projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'pci-v4.4-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci
[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
138 #include <trace/events/sock.h>
139
140 #ifdef CONFIG_INET
141 #include <net/tcp.h>
142 #endif
143
144 #include <net/busy_poll.h>
145
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
148
149 /**
150 * sk_ns_capable - General socket capability test
151 * @sk: Socket to use a capability on or through
152 * @user_ns: The user namespace of the capability to use
153 * @cap: The capability to use
154 *
155 * Test to see if the opener of the socket had when the socket was
156 * created and the current process has the capability @cap in the user
157 * namespace @user_ns.
158 */
159 bool sk_ns_capable(const struct sock *sk,
160 struct user_namespace *user_ns, int cap)
161 {
162 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
163 ns_capable(user_ns, cap);
164 }
165 EXPORT_SYMBOL(sk_ns_capable);
166
167 /**
168 * sk_capable - Socket global capability test
169 * @sk: Socket to use a capability on or through
170 * @cap: The global capability to use
171 *
172 * Test to see if the opener of the socket had when the socket was
173 * created and the current process has the capability @cap in all user
174 * namespaces.
175 */
176 bool sk_capable(const struct sock *sk, int cap)
177 {
178 return sk_ns_capable(sk, &init_user_ns, cap);
179 }
180 EXPORT_SYMBOL(sk_capable);
181
182 /**
183 * sk_net_capable - Network namespace socket capability test
184 * @sk: Socket to use a capability on or through
185 * @cap: The capability to use
186 *
187 * Test to see if the opener of the socket had when the socket was created
188 * and the current process has the capability @cap over the network namespace
189 * the socket is a member of.
190 */
191 bool sk_net_capable(const struct sock *sk, int cap)
192 {
193 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
194 }
195 EXPORT_SYMBOL(sk_net_capable);
196
197
198 #ifdef CONFIG_MEMCG_KMEM
199 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
200 {
201 struct proto *proto;
202 int ret = 0;
203
204 mutex_lock(&proto_list_mutex);
205 list_for_each_entry(proto, &proto_list, node) {
206 if (proto->init_cgroup) {
207 ret = proto->init_cgroup(memcg, ss);
208 if (ret)
209 goto out;
210 }
211 }
212
213 mutex_unlock(&proto_list_mutex);
214 return ret;
215 out:
216 list_for_each_entry_continue_reverse(proto, &proto_list, node)
217 if (proto->destroy_cgroup)
218 proto->destroy_cgroup(memcg);
219 mutex_unlock(&proto_list_mutex);
220 return ret;
221 }
222
223 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
224 {
225 struct proto *proto;
226
227 mutex_lock(&proto_list_mutex);
228 list_for_each_entry_reverse(proto, &proto_list, node)
229 if (proto->destroy_cgroup)
230 proto->destroy_cgroup(memcg);
231 mutex_unlock(&proto_list_mutex);
232 }
233 #endif
234
235 /*
236 * Each address family might have different locking rules, so we have
237 * one slock key per address family:
238 */
239 static struct lock_class_key af_family_keys[AF_MAX];
240 static struct lock_class_key af_family_slock_keys[AF_MAX];
241
242 #if defined(CONFIG_MEMCG_KMEM)
243 struct static_key memcg_socket_limit_enabled;
244 EXPORT_SYMBOL(memcg_socket_limit_enabled);
245 #endif
246
247 /*
248 * Make lock validator output more readable. (we pre-construct these
249 * strings build-time, so that runtime initialization of socket
250 * locks is fast):
251 */
252 static const char *const af_family_key_strings[AF_MAX+1] = {
253 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
254 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
255 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
256 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
257 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
258 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
259 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
260 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
261 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
262 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
263 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
264 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
265 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
266 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
267 };
268 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
269 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
270 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
271 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
272 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
273 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
274 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
275 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
276 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
277 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
278 "slock-27" , "slock-28" , "slock-AF_CAN" ,
279 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
280 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
281 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
282 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
283 };
284 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
285 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
286 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
287 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
288 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
289 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
290 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
291 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
292 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
293 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
294 "clock-27" , "clock-28" , "clock-AF_CAN" ,
295 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
296 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
297 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
298 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
299 };
300
301 /*
302 * sk_callback_lock locking rules are per-address-family,
303 * so split the lock classes by using a per-AF key:
304 */
305 static struct lock_class_key af_callback_keys[AF_MAX];
306
307 /* Take into consideration the size of the struct sk_buff overhead in the
308 * determination of these values, since that is non-constant across
309 * platforms. This makes socket queueing behavior and performance
310 * not depend upon such differences.
311 */
312 #define _SK_MEM_PACKETS 256
313 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
314 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
316
317 /* Run time adjustable parameters. */
318 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
319 EXPORT_SYMBOL(sysctl_wmem_max);
320 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
321 EXPORT_SYMBOL(sysctl_rmem_max);
322 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
323 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
324
325 /* Maximal space eaten by iovec or ancillary data plus some space */
326 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
327 EXPORT_SYMBOL(sysctl_optmem_max);
328
329 int sysctl_tstamp_allow_data __read_mostly = 1;
330
331 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
332 EXPORT_SYMBOL_GPL(memalloc_socks);
333
334 /**
335 * sk_set_memalloc - sets %SOCK_MEMALLOC
336 * @sk: socket to set it on
337 *
338 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
339 * It's the responsibility of the admin to adjust min_free_kbytes
340 * to meet the requirements
341 */
342 void sk_set_memalloc(struct sock *sk)
343 {
344 sock_set_flag(sk, SOCK_MEMALLOC);
345 sk->sk_allocation |= __GFP_MEMALLOC;
346 static_key_slow_inc(&memalloc_socks);
347 }
348 EXPORT_SYMBOL_GPL(sk_set_memalloc);
349
350 void sk_clear_memalloc(struct sock *sk)
351 {
352 sock_reset_flag(sk, SOCK_MEMALLOC);
353 sk->sk_allocation &= ~__GFP_MEMALLOC;
354 static_key_slow_dec(&memalloc_socks);
355
356 /*
357 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
358 * progress of swapping. SOCK_MEMALLOC may be cleared while
359 * it has rmem allocations due to the last swapfile being deactivated
360 * but there is a risk that the socket is unusable due to exceeding
361 * the rmem limits. Reclaim the reserves and obey rmem limits again.
362 */
363 sk_mem_reclaim(sk);
364 }
365 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
366
367 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
368 {
369 int ret;
370 unsigned long pflags = current->flags;
371
372 /* these should have been dropped before queueing */
373 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
374
375 current->flags |= PF_MEMALLOC;
376 ret = sk->sk_backlog_rcv(sk, skb);
377 tsk_restore_flags(current, pflags, PF_MEMALLOC);
378
379 return ret;
380 }
381 EXPORT_SYMBOL(__sk_backlog_rcv);
382
383 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
384 {
385 struct timeval tv;
386
387 if (optlen < sizeof(tv))
388 return -EINVAL;
389 if (copy_from_user(&tv, optval, sizeof(tv)))
390 return -EFAULT;
391 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
392 return -EDOM;
393
394 if (tv.tv_sec < 0) {
395 static int warned __read_mostly;
396
397 *timeo_p = 0;
398 if (warned < 10 && net_ratelimit()) {
399 warned++;
400 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
401 __func__, current->comm, task_pid_nr(current));
402 }
403 return 0;
404 }
405 *timeo_p = MAX_SCHEDULE_TIMEOUT;
406 if (tv.tv_sec == 0 && tv.tv_usec == 0)
407 return 0;
408 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
409 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
410 return 0;
411 }
412
413 static void sock_warn_obsolete_bsdism(const char *name)
414 {
415 static int warned;
416 static char warncomm[TASK_COMM_LEN];
417 if (strcmp(warncomm, current->comm) && warned < 5) {
418 strcpy(warncomm, current->comm);
419 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
420 warncomm, name);
421 warned++;
422 }
423 }
424
425 static bool sock_needs_netstamp(const struct sock *sk)
426 {
427 switch (sk->sk_family) {
428 case AF_UNSPEC:
429 case AF_UNIX:
430 return false;
431 default:
432 return true;
433 }
434 }
435
436 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
437
438 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
439 {
440 if (sk->sk_flags & flags) {
441 sk->sk_flags &= ~flags;
442 if (sock_needs_netstamp(sk) &&
443 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
444 net_disable_timestamp();
445 }
446 }
447
448
449 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
450 {
451 int err;
452 unsigned long flags;
453 struct sk_buff_head *list = &sk->sk_receive_queue;
454
455 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
456 atomic_inc(&sk->sk_drops);
457 trace_sock_rcvqueue_full(sk, skb);
458 return -ENOMEM;
459 }
460
461 err = sk_filter(sk, skb);
462 if (err)
463 return err;
464
465 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
466 atomic_inc(&sk->sk_drops);
467 return -ENOBUFS;
468 }
469
470 skb->dev = NULL;
471 skb_set_owner_r(skb, sk);
472
473 /* we escape from rcu protected region, make sure we dont leak
474 * a norefcounted dst
475 */
476 skb_dst_force(skb);
477
478 spin_lock_irqsave(&list->lock, flags);
479 sock_skb_set_dropcount(sk, skb);
480 __skb_queue_tail(list, skb);
481 spin_unlock_irqrestore(&list->lock, flags);
482
483 if (!sock_flag(sk, SOCK_DEAD))
484 sk->sk_data_ready(sk);
485 return 0;
486 }
487 EXPORT_SYMBOL(sock_queue_rcv_skb);
488
489 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
490 {
491 int rc = NET_RX_SUCCESS;
492
493 if (sk_filter(sk, skb))
494 goto discard_and_relse;
495
496 skb->dev = NULL;
497
498 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
499 atomic_inc(&sk->sk_drops);
500 goto discard_and_relse;
501 }
502 if (nested)
503 bh_lock_sock_nested(sk);
504 else
505 bh_lock_sock(sk);
506 if (!sock_owned_by_user(sk)) {
507 /*
508 * trylock + unlock semantics:
509 */
510 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
511
512 rc = sk_backlog_rcv(sk, skb);
513
514 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
515 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
516 bh_unlock_sock(sk);
517 atomic_inc(&sk->sk_drops);
518 goto discard_and_relse;
519 }
520
521 bh_unlock_sock(sk);
522 out:
523 sock_put(sk);
524 return rc;
525 discard_and_relse:
526 kfree_skb(skb);
527 goto out;
528 }
529 EXPORT_SYMBOL(sk_receive_skb);
530
531 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
532 {
533 struct dst_entry *dst = __sk_dst_get(sk);
534
535 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
536 sk_tx_queue_clear(sk);
537 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
538 dst_release(dst);
539 return NULL;
540 }
541
542 return dst;
543 }
544 EXPORT_SYMBOL(__sk_dst_check);
545
546 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
547 {
548 struct dst_entry *dst = sk_dst_get(sk);
549
550 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
551 sk_dst_reset(sk);
552 dst_release(dst);
553 return NULL;
554 }
555
556 return dst;
557 }
558 EXPORT_SYMBOL(sk_dst_check);
559
560 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
561 int optlen)
562 {
563 int ret = -ENOPROTOOPT;
564 #ifdef CONFIG_NETDEVICES
565 struct net *net = sock_net(sk);
566 char devname[IFNAMSIZ];
567 int index;
568
569 /* Sorry... */
570 ret = -EPERM;
571 if (!ns_capable(net->user_ns, CAP_NET_RAW))
572 goto out;
573
574 ret = -EINVAL;
575 if (optlen < 0)
576 goto out;
577
578 /* Bind this socket to a particular device like "eth0",
579 * as specified in the passed interface name. If the
580 * name is "" or the option length is zero the socket
581 * is not bound.
582 */
583 if (optlen > IFNAMSIZ - 1)
584 optlen = IFNAMSIZ - 1;
585 memset(devname, 0, sizeof(devname));
586
587 ret = -EFAULT;
588 if (copy_from_user(devname, optval, optlen))
589 goto out;
590
591 index = 0;
592 if (devname[0] != '\0') {
593 struct net_device *dev;
594
595 rcu_read_lock();
596 dev = dev_get_by_name_rcu(net, devname);
597 if (dev)
598 index = dev->ifindex;
599 rcu_read_unlock();
600 ret = -ENODEV;
601 if (!dev)
602 goto out;
603 }
604
605 lock_sock(sk);
606 sk->sk_bound_dev_if = index;
607 sk_dst_reset(sk);
608 release_sock(sk);
609
610 ret = 0;
611
612 out:
613 #endif
614
615 return ret;
616 }
617
618 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
619 int __user *optlen, int len)
620 {
621 int ret = -ENOPROTOOPT;
622 #ifdef CONFIG_NETDEVICES
623 struct net *net = sock_net(sk);
624 char devname[IFNAMSIZ];
625
626 if (sk->sk_bound_dev_if == 0) {
627 len = 0;
628 goto zero;
629 }
630
631 ret = -EINVAL;
632 if (len < IFNAMSIZ)
633 goto out;
634
635 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
636 if (ret)
637 goto out;
638
639 len = strlen(devname) + 1;
640
641 ret = -EFAULT;
642 if (copy_to_user(optval, devname, len))
643 goto out;
644
645 zero:
646 ret = -EFAULT;
647 if (put_user(len, optlen))
648 goto out;
649
650 ret = 0;
651
652 out:
653 #endif
654
655 return ret;
656 }
657
658 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
659 {
660 if (valbool)
661 sock_set_flag(sk, bit);
662 else
663 sock_reset_flag(sk, bit);
664 }
665
666 bool sk_mc_loop(struct sock *sk)
667 {
668 if (dev_recursion_level())
669 return false;
670 if (!sk)
671 return true;
672 switch (sk->sk_family) {
673 case AF_INET:
674 return inet_sk(sk)->mc_loop;
675 #if IS_ENABLED(CONFIG_IPV6)
676 case AF_INET6:
677 return inet6_sk(sk)->mc_loop;
678 #endif
679 }
680 WARN_ON(1);
681 return true;
682 }
683 EXPORT_SYMBOL(sk_mc_loop);
684
685 /*
686 * This is meant for all protocols to use and covers goings on
687 * at the socket level. Everything here is generic.
688 */
689
690 int sock_setsockopt(struct socket *sock, int level, int optname,
691 char __user *optval, unsigned int optlen)
692 {
693 struct sock *sk = sock->sk;
694 int val;
695 int valbool;
696 struct linger ling;
697 int ret = 0;
698
699 /*
700 * Options without arguments
701 */
702
703 if (optname == SO_BINDTODEVICE)
704 return sock_setbindtodevice(sk, optval, optlen);
705
706 if (optlen < sizeof(int))
707 return -EINVAL;
708
709 if (get_user(val, (int __user *)optval))
710 return -EFAULT;
711
712 valbool = val ? 1 : 0;
713
714 lock_sock(sk);
715
716 switch (optname) {
717 case SO_DEBUG:
718 if (val && !capable(CAP_NET_ADMIN))
719 ret = -EACCES;
720 else
721 sock_valbool_flag(sk, SOCK_DBG, valbool);
722 break;
723 case SO_REUSEADDR:
724 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
725 break;
726 case SO_REUSEPORT:
727 sk->sk_reuseport = valbool;
728 break;
729 case SO_TYPE:
730 case SO_PROTOCOL:
731 case SO_DOMAIN:
732 case SO_ERROR:
733 ret = -ENOPROTOOPT;
734 break;
735 case SO_DONTROUTE:
736 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
737 break;
738 case SO_BROADCAST:
739 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
740 break;
741 case SO_SNDBUF:
742 /* Don't error on this BSD doesn't and if you think
743 * about it this is right. Otherwise apps have to
744 * play 'guess the biggest size' games. RCVBUF/SNDBUF
745 * are treated in BSD as hints
746 */
747 val = min_t(u32, val, sysctl_wmem_max);
748 set_sndbuf:
749 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
750 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
751 /* Wake up sending tasks if we upped the value. */
752 sk->sk_write_space(sk);
753 break;
754
755 case SO_SNDBUFFORCE:
756 if (!capable(CAP_NET_ADMIN)) {
757 ret = -EPERM;
758 break;
759 }
760 goto set_sndbuf;
761
762 case SO_RCVBUF:
763 /* Don't error on this BSD doesn't and if you think
764 * about it this is right. Otherwise apps have to
765 * play 'guess the biggest size' games. RCVBUF/SNDBUF
766 * are treated in BSD as hints
767 */
768 val = min_t(u32, val, sysctl_rmem_max);
769 set_rcvbuf:
770 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
771 /*
772 * We double it on the way in to account for
773 * "struct sk_buff" etc. overhead. Applications
774 * assume that the SO_RCVBUF setting they make will
775 * allow that much actual data to be received on that
776 * socket.
777 *
778 * Applications are unaware that "struct sk_buff" and
779 * other overheads allocate from the receive buffer
780 * during socket buffer allocation.
781 *
782 * And after considering the possible alternatives,
783 * returning the value we actually used in getsockopt
784 * is the most desirable behavior.
785 */
786 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
787 break;
788
789 case SO_RCVBUFFORCE:
790 if (!capable(CAP_NET_ADMIN)) {
791 ret = -EPERM;
792 break;
793 }
794 goto set_rcvbuf;
795
796 case SO_KEEPALIVE:
797 #ifdef CONFIG_INET
798 if (sk->sk_protocol == IPPROTO_TCP &&
799 sk->sk_type == SOCK_STREAM)
800 tcp_set_keepalive(sk, valbool);
801 #endif
802 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
803 break;
804
805 case SO_OOBINLINE:
806 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
807 break;
808
809 case SO_NO_CHECK:
810 sk->sk_no_check_tx = valbool;
811 break;
812
813 case SO_PRIORITY:
814 if ((val >= 0 && val <= 6) ||
815 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
816 sk->sk_priority = val;
817 else
818 ret = -EPERM;
819 break;
820
821 case SO_LINGER:
822 if (optlen < sizeof(ling)) {
823 ret = -EINVAL; /* 1003.1g */
824 break;
825 }
826 if (copy_from_user(&ling, optval, sizeof(ling))) {
827 ret = -EFAULT;
828 break;
829 }
830 if (!ling.l_onoff)
831 sock_reset_flag(sk, SOCK_LINGER);
832 else {
833 #if (BITS_PER_LONG == 32)
834 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
835 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
836 else
837 #endif
838 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
839 sock_set_flag(sk, SOCK_LINGER);
840 }
841 break;
842
843 case SO_BSDCOMPAT:
844 sock_warn_obsolete_bsdism("setsockopt");
845 break;
846
847 case SO_PASSCRED:
848 if (valbool)
849 set_bit(SOCK_PASSCRED, &sock->flags);
850 else
851 clear_bit(SOCK_PASSCRED, &sock->flags);
852 break;
853
854 case SO_TIMESTAMP:
855 case SO_TIMESTAMPNS:
856 if (valbool) {
857 if (optname == SO_TIMESTAMP)
858 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
859 else
860 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
861 sock_set_flag(sk, SOCK_RCVTSTAMP);
862 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
863 } else {
864 sock_reset_flag(sk, SOCK_RCVTSTAMP);
865 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
866 }
867 break;
868
869 case SO_TIMESTAMPING:
870 if (val & ~SOF_TIMESTAMPING_MASK) {
871 ret = -EINVAL;
872 break;
873 }
874
875 if (val & SOF_TIMESTAMPING_OPT_ID &&
876 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
877 if (sk->sk_protocol == IPPROTO_TCP) {
878 if (sk->sk_state != TCP_ESTABLISHED) {
879 ret = -EINVAL;
880 break;
881 }
882 sk->sk_tskey = tcp_sk(sk)->snd_una;
883 } else {
884 sk->sk_tskey = 0;
885 }
886 }
887 sk->sk_tsflags = val;
888 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
889 sock_enable_timestamp(sk,
890 SOCK_TIMESTAMPING_RX_SOFTWARE);
891 else
892 sock_disable_timestamp(sk,
893 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
894 break;
895
896 case SO_RCVLOWAT:
897 if (val < 0)
898 val = INT_MAX;
899 sk->sk_rcvlowat = val ? : 1;
900 break;
901
902 case SO_RCVTIMEO:
903 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
904 break;
905
906 case SO_SNDTIMEO:
907 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
908 break;
909
910 case SO_ATTACH_FILTER:
911 ret = -EINVAL;
912 if (optlen == sizeof(struct sock_fprog)) {
913 struct sock_fprog fprog;
914
915 ret = -EFAULT;
916 if (copy_from_user(&fprog, optval, sizeof(fprog)))
917 break;
918
919 ret = sk_attach_filter(&fprog, sk);
920 }
921 break;
922
923 case SO_ATTACH_BPF:
924 ret = -EINVAL;
925 if (optlen == sizeof(u32)) {
926 u32 ufd;
927
928 ret = -EFAULT;
929 if (copy_from_user(&ufd, optval, sizeof(ufd)))
930 break;
931
932 ret = sk_attach_bpf(ufd, sk);
933 }
934 break;
935
936 case SO_DETACH_FILTER:
937 ret = sk_detach_filter(sk);
938 break;
939
940 case SO_LOCK_FILTER:
941 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
942 ret = -EPERM;
943 else
944 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
945 break;
946
947 case SO_PASSSEC:
948 if (valbool)
949 set_bit(SOCK_PASSSEC, &sock->flags);
950 else
951 clear_bit(SOCK_PASSSEC, &sock->flags);
952 break;
953 case SO_MARK:
954 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
955 ret = -EPERM;
956 else
957 sk->sk_mark = val;
958 break;
959
960 case SO_RXQ_OVFL:
961 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
962 break;
963
964 case SO_WIFI_STATUS:
965 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
966 break;
967
968 case SO_PEEK_OFF:
969 if (sock->ops->set_peek_off)
970 ret = sock->ops->set_peek_off(sk, val);
971 else
972 ret = -EOPNOTSUPP;
973 break;
974
975 case SO_NOFCS:
976 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
977 break;
978
979 case SO_SELECT_ERR_QUEUE:
980 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
981 break;
982
983 #ifdef CONFIG_NET_RX_BUSY_POLL
984 case SO_BUSY_POLL:
985 /* allow unprivileged users to decrease the value */
986 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
987 ret = -EPERM;
988 else {
989 if (val < 0)
990 ret = -EINVAL;
991 else
992 sk->sk_ll_usec = val;
993 }
994 break;
995 #endif
996
997 case SO_MAX_PACING_RATE:
998 sk->sk_max_pacing_rate = val;
999 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1000 sk->sk_max_pacing_rate);
1001 break;
1002
1003 case SO_INCOMING_CPU:
1004 sk->sk_incoming_cpu = val;
1005 break;
1006
1007 default:
1008 ret = -ENOPROTOOPT;
1009 break;
1010 }
1011 release_sock(sk);
1012 return ret;
1013 }
1014 EXPORT_SYMBOL(sock_setsockopt);
1015
1016
1017 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1018 struct ucred *ucred)
1019 {
1020 ucred->pid = pid_vnr(pid);
1021 ucred->uid = ucred->gid = -1;
1022 if (cred) {
1023 struct user_namespace *current_ns = current_user_ns();
1024
1025 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1026 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1027 }
1028 }
1029
1030 int sock_getsockopt(struct socket *sock, int level, int optname,
1031 char __user *optval, int __user *optlen)
1032 {
1033 struct sock *sk = sock->sk;
1034
1035 union {
1036 int val;
1037 struct linger ling;
1038 struct timeval tm;
1039 } v;
1040
1041 int lv = sizeof(int);
1042 int len;
1043
1044 if (get_user(len, optlen))
1045 return -EFAULT;
1046 if (len < 0)
1047 return -EINVAL;
1048
1049 memset(&v, 0, sizeof(v));
1050
1051 switch (optname) {
1052 case SO_DEBUG:
1053 v.val = sock_flag(sk, SOCK_DBG);
1054 break;
1055
1056 case SO_DONTROUTE:
1057 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1058 break;
1059
1060 case SO_BROADCAST:
1061 v.val = sock_flag(sk, SOCK_BROADCAST);
1062 break;
1063
1064 case SO_SNDBUF:
1065 v.val = sk->sk_sndbuf;
1066 break;
1067
1068 case SO_RCVBUF:
1069 v.val = sk->sk_rcvbuf;
1070 break;
1071
1072 case SO_REUSEADDR:
1073 v.val = sk->sk_reuse;
1074 break;
1075
1076 case SO_REUSEPORT:
1077 v.val = sk->sk_reuseport;
1078 break;
1079
1080 case SO_KEEPALIVE:
1081 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1082 break;
1083
1084 case SO_TYPE:
1085 v.val = sk->sk_type;
1086 break;
1087
1088 case SO_PROTOCOL:
1089 v.val = sk->sk_protocol;
1090 break;
1091
1092 case SO_DOMAIN:
1093 v.val = sk->sk_family;
1094 break;
1095
1096 case SO_ERROR:
1097 v.val = -sock_error(sk);
1098 if (v.val == 0)
1099 v.val = xchg(&sk->sk_err_soft, 0);
1100 break;
1101
1102 case SO_OOBINLINE:
1103 v.val = sock_flag(sk, SOCK_URGINLINE);
1104 break;
1105
1106 case SO_NO_CHECK:
1107 v.val = sk->sk_no_check_tx;
1108 break;
1109
1110 case SO_PRIORITY:
1111 v.val = sk->sk_priority;
1112 break;
1113
1114 case SO_LINGER:
1115 lv = sizeof(v.ling);
1116 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1117 v.ling.l_linger = sk->sk_lingertime / HZ;
1118 break;
1119
1120 case SO_BSDCOMPAT:
1121 sock_warn_obsolete_bsdism("getsockopt");
1122 break;
1123
1124 case SO_TIMESTAMP:
1125 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1126 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1127 break;
1128
1129 case SO_TIMESTAMPNS:
1130 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1131 break;
1132
1133 case SO_TIMESTAMPING:
1134 v.val = sk->sk_tsflags;
1135 break;
1136
1137 case SO_RCVTIMEO:
1138 lv = sizeof(struct timeval);
1139 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1140 v.tm.tv_sec = 0;
1141 v.tm.tv_usec = 0;
1142 } else {
1143 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1144 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1145 }
1146 break;
1147
1148 case SO_SNDTIMEO:
1149 lv = sizeof(struct timeval);
1150 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1151 v.tm.tv_sec = 0;
1152 v.tm.tv_usec = 0;
1153 } else {
1154 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1155 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1156 }
1157 break;
1158
1159 case SO_RCVLOWAT:
1160 v.val = sk->sk_rcvlowat;
1161 break;
1162
1163 case SO_SNDLOWAT:
1164 v.val = 1;
1165 break;
1166
1167 case SO_PASSCRED:
1168 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1169 break;
1170
1171 case SO_PEERCRED:
1172 {
1173 struct ucred peercred;
1174 if (len > sizeof(peercred))
1175 len = sizeof(peercred);
1176 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1177 if (copy_to_user(optval, &peercred, len))
1178 return -EFAULT;
1179 goto lenout;
1180 }
1181
1182 case SO_PEERNAME:
1183 {
1184 char address[128];
1185
1186 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1187 return -ENOTCONN;
1188 if (lv < len)
1189 return -EINVAL;
1190 if (copy_to_user(optval, address, len))
1191 return -EFAULT;
1192 goto lenout;
1193 }
1194
1195 /* Dubious BSD thing... Probably nobody even uses it, but
1196 * the UNIX standard wants it for whatever reason... -DaveM
1197 */
1198 case SO_ACCEPTCONN:
1199 v.val = sk->sk_state == TCP_LISTEN;
1200 break;
1201
1202 case SO_PASSSEC:
1203 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1204 break;
1205
1206 case SO_PEERSEC:
1207 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1208
1209 case SO_MARK:
1210 v.val = sk->sk_mark;
1211 break;
1212
1213 case SO_RXQ_OVFL:
1214 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1215 break;
1216
1217 case SO_WIFI_STATUS:
1218 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1219 break;
1220
1221 case SO_PEEK_OFF:
1222 if (!sock->ops->set_peek_off)
1223 return -EOPNOTSUPP;
1224
1225 v.val = sk->sk_peek_off;
1226 break;
1227 case SO_NOFCS:
1228 v.val = sock_flag(sk, SOCK_NOFCS);
1229 break;
1230
1231 case SO_BINDTODEVICE:
1232 return sock_getbindtodevice(sk, optval, optlen, len);
1233
1234 case SO_GET_FILTER:
1235 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1236 if (len < 0)
1237 return len;
1238
1239 goto lenout;
1240
1241 case SO_LOCK_FILTER:
1242 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1243 break;
1244
1245 case SO_BPF_EXTENSIONS:
1246 v.val = bpf_tell_extensions();
1247 break;
1248
1249 case SO_SELECT_ERR_QUEUE:
1250 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1251 break;
1252
1253 #ifdef CONFIG_NET_RX_BUSY_POLL
1254 case SO_BUSY_POLL:
1255 v.val = sk->sk_ll_usec;
1256 break;
1257 #endif
1258
1259 case SO_MAX_PACING_RATE:
1260 v.val = sk->sk_max_pacing_rate;
1261 break;
1262
1263 case SO_INCOMING_CPU:
1264 v.val = sk->sk_incoming_cpu;
1265 break;
1266
1267 default:
1268 /* We implement the SO_SNDLOWAT etc to not be settable
1269 * (1003.1g 7).
1270 */
1271 return -ENOPROTOOPT;
1272 }
1273
1274 if (len > lv)
1275 len = lv;
1276 if (copy_to_user(optval, &v, len))
1277 return -EFAULT;
1278 lenout:
1279 if (put_user(len, optlen))
1280 return -EFAULT;
1281 return 0;
1282 }
1283
1284 /*
1285 * Initialize an sk_lock.
1286 *
1287 * (We also register the sk_lock with the lock validator.)
1288 */
1289 static inline void sock_lock_init(struct sock *sk)
1290 {
1291 sock_lock_init_class_and_name(sk,
1292 af_family_slock_key_strings[sk->sk_family],
1293 af_family_slock_keys + sk->sk_family,
1294 af_family_key_strings[sk->sk_family],
1295 af_family_keys + sk->sk_family);
1296 }
1297
1298 /*
1299 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1300 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1301 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1302 */
1303 static void sock_copy(struct sock *nsk, const struct sock *osk)
1304 {
1305 #ifdef CONFIG_SECURITY_NETWORK
1306 void *sptr = nsk->sk_security;
1307 #endif
1308 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1309
1310 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1311 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1312
1313 #ifdef CONFIG_SECURITY_NETWORK
1314 nsk->sk_security = sptr;
1315 security_sk_clone(osk, nsk);
1316 #endif
1317 }
1318
1319 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1320 {
1321 unsigned long nulls1, nulls2;
1322
1323 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1324 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1325 if (nulls1 > nulls2)
1326 swap(nulls1, nulls2);
1327
1328 if (nulls1 != 0)
1329 memset((char *)sk, 0, nulls1);
1330 memset((char *)sk + nulls1 + sizeof(void *), 0,
1331 nulls2 - nulls1 - sizeof(void *));
1332 memset((char *)sk + nulls2 + sizeof(void *), 0,
1333 size - nulls2 - sizeof(void *));
1334 }
1335 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1336
1337 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1338 int family)
1339 {
1340 struct sock *sk;
1341 struct kmem_cache *slab;
1342
1343 slab = prot->slab;
1344 if (slab != NULL) {
1345 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1346 if (!sk)
1347 return sk;
1348 if (priority & __GFP_ZERO) {
1349 if (prot->clear_sk)
1350 prot->clear_sk(sk, prot->obj_size);
1351 else
1352 sk_prot_clear_nulls(sk, prot->obj_size);
1353 }
1354 } else
1355 sk = kmalloc(prot->obj_size, priority);
1356
1357 if (sk != NULL) {
1358 kmemcheck_annotate_bitfield(sk, flags);
1359
1360 if (security_sk_alloc(sk, family, priority))
1361 goto out_free;
1362
1363 if (!try_module_get(prot->owner))
1364 goto out_free_sec;
1365 sk_tx_queue_clear(sk);
1366 }
1367
1368 return sk;
1369
1370 out_free_sec:
1371 security_sk_free(sk);
1372 out_free:
1373 if (slab != NULL)
1374 kmem_cache_free(slab, sk);
1375 else
1376 kfree(sk);
1377 return NULL;
1378 }
1379
1380 static void sk_prot_free(struct proto *prot, struct sock *sk)
1381 {
1382 struct kmem_cache *slab;
1383 struct module *owner;
1384
1385 owner = prot->owner;
1386 slab = prot->slab;
1387
1388 security_sk_free(sk);
1389 if (slab != NULL)
1390 kmem_cache_free(slab, sk);
1391 else
1392 kfree(sk);
1393 module_put(owner);
1394 }
1395
1396 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1397 void sock_update_netprioidx(struct sock *sk)
1398 {
1399 if (in_interrupt())
1400 return;
1401
1402 sk->sk_cgrp_prioidx = task_netprioidx(current);
1403 }
1404 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1405 #endif
1406
1407 /**
1408 * sk_alloc - All socket objects are allocated here
1409 * @net: the applicable net namespace
1410 * @family: protocol family
1411 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1412 * @prot: struct proto associated with this new sock instance
1413 * @kern: is this to be a kernel socket?
1414 */
1415 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1416 struct proto *prot, int kern)
1417 {
1418 struct sock *sk;
1419
1420 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1421 if (sk) {
1422 sk->sk_family = family;
1423 /*
1424 * See comment in struct sock definition to understand
1425 * why we need sk_prot_creator -acme
1426 */
1427 sk->sk_prot = sk->sk_prot_creator = prot;
1428 sock_lock_init(sk);
1429 sk->sk_net_refcnt = kern ? 0 : 1;
1430 if (likely(sk->sk_net_refcnt))
1431 get_net(net);
1432 sock_net_set(sk, net);
1433 atomic_set(&sk->sk_wmem_alloc, 1);
1434
1435 sock_update_classid(sk);
1436 sock_update_netprioidx(sk);
1437 }
1438
1439 return sk;
1440 }
1441 EXPORT_SYMBOL(sk_alloc);
1442
1443 void sk_destruct(struct sock *sk)
1444 {
1445 struct sk_filter *filter;
1446
1447 if (sk->sk_destruct)
1448 sk->sk_destruct(sk);
1449
1450 filter = rcu_dereference_check(sk->sk_filter,
1451 atomic_read(&sk->sk_wmem_alloc) == 0);
1452 if (filter) {
1453 sk_filter_uncharge(sk, filter);
1454 RCU_INIT_POINTER(sk->sk_filter, NULL);
1455 }
1456
1457 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1458
1459 if (atomic_read(&sk->sk_omem_alloc))
1460 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1461 __func__, atomic_read(&sk->sk_omem_alloc));
1462
1463 if (sk->sk_peer_cred)
1464 put_cred(sk->sk_peer_cred);
1465 put_pid(sk->sk_peer_pid);
1466 if (likely(sk->sk_net_refcnt))
1467 put_net(sock_net(sk));
1468 sk_prot_free(sk->sk_prot_creator, sk);
1469 }
1470
1471 static void __sk_free(struct sock *sk)
1472 {
1473 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1474 sock_diag_broadcast_destroy(sk);
1475 else
1476 sk_destruct(sk);
1477 }
1478
1479 void sk_free(struct sock *sk)
1480 {
1481 /*
1482 * We subtract one from sk_wmem_alloc and can know if
1483 * some packets are still in some tx queue.
1484 * If not null, sock_wfree() will call __sk_free(sk) later
1485 */
1486 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1487 __sk_free(sk);
1488 }
1489 EXPORT_SYMBOL(sk_free);
1490
1491 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1492 {
1493 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1494 sock_update_memcg(newsk);
1495 }
1496
1497 /**
1498 * sk_clone_lock - clone a socket, and lock its clone
1499 * @sk: the socket to clone
1500 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1501 *
1502 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1503 */
1504 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1505 {
1506 struct sock *newsk;
1507 bool is_charged = true;
1508
1509 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1510 if (newsk != NULL) {
1511 struct sk_filter *filter;
1512
1513 sock_copy(newsk, sk);
1514
1515 /* SANITY */
1516 if (likely(newsk->sk_net_refcnt))
1517 get_net(sock_net(newsk));
1518 sk_node_init(&newsk->sk_node);
1519 sock_lock_init(newsk);
1520 bh_lock_sock(newsk);
1521 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1522 newsk->sk_backlog.len = 0;
1523
1524 atomic_set(&newsk->sk_rmem_alloc, 0);
1525 /*
1526 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1527 */
1528 atomic_set(&newsk->sk_wmem_alloc, 1);
1529 atomic_set(&newsk->sk_omem_alloc, 0);
1530 skb_queue_head_init(&newsk->sk_receive_queue);
1531 skb_queue_head_init(&newsk->sk_write_queue);
1532
1533 rwlock_init(&newsk->sk_callback_lock);
1534 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1535 af_callback_keys + newsk->sk_family,
1536 af_family_clock_key_strings[newsk->sk_family]);
1537
1538 newsk->sk_dst_cache = NULL;
1539 newsk->sk_wmem_queued = 0;
1540 newsk->sk_forward_alloc = 0;
1541 newsk->sk_send_head = NULL;
1542 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1543
1544 sock_reset_flag(newsk, SOCK_DONE);
1545 skb_queue_head_init(&newsk->sk_error_queue);
1546
1547 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1548 if (filter != NULL)
1549 /* though it's an empty new sock, the charging may fail
1550 * if sysctl_optmem_max was changed between creation of
1551 * original socket and cloning
1552 */
1553 is_charged = sk_filter_charge(newsk, filter);
1554
1555 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1556 /* It is still raw copy of parent, so invalidate
1557 * destructor and make plain sk_free() */
1558 newsk->sk_destruct = NULL;
1559 bh_unlock_sock(newsk);
1560 sk_free(newsk);
1561 newsk = NULL;
1562 goto out;
1563 }
1564
1565 newsk->sk_err = 0;
1566 newsk->sk_priority = 0;
1567 newsk->sk_incoming_cpu = raw_smp_processor_id();
1568 atomic64_set(&newsk->sk_cookie, 0);
1569 /*
1570 * Before updating sk_refcnt, we must commit prior changes to memory
1571 * (Documentation/RCU/rculist_nulls.txt for details)
1572 */
1573 smp_wmb();
1574 atomic_set(&newsk->sk_refcnt, 2);
1575
1576 /*
1577 * Increment the counter in the same struct proto as the master
1578 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1579 * is the same as sk->sk_prot->socks, as this field was copied
1580 * with memcpy).
1581 *
1582 * This _changes_ the previous behaviour, where
1583 * tcp_create_openreq_child always was incrementing the
1584 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1585 * to be taken into account in all callers. -acme
1586 */
1587 sk_refcnt_debug_inc(newsk);
1588 sk_set_socket(newsk, NULL);
1589 newsk->sk_wq = NULL;
1590
1591 sk_update_clone(sk, newsk);
1592
1593 if (newsk->sk_prot->sockets_allocated)
1594 sk_sockets_allocated_inc(newsk);
1595
1596 if (sock_needs_netstamp(sk) &&
1597 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1598 net_enable_timestamp();
1599 }
1600 out:
1601 return newsk;
1602 }
1603 EXPORT_SYMBOL_GPL(sk_clone_lock);
1604
1605 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1606 {
1607 u32 max_segs = 1;
1608
1609 sk_dst_set(sk, dst);
1610 sk->sk_route_caps = dst->dev->features;
1611 if (sk->sk_route_caps & NETIF_F_GSO)
1612 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1613 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1614 if (sk_can_gso(sk)) {
1615 if (dst->header_len) {
1616 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1617 } else {
1618 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1619 sk->sk_gso_max_size = dst->dev->gso_max_size;
1620 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1621 }
1622 }
1623 sk->sk_gso_max_segs = max_segs;
1624 }
1625 EXPORT_SYMBOL_GPL(sk_setup_caps);
1626
1627 /*
1628 * Simple resource managers for sockets.
1629 */
1630
1631
1632 /*
1633 * Write buffer destructor automatically called from kfree_skb.
1634 */
1635 void sock_wfree(struct sk_buff *skb)
1636 {
1637 struct sock *sk = skb->sk;
1638 unsigned int len = skb->truesize;
1639
1640 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1641 /*
1642 * Keep a reference on sk_wmem_alloc, this will be released
1643 * after sk_write_space() call
1644 */
1645 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1646 sk->sk_write_space(sk);
1647 len = 1;
1648 }
1649 /*
1650 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1651 * could not do because of in-flight packets
1652 */
1653 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1654 __sk_free(sk);
1655 }
1656 EXPORT_SYMBOL(sock_wfree);
1657
1658 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1659 {
1660 skb_orphan(skb);
1661 skb->sk = sk;
1662 #ifdef CONFIG_INET
1663 if (unlikely(!sk_fullsock(sk))) {
1664 skb->destructor = sock_edemux;
1665 sock_hold(sk);
1666 return;
1667 }
1668 #endif
1669 skb->destructor = sock_wfree;
1670 skb_set_hash_from_sk(skb, sk);
1671 /*
1672 * We used to take a refcount on sk, but following operation
1673 * is enough to guarantee sk_free() wont free this sock until
1674 * all in-flight packets are completed
1675 */
1676 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1677 }
1678 EXPORT_SYMBOL(skb_set_owner_w);
1679
1680 void skb_orphan_partial(struct sk_buff *skb)
1681 {
1682 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1683 * so we do not completely orphan skb, but transfert all
1684 * accounted bytes but one, to avoid unexpected reorders.
1685 */
1686 if (skb->destructor == sock_wfree
1687 #ifdef CONFIG_INET
1688 || skb->destructor == tcp_wfree
1689 #endif
1690 ) {
1691 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1692 skb->truesize = 1;
1693 } else {
1694 skb_orphan(skb);
1695 }
1696 }
1697 EXPORT_SYMBOL(skb_orphan_partial);
1698
1699 /*
1700 * Read buffer destructor automatically called from kfree_skb.
1701 */
1702 void sock_rfree(struct sk_buff *skb)
1703 {
1704 struct sock *sk = skb->sk;
1705 unsigned int len = skb->truesize;
1706
1707 atomic_sub(len, &sk->sk_rmem_alloc);
1708 sk_mem_uncharge(sk, len);
1709 }
1710 EXPORT_SYMBOL(sock_rfree);
1711
1712 /*
1713 * Buffer destructor for skbs that are not used directly in read or write
1714 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1715 */
1716 void sock_efree(struct sk_buff *skb)
1717 {
1718 sock_put(skb->sk);
1719 }
1720 EXPORT_SYMBOL(sock_efree);
1721
1722 kuid_t sock_i_uid(struct sock *sk)
1723 {
1724 kuid_t uid;
1725
1726 read_lock_bh(&sk->sk_callback_lock);
1727 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1728 read_unlock_bh(&sk->sk_callback_lock);
1729 return uid;
1730 }
1731 EXPORT_SYMBOL(sock_i_uid);
1732
1733 unsigned long sock_i_ino(struct sock *sk)
1734 {
1735 unsigned long ino;
1736
1737 read_lock_bh(&sk->sk_callback_lock);
1738 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1739 read_unlock_bh(&sk->sk_callback_lock);
1740 return ino;
1741 }
1742 EXPORT_SYMBOL(sock_i_ino);
1743
1744 /*
1745 * Allocate a skb from the socket's send buffer.
1746 */
1747 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1748 gfp_t priority)
1749 {
1750 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1751 struct sk_buff *skb = alloc_skb(size, priority);
1752 if (skb) {
1753 skb_set_owner_w(skb, sk);
1754 return skb;
1755 }
1756 }
1757 return NULL;
1758 }
1759 EXPORT_SYMBOL(sock_wmalloc);
1760
1761 /*
1762 * Allocate a memory block from the socket's option memory buffer.
1763 */
1764 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1765 {
1766 if ((unsigned int)size <= sysctl_optmem_max &&
1767 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1768 void *mem;
1769 /* First do the add, to avoid the race if kmalloc
1770 * might sleep.
1771 */
1772 atomic_add(size, &sk->sk_omem_alloc);
1773 mem = kmalloc(size, priority);
1774 if (mem)
1775 return mem;
1776 atomic_sub(size, &sk->sk_omem_alloc);
1777 }
1778 return NULL;
1779 }
1780 EXPORT_SYMBOL(sock_kmalloc);
1781
1782 /* Free an option memory block. Note, we actually want the inline
1783 * here as this allows gcc to detect the nullify and fold away the
1784 * condition entirely.
1785 */
1786 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1787 const bool nullify)
1788 {
1789 if (WARN_ON_ONCE(!mem))
1790 return;
1791 if (nullify)
1792 kzfree(mem);
1793 else
1794 kfree(mem);
1795 atomic_sub(size, &sk->sk_omem_alloc);
1796 }
1797
1798 void sock_kfree_s(struct sock *sk, void *mem, int size)
1799 {
1800 __sock_kfree_s(sk, mem, size, false);
1801 }
1802 EXPORT_SYMBOL(sock_kfree_s);
1803
1804 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1805 {
1806 __sock_kfree_s(sk, mem, size, true);
1807 }
1808 EXPORT_SYMBOL(sock_kzfree_s);
1809
1810 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1811 I think, these locks should be removed for datagram sockets.
1812 */
1813 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1814 {
1815 DEFINE_WAIT(wait);
1816
1817 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1818 for (;;) {
1819 if (!timeo)
1820 break;
1821 if (signal_pending(current))
1822 break;
1823 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1824 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1825 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1826 break;
1827 if (sk->sk_shutdown & SEND_SHUTDOWN)
1828 break;
1829 if (sk->sk_err)
1830 break;
1831 timeo = schedule_timeout(timeo);
1832 }
1833 finish_wait(sk_sleep(sk), &wait);
1834 return timeo;
1835 }
1836
1837
1838 /*
1839 * Generic send/receive buffer handlers
1840 */
1841
1842 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1843 unsigned long data_len, int noblock,
1844 int *errcode, int max_page_order)
1845 {
1846 struct sk_buff *skb;
1847 long timeo;
1848 int err;
1849
1850 timeo = sock_sndtimeo(sk, noblock);
1851 for (;;) {
1852 err = sock_error(sk);
1853 if (err != 0)
1854 goto failure;
1855
1856 err = -EPIPE;
1857 if (sk->sk_shutdown & SEND_SHUTDOWN)
1858 goto failure;
1859
1860 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1861 break;
1862
1863 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1864 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1865 err = -EAGAIN;
1866 if (!timeo)
1867 goto failure;
1868 if (signal_pending(current))
1869 goto interrupted;
1870 timeo = sock_wait_for_wmem(sk, timeo);
1871 }
1872 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1873 errcode, sk->sk_allocation);
1874 if (skb)
1875 skb_set_owner_w(skb, sk);
1876 return skb;
1877
1878 interrupted:
1879 err = sock_intr_errno(timeo);
1880 failure:
1881 *errcode = err;
1882 return NULL;
1883 }
1884 EXPORT_SYMBOL(sock_alloc_send_pskb);
1885
1886 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1887 int noblock, int *errcode)
1888 {
1889 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1890 }
1891 EXPORT_SYMBOL(sock_alloc_send_skb);
1892
1893 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1894 struct sockcm_cookie *sockc)
1895 {
1896 struct cmsghdr *cmsg;
1897
1898 for_each_cmsghdr(cmsg, msg) {
1899 if (!CMSG_OK(msg, cmsg))
1900 return -EINVAL;
1901 if (cmsg->cmsg_level != SOL_SOCKET)
1902 continue;
1903 switch (cmsg->cmsg_type) {
1904 case SO_MARK:
1905 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1906 return -EPERM;
1907 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1908 return -EINVAL;
1909 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1910 break;
1911 default:
1912 return -EINVAL;
1913 }
1914 }
1915 return 0;
1916 }
1917 EXPORT_SYMBOL(sock_cmsg_send);
1918
1919 /* On 32bit arches, an skb frag is limited to 2^15 */
1920 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1921
1922 /**
1923 * skb_page_frag_refill - check that a page_frag contains enough room
1924 * @sz: minimum size of the fragment we want to get
1925 * @pfrag: pointer to page_frag
1926 * @gfp: priority for memory allocation
1927 *
1928 * Note: While this allocator tries to use high order pages, there is
1929 * no guarantee that allocations succeed. Therefore, @sz MUST be
1930 * less or equal than PAGE_SIZE.
1931 */
1932 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1933 {
1934 if (pfrag->page) {
1935 if (atomic_read(&pfrag->page->_count) == 1) {
1936 pfrag->offset = 0;
1937 return true;
1938 }
1939 if (pfrag->offset + sz <= pfrag->size)
1940 return true;
1941 put_page(pfrag->page);
1942 }
1943
1944 pfrag->offset = 0;
1945 if (SKB_FRAG_PAGE_ORDER) {
1946 /* Avoid direct reclaim but allow kswapd to wake */
1947 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1948 __GFP_COMP | __GFP_NOWARN |
1949 __GFP_NORETRY,
1950 SKB_FRAG_PAGE_ORDER);
1951 if (likely(pfrag->page)) {
1952 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1953 return true;
1954 }
1955 }
1956 pfrag->page = alloc_page(gfp);
1957 if (likely(pfrag->page)) {
1958 pfrag->size = PAGE_SIZE;
1959 return true;
1960 }
1961 return false;
1962 }
1963 EXPORT_SYMBOL(skb_page_frag_refill);
1964
1965 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1966 {
1967 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1968 return true;
1969
1970 sk_enter_memory_pressure(sk);
1971 sk_stream_moderate_sndbuf(sk);
1972 return false;
1973 }
1974 EXPORT_SYMBOL(sk_page_frag_refill);
1975
1976 static void __lock_sock(struct sock *sk)
1977 __releases(&sk->sk_lock.slock)
1978 __acquires(&sk->sk_lock.slock)
1979 {
1980 DEFINE_WAIT(wait);
1981
1982 for (;;) {
1983 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1984 TASK_UNINTERRUPTIBLE);
1985 spin_unlock_bh(&sk->sk_lock.slock);
1986 schedule();
1987 spin_lock_bh(&sk->sk_lock.slock);
1988 if (!sock_owned_by_user(sk))
1989 break;
1990 }
1991 finish_wait(&sk->sk_lock.wq, &wait);
1992 }
1993
1994 static void __release_sock(struct sock *sk)
1995 __releases(&sk->sk_lock.slock)
1996 __acquires(&sk->sk_lock.slock)
1997 {
1998 struct sk_buff *skb = sk->sk_backlog.head;
1999
2000 do {
2001 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2002 bh_unlock_sock(sk);
2003
2004 do {
2005 struct sk_buff *next = skb->next;
2006
2007 prefetch(next);
2008 WARN_ON_ONCE(skb_dst_is_noref(skb));
2009 skb->next = NULL;
2010 sk_backlog_rcv(sk, skb);
2011
2012 /*
2013 * We are in process context here with softirqs
2014 * disabled, use cond_resched_softirq() to preempt.
2015 * This is safe to do because we've taken the backlog
2016 * queue private:
2017 */
2018 cond_resched_softirq();
2019
2020 skb = next;
2021 } while (skb != NULL);
2022
2023 bh_lock_sock(sk);
2024 } while ((skb = sk->sk_backlog.head) != NULL);
2025
2026 /*
2027 * Doing the zeroing here guarantee we can not loop forever
2028 * while a wild producer attempts to flood us.
2029 */
2030 sk->sk_backlog.len = 0;
2031 }
2032
2033 /**
2034 * sk_wait_data - wait for data to arrive at sk_receive_queue
2035 * @sk: sock to wait on
2036 * @timeo: for how long
2037 * @skb: last skb seen on sk_receive_queue
2038 *
2039 * Now socket state including sk->sk_err is changed only under lock,
2040 * hence we may omit checks after joining wait queue.
2041 * We check receive queue before schedule() only as optimization;
2042 * it is very likely that release_sock() added new data.
2043 */
2044 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2045 {
2046 int rc;
2047 DEFINE_WAIT(wait);
2048
2049 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2050 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2051 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2052 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2053 finish_wait(sk_sleep(sk), &wait);
2054 return rc;
2055 }
2056 EXPORT_SYMBOL(sk_wait_data);
2057
2058 /**
2059 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2060 * @sk: socket
2061 * @size: memory size to allocate
2062 * @kind: allocation type
2063 *
2064 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2065 * rmem allocation. This function assumes that protocols which have
2066 * memory_pressure use sk_wmem_queued as write buffer accounting.
2067 */
2068 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2069 {
2070 struct proto *prot = sk->sk_prot;
2071 int amt = sk_mem_pages(size);
2072 long allocated;
2073 int parent_status = UNDER_LIMIT;
2074
2075 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2076
2077 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2078
2079 /* Under limit. */
2080 if (parent_status == UNDER_LIMIT &&
2081 allocated <= sk_prot_mem_limits(sk, 0)) {
2082 sk_leave_memory_pressure(sk);
2083 return 1;
2084 }
2085
2086 /* Under pressure. (we or our parents) */
2087 if ((parent_status > SOFT_LIMIT) ||
2088 allocated > sk_prot_mem_limits(sk, 1))
2089 sk_enter_memory_pressure(sk);
2090
2091 /* Over hard limit (we or our parents) */
2092 if ((parent_status == OVER_LIMIT) ||
2093 (allocated > sk_prot_mem_limits(sk, 2)))
2094 goto suppress_allocation;
2095
2096 /* guarantee minimum buffer size under pressure */
2097 if (kind == SK_MEM_RECV) {
2098 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2099 return 1;
2100
2101 } else { /* SK_MEM_SEND */
2102 if (sk->sk_type == SOCK_STREAM) {
2103 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2104 return 1;
2105 } else if (atomic_read(&sk->sk_wmem_alloc) <
2106 prot->sysctl_wmem[0])
2107 return 1;
2108 }
2109
2110 if (sk_has_memory_pressure(sk)) {
2111 int alloc;
2112
2113 if (!sk_under_memory_pressure(sk))
2114 return 1;
2115 alloc = sk_sockets_allocated_read_positive(sk);
2116 if (sk_prot_mem_limits(sk, 2) > alloc *
2117 sk_mem_pages(sk->sk_wmem_queued +
2118 atomic_read(&sk->sk_rmem_alloc) +
2119 sk->sk_forward_alloc))
2120 return 1;
2121 }
2122
2123 suppress_allocation:
2124
2125 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2126 sk_stream_moderate_sndbuf(sk);
2127
2128 /* Fail only if socket is _under_ its sndbuf.
2129 * In this case we cannot block, so that we have to fail.
2130 */
2131 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2132 return 1;
2133 }
2134
2135 trace_sock_exceed_buf_limit(sk, prot, allocated);
2136
2137 /* Alas. Undo changes. */
2138 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2139
2140 sk_memory_allocated_sub(sk, amt);
2141
2142 return 0;
2143 }
2144 EXPORT_SYMBOL(__sk_mem_schedule);
2145
2146 /**
2147 * __sk_mem_reclaim - reclaim memory_allocated
2148 * @sk: socket
2149 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2150 */
2151 void __sk_mem_reclaim(struct sock *sk, int amount)
2152 {
2153 amount >>= SK_MEM_QUANTUM_SHIFT;
2154 sk_memory_allocated_sub(sk, amount);
2155 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2156
2157 if (sk_under_memory_pressure(sk) &&
2158 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2159 sk_leave_memory_pressure(sk);
2160 }
2161 EXPORT_SYMBOL(__sk_mem_reclaim);
2162
2163
2164 /*
2165 * Set of default routines for initialising struct proto_ops when
2166 * the protocol does not support a particular function. In certain
2167 * cases where it makes no sense for a protocol to have a "do nothing"
2168 * function, some default processing is provided.
2169 */
2170
2171 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2172 {
2173 return -EOPNOTSUPP;
2174 }
2175 EXPORT_SYMBOL(sock_no_bind);
2176
2177 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2178 int len, int flags)
2179 {
2180 return -EOPNOTSUPP;
2181 }
2182 EXPORT_SYMBOL(sock_no_connect);
2183
2184 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2185 {
2186 return -EOPNOTSUPP;
2187 }
2188 EXPORT_SYMBOL(sock_no_socketpair);
2189
2190 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2191 {
2192 return -EOPNOTSUPP;
2193 }
2194 EXPORT_SYMBOL(sock_no_accept);
2195
2196 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2197 int *len, int peer)
2198 {
2199 return -EOPNOTSUPP;
2200 }
2201 EXPORT_SYMBOL(sock_no_getname);
2202
2203 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2204 {
2205 return 0;
2206 }
2207 EXPORT_SYMBOL(sock_no_poll);
2208
2209 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2210 {
2211 return -EOPNOTSUPP;
2212 }
2213 EXPORT_SYMBOL(sock_no_ioctl);
2214
2215 int sock_no_listen(struct socket *sock, int backlog)
2216 {
2217 return -EOPNOTSUPP;
2218 }
2219 EXPORT_SYMBOL(sock_no_listen);
2220
2221 int sock_no_shutdown(struct socket *sock, int how)
2222 {
2223 return -EOPNOTSUPP;
2224 }
2225 EXPORT_SYMBOL(sock_no_shutdown);
2226
2227 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2228 char __user *optval, unsigned int optlen)
2229 {
2230 return -EOPNOTSUPP;
2231 }
2232 EXPORT_SYMBOL(sock_no_setsockopt);
2233
2234 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2235 char __user *optval, int __user *optlen)
2236 {
2237 return -EOPNOTSUPP;
2238 }
2239 EXPORT_SYMBOL(sock_no_getsockopt);
2240
2241 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2242 {
2243 return -EOPNOTSUPP;
2244 }
2245 EXPORT_SYMBOL(sock_no_sendmsg);
2246
2247 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2248 int flags)
2249 {
2250 return -EOPNOTSUPP;
2251 }
2252 EXPORT_SYMBOL(sock_no_recvmsg);
2253
2254 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2255 {
2256 /* Mirror missing mmap method error code */
2257 return -ENODEV;
2258 }
2259 EXPORT_SYMBOL(sock_no_mmap);
2260
2261 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2262 {
2263 ssize_t res;
2264 struct msghdr msg = {.msg_flags = flags};
2265 struct kvec iov;
2266 char *kaddr = kmap(page);
2267 iov.iov_base = kaddr + offset;
2268 iov.iov_len = size;
2269 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2270 kunmap(page);
2271 return res;
2272 }
2273 EXPORT_SYMBOL(sock_no_sendpage);
2274
2275 /*
2276 * Default Socket Callbacks
2277 */
2278
2279 static void sock_def_wakeup(struct sock *sk)
2280 {
2281 struct socket_wq *wq;
2282
2283 rcu_read_lock();
2284 wq = rcu_dereference(sk->sk_wq);
2285 if (wq_has_sleeper(wq))
2286 wake_up_interruptible_all(&wq->wait);
2287 rcu_read_unlock();
2288 }
2289
2290 static void sock_def_error_report(struct sock *sk)
2291 {
2292 struct socket_wq *wq;
2293
2294 rcu_read_lock();
2295 wq = rcu_dereference(sk->sk_wq);
2296 if (wq_has_sleeper(wq))
2297 wake_up_interruptible_poll(&wq->wait, POLLERR);
2298 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2299 rcu_read_unlock();
2300 }
2301
2302 static void sock_def_readable(struct sock *sk)
2303 {
2304 struct socket_wq *wq;
2305
2306 rcu_read_lock();
2307 wq = rcu_dereference(sk->sk_wq);
2308 if (wq_has_sleeper(wq))
2309 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2310 POLLRDNORM | POLLRDBAND);
2311 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2312 rcu_read_unlock();
2313 }
2314
2315 static void sock_def_write_space(struct sock *sk)
2316 {
2317 struct socket_wq *wq;
2318
2319 rcu_read_lock();
2320
2321 /* Do not wake up a writer until he can make "significant"
2322 * progress. --DaveM
2323 */
2324 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2325 wq = rcu_dereference(sk->sk_wq);
2326 if (wq_has_sleeper(wq))
2327 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2328 POLLWRNORM | POLLWRBAND);
2329
2330 /* Should agree with poll, otherwise some programs break */
2331 if (sock_writeable(sk))
2332 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2333 }
2334
2335 rcu_read_unlock();
2336 }
2337
2338 static void sock_def_destruct(struct sock *sk)
2339 {
2340 }
2341
2342 void sk_send_sigurg(struct sock *sk)
2343 {
2344 if (sk->sk_socket && sk->sk_socket->file)
2345 if (send_sigurg(&sk->sk_socket->file->f_owner))
2346 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2347 }
2348 EXPORT_SYMBOL(sk_send_sigurg);
2349
2350 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2351 unsigned long expires)
2352 {
2353 if (!mod_timer(timer, expires))
2354 sock_hold(sk);
2355 }
2356 EXPORT_SYMBOL(sk_reset_timer);
2357
2358 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2359 {
2360 if (del_timer(timer))
2361 __sock_put(sk);
2362 }
2363 EXPORT_SYMBOL(sk_stop_timer);
2364
2365 void sock_init_data(struct socket *sock, struct sock *sk)
2366 {
2367 skb_queue_head_init(&sk->sk_receive_queue);
2368 skb_queue_head_init(&sk->sk_write_queue);
2369 skb_queue_head_init(&sk->sk_error_queue);
2370
2371 sk->sk_send_head = NULL;
2372
2373 init_timer(&sk->sk_timer);
2374
2375 sk->sk_allocation = GFP_KERNEL;
2376 sk->sk_rcvbuf = sysctl_rmem_default;
2377 sk->sk_sndbuf = sysctl_wmem_default;
2378 sk->sk_state = TCP_CLOSE;
2379 sk_set_socket(sk, sock);
2380
2381 sock_set_flag(sk, SOCK_ZAPPED);
2382
2383 if (sock) {
2384 sk->sk_type = sock->type;
2385 sk->sk_wq = sock->wq;
2386 sock->sk = sk;
2387 } else
2388 sk->sk_wq = NULL;
2389
2390 rwlock_init(&sk->sk_callback_lock);
2391 lockdep_set_class_and_name(&sk->sk_callback_lock,
2392 af_callback_keys + sk->sk_family,
2393 af_family_clock_key_strings[sk->sk_family]);
2394
2395 sk->sk_state_change = sock_def_wakeup;
2396 sk->sk_data_ready = sock_def_readable;
2397 sk->sk_write_space = sock_def_write_space;
2398 sk->sk_error_report = sock_def_error_report;
2399 sk->sk_destruct = sock_def_destruct;
2400
2401 sk->sk_frag.page = NULL;
2402 sk->sk_frag.offset = 0;
2403 sk->sk_peek_off = -1;
2404
2405 sk->sk_peer_pid = NULL;
2406 sk->sk_peer_cred = NULL;
2407 sk->sk_write_pending = 0;
2408 sk->sk_rcvlowat = 1;
2409 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2410 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2411
2412 sk->sk_stamp = ktime_set(-1L, 0);
2413
2414 #ifdef CONFIG_NET_RX_BUSY_POLL
2415 sk->sk_napi_id = 0;
2416 sk->sk_ll_usec = sysctl_net_busy_read;
2417 #endif
2418
2419 sk->sk_max_pacing_rate = ~0U;
2420 sk->sk_pacing_rate = ~0U;
2421 sk->sk_incoming_cpu = -1;
2422 /*
2423 * Before updating sk_refcnt, we must commit prior changes to memory
2424 * (Documentation/RCU/rculist_nulls.txt for details)
2425 */
2426 smp_wmb();
2427 atomic_set(&sk->sk_refcnt, 1);
2428 atomic_set(&sk->sk_drops, 0);
2429 }
2430 EXPORT_SYMBOL(sock_init_data);
2431
2432 void lock_sock_nested(struct sock *sk, int subclass)
2433 {
2434 might_sleep();
2435 spin_lock_bh(&sk->sk_lock.slock);
2436 if (sk->sk_lock.owned)
2437 __lock_sock(sk);
2438 sk->sk_lock.owned = 1;
2439 spin_unlock(&sk->sk_lock.slock);
2440 /*
2441 * The sk_lock has mutex_lock() semantics here:
2442 */
2443 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2444 local_bh_enable();
2445 }
2446 EXPORT_SYMBOL(lock_sock_nested);
2447
2448 void release_sock(struct sock *sk)
2449 {
2450 /*
2451 * The sk_lock has mutex_unlock() semantics:
2452 */
2453 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2454
2455 spin_lock_bh(&sk->sk_lock.slock);
2456 if (sk->sk_backlog.tail)
2457 __release_sock(sk);
2458
2459 /* Warning : release_cb() might need to release sk ownership,
2460 * ie call sock_release_ownership(sk) before us.
2461 */
2462 if (sk->sk_prot->release_cb)
2463 sk->sk_prot->release_cb(sk);
2464
2465 sock_release_ownership(sk);
2466 if (waitqueue_active(&sk->sk_lock.wq))
2467 wake_up(&sk->sk_lock.wq);
2468 spin_unlock_bh(&sk->sk_lock.slock);
2469 }
2470 EXPORT_SYMBOL(release_sock);
2471
2472 /**
2473 * lock_sock_fast - fast version of lock_sock
2474 * @sk: socket
2475 *
2476 * This version should be used for very small section, where process wont block
2477 * return false if fast path is taken
2478 * sk_lock.slock locked, owned = 0, BH disabled
2479 * return true if slow path is taken
2480 * sk_lock.slock unlocked, owned = 1, BH enabled
2481 */
2482 bool lock_sock_fast(struct sock *sk)
2483 {
2484 might_sleep();
2485 spin_lock_bh(&sk->sk_lock.slock);
2486
2487 if (!sk->sk_lock.owned)
2488 /*
2489 * Note : We must disable BH
2490 */
2491 return false;
2492
2493 __lock_sock(sk);
2494 sk->sk_lock.owned = 1;
2495 spin_unlock(&sk->sk_lock.slock);
2496 /*
2497 * The sk_lock has mutex_lock() semantics here:
2498 */
2499 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2500 local_bh_enable();
2501 return true;
2502 }
2503 EXPORT_SYMBOL(lock_sock_fast);
2504
2505 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2506 {
2507 struct timeval tv;
2508 if (!sock_flag(sk, SOCK_TIMESTAMP))
2509 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2510 tv = ktime_to_timeval(sk->sk_stamp);
2511 if (tv.tv_sec == -1)
2512 return -ENOENT;
2513 if (tv.tv_sec == 0) {
2514 sk->sk_stamp = ktime_get_real();
2515 tv = ktime_to_timeval(sk->sk_stamp);
2516 }
2517 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2518 }
2519 EXPORT_SYMBOL(sock_get_timestamp);
2520
2521 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2522 {
2523 struct timespec ts;
2524 if (!sock_flag(sk, SOCK_TIMESTAMP))
2525 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2526 ts = ktime_to_timespec(sk->sk_stamp);
2527 if (ts.tv_sec == -1)
2528 return -ENOENT;
2529 if (ts.tv_sec == 0) {
2530 sk->sk_stamp = ktime_get_real();
2531 ts = ktime_to_timespec(sk->sk_stamp);
2532 }
2533 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2534 }
2535 EXPORT_SYMBOL(sock_get_timestampns);
2536
2537 void sock_enable_timestamp(struct sock *sk, int flag)
2538 {
2539 if (!sock_flag(sk, flag)) {
2540 unsigned long previous_flags = sk->sk_flags;
2541
2542 sock_set_flag(sk, flag);
2543 /*
2544 * we just set one of the two flags which require net
2545 * time stamping, but time stamping might have been on
2546 * already because of the other one
2547 */
2548 if (sock_needs_netstamp(sk) &&
2549 !(previous_flags & SK_FLAGS_TIMESTAMP))
2550 net_enable_timestamp();
2551 }
2552 }
2553
2554 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2555 int level, int type)
2556 {
2557 struct sock_exterr_skb *serr;
2558 struct sk_buff *skb;
2559 int copied, err;
2560
2561 err = -EAGAIN;
2562 skb = sock_dequeue_err_skb(sk);
2563 if (skb == NULL)
2564 goto out;
2565
2566 copied = skb->len;
2567 if (copied > len) {
2568 msg->msg_flags |= MSG_TRUNC;
2569 copied = len;
2570 }
2571 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2572 if (err)
2573 goto out_free_skb;
2574
2575 sock_recv_timestamp(msg, sk, skb);
2576
2577 serr = SKB_EXT_ERR(skb);
2578 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2579
2580 msg->msg_flags |= MSG_ERRQUEUE;
2581 err = copied;
2582
2583 out_free_skb:
2584 kfree_skb(skb);
2585 out:
2586 return err;
2587 }
2588 EXPORT_SYMBOL(sock_recv_errqueue);
2589
2590 /*
2591 * Get a socket option on an socket.
2592 *
2593 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2594 * asynchronous errors should be reported by getsockopt. We assume
2595 * this means if you specify SO_ERROR (otherwise whats the point of it).
2596 */
2597 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2598 char __user *optval, int __user *optlen)
2599 {
2600 struct sock *sk = sock->sk;
2601
2602 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2603 }
2604 EXPORT_SYMBOL(sock_common_getsockopt);
2605
2606 #ifdef CONFIG_COMPAT
2607 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2608 char __user *optval, int __user *optlen)
2609 {
2610 struct sock *sk = sock->sk;
2611
2612 if (sk->sk_prot->compat_getsockopt != NULL)
2613 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2614 optval, optlen);
2615 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2616 }
2617 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2618 #endif
2619
2620 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2621 int flags)
2622 {
2623 struct sock *sk = sock->sk;
2624 int addr_len = 0;
2625 int err;
2626
2627 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2628 flags & ~MSG_DONTWAIT, &addr_len);
2629 if (err >= 0)
2630 msg->msg_namelen = addr_len;
2631 return err;
2632 }
2633 EXPORT_SYMBOL(sock_common_recvmsg);
2634
2635 /*
2636 * Set socket options on an inet socket.
2637 */
2638 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2639 char __user *optval, unsigned int optlen)
2640 {
2641 struct sock *sk = sock->sk;
2642
2643 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2644 }
2645 EXPORT_SYMBOL(sock_common_setsockopt);
2646
2647 #ifdef CONFIG_COMPAT
2648 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2649 char __user *optval, unsigned int optlen)
2650 {
2651 struct sock *sk = sock->sk;
2652
2653 if (sk->sk_prot->compat_setsockopt != NULL)
2654 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2655 optval, optlen);
2656 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2657 }
2658 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2659 #endif
2660
2661 void sk_common_release(struct sock *sk)
2662 {
2663 if (sk->sk_prot->destroy)
2664 sk->sk_prot->destroy(sk);
2665
2666 /*
2667 * Observation: when sock_common_release is called, processes have
2668 * no access to socket. But net still has.
2669 * Step one, detach it from networking:
2670 *
2671 * A. Remove from hash tables.
2672 */
2673
2674 sk->sk_prot->unhash(sk);
2675
2676 /*
2677 * In this point socket cannot receive new packets, but it is possible
2678 * that some packets are in flight because some CPU runs receiver and
2679 * did hash table lookup before we unhashed socket. They will achieve
2680 * receive queue and will be purged by socket destructor.
2681 *
2682 * Also we still have packets pending on receive queue and probably,
2683 * our own packets waiting in device queues. sock_destroy will drain
2684 * receive queue, but transmitted packets will delay socket destruction
2685 * until the last reference will be released.
2686 */
2687
2688 sock_orphan(sk);
2689
2690 xfrm_sk_free_policy(sk);
2691
2692 sk_refcnt_debug_release(sk);
2693
2694 if (sk->sk_frag.page) {
2695 put_page(sk->sk_frag.page);
2696 sk->sk_frag.page = NULL;
2697 }
2698
2699 sock_put(sk);
2700 }
2701 EXPORT_SYMBOL(sk_common_release);
2702
2703 #ifdef CONFIG_PROC_FS
2704 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2705 struct prot_inuse {
2706 int val[PROTO_INUSE_NR];
2707 };
2708
2709 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2710
2711 #ifdef CONFIG_NET_NS
2712 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2713 {
2714 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2715 }
2716 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2717
2718 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2719 {
2720 int cpu, idx = prot->inuse_idx;
2721 int res = 0;
2722
2723 for_each_possible_cpu(cpu)
2724 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2725
2726 return res >= 0 ? res : 0;
2727 }
2728 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2729
2730 static int __net_init sock_inuse_init_net(struct net *net)
2731 {
2732 net->core.inuse = alloc_percpu(struct prot_inuse);
2733 return net->core.inuse ? 0 : -ENOMEM;
2734 }
2735
2736 static void __net_exit sock_inuse_exit_net(struct net *net)
2737 {
2738 free_percpu(net->core.inuse);
2739 }
2740
2741 static struct pernet_operations net_inuse_ops = {
2742 .init = sock_inuse_init_net,
2743 .exit = sock_inuse_exit_net,
2744 };
2745
2746 static __init int net_inuse_init(void)
2747 {
2748 if (register_pernet_subsys(&net_inuse_ops))
2749 panic("Cannot initialize net inuse counters");
2750
2751 return 0;
2752 }
2753
2754 core_initcall(net_inuse_init);
2755 #else
2756 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2757
2758 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2759 {
2760 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2761 }
2762 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2763
2764 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2765 {
2766 int cpu, idx = prot->inuse_idx;
2767 int res = 0;
2768
2769 for_each_possible_cpu(cpu)
2770 res += per_cpu(prot_inuse, cpu).val[idx];
2771
2772 return res >= 0 ? res : 0;
2773 }
2774 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2775 #endif
2776
2777 static void assign_proto_idx(struct proto *prot)
2778 {
2779 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2780
2781 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2782 pr_err("PROTO_INUSE_NR exhausted\n");
2783 return;
2784 }
2785
2786 set_bit(prot->inuse_idx, proto_inuse_idx);
2787 }
2788
2789 static void release_proto_idx(struct proto *prot)
2790 {
2791 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2792 clear_bit(prot->inuse_idx, proto_inuse_idx);
2793 }
2794 #else
2795 static inline void assign_proto_idx(struct proto *prot)
2796 {
2797 }
2798
2799 static inline void release_proto_idx(struct proto *prot)
2800 {
2801 }
2802 #endif
2803
2804 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2805 {
2806 if (!rsk_prot)
2807 return;
2808 kfree(rsk_prot->slab_name);
2809 rsk_prot->slab_name = NULL;
2810 kmem_cache_destroy(rsk_prot->slab);
2811 rsk_prot->slab = NULL;
2812 }
2813
2814 static int req_prot_init(const struct proto *prot)
2815 {
2816 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2817
2818 if (!rsk_prot)
2819 return 0;
2820
2821 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2822 prot->name);
2823 if (!rsk_prot->slab_name)
2824 return -ENOMEM;
2825
2826 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2827 rsk_prot->obj_size, 0,
2828 prot->slab_flags, NULL);
2829
2830 if (!rsk_prot->slab) {
2831 pr_crit("%s: Can't create request sock SLAB cache!\n",
2832 prot->name);
2833 return -ENOMEM;
2834 }
2835 return 0;
2836 }
2837
2838 int proto_register(struct proto *prot, int alloc_slab)
2839 {
2840 if (alloc_slab) {
2841 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2842 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2843 NULL);
2844
2845 if (prot->slab == NULL) {
2846 pr_crit("%s: Can't create sock SLAB cache!\n",
2847 prot->name);
2848 goto out;
2849 }
2850
2851 if (req_prot_init(prot))
2852 goto out_free_request_sock_slab;
2853
2854 if (prot->twsk_prot != NULL) {
2855 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2856
2857 if (prot->twsk_prot->twsk_slab_name == NULL)
2858 goto out_free_request_sock_slab;
2859
2860 prot->twsk_prot->twsk_slab =
2861 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2862 prot->twsk_prot->twsk_obj_size,
2863 0,
2864 prot->slab_flags,
2865 NULL);
2866 if (prot->twsk_prot->twsk_slab == NULL)
2867 goto out_free_timewait_sock_slab_name;
2868 }
2869 }
2870
2871 mutex_lock(&proto_list_mutex);
2872 list_add(&prot->node, &proto_list);
2873 assign_proto_idx(prot);
2874 mutex_unlock(&proto_list_mutex);
2875 return 0;
2876
2877 out_free_timewait_sock_slab_name:
2878 kfree(prot->twsk_prot->twsk_slab_name);
2879 out_free_request_sock_slab:
2880 req_prot_cleanup(prot->rsk_prot);
2881
2882 kmem_cache_destroy(prot->slab);
2883 prot->slab = NULL;
2884 out:
2885 return -ENOBUFS;
2886 }
2887 EXPORT_SYMBOL(proto_register);
2888
2889 void proto_unregister(struct proto *prot)
2890 {
2891 mutex_lock(&proto_list_mutex);
2892 release_proto_idx(prot);
2893 list_del(&prot->node);
2894 mutex_unlock(&proto_list_mutex);
2895
2896 kmem_cache_destroy(prot->slab);
2897 prot->slab = NULL;
2898
2899 req_prot_cleanup(prot->rsk_prot);
2900
2901 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2902 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2903 kfree(prot->twsk_prot->twsk_slab_name);
2904 prot->twsk_prot->twsk_slab = NULL;
2905 }
2906 }
2907 EXPORT_SYMBOL(proto_unregister);
2908
2909 #ifdef CONFIG_PROC_FS
2910 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2911 __acquires(proto_list_mutex)
2912 {
2913 mutex_lock(&proto_list_mutex);
2914 return seq_list_start_head(&proto_list, *pos);
2915 }
2916
2917 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2918 {
2919 return seq_list_next(v, &proto_list, pos);
2920 }
2921
2922 static void proto_seq_stop(struct seq_file *seq, void *v)
2923 __releases(proto_list_mutex)
2924 {
2925 mutex_unlock(&proto_list_mutex);
2926 }
2927
2928 static char proto_method_implemented(const void *method)
2929 {
2930 return method == NULL ? 'n' : 'y';
2931 }
2932 static long sock_prot_memory_allocated(struct proto *proto)
2933 {
2934 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2935 }
2936
2937 static char *sock_prot_memory_pressure(struct proto *proto)
2938 {
2939 return proto->memory_pressure != NULL ?
2940 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2941 }
2942
2943 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2944 {
2945
2946 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2947 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2948 proto->name,
2949 proto->obj_size,
2950 sock_prot_inuse_get(seq_file_net(seq), proto),
2951 sock_prot_memory_allocated(proto),
2952 sock_prot_memory_pressure(proto),
2953 proto->max_header,
2954 proto->slab == NULL ? "no" : "yes",
2955 module_name(proto->owner),
2956 proto_method_implemented(proto->close),
2957 proto_method_implemented(proto->connect),
2958 proto_method_implemented(proto->disconnect),
2959 proto_method_implemented(proto->accept),
2960 proto_method_implemented(proto->ioctl),
2961 proto_method_implemented(proto->init),
2962 proto_method_implemented(proto->destroy),
2963 proto_method_implemented(proto->shutdown),
2964 proto_method_implemented(proto->setsockopt),
2965 proto_method_implemented(proto->getsockopt),
2966 proto_method_implemented(proto->sendmsg),
2967 proto_method_implemented(proto->recvmsg),
2968 proto_method_implemented(proto->sendpage),
2969 proto_method_implemented(proto->bind),
2970 proto_method_implemented(proto->backlog_rcv),
2971 proto_method_implemented(proto->hash),
2972 proto_method_implemented(proto->unhash),
2973 proto_method_implemented(proto->get_port),
2974 proto_method_implemented(proto->enter_memory_pressure));
2975 }
2976
2977 static int proto_seq_show(struct seq_file *seq, void *v)
2978 {
2979 if (v == &proto_list)
2980 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2981 "protocol",
2982 "size",
2983 "sockets",
2984 "memory",
2985 "press",
2986 "maxhdr",
2987 "slab",
2988 "module",
2989 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2990 else
2991 proto_seq_printf(seq, list_entry(v, struct proto, node));
2992 return 0;
2993 }
2994
2995 static const struct seq_operations proto_seq_ops = {
2996 .start = proto_seq_start,
2997 .next = proto_seq_next,
2998 .stop = proto_seq_stop,
2999 .show = proto_seq_show,
3000 };
3001
3002 static int proto_seq_open(struct inode *inode, struct file *file)
3003 {
3004 return seq_open_net(inode, file, &proto_seq_ops,
3005 sizeof(struct seq_net_private));
3006 }
3007
3008 static const struct file_operations proto_seq_fops = {
3009 .owner = THIS_MODULE,
3010 .open = proto_seq_open,
3011 .read = seq_read,
3012 .llseek = seq_lseek,
3013 .release = seq_release_net,
3014 };
3015
3016 static __net_init int proto_init_net(struct net *net)
3017 {
3018 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3019 return -ENOMEM;
3020
3021 return 0;
3022 }
3023
3024 static __net_exit void proto_exit_net(struct net *net)
3025 {
3026 remove_proc_entry("protocols", net->proc_net);
3027 }
3028
3029
3030 static __net_initdata struct pernet_operations proto_net_ops = {
3031 .init = proto_init_net,
3032 .exit = proto_exit_net,
3033 };
3034
3035 static int __init proto_init(void)
3036 {
3037 return register_pernet_subsys(&proto_net_ops);
3038 }
3039
3040 subsys_initcall(proto_init);
3041
3042 #endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.173874 seconds and 6 git commands to generate.