]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - net/core/sock.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: properly release sk_frag.page
[mirror_ubuntu-zesty-kernel.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 <linux/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_KCM" ,
225 "sk_lock-AF_QIPCRTR", "sk_lock-AF_MAX"
226 };
227 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
228 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
229 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
230 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
231 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
232 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
233 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
234 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
235 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
236 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
237 "slock-27" , "slock-28" , "slock-AF_CAN" ,
238 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
239 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
240 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
241 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" ,
242 "slock-AF_QIPCRTR", "slock-AF_MAX"
243 };
244 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
245 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
246 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
247 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
248 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
249 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
250 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
251 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
252 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
253 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
254 "clock-27" , "clock-28" , "clock-AF_CAN" ,
255 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
256 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
257 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
258 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" ,
259 "clock-AF_QIPCRTR", "clock-AF_MAX"
260 };
261
262 /*
263 * sk_callback_lock locking rules are per-address-family,
264 * so split the lock classes by using a per-AF key:
265 */
266 static struct lock_class_key af_callback_keys[AF_MAX];
267
268 /* Take into consideration the size of the struct sk_buff overhead in the
269 * determination of these values, since that is non-constant across
270 * platforms. This makes socket queueing behavior and performance
271 * not depend upon such differences.
272 */
273 #define _SK_MEM_PACKETS 256
274 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
275 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
276 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
277
278 /* Run time adjustable parameters. */
279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
280 EXPORT_SYMBOL(sysctl_wmem_max);
281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
282 EXPORT_SYMBOL(sysctl_rmem_max);
283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
285
286 /* Maximal space eaten by iovec or ancillary data plus some space */
287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
288 EXPORT_SYMBOL(sysctl_optmem_max);
289
290 int sysctl_tstamp_allow_data __read_mostly = 1;
291
292 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
293 EXPORT_SYMBOL_GPL(memalloc_socks);
294
295 /**
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
298 *
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
302 */
303 void sk_set_memalloc(struct sock *sk)
304 {
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
307 static_key_slow_inc(&memalloc_socks);
308 }
309 EXPORT_SYMBOL_GPL(sk_set_memalloc);
310
311 void sk_clear_memalloc(struct sock *sk)
312 {
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
315 static_key_slow_dec(&memalloc_socks);
316
317 /*
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
323 */
324 sk_mem_reclaim(sk);
325 }
326 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
327
328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
329 {
330 int ret;
331 unsigned long pflags = current->flags;
332
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
335
336 current->flags |= PF_MEMALLOC;
337 ret = sk->sk_backlog_rcv(sk, skb);
338 tsk_restore_flags(current, pflags, PF_MEMALLOC);
339
340 return ret;
341 }
342 EXPORT_SYMBOL(__sk_backlog_rcv);
343
344 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
345 {
346 struct timeval tv;
347
348 if (optlen < sizeof(tv))
349 return -EINVAL;
350 if (copy_from_user(&tv, optval, sizeof(tv)))
351 return -EFAULT;
352 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
353 return -EDOM;
354
355 if (tv.tv_sec < 0) {
356 static int warned __read_mostly;
357
358 *timeo_p = 0;
359 if (warned < 10 && net_ratelimit()) {
360 warned++;
361 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
362 __func__, current->comm, task_pid_nr(current));
363 }
364 return 0;
365 }
366 *timeo_p = MAX_SCHEDULE_TIMEOUT;
367 if (tv.tv_sec == 0 && tv.tv_usec == 0)
368 return 0;
369 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
370 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
371 return 0;
372 }
373
374 static void sock_warn_obsolete_bsdism(const char *name)
375 {
376 static int warned;
377 static char warncomm[TASK_COMM_LEN];
378 if (strcmp(warncomm, current->comm) && warned < 5) {
379 strcpy(warncomm, current->comm);
380 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
381 warncomm, name);
382 warned++;
383 }
384 }
385
386 static bool sock_needs_netstamp(const struct sock *sk)
387 {
388 switch (sk->sk_family) {
389 case AF_UNSPEC:
390 case AF_UNIX:
391 return false;
392 default:
393 return true;
394 }
395 }
396
397 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
398 {
399 if (sk->sk_flags & flags) {
400 sk->sk_flags &= ~flags;
401 if (sock_needs_netstamp(sk) &&
402 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
403 net_disable_timestamp();
404 }
405 }
406
407
408 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
409 {
410 unsigned long flags;
411 struct sk_buff_head *list = &sk->sk_receive_queue;
412
413 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
414 atomic_inc(&sk->sk_drops);
415 trace_sock_rcvqueue_full(sk, skb);
416 return -ENOMEM;
417 }
418
419 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
420 atomic_inc(&sk->sk_drops);
421 return -ENOBUFS;
422 }
423
424 skb->dev = NULL;
425 skb_set_owner_r(skb, sk);
426
427 /* we escape from rcu protected region, make sure we dont leak
428 * a norefcounted dst
429 */
430 skb_dst_force(skb);
431
432 spin_lock_irqsave(&list->lock, flags);
433 sock_skb_set_dropcount(sk, skb);
434 __skb_queue_tail(list, skb);
435 spin_unlock_irqrestore(&list->lock, flags);
436
437 if (!sock_flag(sk, SOCK_DEAD))
438 sk->sk_data_ready(sk);
439 return 0;
440 }
441 EXPORT_SYMBOL(__sock_queue_rcv_skb);
442
443 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
444 {
445 int err;
446
447 err = sk_filter(sk, skb);
448 if (err)
449 return err;
450
451 return __sock_queue_rcv_skb(sk, skb);
452 }
453 EXPORT_SYMBOL(sock_queue_rcv_skb);
454
455 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
456 const int nested, unsigned int trim_cap, bool refcounted)
457 {
458 int rc = NET_RX_SUCCESS;
459
460 if (sk_filter_trim_cap(sk, skb, trim_cap))
461 goto discard_and_relse;
462
463 skb->dev = NULL;
464
465 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
466 atomic_inc(&sk->sk_drops);
467 goto discard_and_relse;
468 }
469 if (nested)
470 bh_lock_sock_nested(sk);
471 else
472 bh_lock_sock(sk);
473 if (!sock_owned_by_user(sk)) {
474 /*
475 * trylock + unlock semantics:
476 */
477 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
478
479 rc = sk_backlog_rcv(sk, skb);
480
481 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
482 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
483 bh_unlock_sock(sk);
484 atomic_inc(&sk->sk_drops);
485 goto discard_and_relse;
486 }
487
488 bh_unlock_sock(sk);
489 out:
490 if (refcounted)
491 sock_put(sk);
492 return rc;
493 discard_and_relse:
494 kfree_skb(skb);
495 goto out;
496 }
497 EXPORT_SYMBOL(__sk_receive_skb);
498
499 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
500 {
501 struct dst_entry *dst = __sk_dst_get(sk);
502
503 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
504 sk_tx_queue_clear(sk);
505 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
506 dst_release(dst);
507 return NULL;
508 }
509
510 return dst;
511 }
512 EXPORT_SYMBOL(__sk_dst_check);
513
514 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
515 {
516 struct dst_entry *dst = sk_dst_get(sk);
517
518 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
519 sk_dst_reset(sk);
520 dst_release(dst);
521 return NULL;
522 }
523
524 return dst;
525 }
526 EXPORT_SYMBOL(sk_dst_check);
527
528 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
529 int optlen)
530 {
531 int ret = -ENOPROTOOPT;
532 #ifdef CONFIG_NETDEVICES
533 struct net *net = sock_net(sk);
534 char devname[IFNAMSIZ];
535 int index;
536
537 /* Sorry... */
538 ret = -EPERM;
539 if (!ns_capable(net->user_ns, CAP_NET_RAW))
540 goto out;
541
542 ret = -EINVAL;
543 if (optlen < 0)
544 goto out;
545
546 /* Bind this socket to a particular device like "eth0",
547 * as specified in the passed interface name. If the
548 * name is "" or the option length is zero the socket
549 * is not bound.
550 */
551 if (optlen > IFNAMSIZ - 1)
552 optlen = IFNAMSIZ - 1;
553 memset(devname, 0, sizeof(devname));
554
555 ret = -EFAULT;
556 if (copy_from_user(devname, optval, optlen))
557 goto out;
558
559 index = 0;
560 if (devname[0] != '\0') {
561 struct net_device *dev;
562
563 rcu_read_lock();
564 dev = dev_get_by_name_rcu(net, devname);
565 if (dev)
566 index = dev->ifindex;
567 rcu_read_unlock();
568 ret = -ENODEV;
569 if (!dev)
570 goto out;
571 }
572
573 lock_sock(sk);
574 sk->sk_bound_dev_if = index;
575 sk_dst_reset(sk);
576 release_sock(sk);
577
578 ret = 0;
579
580 out:
581 #endif
582
583 return ret;
584 }
585
586 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
587 int __user *optlen, int len)
588 {
589 int ret = -ENOPROTOOPT;
590 #ifdef CONFIG_NETDEVICES
591 struct net *net = sock_net(sk);
592 char devname[IFNAMSIZ];
593
594 if (sk->sk_bound_dev_if == 0) {
595 len = 0;
596 goto zero;
597 }
598
599 ret = -EINVAL;
600 if (len < IFNAMSIZ)
601 goto out;
602
603 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
604 if (ret)
605 goto out;
606
607 len = strlen(devname) + 1;
608
609 ret = -EFAULT;
610 if (copy_to_user(optval, devname, len))
611 goto out;
612
613 zero:
614 ret = -EFAULT;
615 if (put_user(len, optlen))
616 goto out;
617
618 ret = 0;
619
620 out:
621 #endif
622
623 return ret;
624 }
625
626 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
627 {
628 if (valbool)
629 sock_set_flag(sk, bit);
630 else
631 sock_reset_flag(sk, bit);
632 }
633
634 bool sk_mc_loop(struct sock *sk)
635 {
636 if (dev_recursion_level())
637 return false;
638 if (!sk)
639 return true;
640 switch (sk->sk_family) {
641 case AF_INET:
642 return inet_sk(sk)->mc_loop;
643 #if IS_ENABLED(CONFIG_IPV6)
644 case AF_INET6:
645 return inet6_sk(sk)->mc_loop;
646 #endif
647 }
648 WARN_ON(1);
649 return true;
650 }
651 EXPORT_SYMBOL(sk_mc_loop);
652
653 /*
654 * This is meant for all protocols to use and covers goings on
655 * at the socket level. Everything here is generic.
656 */
657
658 int sock_setsockopt(struct socket *sock, int level, int optname,
659 char __user *optval, unsigned int optlen)
660 {
661 struct sock *sk = sock->sk;
662 int val;
663 int valbool;
664 struct linger ling;
665 int ret = 0;
666
667 /*
668 * Options without arguments
669 */
670
671 if (optname == SO_BINDTODEVICE)
672 return sock_setbindtodevice(sk, optval, optlen);
673
674 if (optlen < sizeof(int))
675 return -EINVAL;
676
677 if (get_user(val, (int __user *)optval))
678 return -EFAULT;
679
680 valbool = val ? 1 : 0;
681
682 lock_sock(sk);
683
684 switch (optname) {
685 case SO_DEBUG:
686 if (val && !capable(CAP_NET_ADMIN))
687 ret = -EACCES;
688 else
689 sock_valbool_flag(sk, SOCK_DBG, valbool);
690 break;
691 case SO_REUSEADDR:
692 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
693 break;
694 case SO_REUSEPORT:
695 sk->sk_reuseport = valbool;
696 break;
697 case SO_TYPE:
698 case SO_PROTOCOL:
699 case SO_DOMAIN:
700 case SO_ERROR:
701 ret = -ENOPROTOOPT;
702 break;
703 case SO_DONTROUTE:
704 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
705 break;
706 case SO_BROADCAST:
707 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
708 break;
709 case SO_SNDBUF:
710 /* Don't error on this BSD doesn't and if you think
711 * about it this is right. Otherwise apps have to
712 * play 'guess the biggest size' games. RCVBUF/SNDBUF
713 * are treated in BSD as hints
714 */
715 val = min_t(u32, val, sysctl_wmem_max);
716 set_sndbuf:
717 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
718 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
719 /* Wake up sending tasks if we upped the value. */
720 sk->sk_write_space(sk);
721 break;
722
723 case SO_SNDBUFFORCE:
724 if (!capable(CAP_NET_ADMIN)) {
725 ret = -EPERM;
726 break;
727 }
728 goto set_sndbuf;
729
730 case SO_RCVBUF:
731 /* Don't error on this BSD doesn't and if you think
732 * about it this is right. Otherwise apps have to
733 * play 'guess the biggest size' games. RCVBUF/SNDBUF
734 * are treated in BSD as hints
735 */
736 val = min_t(u32, val, sysctl_rmem_max);
737 set_rcvbuf:
738 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
739 /*
740 * We double it on the way in to account for
741 * "struct sk_buff" etc. overhead. Applications
742 * assume that the SO_RCVBUF setting they make will
743 * allow that much actual data to be received on that
744 * socket.
745 *
746 * Applications are unaware that "struct sk_buff" and
747 * other overheads allocate from the receive buffer
748 * during socket buffer allocation.
749 *
750 * And after considering the possible alternatives,
751 * returning the value we actually used in getsockopt
752 * is the most desirable behavior.
753 */
754 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
755 break;
756
757 case SO_RCVBUFFORCE:
758 if (!capable(CAP_NET_ADMIN)) {
759 ret = -EPERM;
760 break;
761 }
762 goto set_rcvbuf;
763
764 case SO_KEEPALIVE:
765 #ifdef CONFIG_INET
766 if (sk->sk_protocol == IPPROTO_TCP &&
767 sk->sk_type == SOCK_STREAM)
768 tcp_set_keepalive(sk, valbool);
769 #endif
770 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
771 break;
772
773 case SO_OOBINLINE:
774 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
775 break;
776
777 case SO_NO_CHECK:
778 sk->sk_no_check_tx = valbool;
779 break;
780
781 case SO_PRIORITY:
782 if ((val >= 0 && val <= 6) ||
783 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
784 sk->sk_priority = val;
785 else
786 ret = -EPERM;
787 break;
788
789 case SO_LINGER:
790 if (optlen < sizeof(ling)) {
791 ret = -EINVAL; /* 1003.1g */
792 break;
793 }
794 if (copy_from_user(&ling, optval, sizeof(ling))) {
795 ret = -EFAULT;
796 break;
797 }
798 if (!ling.l_onoff)
799 sock_reset_flag(sk, SOCK_LINGER);
800 else {
801 #if (BITS_PER_LONG == 32)
802 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
803 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
804 else
805 #endif
806 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
807 sock_set_flag(sk, SOCK_LINGER);
808 }
809 break;
810
811 case SO_BSDCOMPAT:
812 sock_warn_obsolete_bsdism("setsockopt");
813 break;
814
815 case SO_PASSCRED:
816 if (valbool)
817 set_bit(SOCK_PASSCRED, &sock->flags);
818 else
819 clear_bit(SOCK_PASSCRED, &sock->flags);
820 break;
821
822 case SO_TIMESTAMP:
823 case SO_TIMESTAMPNS:
824 if (valbool) {
825 if (optname == SO_TIMESTAMP)
826 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
827 else
828 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
829 sock_set_flag(sk, SOCK_RCVTSTAMP);
830 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
831 } else {
832 sock_reset_flag(sk, SOCK_RCVTSTAMP);
833 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
834 }
835 break;
836
837 case SO_TIMESTAMPING:
838 if (val & ~SOF_TIMESTAMPING_MASK) {
839 ret = -EINVAL;
840 break;
841 }
842
843 if (val & SOF_TIMESTAMPING_OPT_ID &&
844 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
845 if (sk->sk_protocol == IPPROTO_TCP &&
846 sk->sk_type == SOCK_STREAM) {
847 if ((1 << sk->sk_state) &
848 (TCPF_CLOSE | TCPF_LISTEN)) {
849 ret = -EINVAL;
850 break;
851 }
852 sk->sk_tskey = tcp_sk(sk)->snd_una;
853 } else {
854 sk->sk_tskey = 0;
855 }
856 }
857
858 if (val & SOF_TIMESTAMPING_OPT_STATS &&
859 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
860 ret = -EINVAL;
861 break;
862 }
863
864 sk->sk_tsflags = val;
865 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
866 sock_enable_timestamp(sk,
867 SOCK_TIMESTAMPING_RX_SOFTWARE);
868 else
869 sock_disable_timestamp(sk,
870 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
871 break;
872
873 case SO_RCVLOWAT:
874 if (val < 0)
875 val = INT_MAX;
876 sk->sk_rcvlowat = val ? : 1;
877 break;
878
879 case SO_RCVTIMEO:
880 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
881 break;
882
883 case SO_SNDTIMEO:
884 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
885 break;
886
887 case SO_ATTACH_FILTER:
888 ret = -EINVAL;
889 if (optlen == sizeof(struct sock_fprog)) {
890 struct sock_fprog fprog;
891
892 ret = -EFAULT;
893 if (copy_from_user(&fprog, optval, sizeof(fprog)))
894 break;
895
896 ret = sk_attach_filter(&fprog, sk);
897 }
898 break;
899
900 case SO_ATTACH_BPF:
901 ret = -EINVAL;
902 if (optlen == sizeof(u32)) {
903 u32 ufd;
904
905 ret = -EFAULT;
906 if (copy_from_user(&ufd, optval, sizeof(ufd)))
907 break;
908
909 ret = sk_attach_bpf(ufd, sk);
910 }
911 break;
912
913 case SO_ATTACH_REUSEPORT_CBPF:
914 ret = -EINVAL;
915 if (optlen == sizeof(struct sock_fprog)) {
916 struct sock_fprog fprog;
917
918 ret = -EFAULT;
919 if (copy_from_user(&fprog, optval, sizeof(fprog)))
920 break;
921
922 ret = sk_reuseport_attach_filter(&fprog, sk);
923 }
924 break;
925
926 case SO_ATTACH_REUSEPORT_EBPF:
927 ret = -EINVAL;
928 if (optlen == sizeof(u32)) {
929 u32 ufd;
930
931 ret = -EFAULT;
932 if (copy_from_user(&ufd, optval, sizeof(ufd)))
933 break;
934
935 ret = sk_reuseport_attach_bpf(ufd, sk);
936 }
937 break;
938
939 case SO_DETACH_FILTER:
940 ret = sk_detach_filter(sk);
941 break;
942
943 case SO_LOCK_FILTER:
944 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
945 ret = -EPERM;
946 else
947 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
948 break;
949
950 case SO_PASSSEC:
951 if (valbool)
952 set_bit(SOCK_PASSSEC, &sock->flags);
953 else
954 clear_bit(SOCK_PASSSEC, &sock->flags);
955 break;
956 case SO_MARK:
957 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
958 ret = -EPERM;
959 else
960 sk->sk_mark = val;
961 break;
962
963 case SO_RXQ_OVFL:
964 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
965 break;
966
967 case SO_WIFI_STATUS:
968 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
969 break;
970
971 case SO_PEEK_OFF:
972 if (sock->ops->set_peek_off)
973 ret = sock->ops->set_peek_off(sk, val);
974 else
975 ret = -EOPNOTSUPP;
976 break;
977
978 case SO_NOFCS:
979 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
980 break;
981
982 case SO_SELECT_ERR_QUEUE:
983 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
984 break;
985
986 #ifdef CONFIG_NET_RX_BUSY_POLL
987 case SO_BUSY_POLL:
988 /* allow unprivileged users to decrease the value */
989 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
990 ret = -EPERM;
991 else {
992 if (val < 0)
993 ret = -EINVAL;
994 else
995 sk->sk_ll_usec = val;
996 }
997 break;
998 #endif
999
1000 case SO_MAX_PACING_RATE:
1001 sk->sk_max_pacing_rate = val;
1002 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1003 sk->sk_max_pacing_rate);
1004 break;
1005
1006 case SO_INCOMING_CPU:
1007 sk->sk_incoming_cpu = val;
1008 break;
1009
1010 case SO_CNX_ADVICE:
1011 if (val == 1)
1012 dst_negative_advice(sk);
1013 break;
1014 default:
1015 ret = -ENOPROTOOPT;
1016 break;
1017 }
1018 release_sock(sk);
1019 return ret;
1020 }
1021 EXPORT_SYMBOL(sock_setsockopt);
1022
1023
1024 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1025 struct ucred *ucred)
1026 {
1027 ucred->pid = pid_vnr(pid);
1028 ucred->uid = ucred->gid = -1;
1029 if (cred) {
1030 struct user_namespace *current_ns = current_user_ns();
1031
1032 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1033 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1034 }
1035 }
1036
1037 int sock_getsockopt(struct socket *sock, int level, int optname,
1038 char __user *optval, int __user *optlen)
1039 {
1040 struct sock *sk = sock->sk;
1041
1042 union {
1043 int val;
1044 struct linger ling;
1045 struct timeval tm;
1046 } v;
1047
1048 int lv = sizeof(int);
1049 int len;
1050
1051 if (get_user(len, optlen))
1052 return -EFAULT;
1053 if (len < 0)
1054 return -EINVAL;
1055
1056 memset(&v, 0, sizeof(v));
1057
1058 switch (optname) {
1059 case SO_DEBUG:
1060 v.val = sock_flag(sk, SOCK_DBG);
1061 break;
1062
1063 case SO_DONTROUTE:
1064 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1065 break;
1066
1067 case SO_BROADCAST:
1068 v.val = sock_flag(sk, SOCK_BROADCAST);
1069 break;
1070
1071 case SO_SNDBUF:
1072 v.val = sk->sk_sndbuf;
1073 break;
1074
1075 case SO_RCVBUF:
1076 v.val = sk->sk_rcvbuf;
1077 break;
1078
1079 case SO_REUSEADDR:
1080 v.val = sk->sk_reuse;
1081 break;
1082
1083 case SO_REUSEPORT:
1084 v.val = sk->sk_reuseport;
1085 break;
1086
1087 case SO_KEEPALIVE:
1088 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1089 break;
1090
1091 case SO_TYPE:
1092 v.val = sk->sk_type;
1093 break;
1094
1095 case SO_PROTOCOL:
1096 v.val = sk->sk_protocol;
1097 break;
1098
1099 case SO_DOMAIN:
1100 v.val = sk->sk_family;
1101 break;
1102
1103 case SO_ERROR:
1104 v.val = -sock_error(sk);
1105 if (v.val == 0)
1106 v.val = xchg(&sk->sk_err_soft, 0);
1107 break;
1108
1109 case SO_OOBINLINE:
1110 v.val = sock_flag(sk, SOCK_URGINLINE);
1111 break;
1112
1113 case SO_NO_CHECK:
1114 v.val = sk->sk_no_check_tx;
1115 break;
1116
1117 case SO_PRIORITY:
1118 v.val = sk->sk_priority;
1119 break;
1120
1121 case SO_LINGER:
1122 lv = sizeof(v.ling);
1123 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1124 v.ling.l_linger = sk->sk_lingertime / HZ;
1125 break;
1126
1127 case SO_BSDCOMPAT:
1128 sock_warn_obsolete_bsdism("getsockopt");
1129 break;
1130
1131 case SO_TIMESTAMP:
1132 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1133 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1134 break;
1135
1136 case SO_TIMESTAMPNS:
1137 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1138 break;
1139
1140 case SO_TIMESTAMPING:
1141 v.val = sk->sk_tsflags;
1142 break;
1143
1144 case SO_RCVTIMEO:
1145 lv = sizeof(struct timeval);
1146 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1147 v.tm.tv_sec = 0;
1148 v.tm.tv_usec = 0;
1149 } else {
1150 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1151 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1152 }
1153 break;
1154
1155 case SO_SNDTIMEO:
1156 lv = sizeof(struct timeval);
1157 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1158 v.tm.tv_sec = 0;
1159 v.tm.tv_usec = 0;
1160 } else {
1161 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1162 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1163 }
1164 break;
1165
1166 case SO_RCVLOWAT:
1167 v.val = sk->sk_rcvlowat;
1168 break;
1169
1170 case SO_SNDLOWAT:
1171 v.val = 1;
1172 break;
1173
1174 case SO_PASSCRED:
1175 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1176 break;
1177
1178 case SO_PEERCRED:
1179 {
1180 struct ucred peercred;
1181 if (len > sizeof(peercred))
1182 len = sizeof(peercred);
1183 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1184 if (copy_to_user(optval, &peercred, len))
1185 return -EFAULT;
1186 goto lenout;
1187 }
1188
1189 case SO_PEERNAME:
1190 {
1191 char address[128];
1192
1193 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1194 return -ENOTCONN;
1195 if (lv < len)
1196 return -EINVAL;
1197 if (copy_to_user(optval, address, len))
1198 return -EFAULT;
1199 goto lenout;
1200 }
1201
1202 /* Dubious BSD thing... Probably nobody even uses it, but
1203 * the UNIX standard wants it for whatever reason... -DaveM
1204 */
1205 case SO_ACCEPTCONN:
1206 v.val = sk->sk_state == TCP_LISTEN;
1207 break;
1208
1209 case SO_PASSSEC:
1210 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1211 break;
1212
1213 case SO_PEERSEC:
1214 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1215
1216 case SO_MARK:
1217 v.val = sk->sk_mark;
1218 break;
1219
1220 case SO_RXQ_OVFL:
1221 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1222 break;
1223
1224 case SO_WIFI_STATUS:
1225 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1226 break;
1227
1228 case SO_PEEK_OFF:
1229 if (!sock->ops->set_peek_off)
1230 return -EOPNOTSUPP;
1231
1232 v.val = sk->sk_peek_off;
1233 break;
1234 case SO_NOFCS:
1235 v.val = sock_flag(sk, SOCK_NOFCS);
1236 break;
1237
1238 case SO_BINDTODEVICE:
1239 return sock_getbindtodevice(sk, optval, optlen, len);
1240
1241 case SO_GET_FILTER:
1242 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1243 if (len < 0)
1244 return len;
1245
1246 goto lenout;
1247
1248 case SO_LOCK_FILTER:
1249 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1250 break;
1251
1252 case SO_BPF_EXTENSIONS:
1253 v.val = bpf_tell_extensions();
1254 break;
1255
1256 case SO_SELECT_ERR_QUEUE:
1257 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1258 break;
1259
1260 #ifdef CONFIG_NET_RX_BUSY_POLL
1261 case SO_BUSY_POLL:
1262 v.val = sk->sk_ll_usec;
1263 break;
1264 #endif
1265
1266 case SO_MAX_PACING_RATE:
1267 v.val = sk->sk_max_pacing_rate;
1268 break;
1269
1270 case SO_INCOMING_CPU:
1271 v.val = sk->sk_incoming_cpu;
1272 break;
1273
1274 default:
1275 /* We implement the SO_SNDLOWAT etc to not be settable
1276 * (1003.1g 7).
1277 */
1278 return -ENOPROTOOPT;
1279 }
1280
1281 if (len > lv)
1282 len = lv;
1283 if (copy_to_user(optval, &v, len))
1284 return -EFAULT;
1285 lenout:
1286 if (put_user(len, optlen))
1287 return -EFAULT;
1288 return 0;
1289 }
1290
1291 /*
1292 * Initialize an sk_lock.
1293 *
1294 * (We also register the sk_lock with the lock validator.)
1295 */
1296 static inline void sock_lock_init(struct sock *sk)
1297 {
1298 sock_lock_init_class_and_name(sk,
1299 af_family_slock_key_strings[sk->sk_family],
1300 af_family_slock_keys + sk->sk_family,
1301 af_family_key_strings[sk->sk_family],
1302 af_family_keys + sk->sk_family);
1303 }
1304
1305 /*
1306 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1307 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1308 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1309 */
1310 static void sock_copy(struct sock *nsk, const struct sock *osk)
1311 {
1312 #ifdef CONFIG_SECURITY_NETWORK
1313 void *sptr = nsk->sk_security;
1314 #endif
1315 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1316
1317 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1318 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1319
1320 #ifdef CONFIG_SECURITY_NETWORK
1321 nsk->sk_security = sptr;
1322 security_sk_clone(osk, nsk);
1323 #endif
1324 }
1325
1326 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1327 int family)
1328 {
1329 struct sock *sk;
1330 struct kmem_cache *slab;
1331
1332 slab = prot->slab;
1333 if (slab != NULL) {
1334 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1335 if (!sk)
1336 return sk;
1337 if (priority & __GFP_ZERO)
1338 sk_prot_clear_nulls(sk, prot->obj_size);
1339 } else
1340 sk = kmalloc(prot->obj_size, priority);
1341
1342 if (sk != NULL) {
1343 kmemcheck_annotate_bitfield(sk, flags);
1344
1345 if (security_sk_alloc(sk, family, priority))
1346 goto out_free;
1347
1348 if (!try_module_get(prot->owner))
1349 goto out_free_sec;
1350 sk_tx_queue_clear(sk);
1351 }
1352
1353 return sk;
1354
1355 out_free_sec:
1356 security_sk_free(sk);
1357 out_free:
1358 if (slab != NULL)
1359 kmem_cache_free(slab, sk);
1360 else
1361 kfree(sk);
1362 return NULL;
1363 }
1364
1365 static void sk_prot_free(struct proto *prot, struct sock *sk)
1366 {
1367 struct kmem_cache *slab;
1368 struct module *owner;
1369
1370 owner = prot->owner;
1371 slab = prot->slab;
1372
1373 cgroup_sk_free(&sk->sk_cgrp_data);
1374 mem_cgroup_sk_free(sk);
1375 security_sk_free(sk);
1376 if (slab != NULL)
1377 kmem_cache_free(slab, sk);
1378 else
1379 kfree(sk);
1380 module_put(owner);
1381 }
1382
1383 /**
1384 * sk_alloc - All socket objects are allocated here
1385 * @net: the applicable net namespace
1386 * @family: protocol family
1387 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1388 * @prot: struct proto associated with this new sock instance
1389 * @kern: is this to be a kernel socket?
1390 */
1391 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1392 struct proto *prot, int kern)
1393 {
1394 struct sock *sk;
1395
1396 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1397 if (sk) {
1398 sk->sk_family = family;
1399 /*
1400 * See comment in struct sock definition to understand
1401 * why we need sk_prot_creator -acme
1402 */
1403 sk->sk_prot = sk->sk_prot_creator = prot;
1404 sock_lock_init(sk);
1405 sk->sk_net_refcnt = kern ? 0 : 1;
1406 if (likely(sk->sk_net_refcnt))
1407 get_net(net);
1408 sock_net_set(sk, net);
1409 atomic_set(&sk->sk_wmem_alloc, 1);
1410
1411 mem_cgroup_sk_alloc(sk);
1412 cgroup_sk_alloc(&sk->sk_cgrp_data);
1413 sock_update_classid(&sk->sk_cgrp_data);
1414 sock_update_netprioidx(&sk->sk_cgrp_data);
1415 }
1416
1417 return sk;
1418 }
1419 EXPORT_SYMBOL(sk_alloc);
1420
1421 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1422 * grace period. This is the case for UDP sockets and TCP listeners.
1423 */
1424 static void __sk_destruct(struct rcu_head *head)
1425 {
1426 struct sock *sk = container_of(head, struct sock, sk_rcu);
1427 struct sk_filter *filter;
1428
1429 if (sk->sk_destruct)
1430 sk->sk_destruct(sk);
1431
1432 filter = rcu_dereference_check(sk->sk_filter,
1433 atomic_read(&sk->sk_wmem_alloc) == 0);
1434 if (filter) {
1435 sk_filter_uncharge(sk, filter);
1436 RCU_INIT_POINTER(sk->sk_filter, NULL);
1437 }
1438 if (rcu_access_pointer(sk->sk_reuseport_cb))
1439 reuseport_detach_sock(sk);
1440
1441 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1442
1443 if (atomic_read(&sk->sk_omem_alloc))
1444 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1445 __func__, atomic_read(&sk->sk_omem_alloc));
1446
1447 if (sk->sk_frag.page) {
1448 put_page(sk->sk_frag.page);
1449 sk->sk_frag.page = NULL;
1450 }
1451
1452 if (sk->sk_peer_cred)
1453 put_cred(sk->sk_peer_cred);
1454 put_pid(sk->sk_peer_pid);
1455 if (likely(sk->sk_net_refcnt))
1456 put_net(sock_net(sk));
1457 sk_prot_free(sk->sk_prot_creator, sk);
1458 }
1459
1460 void sk_destruct(struct sock *sk)
1461 {
1462 if (sock_flag(sk, SOCK_RCU_FREE))
1463 call_rcu(&sk->sk_rcu, __sk_destruct);
1464 else
1465 __sk_destruct(&sk->sk_rcu);
1466 }
1467
1468 static void __sk_free(struct sock *sk)
1469 {
1470 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1471 sock_diag_broadcast_destroy(sk);
1472 else
1473 sk_destruct(sk);
1474 }
1475
1476 void sk_free(struct sock *sk)
1477 {
1478 /*
1479 * We subtract one from sk_wmem_alloc and can know if
1480 * some packets are still in some tx queue.
1481 * If not null, sock_wfree() will call __sk_free(sk) later
1482 */
1483 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1484 __sk_free(sk);
1485 }
1486 EXPORT_SYMBOL(sk_free);
1487
1488 /**
1489 * sk_clone_lock - clone a socket, and lock its clone
1490 * @sk: the socket to clone
1491 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1492 *
1493 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1494 */
1495 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1496 {
1497 struct sock *newsk;
1498 bool is_charged = true;
1499
1500 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1501 if (newsk != NULL) {
1502 struct sk_filter *filter;
1503
1504 sock_copy(newsk, sk);
1505
1506 /* SANITY */
1507 if (likely(newsk->sk_net_refcnt))
1508 get_net(sock_net(newsk));
1509 sk_node_init(&newsk->sk_node);
1510 sock_lock_init(newsk);
1511 bh_lock_sock(newsk);
1512 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1513 newsk->sk_backlog.len = 0;
1514
1515 atomic_set(&newsk->sk_rmem_alloc, 0);
1516 /*
1517 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1518 */
1519 atomic_set(&newsk->sk_wmem_alloc, 1);
1520 atomic_set(&newsk->sk_omem_alloc, 0);
1521 skb_queue_head_init(&newsk->sk_receive_queue);
1522 skb_queue_head_init(&newsk->sk_write_queue);
1523
1524 rwlock_init(&newsk->sk_callback_lock);
1525 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1526 af_callback_keys + newsk->sk_family,
1527 af_family_clock_key_strings[newsk->sk_family]);
1528
1529 newsk->sk_dst_cache = NULL;
1530 newsk->sk_wmem_queued = 0;
1531 newsk->sk_forward_alloc = 0;
1532 atomic_set(&newsk->sk_drops, 0);
1533 newsk->sk_send_head = NULL;
1534 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1535
1536 sock_reset_flag(newsk, SOCK_DONE);
1537 skb_queue_head_init(&newsk->sk_error_queue);
1538
1539 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1540 if (filter != NULL)
1541 /* though it's an empty new sock, the charging may fail
1542 * if sysctl_optmem_max was changed between creation of
1543 * original socket and cloning
1544 */
1545 is_charged = sk_filter_charge(newsk, filter);
1546
1547 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1548 /* It is still raw copy of parent, so invalidate
1549 * destructor and make plain sk_free() */
1550 newsk->sk_destruct = NULL;
1551 bh_unlock_sock(newsk);
1552 sk_free(newsk);
1553 newsk = NULL;
1554 goto out;
1555 }
1556 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1557
1558 newsk->sk_err = 0;
1559 newsk->sk_err_soft = 0;
1560 newsk->sk_priority = 0;
1561 newsk->sk_incoming_cpu = raw_smp_processor_id();
1562 atomic64_set(&newsk->sk_cookie, 0);
1563
1564 mem_cgroup_sk_alloc(newsk);
1565 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1566
1567 /*
1568 * Before updating sk_refcnt, we must commit prior changes to memory
1569 * (Documentation/RCU/rculist_nulls.txt for details)
1570 */
1571 smp_wmb();
1572 atomic_set(&newsk->sk_refcnt, 2);
1573
1574 /*
1575 * Increment the counter in the same struct proto as the master
1576 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1577 * is the same as sk->sk_prot->socks, as this field was copied
1578 * with memcpy).
1579 *
1580 * This _changes_ the previous behaviour, where
1581 * tcp_create_openreq_child always was incrementing the
1582 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1583 * to be taken into account in all callers. -acme
1584 */
1585 sk_refcnt_debug_inc(newsk);
1586 sk_set_socket(newsk, NULL);
1587 newsk->sk_wq = NULL;
1588
1589 if (newsk->sk_prot->sockets_allocated)
1590 sk_sockets_allocated_inc(newsk);
1591
1592 if (sock_needs_netstamp(sk) &&
1593 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1594 net_enable_timestamp();
1595 }
1596 out:
1597 return newsk;
1598 }
1599 EXPORT_SYMBOL_GPL(sk_clone_lock);
1600
1601 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1602 {
1603 u32 max_segs = 1;
1604
1605 sk_dst_set(sk, dst);
1606 sk->sk_route_caps = dst->dev->features;
1607 if (sk->sk_route_caps & NETIF_F_GSO)
1608 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1609 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1610 if (sk_can_gso(sk)) {
1611 if (dst->header_len) {
1612 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1613 } else {
1614 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1615 sk->sk_gso_max_size = dst->dev->gso_max_size;
1616 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1617 }
1618 }
1619 sk->sk_gso_max_segs = max_segs;
1620 }
1621 EXPORT_SYMBOL_GPL(sk_setup_caps);
1622
1623 /*
1624 * Simple resource managers for sockets.
1625 */
1626
1627
1628 /*
1629 * Write buffer destructor automatically called from kfree_skb.
1630 */
1631 void sock_wfree(struct sk_buff *skb)
1632 {
1633 struct sock *sk = skb->sk;
1634 unsigned int len = skb->truesize;
1635
1636 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1637 /*
1638 * Keep a reference on sk_wmem_alloc, this will be released
1639 * after sk_write_space() call
1640 */
1641 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1642 sk->sk_write_space(sk);
1643 len = 1;
1644 }
1645 /*
1646 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1647 * could not do because of in-flight packets
1648 */
1649 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1650 __sk_free(sk);
1651 }
1652 EXPORT_SYMBOL(sock_wfree);
1653
1654 /* This variant of sock_wfree() is used by TCP,
1655 * since it sets SOCK_USE_WRITE_QUEUE.
1656 */
1657 void __sock_wfree(struct sk_buff *skb)
1658 {
1659 struct sock *sk = skb->sk;
1660
1661 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1662 __sk_free(sk);
1663 }
1664
1665 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1666 {
1667 skb_orphan(skb);
1668 skb->sk = sk;
1669 #ifdef CONFIG_INET
1670 if (unlikely(!sk_fullsock(sk))) {
1671 skb->destructor = sock_edemux;
1672 sock_hold(sk);
1673 return;
1674 }
1675 #endif
1676 skb->destructor = sock_wfree;
1677 skb_set_hash_from_sk(skb, sk);
1678 /*
1679 * We used to take a refcount on sk, but following operation
1680 * is enough to guarantee sk_free() wont free this sock until
1681 * all in-flight packets are completed
1682 */
1683 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1684 }
1685 EXPORT_SYMBOL(skb_set_owner_w);
1686
1687 /* This helper is used by netem, as it can hold packets in its
1688 * delay queue. We want to allow the owner socket to send more
1689 * packets, as if they were already TX completed by a typical driver.
1690 * But we also want to keep skb->sk set because some packet schedulers
1691 * rely on it (sch_fq for example). So we set skb->truesize to a small
1692 * amount (1) and decrease sk_wmem_alloc accordingly.
1693 */
1694 void skb_orphan_partial(struct sk_buff *skb)
1695 {
1696 /* If this skb is a TCP pure ACK or already went here,
1697 * we have nothing to do. 2 is already a very small truesize.
1698 */
1699 if (skb->truesize <= 2)
1700 return;
1701
1702 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1703 * so we do not completely orphan skb, but transfert all
1704 * accounted bytes but one, to avoid unexpected reorders.
1705 */
1706 if (skb->destructor == sock_wfree
1707 #ifdef CONFIG_INET
1708 || skb->destructor == tcp_wfree
1709 #endif
1710 ) {
1711 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1712 skb->truesize = 1;
1713 } else {
1714 skb_orphan(skb);
1715 }
1716 }
1717 EXPORT_SYMBOL(skb_orphan_partial);
1718
1719 /*
1720 * Read buffer destructor automatically called from kfree_skb.
1721 */
1722 void sock_rfree(struct sk_buff *skb)
1723 {
1724 struct sock *sk = skb->sk;
1725 unsigned int len = skb->truesize;
1726
1727 atomic_sub(len, &sk->sk_rmem_alloc);
1728 sk_mem_uncharge(sk, len);
1729 }
1730 EXPORT_SYMBOL(sock_rfree);
1731
1732 /*
1733 * Buffer destructor for skbs that are not used directly in read or write
1734 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1735 */
1736 void sock_efree(struct sk_buff *skb)
1737 {
1738 sock_put(skb->sk);
1739 }
1740 EXPORT_SYMBOL(sock_efree);
1741
1742 kuid_t sock_i_uid(struct sock *sk)
1743 {
1744 kuid_t uid;
1745
1746 read_lock_bh(&sk->sk_callback_lock);
1747 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1748 read_unlock_bh(&sk->sk_callback_lock);
1749 return uid;
1750 }
1751 EXPORT_SYMBOL(sock_i_uid);
1752
1753 unsigned long sock_i_ino(struct sock *sk)
1754 {
1755 unsigned long ino;
1756
1757 read_lock_bh(&sk->sk_callback_lock);
1758 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1759 read_unlock_bh(&sk->sk_callback_lock);
1760 return ino;
1761 }
1762 EXPORT_SYMBOL(sock_i_ino);
1763
1764 /*
1765 * Allocate a skb from the socket's send buffer.
1766 */
1767 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1768 gfp_t priority)
1769 {
1770 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1771 struct sk_buff *skb = alloc_skb(size, priority);
1772 if (skb) {
1773 skb_set_owner_w(skb, sk);
1774 return skb;
1775 }
1776 }
1777 return NULL;
1778 }
1779 EXPORT_SYMBOL(sock_wmalloc);
1780
1781 /*
1782 * Allocate a memory block from the socket's option memory buffer.
1783 */
1784 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1785 {
1786 if ((unsigned int)size <= sysctl_optmem_max &&
1787 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1788 void *mem;
1789 /* First do the add, to avoid the race if kmalloc
1790 * might sleep.
1791 */
1792 atomic_add(size, &sk->sk_omem_alloc);
1793 mem = kmalloc(size, priority);
1794 if (mem)
1795 return mem;
1796 atomic_sub(size, &sk->sk_omem_alloc);
1797 }
1798 return NULL;
1799 }
1800 EXPORT_SYMBOL(sock_kmalloc);
1801
1802 /* Free an option memory block. Note, we actually want the inline
1803 * here as this allows gcc to detect the nullify and fold away the
1804 * condition entirely.
1805 */
1806 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1807 const bool nullify)
1808 {
1809 if (WARN_ON_ONCE(!mem))
1810 return;
1811 if (nullify)
1812 kzfree(mem);
1813 else
1814 kfree(mem);
1815 atomic_sub(size, &sk->sk_omem_alloc);
1816 }
1817
1818 void sock_kfree_s(struct sock *sk, void *mem, int size)
1819 {
1820 __sock_kfree_s(sk, mem, size, false);
1821 }
1822 EXPORT_SYMBOL(sock_kfree_s);
1823
1824 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1825 {
1826 __sock_kfree_s(sk, mem, size, true);
1827 }
1828 EXPORT_SYMBOL(sock_kzfree_s);
1829
1830 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1831 I think, these locks should be removed for datagram sockets.
1832 */
1833 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1834 {
1835 DEFINE_WAIT(wait);
1836
1837 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1838 for (;;) {
1839 if (!timeo)
1840 break;
1841 if (signal_pending(current))
1842 break;
1843 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1844 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1845 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1846 break;
1847 if (sk->sk_shutdown & SEND_SHUTDOWN)
1848 break;
1849 if (sk->sk_err)
1850 break;
1851 timeo = schedule_timeout(timeo);
1852 }
1853 finish_wait(sk_sleep(sk), &wait);
1854 return timeo;
1855 }
1856
1857
1858 /*
1859 * Generic send/receive buffer handlers
1860 */
1861
1862 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1863 unsigned long data_len, int noblock,
1864 int *errcode, int max_page_order)
1865 {
1866 struct sk_buff *skb;
1867 long timeo;
1868 int err;
1869
1870 timeo = sock_sndtimeo(sk, noblock);
1871 for (;;) {
1872 err = sock_error(sk);
1873 if (err != 0)
1874 goto failure;
1875
1876 err = -EPIPE;
1877 if (sk->sk_shutdown & SEND_SHUTDOWN)
1878 goto failure;
1879
1880 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1881 break;
1882
1883 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1884 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1885 err = -EAGAIN;
1886 if (!timeo)
1887 goto failure;
1888 if (signal_pending(current))
1889 goto interrupted;
1890 timeo = sock_wait_for_wmem(sk, timeo);
1891 }
1892 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1893 errcode, sk->sk_allocation);
1894 if (skb)
1895 skb_set_owner_w(skb, sk);
1896 return skb;
1897
1898 interrupted:
1899 err = sock_intr_errno(timeo);
1900 failure:
1901 *errcode = err;
1902 return NULL;
1903 }
1904 EXPORT_SYMBOL(sock_alloc_send_pskb);
1905
1906 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1907 int noblock, int *errcode)
1908 {
1909 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1910 }
1911 EXPORT_SYMBOL(sock_alloc_send_skb);
1912
1913 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1914 struct sockcm_cookie *sockc)
1915 {
1916 u32 tsflags;
1917
1918 switch (cmsg->cmsg_type) {
1919 case SO_MARK:
1920 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1921 return -EPERM;
1922 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1923 return -EINVAL;
1924 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1925 break;
1926 case SO_TIMESTAMPING:
1927 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1928 return -EINVAL;
1929
1930 tsflags = *(u32 *)CMSG_DATA(cmsg);
1931 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1932 return -EINVAL;
1933
1934 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1935 sockc->tsflags |= tsflags;
1936 break;
1937 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1938 case SCM_RIGHTS:
1939 case SCM_CREDENTIALS:
1940 break;
1941 default:
1942 return -EINVAL;
1943 }
1944 return 0;
1945 }
1946 EXPORT_SYMBOL(__sock_cmsg_send);
1947
1948 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1949 struct sockcm_cookie *sockc)
1950 {
1951 struct cmsghdr *cmsg;
1952 int ret;
1953
1954 for_each_cmsghdr(cmsg, msg) {
1955 if (!CMSG_OK(msg, cmsg))
1956 return -EINVAL;
1957 if (cmsg->cmsg_level != SOL_SOCKET)
1958 continue;
1959 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1960 if (ret)
1961 return ret;
1962 }
1963 return 0;
1964 }
1965 EXPORT_SYMBOL(sock_cmsg_send);
1966
1967 /* On 32bit arches, an skb frag is limited to 2^15 */
1968 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1969
1970 /**
1971 * skb_page_frag_refill - check that a page_frag contains enough room
1972 * @sz: minimum size of the fragment we want to get
1973 * @pfrag: pointer to page_frag
1974 * @gfp: priority for memory allocation
1975 *
1976 * Note: While this allocator tries to use high order pages, there is
1977 * no guarantee that allocations succeed. Therefore, @sz MUST be
1978 * less or equal than PAGE_SIZE.
1979 */
1980 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1981 {
1982 if (pfrag->page) {
1983 if (page_ref_count(pfrag->page) == 1) {
1984 pfrag->offset = 0;
1985 return true;
1986 }
1987 if (pfrag->offset + sz <= pfrag->size)
1988 return true;
1989 put_page(pfrag->page);
1990 }
1991
1992 pfrag->offset = 0;
1993 if (SKB_FRAG_PAGE_ORDER) {
1994 /* Avoid direct reclaim but allow kswapd to wake */
1995 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1996 __GFP_COMP | __GFP_NOWARN |
1997 __GFP_NORETRY,
1998 SKB_FRAG_PAGE_ORDER);
1999 if (likely(pfrag->page)) {
2000 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2001 return true;
2002 }
2003 }
2004 pfrag->page = alloc_page(gfp);
2005 if (likely(pfrag->page)) {
2006 pfrag->size = PAGE_SIZE;
2007 return true;
2008 }
2009 return false;
2010 }
2011 EXPORT_SYMBOL(skb_page_frag_refill);
2012
2013 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2014 {
2015 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2016 return true;
2017
2018 sk_enter_memory_pressure(sk);
2019 sk_stream_moderate_sndbuf(sk);
2020 return false;
2021 }
2022 EXPORT_SYMBOL(sk_page_frag_refill);
2023
2024 static void __lock_sock(struct sock *sk)
2025 __releases(&sk->sk_lock.slock)
2026 __acquires(&sk->sk_lock.slock)
2027 {
2028 DEFINE_WAIT(wait);
2029
2030 for (;;) {
2031 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2032 TASK_UNINTERRUPTIBLE);
2033 spin_unlock_bh(&sk->sk_lock.slock);
2034 schedule();
2035 spin_lock_bh(&sk->sk_lock.slock);
2036 if (!sock_owned_by_user(sk))
2037 break;
2038 }
2039 finish_wait(&sk->sk_lock.wq, &wait);
2040 }
2041
2042 static void __release_sock(struct sock *sk)
2043 __releases(&sk->sk_lock.slock)
2044 __acquires(&sk->sk_lock.slock)
2045 {
2046 struct sk_buff *skb, *next;
2047
2048 while ((skb = sk->sk_backlog.head) != NULL) {
2049 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2050
2051 spin_unlock_bh(&sk->sk_lock.slock);
2052
2053 do {
2054 next = skb->next;
2055 prefetch(next);
2056 WARN_ON_ONCE(skb_dst_is_noref(skb));
2057 skb->next = NULL;
2058 sk_backlog_rcv(sk, skb);
2059
2060 cond_resched();
2061
2062 skb = next;
2063 } while (skb != NULL);
2064
2065 spin_lock_bh(&sk->sk_lock.slock);
2066 }
2067
2068 /*
2069 * Doing the zeroing here guarantee we can not loop forever
2070 * while a wild producer attempts to flood us.
2071 */
2072 sk->sk_backlog.len = 0;
2073 }
2074
2075 void __sk_flush_backlog(struct sock *sk)
2076 {
2077 spin_lock_bh(&sk->sk_lock.slock);
2078 __release_sock(sk);
2079 spin_unlock_bh(&sk->sk_lock.slock);
2080 }
2081
2082 /**
2083 * sk_wait_data - wait for data to arrive at sk_receive_queue
2084 * @sk: sock to wait on
2085 * @timeo: for how long
2086 * @skb: last skb seen on sk_receive_queue
2087 *
2088 * Now socket state including sk->sk_err is changed only under lock,
2089 * hence we may omit checks after joining wait queue.
2090 * We check receive queue before schedule() only as optimization;
2091 * it is very likely that release_sock() added new data.
2092 */
2093 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2094 {
2095 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2096 int rc;
2097
2098 add_wait_queue(sk_sleep(sk), &wait);
2099 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2100 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2101 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2102 remove_wait_queue(sk_sleep(sk), &wait);
2103 return rc;
2104 }
2105 EXPORT_SYMBOL(sk_wait_data);
2106
2107 /**
2108 * __sk_mem_raise_allocated - increase memory_allocated
2109 * @sk: socket
2110 * @size: memory size to allocate
2111 * @amt: pages to allocate
2112 * @kind: allocation type
2113 *
2114 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2115 */
2116 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2117 {
2118 struct proto *prot = sk->sk_prot;
2119 long allocated = sk_memory_allocated_add(sk, amt);
2120
2121 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2122 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2123 goto suppress_allocation;
2124
2125 /* Under limit. */
2126 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2127 sk_leave_memory_pressure(sk);
2128 return 1;
2129 }
2130
2131 /* Under pressure. */
2132 if (allocated > sk_prot_mem_limits(sk, 1))
2133 sk_enter_memory_pressure(sk);
2134
2135 /* Over hard limit. */
2136 if (allocated > sk_prot_mem_limits(sk, 2))
2137 goto suppress_allocation;
2138
2139 /* guarantee minimum buffer size under pressure */
2140 if (kind == SK_MEM_RECV) {
2141 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2142 return 1;
2143
2144 } else { /* SK_MEM_SEND */
2145 if (sk->sk_type == SOCK_STREAM) {
2146 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2147 return 1;
2148 } else if (atomic_read(&sk->sk_wmem_alloc) <
2149 prot->sysctl_wmem[0])
2150 return 1;
2151 }
2152
2153 if (sk_has_memory_pressure(sk)) {
2154 int alloc;
2155
2156 if (!sk_under_memory_pressure(sk))
2157 return 1;
2158 alloc = sk_sockets_allocated_read_positive(sk);
2159 if (sk_prot_mem_limits(sk, 2) > alloc *
2160 sk_mem_pages(sk->sk_wmem_queued +
2161 atomic_read(&sk->sk_rmem_alloc) +
2162 sk->sk_forward_alloc))
2163 return 1;
2164 }
2165
2166 suppress_allocation:
2167
2168 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2169 sk_stream_moderate_sndbuf(sk);
2170
2171 /* Fail only if socket is _under_ its sndbuf.
2172 * In this case we cannot block, so that we have to fail.
2173 */
2174 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2175 return 1;
2176 }
2177
2178 trace_sock_exceed_buf_limit(sk, prot, allocated);
2179
2180 sk_memory_allocated_sub(sk, amt);
2181
2182 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2183 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2184
2185 return 0;
2186 }
2187 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2188
2189 /**
2190 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2191 * @sk: socket
2192 * @size: memory size to allocate
2193 * @kind: allocation type
2194 *
2195 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2196 * rmem allocation. This function assumes that protocols which have
2197 * memory_pressure use sk_wmem_queued as write buffer accounting.
2198 */
2199 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2200 {
2201 int ret, amt = sk_mem_pages(size);
2202
2203 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2204 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2205 if (!ret)
2206 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2207 return ret;
2208 }
2209 EXPORT_SYMBOL(__sk_mem_schedule);
2210
2211 /**
2212 * __sk_mem_reduce_allocated - reclaim memory_allocated
2213 * @sk: socket
2214 * @amount: number of quanta
2215 *
2216 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2217 */
2218 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2219 {
2220 sk_memory_allocated_sub(sk, amount);
2221
2222 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2223 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2224
2225 if (sk_under_memory_pressure(sk) &&
2226 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2227 sk_leave_memory_pressure(sk);
2228 }
2229 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2230
2231 /**
2232 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2233 * @sk: socket
2234 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2235 */
2236 void __sk_mem_reclaim(struct sock *sk, int amount)
2237 {
2238 amount >>= SK_MEM_QUANTUM_SHIFT;
2239 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2240 __sk_mem_reduce_allocated(sk, amount);
2241 }
2242 EXPORT_SYMBOL(__sk_mem_reclaim);
2243
2244 int sk_set_peek_off(struct sock *sk, int val)
2245 {
2246 if (val < 0)
2247 return -EINVAL;
2248
2249 sk->sk_peek_off = val;
2250 return 0;
2251 }
2252 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2253
2254 /*
2255 * Set of default routines for initialising struct proto_ops when
2256 * the protocol does not support a particular function. In certain
2257 * cases where it makes no sense for a protocol to have a "do nothing"
2258 * function, some default processing is provided.
2259 */
2260
2261 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2262 {
2263 return -EOPNOTSUPP;
2264 }
2265 EXPORT_SYMBOL(sock_no_bind);
2266
2267 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2268 int len, int flags)
2269 {
2270 return -EOPNOTSUPP;
2271 }
2272 EXPORT_SYMBOL(sock_no_connect);
2273
2274 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2275 {
2276 return -EOPNOTSUPP;
2277 }
2278 EXPORT_SYMBOL(sock_no_socketpair);
2279
2280 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2281 {
2282 return -EOPNOTSUPP;
2283 }
2284 EXPORT_SYMBOL(sock_no_accept);
2285
2286 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2287 int *len, int peer)
2288 {
2289 return -EOPNOTSUPP;
2290 }
2291 EXPORT_SYMBOL(sock_no_getname);
2292
2293 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2294 {
2295 return 0;
2296 }
2297 EXPORT_SYMBOL(sock_no_poll);
2298
2299 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2300 {
2301 return -EOPNOTSUPP;
2302 }
2303 EXPORT_SYMBOL(sock_no_ioctl);
2304
2305 int sock_no_listen(struct socket *sock, int backlog)
2306 {
2307 return -EOPNOTSUPP;
2308 }
2309 EXPORT_SYMBOL(sock_no_listen);
2310
2311 int sock_no_shutdown(struct socket *sock, int how)
2312 {
2313 return -EOPNOTSUPP;
2314 }
2315 EXPORT_SYMBOL(sock_no_shutdown);
2316
2317 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2318 char __user *optval, unsigned int optlen)
2319 {
2320 return -EOPNOTSUPP;
2321 }
2322 EXPORT_SYMBOL(sock_no_setsockopt);
2323
2324 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2325 char __user *optval, int __user *optlen)
2326 {
2327 return -EOPNOTSUPP;
2328 }
2329 EXPORT_SYMBOL(sock_no_getsockopt);
2330
2331 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2332 {
2333 return -EOPNOTSUPP;
2334 }
2335 EXPORT_SYMBOL(sock_no_sendmsg);
2336
2337 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2338 int flags)
2339 {
2340 return -EOPNOTSUPP;
2341 }
2342 EXPORT_SYMBOL(sock_no_recvmsg);
2343
2344 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2345 {
2346 /* Mirror missing mmap method error code */
2347 return -ENODEV;
2348 }
2349 EXPORT_SYMBOL(sock_no_mmap);
2350
2351 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2352 {
2353 ssize_t res;
2354 struct msghdr msg = {.msg_flags = flags};
2355 struct kvec iov;
2356 char *kaddr = kmap(page);
2357 iov.iov_base = kaddr + offset;
2358 iov.iov_len = size;
2359 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2360 kunmap(page);
2361 return res;
2362 }
2363 EXPORT_SYMBOL(sock_no_sendpage);
2364
2365 /*
2366 * Default Socket Callbacks
2367 */
2368
2369 static void sock_def_wakeup(struct sock *sk)
2370 {
2371 struct socket_wq *wq;
2372
2373 rcu_read_lock();
2374 wq = rcu_dereference(sk->sk_wq);
2375 if (skwq_has_sleeper(wq))
2376 wake_up_interruptible_all(&wq->wait);
2377 rcu_read_unlock();
2378 }
2379
2380 static void sock_def_error_report(struct sock *sk)
2381 {
2382 struct socket_wq *wq;
2383
2384 rcu_read_lock();
2385 wq = rcu_dereference(sk->sk_wq);
2386 if (skwq_has_sleeper(wq))
2387 wake_up_interruptible_poll(&wq->wait, POLLERR);
2388 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2389 rcu_read_unlock();
2390 }
2391
2392 static void sock_def_readable(struct sock *sk)
2393 {
2394 struct socket_wq *wq;
2395
2396 rcu_read_lock();
2397 wq = rcu_dereference(sk->sk_wq);
2398 if (skwq_has_sleeper(wq))
2399 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2400 POLLRDNORM | POLLRDBAND);
2401 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2402 rcu_read_unlock();
2403 }
2404
2405 static void sock_def_write_space(struct sock *sk)
2406 {
2407 struct socket_wq *wq;
2408
2409 rcu_read_lock();
2410
2411 /* Do not wake up a writer until he can make "significant"
2412 * progress. --DaveM
2413 */
2414 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2415 wq = rcu_dereference(sk->sk_wq);
2416 if (skwq_has_sleeper(wq))
2417 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2418 POLLWRNORM | POLLWRBAND);
2419
2420 /* Should agree with poll, otherwise some programs break */
2421 if (sock_writeable(sk))
2422 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2423 }
2424
2425 rcu_read_unlock();
2426 }
2427
2428 static void sock_def_destruct(struct sock *sk)
2429 {
2430 }
2431
2432 void sk_send_sigurg(struct sock *sk)
2433 {
2434 if (sk->sk_socket && sk->sk_socket->file)
2435 if (send_sigurg(&sk->sk_socket->file->f_owner))
2436 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2437 }
2438 EXPORT_SYMBOL(sk_send_sigurg);
2439
2440 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2441 unsigned long expires)
2442 {
2443 if (!mod_timer(timer, expires))
2444 sock_hold(sk);
2445 }
2446 EXPORT_SYMBOL(sk_reset_timer);
2447
2448 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2449 {
2450 if (del_timer(timer))
2451 __sock_put(sk);
2452 }
2453 EXPORT_SYMBOL(sk_stop_timer);
2454
2455 void sock_init_data(struct socket *sock, struct sock *sk)
2456 {
2457 skb_queue_head_init(&sk->sk_receive_queue);
2458 skb_queue_head_init(&sk->sk_write_queue);
2459 skb_queue_head_init(&sk->sk_error_queue);
2460
2461 sk->sk_send_head = NULL;
2462
2463 init_timer(&sk->sk_timer);
2464
2465 sk->sk_allocation = GFP_KERNEL;
2466 sk->sk_rcvbuf = sysctl_rmem_default;
2467 sk->sk_sndbuf = sysctl_wmem_default;
2468 sk->sk_state = TCP_CLOSE;
2469 sk_set_socket(sk, sock);
2470
2471 sock_set_flag(sk, SOCK_ZAPPED);
2472
2473 if (sock) {
2474 sk->sk_type = sock->type;
2475 sk->sk_wq = sock->wq;
2476 sock->sk = sk;
2477 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2478 } else {
2479 sk->sk_wq = NULL;
2480 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2481 }
2482
2483 rwlock_init(&sk->sk_callback_lock);
2484 lockdep_set_class_and_name(&sk->sk_callback_lock,
2485 af_callback_keys + sk->sk_family,
2486 af_family_clock_key_strings[sk->sk_family]);
2487
2488 sk->sk_state_change = sock_def_wakeup;
2489 sk->sk_data_ready = sock_def_readable;
2490 sk->sk_write_space = sock_def_write_space;
2491 sk->sk_error_report = sock_def_error_report;
2492 sk->sk_destruct = sock_def_destruct;
2493
2494 sk->sk_frag.page = NULL;
2495 sk->sk_frag.offset = 0;
2496 sk->sk_peek_off = -1;
2497
2498 sk->sk_peer_pid = NULL;
2499 sk->sk_peer_cred = NULL;
2500 sk->sk_write_pending = 0;
2501 sk->sk_rcvlowat = 1;
2502 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2503 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2504
2505 sk->sk_stamp = ktime_set(-1L, 0);
2506
2507 #ifdef CONFIG_NET_RX_BUSY_POLL
2508 sk->sk_napi_id = 0;
2509 sk->sk_ll_usec = sysctl_net_busy_read;
2510 #endif
2511
2512 sk->sk_max_pacing_rate = ~0U;
2513 sk->sk_pacing_rate = ~0U;
2514 sk->sk_incoming_cpu = -1;
2515 /*
2516 * Before updating sk_refcnt, we must commit prior changes to memory
2517 * (Documentation/RCU/rculist_nulls.txt for details)
2518 */
2519 smp_wmb();
2520 atomic_set(&sk->sk_refcnt, 1);
2521 atomic_set(&sk->sk_drops, 0);
2522 }
2523 EXPORT_SYMBOL(sock_init_data);
2524
2525 void lock_sock_nested(struct sock *sk, int subclass)
2526 {
2527 might_sleep();
2528 spin_lock_bh(&sk->sk_lock.slock);
2529 if (sk->sk_lock.owned)
2530 __lock_sock(sk);
2531 sk->sk_lock.owned = 1;
2532 spin_unlock(&sk->sk_lock.slock);
2533 /*
2534 * The sk_lock has mutex_lock() semantics here:
2535 */
2536 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2537 local_bh_enable();
2538 }
2539 EXPORT_SYMBOL(lock_sock_nested);
2540
2541 void release_sock(struct sock *sk)
2542 {
2543 spin_lock_bh(&sk->sk_lock.slock);
2544 if (sk->sk_backlog.tail)
2545 __release_sock(sk);
2546
2547 /* Warning : release_cb() might need to release sk ownership,
2548 * ie call sock_release_ownership(sk) before us.
2549 */
2550 if (sk->sk_prot->release_cb)
2551 sk->sk_prot->release_cb(sk);
2552
2553 sock_release_ownership(sk);
2554 if (waitqueue_active(&sk->sk_lock.wq))
2555 wake_up(&sk->sk_lock.wq);
2556 spin_unlock_bh(&sk->sk_lock.slock);
2557 }
2558 EXPORT_SYMBOL(release_sock);
2559
2560 /**
2561 * lock_sock_fast - fast version of lock_sock
2562 * @sk: socket
2563 *
2564 * This version should be used for very small section, where process wont block
2565 * return false if fast path is taken
2566 * sk_lock.slock locked, owned = 0, BH disabled
2567 * return true if slow path is taken
2568 * sk_lock.slock unlocked, owned = 1, BH enabled
2569 */
2570 bool lock_sock_fast(struct sock *sk)
2571 {
2572 might_sleep();
2573 spin_lock_bh(&sk->sk_lock.slock);
2574
2575 if (!sk->sk_lock.owned)
2576 /*
2577 * Note : We must disable BH
2578 */
2579 return false;
2580
2581 __lock_sock(sk);
2582 sk->sk_lock.owned = 1;
2583 spin_unlock(&sk->sk_lock.slock);
2584 /*
2585 * The sk_lock has mutex_lock() semantics here:
2586 */
2587 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2588 local_bh_enable();
2589 return true;
2590 }
2591 EXPORT_SYMBOL(lock_sock_fast);
2592
2593 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2594 {
2595 struct timeval tv;
2596 if (!sock_flag(sk, SOCK_TIMESTAMP))
2597 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2598 tv = ktime_to_timeval(sk->sk_stamp);
2599 if (tv.tv_sec == -1)
2600 return -ENOENT;
2601 if (tv.tv_sec == 0) {
2602 sk->sk_stamp = ktime_get_real();
2603 tv = ktime_to_timeval(sk->sk_stamp);
2604 }
2605 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2606 }
2607 EXPORT_SYMBOL(sock_get_timestamp);
2608
2609 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2610 {
2611 struct timespec ts;
2612 if (!sock_flag(sk, SOCK_TIMESTAMP))
2613 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2614 ts = ktime_to_timespec(sk->sk_stamp);
2615 if (ts.tv_sec == -1)
2616 return -ENOENT;
2617 if (ts.tv_sec == 0) {
2618 sk->sk_stamp = ktime_get_real();
2619 ts = ktime_to_timespec(sk->sk_stamp);
2620 }
2621 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2622 }
2623 EXPORT_SYMBOL(sock_get_timestampns);
2624
2625 void sock_enable_timestamp(struct sock *sk, int flag)
2626 {
2627 if (!sock_flag(sk, flag)) {
2628 unsigned long previous_flags = sk->sk_flags;
2629
2630 sock_set_flag(sk, flag);
2631 /*
2632 * we just set one of the two flags which require net
2633 * time stamping, but time stamping might have been on
2634 * already because of the other one
2635 */
2636 if (sock_needs_netstamp(sk) &&
2637 !(previous_flags & SK_FLAGS_TIMESTAMP))
2638 net_enable_timestamp();
2639 }
2640 }
2641
2642 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2643 int level, int type)
2644 {
2645 struct sock_exterr_skb *serr;
2646 struct sk_buff *skb;
2647 int copied, err;
2648
2649 err = -EAGAIN;
2650 skb = sock_dequeue_err_skb(sk);
2651 if (skb == NULL)
2652 goto out;
2653
2654 copied = skb->len;
2655 if (copied > len) {
2656 msg->msg_flags |= MSG_TRUNC;
2657 copied = len;
2658 }
2659 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2660 if (err)
2661 goto out_free_skb;
2662
2663 sock_recv_timestamp(msg, sk, skb);
2664
2665 serr = SKB_EXT_ERR(skb);
2666 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2667
2668 msg->msg_flags |= MSG_ERRQUEUE;
2669 err = copied;
2670
2671 out_free_skb:
2672 kfree_skb(skb);
2673 out:
2674 return err;
2675 }
2676 EXPORT_SYMBOL(sock_recv_errqueue);
2677
2678 /*
2679 * Get a socket option on an socket.
2680 *
2681 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2682 * asynchronous errors should be reported by getsockopt. We assume
2683 * this means if you specify SO_ERROR (otherwise whats the point of it).
2684 */
2685 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2686 char __user *optval, int __user *optlen)
2687 {
2688 struct sock *sk = sock->sk;
2689
2690 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2691 }
2692 EXPORT_SYMBOL(sock_common_getsockopt);
2693
2694 #ifdef CONFIG_COMPAT
2695 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2696 char __user *optval, int __user *optlen)
2697 {
2698 struct sock *sk = sock->sk;
2699
2700 if (sk->sk_prot->compat_getsockopt != NULL)
2701 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2702 optval, optlen);
2703 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2704 }
2705 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2706 #endif
2707
2708 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2709 int flags)
2710 {
2711 struct sock *sk = sock->sk;
2712 int addr_len = 0;
2713 int err;
2714
2715 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2716 flags & ~MSG_DONTWAIT, &addr_len);
2717 if (err >= 0)
2718 msg->msg_namelen = addr_len;
2719 return err;
2720 }
2721 EXPORT_SYMBOL(sock_common_recvmsg);
2722
2723 /*
2724 * Set socket options on an inet socket.
2725 */
2726 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2727 char __user *optval, unsigned int optlen)
2728 {
2729 struct sock *sk = sock->sk;
2730
2731 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2732 }
2733 EXPORT_SYMBOL(sock_common_setsockopt);
2734
2735 #ifdef CONFIG_COMPAT
2736 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2737 char __user *optval, unsigned int optlen)
2738 {
2739 struct sock *sk = sock->sk;
2740
2741 if (sk->sk_prot->compat_setsockopt != NULL)
2742 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2743 optval, optlen);
2744 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2745 }
2746 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2747 #endif
2748
2749 void sk_common_release(struct sock *sk)
2750 {
2751 if (sk->sk_prot->destroy)
2752 sk->sk_prot->destroy(sk);
2753
2754 /*
2755 * Observation: when sock_common_release is called, processes have
2756 * no access to socket. But net still has.
2757 * Step one, detach it from networking:
2758 *
2759 * A. Remove from hash tables.
2760 */
2761
2762 sk->sk_prot->unhash(sk);
2763
2764 /*
2765 * In this point socket cannot receive new packets, but it is possible
2766 * that some packets are in flight because some CPU runs receiver and
2767 * did hash table lookup before we unhashed socket. They will achieve
2768 * receive queue and will be purged by socket destructor.
2769 *
2770 * Also we still have packets pending on receive queue and probably,
2771 * our own packets waiting in device queues. sock_destroy will drain
2772 * receive queue, but transmitted packets will delay socket destruction
2773 * until the last reference will be released.
2774 */
2775
2776 sock_orphan(sk);
2777
2778 xfrm_sk_free_policy(sk);
2779
2780 sk_refcnt_debug_release(sk);
2781
2782 sock_put(sk);
2783 }
2784 EXPORT_SYMBOL(sk_common_release);
2785
2786 #ifdef CONFIG_PROC_FS
2787 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2788 struct prot_inuse {
2789 int val[PROTO_INUSE_NR];
2790 };
2791
2792 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2793
2794 #ifdef CONFIG_NET_NS
2795 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2796 {
2797 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2798 }
2799 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2800
2801 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2802 {
2803 int cpu, idx = prot->inuse_idx;
2804 int res = 0;
2805
2806 for_each_possible_cpu(cpu)
2807 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2808
2809 return res >= 0 ? res : 0;
2810 }
2811 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2812
2813 static int __net_init sock_inuse_init_net(struct net *net)
2814 {
2815 net->core.inuse = alloc_percpu(struct prot_inuse);
2816 return net->core.inuse ? 0 : -ENOMEM;
2817 }
2818
2819 static void __net_exit sock_inuse_exit_net(struct net *net)
2820 {
2821 free_percpu(net->core.inuse);
2822 }
2823
2824 static struct pernet_operations net_inuse_ops = {
2825 .init = sock_inuse_init_net,
2826 .exit = sock_inuse_exit_net,
2827 };
2828
2829 static __init int net_inuse_init(void)
2830 {
2831 if (register_pernet_subsys(&net_inuse_ops))
2832 panic("Cannot initialize net inuse counters");
2833
2834 return 0;
2835 }
2836
2837 core_initcall(net_inuse_init);
2838 #else
2839 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2840
2841 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2842 {
2843 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2844 }
2845 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2846
2847 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2848 {
2849 int cpu, idx = prot->inuse_idx;
2850 int res = 0;
2851
2852 for_each_possible_cpu(cpu)
2853 res += per_cpu(prot_inuse, cpu).val[idx];
2854
2855 return res >= 0 ? res : 0;
2856 }
2857 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2858 #endif
2859
2860 static void assign_proto_idx(struct proto *prot)
2861 {
2862 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2863
2864 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2865 pr_err("PROTO_INUSE_NR exhausted\n");
2866 return;
2867 }
2868
2869 set_bit(prot->inuse_idx, proto_inuse_idx);
2870 }
2871
2872 static void release_proto_idx(struct proto *prot)
2873 {
2874 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2875 clear_bit(prot->inuse_idx, proto_inuse_idx);
2876 }
2877 #else
2878 static inline void assign_proto_idx(struct proto *prot)
2879 {
2880 }
2881
2882 static inline void release_proto_idx(struct proto *prot)
2883 {
2884 }
2885 #endif
2886
2887 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2888 {
2889 if (!rsk_prot)
2890 return;
2891 kfree(rsk_prot->slab_name);
2892 rsk_prot->slab_name = NULL;
2893 kmem_cache_destroy(rsk_prot->slab);
2894 rsk_prot->slab = NULL;
2895 }
2896
2897 static int req_prot_init(const struct proto *prot)
2898 {
2899 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2900
2901 if (!rsk_prot)
2902 return 0;
2903
2904 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2905 prot->name);
2906 if (!rsk_prot->slab_name)
2907 return -ENOMEM;
2908
2909 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2910 rsk_prot->obj_size, 0,
2911 prot->slab_flags, NULL);
2912
2913 if (!rsk_prot->slab) {
2914 pr_crit("%s: Can't create request sock SLAB cache!\n",
2915 prot->name);
2916 return -ENOMEM;
2917 }
2918 return 0;
2919 }
2920
2921 int proto_register(struct proto *prot, int alloc_slab)
2922 {
2923 if (alloc_slab) {
2924 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2925 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2926 NULL);
2927
2928 if (prot->slab == NULL) {
2929 pr_crit("%s: Can't create sock SLAB cache!\n",
2930 prot->name);
2931 goto out;
2932 }
2933
2934 if (req_prot_init(prot))
2935 goto out_free_request_sock_slab;
2936
2937 if (prot->twsk_prot != NULL) {
2938 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2939
2940 if (prot->twsk_prot->twsk_slab_name == NULL)
2941 goto out_free_request_sock_slab;
2942
2943 prot->twsk_prot->twsk_slab =
2944 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2945 prot->twsk_prot->twsk_obj_size,
2946 0,
2947 prot->slab_flags,
2948 NULL);
2949 if (prot->twsk_prot->twsk_slab == NULL)
2950 goto out_free_timewait_sock_slab_name;
2951 }
2952 }
2953
2954 mutex_lock(&proto_list_mutex);
2955 list_add(&prot->node, &proto_list);
2956 assign_proto_idx(prot);
2957 mutex_unlock(&proto_list_mutex);
2958 return 0;
2959
2960 out_free_timewait_sock_slab_name:
2961 kfree(prot->twsk_prot->twsk_slab_name);
2962 out_free_request_sock_slab:
2963 req_prot_cleanup(prot->rsk_prot);
2964
2965 kmem_cache_destroy(prot->slab);
2966 prot->slab = NULL;
2967 out:
2968 return -ENOBUFS;
2969 }
2970 EXPORT_SYMBOL(proto_register);
2971
2972 void proto_unregister(struct proto *prot)
2973 {
2974 mutex_lock(&proto_list_mutex);
2975 release_proto_idx(prot);
2976 list_del(&prot->node);
2977 mutex_unlock(&proto_list_mutex);
2978
2979 kmem_cache_destroy(prot->slab);
2980 prot->slab = NULL;
2981
2982 req_prot_cleanup(prot->rsk_prot);
2983
2984 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2985 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2986 kfree(prot->twsk_prot->twsk_slab_name);
2987 prot->twsk_prot->twsk_slab = NULL;
2988 }
2989 }
2990 EXPORT_SYMBOL(proto_unregister);
2991
2992 #ifdef CONFIG_PROC_FS
2993 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2994 __acquires(proto_list_mutex)
2995 {
2996 mutex_lock(&proto_list_mutex);
2997 return seq_list_start_head(&proto_list, *pos);
2998 }
2999
3000 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3001 {
3002 return seq_list_next(v, &proto_list, pos);
3003 }
3004
3005 static void proto_seq_stop(struct seq_file *seq, void *v)
3006 __releases(proto_list_mutex)
3007 {
3008 mutex_unlock(&proto_list_mutex);
3009 }
3010
3011 static char proto_method_implemented(const void *method)
3012 {
3013 return method == NULL ? 'n' : 'y';
3014 }
3015 static long sock_prot_memory_allocated(struct proto *proto)
3016 {
3017 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3018 }
3019
3020 static char *sock_prot_memory_pressure(struct proto *proto)
3021 {
3022 return proto->memory_pressure != NULL ?
3023 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3024 }
3025
3026 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3027 {
3028
3029 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3030 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3031 proto->name,
3032 proto->obj_size,
3033 sock_prot_inuse_get(seq_file_net(seq), proto),
3034 sock_prot_memory_allocated(proto),
3035 sock_prot_memory_pressure(proto),
3036 proto->max_header,
3037 proto->slab == NULL ? "no" : "yes",
3038 module_name(proto->owner),
3039 proto_method_implemented(proto->close),
3040 proto_method_implemented(proto->connect),
3041 proto_method_implemented(proto->disconnect),
3042 proto_method_implemented(proto->accept),
3043 proto_method_implemented(proto->ioctl),
3044 proto_method_implemented(proto->init),
3045 proto_method_implemented(proto->destroy),
3046 proto_method_implemented(proto->shutdown),
3047 proto_method_implemented(proto->setsockopt),
3048 proto_method_implemented(proto->getsockopt),
3049 proto_method_implemented(proto->sendmsg),
3050 proto_method_implemented(proto->recvmsg),
3051 proto_method_implemented(proto->sendpage),
3052 proto_method_implemented(proto->bind),
3053 proto_method_implemented(proto->backlog_rcv),
3054 proto_method_implemented(proto->hash),
3055 proto_method_implemented(proto->unhash),
3056 proto_method_implemented(proto->get_port),
3057 proto_method_implemented(proto->enter_memory_pressure));
3058 }
3059
3060 static int proto_seq_show(struct seq_file *seq, void *v)
3061 {
3062 if (v == &proto_list)
3063 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3064 "protocol",
3065 "size",
3066 "sockets",
3067 "memory",
3068 "press",
3069 "maxhdr",
3070 "slab",
3071 "module",
3072 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3073 else
3074 proto_seq_printf(seq, list_entry(v, struct proto, node));
3075 return 0;
3076 }
3077
3078 static const struct seq_operations proto_seq_ops = {
3079 .start = proto_seq_start,
3080 .next = proto_seq_next,
3081 .stop = proto_seq_stop,
3082 .show = proto_seq_show,
3083 };
3084
3085 static int proto_seq_open(struct inode *inode, struct file *file)
3086 {
3087 return seq_open_net(inode, file, &proto_seq_ops,
3088 sizeof(struct seq_net_private));
3089 }
3090
3091 static const struct file_operations proto_seq_fops = {
3092 .owner = THIS_MODULE,
3093 .open = proto_seq_open,
3094 .read = seq_read,
3095 .llseek = seq_lseek,
3096 .release = seq_release_net,
3097 };
3098
3099 static __net_init int proto_init_net(struct net *net)
3100 {
3101 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3102 return -ENOMEM;
3103
3104 return 0;
3105 }
3106
3107 static __net_exit void proto_exit_net(struct net *net)
3108 {
3109 remove_proc_entry("protocols", net->proc_net);
3110 }
3111
3112
3113 static __net_initdata struct pernet_operations proto_net_ops = {
3114 .init = proto_init_net,
3115 .exit = proto_exit_net,
3116 };
3117
3118 static int __init proto_init(void)
3119 {
3120 return register_pernet_subsys(&proto_net_ops);
3121 }
3122
3123 subsys_initcall(proto_init);
3124
3125 #endif /* PROC_FS */
ubuntu-zesty-kernel mirror