
1639 /*
1640 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1641 * could not do because of in-flight packets
1642 */
1643 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1644 __sk_free(sk);
1645 }
1646 EXPORT_SYMBOL(sock_wfree);
1647
1648 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1649 {
1650 skb_orphan(skb);
1651 skb->sk = sk;
1652 #ifdef CONFIG_INET
1653 if (unlikely(!sk_fullsock(sk))) {
1654 skb->destructor = sock_edemux;
1655 sock_hold(sk);
1656 return;
1657 }
1658 #endif
1659 skb->destructor = sock_wfree;
1660 skb_set_hash_from_sk(skb, sk);
1661 /*
1662 * We used to take a refcount on sk, but following operation
1663 * is enough to guarantee sk_free() wont free this sock until
1664 * all in-flight packets are completed
1665 */
1666 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1667 }
1668 EXPORT_SYMBOL(skb_set_owner_w);
1669
1670 void skb_orphan_partial(struct sk_buff *skb)
1671 {
1672 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1673 * so we do not completely orphan skb, but transfert all
1674 * accounted bytes but one, to avoid unexpected reorders.
1675 */
1676 if (skb->destructor == sock_wfree
1677 #ifdef CONFIG_INET
1678 || skb->destructor == tcp_wfree
1679 #endif
1680 ) {
1681 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1682 skb->truesize = 1;
1683 } else {
1684 skb_orphan(skb);
1685 }
1686 }
1687 EXPORT_SYMBOL(skb_orphan_partial);
1688
1689 /*
1690 * Read buffer destructor automatically called from kfree_skb.
1691 */
1692 void sock_rfree(struct sk_buff *skb)
1693 {
1694 struct sock *sk = skb->sk;
1695 unsigned int len = skb->truesize;
1696
1697 atomic_sub(len, &sk->sk_rmem_alloc);
1698 sk_mem_uncharge(sk, len);
1699 }
1700 EXPORT_SYMBOL(sock_rfree);
1701
1702 /*
1703 * Buffer destructor for skbs that are not used directly in read or write
1704 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1705 */
1706 void sock_efree(struct sk_buff *skb)
1707 {
1708 sock_put(skb->sk);
1709 }
1710 EXPORT_SYMBOL(sock_efree);
1711
1712 kuid_t sock_i_uid(struct sock *sk)
1713 {
1714 kuid_t uid;
1715
1716 read_lock_bh(&sk->sk_callback_lock);
1717 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1718 read_unlock_bh(&sk->sk_callback_lock);
1719 return uid;
1720 }
1721 EXPORT_SYMBOL(sock_i_uid);
1722
1723 unsigned long sock_i_ino(struct sock *sk)
1724 {
1725 unsigned long ino;
1726
1727 read_lock_bh(&sk->sk_callback_lock);
1728 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1729 read_unlock_bh(&sk->sk_callback_lock);
1730 return ino;
1731 }
1732 EXPORT_SYMBOL(sock_i_ino);
1733
1734 /*
1735 * Allocate a skb from the socket's send buffer.
1736 */
1737 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1738 gfp_t priority)
1739 {
1740 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1741 struct sk_buff *skb = alloc_skb(size, priority);
1742 if (skb) {
1743 skb_set_owner_w(skb, sk);
1744 return skb;
1745 }
1746 }
1747 return NULL;
1748 }
1749 EXPORT_SYMBOL(sock_wmalloc);
1750
1751 /*
1752 * Allocate a memory block from the socket's option memory buffer.
1753 */
1754 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1755 {
1756 if ((unsigned int)size <= sysctl_optmem_max &&
1757 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1758 void *mem;
1759 /* First do the add, to avoid the race if kmalloc
1760 * might sleep.
1761 */
1762 atomic_add(size, &sk->sk_omem_alloc);
1763 mem = kmalloc(size, priority);
1764 if (mem)
1765 return mem;
1766 atomic_sub(size, &sk->sk_omem_alloc);
1767 }
1768 return NULL;
1769 }
1770 EXPORT_SYMBOL(sock_kmalloc);
1771
1772 /* Free an option memory block. Note, we actually want the inline
1773 * here as this allows gcc to detect the nullify and fold away the
1774 * condition entirely.
1775 */
1776 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1777 const bool nullify)
1778 {
1779 if (WARN_ON_ONCE(!mem))
1780 return;
1781 if (nullify)
1782 kzfree(mem);
1783 else
1784 kfree(mem);
1785 atomic_sub(size, &sk->sk_omem_alloc);
1786 }
1787
1788 void sock_kfree_s(struct sock *sk, void *mem, int size)
1789 {
1790 __sock_kfree_s(sk, mem, size, false);
1791 }
1792 EXPORT_SYMBOL(sock_kfree_s);
1793
1794 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1795 {
1796 __sock_kfree_s(sk, mem, size, true);
1797 }
1798 EXPORT_SYMBOL(sock_kzfree_s);
1799
1800 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1801 I think, these locks should be removed for datagram sockets.
1802 */
1803 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1804 {
1805 DEFINE_WAIT(wait);
1806
1807 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1808 for (;;) {
1809 if (!timeo)
1810 break;
1811 if (signal_pending(current))
1812 break;
1813 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1814 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1815 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1816 break;
1817 if (sk->sk_shutdown & SEND_SHUTDOWN)
1818 break;
1819 if (sk->sk_err)
1820 break;
1821 timeo = schedule_timeout(timeo);
1822 }
1823 finish_wait(sk_sleep(sk), &wait);
1824 return timeo;
1825 }
1826
1827
1828 /*
1829 * Generic send/receive buffer handlers
1830 */
1831
1832 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1833 unsigned long data_len, int noblock,
1834 int *errcode, int max_page_order)
1835 {
1836 struct sk_buff *skb;
1837 long timeo;
1838 int err;
1839
1840 timeo = sock_sndtimeo(sk, noblock);
1841 for (;;) {
1842 err = sock_error(sk);
1843 if (err != 0)
1844 goto failure;
1845
1846 err = -EPIPE;
1847 if (sk->sk_shutdown & SEND_SHUTDOWN)
1848 goto failure;
1849
1850 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1851 break;
1852
1853 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1854 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1855 err = -EAGAIN;
1856 if (!timeo)
1857 goto failure;
1858 if (signal_pending(current))
1859 goto interrupted;
1860 timeo = sock_wait_for_wmem(sk, timeo);
1861 }
1862 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1863 errcode, sk->sk_allocation);
1864 if (skb)
1865 skb_set_owner_w(skb, sk);
1866 return skb;
1867
1868 interrupted:
1869 err = sock_intr_errno(timeo);
1870 failure:
1871 *errcode = err;
1872 return NULL;
1873 }
1874 EXPORT_SYMBOL(sock_alloc_send_pskb);
1875
1876 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1877 int noblock, int *errcode)
1878 {
1879 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1880 }
1881 EXPORT_SYMBOL(sock_alloc_send_skb);
1882
1883 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1884 struct sockcm_cookie *sockc)
1885 {
1886 u32 tsflags;
1887
1888 switch (cmsg->cmsg_type) {
1889 case SO_MARK:
1890 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1891 return -EPERM;
1892 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1893 return -EINVAL;
1894 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1895 break;
1896 case SO_TIMESTAMPING:
1897 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1898 return -EINVAL;
1899
1900 tsflags = *(u32 *)CMSG_DATA(cmsg);
1901 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1902 return -EINVAL;
1903
1904 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1905 sockc->tsflags |= tsflags;
1906 break;
1907 default:
1908 return -EINVAL;
1909 }
1910 return 0;
1911 }
1912 EXPORT_SYMBOL(__sock_cmsg_send);
1913
1914 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1915 struct sockcm_cookie *sockc)
1916 {
1917 struct cmsghdr *cmsg;
1918 int ret;
1919
1920 for_each_cmsghdr(cmsg, msg) {
1921 if (!CMSG_OK(msg, cmsg))
1922 return -EINVAL;
1923 if (cmsg->cmsg_level != SOL_SOCKET)
1924 continue;
1925 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1926 if (ret)
1927 return ret;
1928 }
1929 return 0;
1930 }
1931 EXPORT_SYMBOL(sock_cmsg_send);
1932
1933 /* On 32bit arches, an skb frag is limited to 2^15 */
1934 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1935
1936 /**
1937 * skb_page_frag_refill - check that a page_frag contains enough room
1938 * @sz: minimum size of the fragment we want to get
1939 * @pfrag: pointer to page_frag
1940 * @gfp: priority for memory allocation
1941 *
1942 * Note: While this allocator tries to use high order pages, there is
1943 * no guarantee that allocations succeed. Therefore, @sz MUST be
1944 * less or equal than PAGE_SIZE.
1945 */
1946 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1947 {
1948 if (pfrag->page) {
1949 if (page_ref_count(pfrag->page) == 1) {
1950 pfrag->offset = 0;
1951 return true;
1952 }
1953 if (pfrag->offset + sz <= pfrag->size)
1954 return true;
1955 put_page(pfrag->page);
1956 }
1957
1958 pfrag->offset = 0;
1959 if (SKB_FRAG_PAGE_ORDER) {
1960 /* Avoid direct reclaim but allow kswapd to wake */
1961 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1962 __GFP_COMP | __GFP_NOWARN |
1963 __GFP_NORETRY,
1964 SKB_FRAG_PAGE_ORDER);
1965 if (likely(pfrag->page)) {
1966 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1967 return true;
1968 }
1969 }
1970 pfrag->page = alloc_page(gfp);
1971 if (likely(pfrag->page)) {
1972 pfrag->size = PAGE_SIZE;
1973 return true;
1974 }
1975 return false;
1976 }
1977 EXPORT_SYMBOL(skb_page_frag_refill);
1978
1979 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1980 {
1981 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1982 return true;
1983
1984 sk_enter_memory_pressure(sk);
1985 sk_stream_moderate_sndbuf(sk);
1986 return false;
1987 }
1988 EXPORT_SYMBOL(sk_page_frag_refill);
1989
1990 static void __lock_sock(struct sock *sk)
1991 __releases(&sk->sk_lock.slock)
1992 __acquires(&sk->sk_lock.slock)
1993 {
1994 DEFINE_WAIT(wait);
1995
1996 for (;;) {
1997 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1998 TASK_UNINTERRUPTIBLE);
1999 spin_unlock_bh(&sk->sk_lock.slock);
2000 schedule();
2001 spin_lock_bh(&sk->sk_lock.slock);
2002 if (!sock_owned_by_user(sk))
2003 break;
2004 }
2005 finish_wait(&sk->sk_lock.wq, &wait);
2006 }
2007
2008 static void __release_sock(struct sock *sk)
2009 __releases(&sk->sk_lock.slock)
2010 __acquires(&sk->sk_lock.slock)
2011 {
2012 struct sk_buff *skb = sk->sk_backlog.head;
2013
2014 do {
2015 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2016 bh_unlock_sock(sk);
2017
2018 do {
2019 struct sk_buff *next = skb->next;
2020
2021 prefetch(next);
2022 WARN_ON_ONCE(skb_dst_is_noref(skb));
2023 skb->next = NULL;
2024 sk_backlog_rcv(sk, skb);
2025
2026 /*
2027 * We are in process context here with softirqs
2028 * disabled, use cond_resched_softirq() to preempt.
2029 * This is safe to do because we've taken the backlog
2030 * queue private:
2031 */
2032 cond_resched_softirq();
2033
2034 skb = next;
2035 } while (skb != NULL);
2036
2037 bh_lock_sock(sk);
2038 } while ((skb = sk->sk_backlog.head) != NULL);
2039
2040 /*
2041 * Doing the zeroing here guarantee we can not loop forever
2042 * while a wild producer attempts to flood us.
2043 */
2044 sk->sk_backlog.len = 0;
2045 }
2046
2047 /**
2048 * sk_wait_data - wait for data to arrive at sk_receive_queue
2049 * @sk: sock to wait on
2050 * @timeo: for how long
2051 * @skb: last skb seen on sk_receive_queue
2052 *
2053 * Now socket state including sk->sk_err is changed only under lock,
2054 * hence we may omit checks after joining wait queue.
2055 * We check receive queue before schedule() only as optimization;
2056 * it is very likely that release_sock() added new data.
2057 */
2058 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2059 {
2060 int rc;
2061 DEFINE_WAIT(wait);
2062
2063 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2064 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2065 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2066 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2067 finish_wait(sk_sleep(sk), &wait);
2068 return rc;
2069 }
2070 EXPORT_SYMBOL(sk_wait_data);
2071
2072 /**
2073 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2074 * @sk: socket
2075 * @size: memory size to allocate
2076 * @kind: allocation type
2077 *
2078 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2079 * rmem allocation. This function assumes that protocols which have
2080 * memory_pressure use sk_wmem_queued as write buffer accounting.
2081 */
2082 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2083 {
2084 struct proto *prot = sk->sk_prot;
2085 int amt = sk_mem_pages(size);
2086 long allocated;
2087
2088 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2089
2090 allocated = sk_memory_allocated_add(sk, amt);
2091
2092 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2093 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2094 goto suppress_allocation;
2095
2096 /* Under limit. */
2097 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2098 sk_leave_memory_pressure(sk);
2099 return 1;
2100 }
2101
2102 /* Under pressure. */
2103 if (allocated > sk_prot_mem_limits(sk, 1))
2104 sk_enter_memory_pressure(sk);
2105
2106 /* Over hard limit. */
2107 if (allocated > sk_prot_mem_limits(sk, 2))
2108 goto suppress_allocation;
2109
2110 /* guarantee minimum buffer size under pressure */
2111 if (kind == SK_MEM_RECV) {
2112 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2113 return 1;
2114
2115 } else { /* SK_MEM_SEND */
2116 if (sk->sk_type == SOCK_STREAM) {
2117 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2118 return 1;
2119 } else if (atomic_read(&sk->sk_wmem_alloc) <
2120 prot->sysctl_wmem[0])
2121 return 1;
2122 }
2123
2124 if (sk_has_memory_pressure(sk)) {
2125 int alloc;
2126
2127 if (!sk_under_memory_pressure(sk))
2128 return 1;
2129 alloc = sk_sockets_allocated_read_positive(sk);
2130 if (sk_prot_mem_limits(sk, 2) > alloc *
2131 sk_mem_pages(sk->sk_wmem_queued +
2132 atomic_read(&sk->sk_rmem_alloc) +
2133 sk->sk_forward_alloc))
2134 return 1;
2135 }
2136
2137 suppress_allocation:
2138
2139 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2140 sk_stream_moderate_sndbuf(sk);
2141
2142 /* Fail only if socket is _under_ its sndbuf.
2143 * In this case we cannot block, so that we have to fail.
2144 */
2145 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2146 return 1;
2147 }
2148
2149 trace_sock_exceed_buf_limit(sk, prot, allocated);
2150
2151 /* Alas. Undo changes. */
2152 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2153
2154 sk_memory_allocated_sub(sk, amt);
2155
2156 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2157 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2158
2159 return 0;
2160 }
2161 EXPORT_SYMBOL(__sk_mem_schedule);
2162
2163 /**
2164 * __sk_mem_reclaim - reclaim memory_allocated
2165 * @sk: socket
2166 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2167 */
2168 void __sk_mem_reclaim(struct sock *sk, int amount)
2169 {
2170 amount >>= SK_MEM_QUANTUM_SHIFT;
2171 sk_memory_allocated_sub(sk, amount);
2172 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2173
2174 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2175 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2176
2177 if (sk_under_memory_pressure(sk) &&
2178 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2179 sk_leave_memory_pressure(sk);
2180 }
2181 EXPORT_SYMBOL(__sk_mem_reclaim);
2182
2183
2184 /*
2185 * Set of default routines for initialising struct proto_ops when
2186 * the protocol does not support a particular function. In certain
2187 * cases where it makes no sense for a protocol to have a "do nothing"
2188 * function, some default processing is provided.
2189 */
2190
2191 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2192 {
2193 return -EOPNOTSUPP;
2194 }
2195 EXPORT_SYMBOL(sock_no_bind);
2196
2197 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2198 int len, int flags)
2199 {
2200 return -EOPNOTSUPP;
2201 }
2202 EXPORT_SYMBOL(sock_no_connect);
2203
2204 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2205 {
2206 return -EOPNOTSUPP;
2207 }
2208 EXPORT_SYMBOL(sock_no_socketpair);
2209
2210 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2211 {
2212 return -EOPNOTSUPP;
2213 }
2214 EXPORT_SYMBOL(sock_no_accept);
2215
2216 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2217 int *len, int peer)
2218 {
2219 return -EOPNOTSUPP;
2220 }
2221 EXPORT_SYMBOL(sock_no_getname);
2222
2223 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2224 {
2225 return 0;
2226 }
2227 EXPORT_SYMBOL(sock_no_poll);
2228
2229 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2230 {
2231 return -EOPNOTSUPP;
2232 }
2233 EXPORT_SYMBOL(sock_no_ioctl);
2234
2235 int sock_no_listen(struct socket *sock, int backlog)
2236 {
2237 return -EOPNOTSUPP;
2238 }
2239 EXPORT_SYMBOL(sock_no_listen);
2240
2241 int sock_no_shutdown(struct socket *sock, int how)
2242 {
2243 return -EOPNOTSUPP;
2244 }
2245 EXPORT_SYMBOL(sock_no_shutdown);
2246
2247 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2248 char __user *optval, unsigned int optlen)
2249 {
2250 return -EOPNOTSUPP;
2251 }
2252 EXPORT_SYMBOL(sock_no_setsockopt);
2253
2254 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2255 char __user *optval, int __user *optlen)
2256 {
2257 return -EOPNOTSUPP;
2258 }
2259 EXPORT_SYMBOL(sock_no_getsockopt);
2260
2261 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2262 {
2263 return -EOPNOTSUPP;
2264 }
2265 EXPORT_SYMBOL(sock_no_sendmsg);
2266
2267 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2268 int flags)
2269 {
2270 return -EOPNOTSUPP;
2271 }
2272 EXPORT_SYMBOL(sock_no_recvmsg);
2273
2274 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2275 {
2276 /* Mirror missing mmap method error code */
2277 return -ENODEV;
2278 }
2279 EXPORT_SYMBOL(sock_no_mmap);
2280
2281 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2282 {
2283 ssize_t res;
2284 struct msghdr msg = {.msg_flags = flags};
2285 struct kvec iov;
2286 char *kaddr = kmap(page);
2287 iov.iov_base = kaddr + offset;
2288 iov.iov_len = size;
2289 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2290 kunmap(page);
2291 return res;
2292 }
2293 EXPORT_SYMBOL(sock_no_sendpage);
2294
2295 /*
2296 * Default Socket Callbacks
2297 */
2298
2299 static void sock_def_wakeup(struct sock *sk)
2300 {
2301 struct socket_wq *wq;
2302
2303 rcu_read_lock();
2304 wq = rcu_dereference(sk->sk_wq);
2305 if (skwq_has_sleeper(wq))
2306 wake_up_interruptible_all(&wq->wait);
2307 rcu_read_unlock();
2308 }
2309
2310 static void sock_def_error_report(struct sock *sk)
2311 {
2312 struct socket_wq *wq;
2313
2314 rcu_read_lock();
2315 wq = rcu_dereference(sk->sk_wq);
2316 if (skwq_has_sleeper(wq))
2317 wake_up_interruptible_poll(&wq->wait, POLLERR);
2318 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2319 rcu_read_unlock();
2320 }
2321
2322 static void sock_def_readable(struct sock *sk)
2323 {
2324 struct socket_wq *wq;
2325
2326 rcu_read_lock();
2327 wq = rcu_dereference(sk->sk_wq);
2328 if (skwq_has_sleeper(wq))
2329 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2330 POLLRDNORM | POLLRDBAND);
2331 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2332 rcu_read_unlock();
2333 }
2334
2335 static void sock_def_write_space(struct sock *sk)
2336 {
2337 struct socket_wq *wq;
2338
2339 rcu_read_lock();
2340
2341 /* Do not wake up a writer until he can make "significant"
2342 * progress. --DaveM
2343 */
2344 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2345 wq = rcu_dereference(sk->sk_wq);
2346 if (skwq_has_sleeper(wq))
2347 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2348 POLLWRNORM | POLLWRBAND);
2349
2350 /* Should agree with poll, otherwise some programs break */
2351 if (sock_writeable(sk))
2352 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2353 }
2354
2355 rcu_read_unlock();
2356 }
2357
2358 static void sock_def_destruct(struct sock *sk)
2359 {
2360 }
2361
2362 void sk_send_sigurg(struct sock *sk)
2363 {
2364 if (sk->sk_socket && sk->sk_socket->file)
2365 if (send_sigurg(&sk->sk_socket->file->f_owner))
2366 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2367 }
2368 EXPORT_SYMBOL(sk_send_sigurg);
2369
2370 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2371 unsigned long expires)
2372 {
2373 if (!mod_timer(timer, expires))
2374 sock_hold(sk);
2375 }
2376 EXPORT_SYMBOL(sk_reset_timer);
2377
2378 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2379 {
2380 if (del_timer(timer))
2381 __sock_put(sk);
2382 }
2383 EXPORT_SYMBOL(sk_stop_timer);
2384
2385 void sock_init_data(struct socket *sock, struct sock *sk)
2386 {
2387 skb_queue_head_init(&sk->sk_receive_queue);
2388 skb_queue_head_init(&sk->sk_write_queue);
2389 skb_queue_head_init(&sk->sk_error_queue);
2390
2391 sk->sk_send_head = NULL;
2392
2393 init_timer(&sk->sk_timer);
2394
2395 sk->sk_allocation = GFP_KERNEL;
2396 sk->sk_rcvbuf = sysctl_rmem_default;
2397 sk->sk_sndbuf = sysctl_wmem_default;
2398 sk->sk_state = TCP_CLOSE;
2399 sk_set_socket(sk, sock);
2400
2401 sock_set_flag(sk, SOCK_ZAPPED);
2402
2403 if (sock) {
2404 sk->sk_type = sock->type;
2405 sk->sk_wq = sock->wq;
2406 sock->sk = sk;
2407 } else
2408 sk->sk_wq = NULL;
2409
2410 rwlock_init(&sk->sk_callback_lock);
2411 lockdep_set_class_and_name(&sk->sk_callback_lock,
2412 af_callback_keys + sk->sk_family,
2413 af_family_clock_key_strings[sk->sk_family]);
2414
2415 sk->sk_state_change = sock_def_wakeup;
2416 sk->sk_data_ready = sock_def_readable;
2417 sk->sk_write_space = sock_def_write_space;
2418 sk->sk_error_report = sock_def_error_report;
2419 sk->sk_destruct = sock_def_destruct;
2420
2421 sk->sk_frag.page = NULL;
2422 sk->sk_frag.offset = 0;
2423 sk->sk_peek_off = -1;
2424
2425 sk->sk_peer_pid = NULL;
2426 sk->sk_peer_cred = NULL;
2427 sk->sk_write_pending = 0;
2428 sk->sk_rcvlowat = 1;
2429 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2430 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2431
2432 sk->sk_stamp = ktime_set(-1L, 0);
2433
2434 #ifdef CONFIG_NET_RX_BUSY_POLL
2435 sk->sk_napi_id = 0;
2436 sk->sk_ll_usec = sysctl_net_busy_read;
2437 #endif
2438
2439 sk->sk_max_pacing_rate = ~0U;
2440 sk->sk_pacing_rate = ~0U;
2441 sk->sk_incoming_cpu = -1;
2442 /*
2443 * Before updating sk_refcnt, we must commit prior changes to memory
2444 * (Documentation/RCU/rculist_nulls.txt for details)
2445 */
2446 smp_wmb();
2447 atomic_set(&sk->sk_refcnt, 1);
2448 atomic_set(&sk->sk_drops, 0);
2449 }
2450 EXPORT_SYMBOL(sock_init_data);
2451
2452 void lock_sock_nested(struct sock *sk, int subclass)
2453 {
2454 might_sleep();
2455 spin_lock_bh(&sk->sk_lock.slock);
2456 if (sk->sk_lock.owned)
2457 __lock_sock(sk);
2458 sk->sk_lock.owned = 1;
2459 spin_unlock(&sk->sk_lock.slock);
2460 /*
2461 * The sk_lock has mutex_lock() semantics here:
2462 */
2463 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2464 local_bh_enable();
2465 }
2466 EXPORT_SYMBOL(lock_sock_nested);
2467
2468 void release_sock(struct sock *sk)
2469 {
2470 /*
2471 * The sk_lock has mutex_unlock() semantics:
2472 */
2473 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2474
2475 spin_lock_bh(&sk->sk_lock.slock);
2476 if (sk->sk_backlog.tail)
2477 __release_sock(sk);
2478
2479 /* Warning : release_cb() might need to release sk ownership,
2480 * ie call sock_release_ownership(sk) before us.
2481 */
2482 if (sk->sk_prot->release_cb)
2483 sk->sk_prot->release_cb(sk);
2484
2485 sock_release_ownership(sk);
2486 if (waitqueue_active(&sk->sk_lock.wq))
2487 wake_up(&sk->sk_lock.wq);
2488 spin_unlock_bh(&sk->sk_lock.slock);
2489 }
2490 EXPORT_SYMBOL(release_sock);
2491
2492 /**
2493 * lock_sock_fast - fast version of lock_sock
2494 * @sk: socket
2495 *
2496 * This version should be used for very small section, where process wont block
2497 * return false if fast path is taken
2498 * sk_lock.slock locked, owned = 0, BH disabled
2499 * return true if slow path is taken
2500 * sk_lock.slock unlocked, owned = 1, BH enabled
2501 */
2502 bool lock_sock_fast(struct sock *sk)
2503 {
2504 might_sleep();
2505 spin_lock_bh(&sk->sk_lock.slock);
2506
2507 if (!sk->sk_lock.owned)
2508 /*
2509 * Note : We must disable BH
2510 */
2511 return false;
2512
2513 __lock_sock(sk);
2514 sk->sk_lock.owned = 1;
2515 spin_unlock(&sk->sk_lock.slock);
2516 /*
2517 * The sk_lock has mutex_lock() semantics here:
2518 */
2519 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2520 local_bh_enable();
2521 return true;
2522 }
2523 EXPORT_SYMBOL(lock_sock_fast);
2524
2525 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2526 {
2527 struct timeval tv;
2528 if (!sock_flag(sk, SOCK_TIMESTAMP))
2529 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2530 tv = ktime_to_timeval(sk->sk_stamp);
2531 if (tv.tv_sec == -1)
2532 return -ENOENT;
2533 if (tv.tv_sec == 0) {
2534 sk->sk_stamp = ktime_get_real();
2535 tv = ktime_to_timeval(sk->sk_stamp);
2536 }
2537 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2538 }
2539 EXPORT_SYMBOL(sock_get_timestamp);
2540
2541 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2542 {
2543 struct timespec ts;
2544 if (!sock_flag(sk, SOCK_TIMESTAMP))
2545 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2546 ts = ktime_to_timespec(sk->sk_stamp);
2547 if (ts.tv_sec == -1)
2548 return -ENOENT;
2549 if (ts.tv_sec == 0) {
2550 sk->sk_stamp = ktime_get_real();
2551 ts = ktime_to_timespec(sk->sk_stamp);
2552 }
2553 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2554 }
2555 EXPORT_SYMBOL(sock_get_timestampns);
2556
2557 void sock_enable_timestamp(struct sock *sk, int flag)
2558 {
2559 if (!sock_flag(sk, flag)) {
2560 unsigned long previous_flags = sk->sk_flags;
2561
2562 sock_set_flag(sk, flag);
2563 /*
2564 * we just set one of the two flags which require net
2565 * time stamping, but time stamping might have been on
2566 * already because of the other one
2567 */
2568 if (sock_needs_netstamp(sk) &&
2569 !(previous_flags & SK_FLAGS_TIMESTAMP))
2570 net_enable_timestamp();
2571 }
2572 }
2573
2574 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2575 int level, int type)
2576 {
2577 struct sock_exterr_skb *serr;
2578 struct sk_buff *skb;
2579 int copied, err;
2580
2581 err = -EAGAIN;
2582 skb = sock_dequeue_err_skb(sk);
2583 if (skb == NULL)
2584 goto out;
2585
2586 copied = skb->len;
2587 if (copied > len) {
2588 msg->msg_flags |= MSG_TRUNC;
2589 copied = len;
2590 }
2591 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2592 if (err)
2593 goto out_free_skb;
2594
2595 sock_recv_timestamp(msg, sk, skb);
2596
2597 serr = SKB_EXT_ERR(skb);
2598 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2599
2600 msg->msg_flags |= MSG_ERRQUEUE;
2601 err = copied;
2602
2603 out_free_skb:
2604 kfree_skb(skb);
2605 out:
2606 return err;
2607 }
2608 EXPORT_SYMBOL(sock_recv_errqueue);
2609
2610 /*
2611 * Get a socket option on an socket.
2612 *
2613 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2614 * asynchronous errors should be reported by getsockopt. We assume
2615 * this means if you specify SO_ERROR (otherwise whats the point of it).
2616 */
2617 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2618 char __user *optval, int __user *optlen)
2619 {
2620 struct sock *sk = sock->sk;
2621
2622 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2623 }
2624 EXPORT_SYMBOL(sock_common_getsockopt);
2625
2626 #ifdef CONFIG_COMPAT
2627 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2628 char __user *optval, int __user *optlen)
2629 {
2630 struct sock *sk = sock->sk;
2631
2632 if (sk->sk_prot->compat_getsockopt != NULL)
2633 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2634 optval, optlen);
2635 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2636 }
2637 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2638 #endif
2639
2640 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2641 int flags)
2642 {
2643 struct sock *sk = sock->sk;
2644 int addr_len = 0;
2645 int err;
2646
2647 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2648 flags & ~MSG_DONTWAIT, &addr_len);
2649 if (err >= 0)
2650 msg->msg_namelen = addr_len;
2651 return err;
2652 }
2653 EXPORT_SYMBOL(sock_common_recvmsg);
2654
2655 /*
2656 * Set socket options on an inet socket.
2657 */
2658 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2659 char __user *optval, unsigned int optlen)
2660 {
2661 struct sock *sk = sock->sk;
2662
2663 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2664 }
2665 EXPORT_SYMBOL(sock_common_setsockopt);
2666
2667 #ifdef CONFIG_COMPAT
2668 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2669 char __user *optval, unsigned int optlen)
2670 {
2671 struct sock *sk = sock->sk;
2672
2673 if (sk->sk_prot->compat_setsockopt != NULL)
2674 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2675 optval, optlen);
2676 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2677 }
2678 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2679 #endif
2680
2681 void sk_common_release(struct sock *sk)
2682 {
2683 if (sk->sk_prot->destroy)
2684 sk->sk_prot->destroy(sk);
2685
2686 /*
2687 * Observation: when sock_common_release is called, processes have
2688 * no access to socket. But net still has.
2689 * Step one, detach it from networking:
2690 *
2691 * A. Remove from hash tables.
2692 */
2693
2694 sk->sk_prot->unhash(sk);
2695
2696 /*
2697 * In this point socket cannot receive new packets, but it is possible
2698 * that some packets are in flight because some CPU runs receiver and
2699 * did hash table lookup before we unhashed socket. They will achieve
2700 * receive queue and will be purged by socket destructor.
2701 *
2702 * Also we still have packets pending on receive queue and probably,
2703 * our own packets waiting in device queues. sock_destroy will drain
2704 * receive queue, but transmitted packets will delay socket destruction
2705 * until the last reference will be released.
2706 */
2707
2708 sock_orphan(sk);
2709
2710 xfrm_sk_free_policy(sk);
2711
2712 sk_refcnt_debug_release(sk);
2713
2714 if (sk->sk_frag.page) {
2715 put_page(sk->sk_frag.page);
2716 sk->sk_frag.page = NULL;
2717 }
2718
2719 sock_put(sk);
2720 }
2721 EXPORT_SYMBOL(sk_common_release);
2722
2723 #ifdef CONFIG_PROC_FS
2724 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2725 struct prot_inuse {
2726 int val[PROTO_INUSE_NR];
2727 };
2728
2729 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2730
2731 #ifdef CONFIG_NET_NS
2732 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2733 {
2734 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2735 }
2736 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2737
2738 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2739 {
2740 int cpu, idx = prot->inuse_idx;
2741 int res = 0;
2742
2743 for_each_possible_cpu(cpu)
2744 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2745
2746 return res >= 0 ? res : 0;
2747 }
2748 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2749
2750 static int __net_init sock_inuse_init_net(struct net *net)
2751 {
2752 net->core.inuse = alloc_percpu(struct prot_inuse);
2753 return net->core.inuse ? 0 : -ENOMEM;
2754 }
2755
2756 static void __net_exit sock_inuse_exit_net(struct net *net)
2757 {
2758 free_percpu(net->core.inuse);
2759 }
2760
2761 static struct pernet_operations net_inuse_ops = {
2762 .init = sock_inuse_init_net,
2763 .exit = sock_inuse_exit_net,
2764 };
2765
2766 static __init int net_inuse_init(void)
2767 {
2768 if (register_pernet_subsys(&net_inuse_ops))
2769 panic("Cannot initialize net inuse counters");
2770
2771 return 0;
2772 }
2773
2774 core_initcall(net_inuse_init);
2775 #else
2776 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2777
2778 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2779 {
2780 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2781 }
2782 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2783
2784 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2785 {
2786 int cpu, idx = prot->inuse_idx;
2787 int res = 0;
2788
2789 for_each_possible_cpu(cpu)
2790 res += per_cpu(prot_inuse, cpu).val[idx];
2791
2792 return res >= 0 ? res : 0;
2793 }
2794 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2795 #endif
2796
2797 static void assign_proto_idx(struct proto *prot)
2798 {
2799 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2800
2801 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2802 pr_err("PROTO_INUSE_NR exhausted\n");
2803 return;
2804 }
2805
2806 set_bit(prot->inuse_idx, proto_inuse_idx);
2807 }
2808
2809 static void release_proto_idx(struct proto *prot)
2810 {
2811 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2812 clear_bit(prot->inuse_idx, proto_inuse_idx);
2813 }
2814 #else
2815 static inline void assign_proto_idx(struct proto *prot)
2816 {
2817 }
2818
2819 static inline void release_proto_idx(struct proto *prot)
2820 {
2821 }
2822 #endif
2823
2824 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2825 {
2826 if (!rsk_prot)
2827 return;
2828 kfree(rsk_prot->slab_name);
2829 rsk_prot->slab_name = NULL;
2830 kmem_cache_destroy(rsk_prot->slab);
2831 rsk_prot->slab = NULL;
2832 }
2833
2834 static int req_prot_init(const struct proto *prot)
2835 {
2836 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2837
2838 if (!rsk_prot)
2839 return 0;
2840
2841 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2842 prot->name);
2843 if (!rsk_prot->slab_name)
2844 return -ENOMEM;
2845
2846 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2847 rsk_prot->obj_size, 0,
2848 prot->slab_flags, NULL);
2849
2850 if (!rsk_prot->slab) {
2851 pr_crit("%s: Can't create request sock SLAB cache!\n",
2852 prot->name);
2853 return -ENOMEM;
2854 }
2855 return 0;
2856 }
2857
2858 int proto_register(struct proto *prot, int alloc_slab)
2859 {
2860 if (alloc_slab) {
2861 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2862 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2863 NULL);
2864
2865 if (prot->slab == NULL) {
2866 pr_crit("%s: Can't create sock SLAB cache!\n",
2867 prot->name);
2868 goto out;
2869 }
2870
2871 if (req_prot_init(prot))
2872 goto out_free_request_sock_slab;
2873
2874 if (prot->twsk_prot != NULL) {
2875 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2876
2877 if (prot->twsk_prot->twsk_slab_name == NULL)
2878 goto out_free_request_sock_slab;
2879
2880 prot->twsk_prot->twsk_slab =
2881 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2882 prot->twsk_prot->twsk_obj_size,
2883 0,
2884 prot->slab_flags,
2885 NULL);
2886 if (prot->twsk_prot->twsk_slab == NULL)
2887 goto out_free_timewait_sock_slab_name;
2888 }
2889 }
2890
2891 mutex_lock(&proto_list_mutex);
2892 list_add(&prot->node, &proto_list);
2893 assign_proto_idx(prot);
2894 mutex_unlock(&proto_list_mutex);
2895 return 0;
2896
2897 out_free_timewait_sock_slab_name:
2898 kfree(prot->twsk_prot->twsk_slab_name);
2899 out_free_request_sock_slab:
2900 req_prot_cleanup(prot->rsk_prot);
2901
2902 kmem_cache_destroy(prot->slab);
2903 prot->slab = NULL;
2904 out:
2905 return -ENOBUFS;
2906 }
2907 EXPORT_SYMBOL(proto_register);
2908
2909 void proto_unregister(struct proto *prot)
2910 {
2911 mutex_lock(&proto_list_mutex);
2912 release_proto_idx(prot);
2913 list_del(&prot->node);
2914 mutex_unlock(&proto_list_mutex);
2915
2916 kmem_cache_destroy(prot->slab);
2917 prot->slab = NULL;
2918
2919 req_prot_cleanup(prot->rsk_prot);
2920
2921 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2922 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2923 kfree(prot->twsk_prot->twsk_slab_name);
2924 prot->twsk_prot->twsk_slab = NULL;
2925 }
2926 }
2927 EXPORT_SYMBOL(proto_unregister);
2928
2929 #ifdef CONFIG_PROC_FS
2930 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2931 __acquires(proto_list_mutex)
2932 {
2933 mutex_lock(&proto_list_mutex);
2934 return seq_list_start_head(&proto_list, *pos);
2935 }
2936
2937 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2938 {
2939 return seq_list_next(v, &proto_list, pos);
2940 }
2941
2942 static void proto_seq_stop(struct seq_file *seq, void *v)
2943 __releases(proto_list_mutex)
2944 {
2945 mutex_unlock(&proto_list_mutex);
2946 }
2947
2948 static char proto_method_implemented(const void *method)
2949 {
2950 return method == NULL ? 'n' : 'y';
2951 }
2952 static long sock_prot_memory_allocated(struct proto *proto)
2953 {
2954 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2955 }
2956
2957 static char *sock_prot_memory_pressure(struct proto *proto)
2958 {
2959 return proto->memory_pressure != NULL ?
2960 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2961 }
2962
2963 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2964 {
2965
2966 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2967 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2968 proto->name,
2969 proto->obj_size,
2970 sock_prot_inuse_get(seq_file_net(seq), proto),
2971 sock_prot_memory_allocated(proto),
2972 sock_prot_memory_pressure(proto),
2973 proto->max_header,
2974 proto->slab == NULL ? "no" : "yes",
2975 module_name(proto->owner),
2976 proto_method_implemented(proto->close),
2977 proto_method_implemented(proto->connect),
2978 proto_method_implemented(proto->disconnect),
2979 proto_method_implemented(proto->accept),
2980 proto_method_implemented(proto->ioctl),
2981 proto_method_implemented(proto->init),
2982 proto_method_implemented(proto->destroy),
2983 proto_method_implemented(proto->shutdown),
2984 proto_method_implemented(proto->setsockopt),
2985 proto_method_implemented(proto->getsockopt),
2986 proto_method_implemented(proto->sendmsg),
2987 proto_method_implemented(proto->recvmsg),
2988 proto_method_implemented(proto->sendpage),
2989 proto_method_implemented(proto->bind),
2990 proto_method_implemented(proto->backlog_rcv),
2991 proto_method_implemented(proto->hash),
2992 proto_method_implemented(proto->unhash),
2993 proto_method_implemented(proto->get_port),
2994 proto_method_implemented(proto->enter_memory_pressure));
2995 }
2996
2997 static int proto_seq_show(struct seq_file *seq, void *v)
2998 {
2999 if (v == &proto_list)
3000 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3001 "protocol",
3002 "size",
3003 "sockets",
3004 "memory",
3005 "press",
3006 "maxhdr",
3007 "slab",
3008 "module",
3009 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3010 else
3011 proto_seq_printf(seq, list_entry(v, struct proto, node));
3012 return 0;
3013 }
3014
3015 static const struct seq_operations proto_seq_ops = {
3016 .start = proto_seq_start,
3017 .next = proto_seq_next,
3018 .stop = proto_seq_stop,
3019 .show = proto_seq_show,
3020 };
3021
3022 static int proto_seq_open(struct inode *inode, struct file *file)
3023 {
3024 return seq_open_net(inode, file, &proto_seq_ops,
3025 sizeof(struct seq_net_private));
3026 }
3027
3028 static const struct file_operations proto_seq_fops = {
3029 .owner = THIS_MODULE,
3030 .open = proto_seq_open,
3031 .read = seq_read,
3032 .llseek = seq_lseek,
3033 .release = seq_release_net,
3034 };
3035
3036 static __net_init int proto_init_net(struct net *net)
3037 {
3038 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3039 return -ENOMEM;
3040
3041 return 0;
3042 }
3043
3044 static __net_exit void proto_exit_net(struct net *net)
3045 {
3046 remove_proc_entry("protocols", net->proc_net);
3047 }
3048
3049
3050 static __net_initdata struct pernet_operations proto_net_ops = {
3051 .init = proto_init_net,
3052 .exit = proto_exit_net,
3053 };
3054
3055 static int __init proto_init(void)
3056 {
3057 return register_pernet_subsys(&proto_net_ops);
3058 }
3059
3060 subsys_initcall(proto_init);
3061
3062 #endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.113551 seconds and 4 git commands to generate.