?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[cris-mirror.git] / net / core / sock.c
blobc501499a04fe973e80e18655b306d762d348ff44
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/sched/mm.h>
106 #include <linux/timer.h>
107 #include <linux/string.h>
108 #include <linux/sockios.h>
109 #include <linux/net.h>
110 #include <linux/mm.h>
111 #include <linux/slab.h>
112 #include <linux/interrupt.h>
113 #include <linux/poll.h>
114 #include <linux/tcp.h>
115 #include <linux/init.h>
116 #include <linux/highmem.h>
117 #include <linux/user_namespace.h>
118 #include <linux/static_key.h>
119 #include <linux/memcontrol.h>
120 #include <linux/prefetch.h>
121
122 #include <linux/uaccess.h>
123
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <net/net_namespace.h>
128 #include <net/request_sock.h>
129 #include <net/sock.h>
130 #include <linux/net_tstamp.h>
131 #include <net/xfrm.h>
132 #include <linux/ipsec.h>
133 #include <net/cls_cgroup.h>
134 #include <net/netprio_cgroup.h>
135 #include <linux/sock_diag.h>
136
137 #include <linux/filter.h>
138 #include <net/sock_reuseport.h>
139
140 #include <trace/events/sock.h>
141
142 #include <net/tcp.h>
143 #include <net/busy_poll.h>
144
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
147
148 static void sock_inuse_add(struct net *net, int val);
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 and separate keys for internal and
201 * userspace sockets.
202 */
203 static struct lock_class_key af_family_keys[AF_MAX];
204 static struct lock_class_key af_family_kern_keys[AF_MAX];
205 static struct lock_class_key af_family_slock_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
207
208 /*
209 * Make lock validator output more readable. (we pre-construct these
210 * strings build-time, so that runtime initialization of socket
211 * locks is fast):
212 */
213
214 #define _sock_locks(x) \
215 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
216 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
217 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
218 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
219 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
220 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
221 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
222 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
223 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
224 x "27" , x "28" , x "AF_CAN" , \
225 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
226 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
227 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
228 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
229 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX"
230
231 static const char *const af_family_key_strings[AF_MAX+1] = {
232 _sock_locks("sk_lock-")
233 };
234 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
235 _sock_locks("slock-")
236 };
237 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
238 _sock_locks("clock-")
239 };
240
241 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
242 _sock_locks("k-sk_lock-")
243 };
244 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
245 _sock_locks("k-slock-")
246 };
247 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
248 _sock_locks("k-clock-")
249 };
250 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
251 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" ,
252 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK",
253 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" ,
254 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" ,
255 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" ,
256 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" ,
257 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" ,
258 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" ,
259 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" ,
260 "rlock-27" , "rlock-28" , "rlock-AF_CAN" ,
261 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" ,
262 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" ,
263 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" ,
264 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" ,
265 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX"
266 };
267 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
268 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" ,
269 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK",
270 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" ,
271 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" ,
272 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" ,
273 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" ,
274 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" ,
275 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" ,
276 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" ,
277 "wlock-27" , "wlock-28" , "wlock-AF_CAN" ,
278 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" ,
279 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" ,
280 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" ,
281 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" ,
282 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX"
283 };
284 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
285 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" ,
286 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK",
287 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" ,
288 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" ,
289 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" ,
290 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" ,
291 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" ,
292 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" ,
293 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" ,
294 "elock-27" , "elock-28" , "elock-AF_CAN" ,
295 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" ,
296 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" ,
297 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" ,
298 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" ,
299 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX"
300 };
301
302 /*
303 * sk_callback_lock and sk queues locking rules are per-address-family,
304 * so split the lock classes by using a per-AF key:
305 */
306 static struct lock_class_key af_callback_keys[AF_MAX];
307 static struct lock_class_key af_rlock_keys[AF_MAX];
308 static struct lock_class_key af_wlock_keys[AF_MAX];
309 static struct lock_class_key af_elock_keys[AF_MAX];
310 static struct lock_class_key af_kern_callback_keys[AF_MAX];
311
312 /* Run time adjustable parameters. */
313 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
314 EXPORT_SYMBOL(sysctl_wmem_max);
315 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
316 EXPORT_SYMBOL(sysctl_rmem_max);
317 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
318 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
319
320 /* Maximal space eaten by iovec or ancillary data plus some space */
321 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
322 EXPORT_SYMBOL(sysctl_optmem_max);
323
324 int sysctl_tstamp_allow_data __read_mostly = 1;
325
326 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
327 EXPORT_SYMBOL_GPL(memalloc_socks);
328
329 /**
330 * sk_set_memalloc - sets %SOCK_MEMALLOC
331 * @sk: socket to set it on
332 *
333 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
334 * It's the responsibility of the admin to adjust min_free_kbytes
335 * to meet the requirements
336 */
337 void sk_set_memalloc(struct sock *sk)
338 {
339 sock_set_flag(sk, SOCK_MEMALLOC);
340 sk->sk_allocation |= __GFP_MEMALLOC;
341 static_key_slow_inc(&memalloc_socks);
342 }
343 EXPORT_SYMBOL_GPL(sk_set_memalloc);
344
345 void sk_clear_memalloc(struct sock *sk)
346 {
347 sock_reset_flag(sk, SOCK_MEMALLOC);
348 sk->sk_allocation &= ~__GFP_MEMALLOC;
349 static_key_slow_dec(&memalloc_socks);
350
351 /*
352 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
353 * progress of swapping. SOCK_MEMALLOC may be cleared while
354 * it has rmem allocations due to the last swapfile being deactivated
355 * but there is a risk that the socket is unusable due to exceeding
356 * the rmem limits. Reclaim the reserves and obey rmem limits again.
357 */
358 sk_mem_reclaim(sk);
359 }
360 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
361
362 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
363 {
364 int ret;
365 unsigned int noreclaim_flag;
366
367 /* these should have been dropped before queueing */
368 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
369
370 noreclaim_flag = memalloc_noreclaim_save();
371 ret = sk->sk_backlog_rcv(sk, skb);
372 memalloc_noreclaim_restore(noreclaim_flag);
373
374 return ret;
375 }
376 EXPORT_SYMBOL(__sk_backlog_rcv);
377
378 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
379 {
380 struct timeval tv;
381
382 if (optlen < sizeof(tv))
383 return -EINVAL;
384 if (copy_from_user(&tv, optval, sizeof(tv)))
385 return -EFAULT;
386 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
387 return -EDOM;
388
389 if (tv.tv_sec < 0) {
390 static int warned __read_mostly;
391
392 *timeo_p = 0;
393 if (warned < 10 && net_ratelimit()) {
394 warned++;
395 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
396 __func__, current->comm, task_pid_nr(current));
397 }
398 return 0;
399 }
400 *timeo_p = MAX_SCHEDULE_TIMEOUT;
401 if (tv.tv_sec == 0 && tv.tv_usec == 0)
402 return 0;
403 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
404 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
405 return 0;
406 }
407
408 static void sock_warn_obsolete_bsdism(const char *name)
409 {
410 static int warned;
411 static char warncomm[TASK_COMM_LEN];
412 if (strcmp(warncomm, current->comm) && warned < 5) {
413 strcpy(warncomm, current->comm);
414 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
415 warncomm, name);
416 warned++;
417 }
418 }
419
420 static bool sock_needs_netstamp(const struct sock *sk)
421 {
422 switch (sk->sk_family) {
423 case AF_UNSPEC:
424 case AF_UNIX:
425 return false;
426 default:
427 return true;
428 }
429 }
430
431 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
432 {
433 if (sk->sk_flags & flags) {
434 sk->sk_flags &= ~flags;
435 if (sock_needs_netstamp(sk) &&
436 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
437 net_disable_timestamp();
438 }
439 }
440
441
442 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
443 {
444 unsigned long flags;
445 struct sk_buff_head *list = &sk->sk_receive_queue;
446
447 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
448 atomic_inc(&sk->sk_drops);
449 trace_sock_rcvqueue_full(sk, skb);
450 return -ENOMEM;
451 }
452
453 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
454 atomic_inc(&sk->sk_drops);
455 return -ENOBUFS;
456 }
457
458 skb->dev = NULL;
459 skb_set_owner_r(skb, sk);
460
461 /* we escape from rcu protected region, make sure we dont leak
462 * a norefcounted dst
463 */
464 skb_dst_force(skb);
465
466 spin_lock_irqsave(&list->lock, flags);
467 sock_skb_set_dropcount(sk, skb);
468 __skb_queue_tail(list, skb);
469 spin_unlock_irqrestore(&list->lock, flags);
470
471 if (!sock_flag(sk, SOCK_DEAD))
472 sk->sk_data_ready(sk);
473 return 0;
474 }
475 EXPORT_SYMBOL(__sock_queue_rcv_skb);
476
477 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
478 {
479 int err;
480
481 err = sk_filter(sk, skb);
482 if (err)
483 return err;
484
485 return __sock_queue_rcv_skb(sk, skb);
486 }
487 EXPORT_SYMBOL(sock_queue_rcv_skb);
488
489 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
490 const int nested, unsigned int trim_cap, bool refcounted)
491 {
492 int rc = NET_RX_SUCCESS;
493
494 if (sk_filter_trim_cap(sk, skb, trim_cap))
495 goto discard_and_relse;
496
497 skb->dev = NULL;
498
499 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
500 atomic_inc(&sk->sk_drops);
501 goto discard_and_relse;
502 }
503 if (nested)
504 bh_lock_sock_nested(sk);
505 else
506 bh_lock_sock(sk);
507 if (!sock_owned_by_user(sk)) {
508 /*
509 * trylock + unlock semantics:
510 */
511 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
512
513 rc = sk_backlog_rcv(sk, skb);
514
515 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
516 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
517 bh_unlock_sock(sk);
518 atomic_inc(&sk->sk_drops);
519 goto discard_and_relse;
520 }
521
522 bh_unlock_sock(sk);
523 out:
524 if (refcounted)
525 sock_put(sk);
526 return rc;
527 discard_and_relse:
528 kfree_skb(skb);
529 goto out;
530 }
531 EXPORT_SYMBOL(__sk_receive_skb);
532
533 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
534 {
535 struct dst_entry *dst = __sk_dst_get(sk);
536
537 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
538 sk_tx_queue_clear(sk);
539 sk->sk_dst_pending_confirm = 0;
540 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
541 dst_release(dst);
542 return NULL;
543 }
544
545 return dst;
546 }
547 EXPORT_SYMBOL(__sk_dst_check);
548
549 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
550 {
551 struct dst_entry *dst = sk_dst_get(sk);
552
553 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
554 sk_dst_reset(sk);
555 dst_release(dst);
556 return NULL;
557 }
558
559 return dst;
560 }
561 EXPORT_SYMBOL(sk_dst_check);
562
563 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
564 int optlen)
565 {
566 int ret = -ENOPROTOOPT;
567 #ifdef CONFIG_NETDEVICES
568 struct net *net = sock_net(sk);
569 char devname[IFNAMSIZ];
570 int index;
571
572 /* Sorry... */
573 ret = -EPERM;
574 if (!ns_capable(net->user_ns, CAP_NET_RAW))
575 goto out;
576
577 ret = -EINVAL;
578 if (optlen < 0)
579 goto out;
580
581 /* Bind this socket to a particular device like "eth0",
582 * as specified in the passed interface name. If the
583 * name is "" or the option length is zero the socket
584 * is not bound.
585 */
586 if (optlen > IFNAMSIZ - 1)
587 optlen = IFNAMSIZ - 1;
588 memset(devname, 0, sizeof(devname));
589
590 ret = -EFAULT;
591 if (copy_from_user(devname, optval, optlen))
592 goto out;
593
594 index = 0;
595 if (devname[0] != '\0') {
596 struct net_device *dev;
597
598 rcu_read_lock();
599 dev = dev_get_by_name_rcu(net, devname);
600 if (dev)
601 index = dev->ifindex;
602 rcu_read_unlock();
603 ret = -ENODEV;
604 if (!dev)
605 goto out;
606 }
607
608 lock_sock(sk);
609 sk->sk_bound_dev_if = index;
610 sk_dst_reset(sk);
611 release_sock(sk);
612
613 ret = 0;
614
615 out:
616 #endif
617
618 return ret;
619 }
620
621 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
622 int __user *optlen, int len)
623 {
624 int ret = -ENOPROTOOPT;
625 #ifdef CONFIG_NETDEVICES
626 struct net *net = sock_net(sk);
627 char devname[IFNAMSIZ];
628
629 if (sk->sk_bound_dev_if == 0) {
630 len = 0;
631 goto zero;
632 }
633
634 ret = -EINVAL;
635 if (len < IFNAMSIZ)
636 goto out;
637
638 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
639 if (ret)
640 goto out;
641
642 len = strlen(devname) + 1;
643
644 ret = -EFAULT;
645 if (copy_to_user(optval, devname, len))
646 goto out;
647
648 zero:
649 ret = -EFAULT;
650 if (put_user(len, optlen))
651 goto out;
652
653 ret = 0;
654
655 out:
656 #endif
657
658 return ret;
659 }
660
661 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
662 {
663 if (valbool)
664 sock_set_flag(sk, bit);
665 else
666 sock_reset_flag(sk, bit);
667 }
668
669 bool sk_mc_loop(struct sock *sk)
670 {
671 if (dev_recursion_level())
672 return false;
673 if (!sk)
674 return true;
675 switch (sk->sk_family) {
676 case AF_INET:
677 return inet_sk(sk)->mc_loop;
678 #if IS_ENABLED(CONFIG_IPV6)
679 case AF_INET6:
680 return inet6_sk(sk)->mc_loop;
681 #endif
682 }
683 WARN_ON(1);
684 return true;
685 }
686 EXPORT_SYMBOL(sk_mc_loop);
687
688 /*
689 * This is meant for all protocols to use and covers goings on
690 * at the socket level. Everything here is generic.
691 */
692
693 int sock_setsockopt(struct socket *sock, int level, int optname,
694 char __user *optval, unsigned int optlen)
695 {
696 struct sock *sk = sock->sk;
697 int val;
698 int valbool;
699 struct linger ling;
700 int ret = 0;
701
702 /*
703 * Options without arguments
704 */
705
706 if (optname == SO_BINDTODEVICE)
707 return sock_setbindtodevice(sk, optval, optlen);
708
709 if (optlen < sizeof(int))
710 return -EINVAL;
711
712 if (get_user(val, (int __user *)optval))
713 return -EFAULT;
714
715 valbool = val ? 1 : 0;
716
717 lock_sock(sk);
718
719 switch (optname) {
720 case SO_DEBUG:
721 if (val && !capable(CAP_NET_ADMIN))
722 ret = -EACCES;
723 else
724 sock_valbool_flag(sk, SOCK_DBG, valbool);
725 break;
726 case SO_REUSEADDR:
727 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
728 break;
729 case SO_REUSEPORT:
730 sk->sk_reuseport = valbool;
731 break;
732 case SO_TYPE:
733 case SO_PROTOCOL:
734 case SO_DOMAIN:
735 case SO_ERROR:
736 ret = -ENOPROTOOPT;
737 break;
738 case SO_DONTROUTE:
739 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
740 break;
741 case SO_BROADCAST:
742 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
743 break;
744 case SO_SNDBUF:
745 /* Don't error on this BSD doesn't and if you think
746 * about it this is right. Otherwise apps have to
747 * play 'guess the biggest size' games. RCVBUF/SNDBUF
748 * are treated in BSD as hints
749 */
750 val = min_t(u32, val, sysctl_wmem_max);
751 set_sndbuf:
752 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
753 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
754 /* Wake up sending tasks if we upped the value. */
755 sk->sk_write_space(sk);
756 break;
757
758 case SO_SNDBUFFORCE:
759 if (!capable(CAP_NET_ADMIN)) {
760 ret = -EPERM;
761 break;
762 }
763 goto set_sndbuf;
764
765 case SO_RCVBUF:
766 /* Don't error on this BSD doesn't and if you think
767 * about it this is right. Otherwise apps have to
768 * play 'guess the biggest size' games. RCVBUF/SNDBUF
769 * are treated in BSD as hints
770 */
771 val = min_t(u32, val, sysctl_rmem_max);
772 set_rcvbuf:
773 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
774 /*
775 * We double it on the way in to account for
776 * "struct sk_buff" etc. overhead. Applications
777 * assume that the SO_RCVBUF setting they make will
778 * allow that much actual data to be received on that
779 * socket.
780 *
781 * Applications are unaware that "struct sk_buff" and
782 * other overheads allocate from the receive buffer
783 * during socket buffer allocation.
784 *
785 * And after considering the possible alternatives,
786 * returning the value we actually used in getsockopt
787 * is the most desirable behavior.
788 */
789 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
790 break;
791
792 case SO_RCVBUFFORCE:
793 if (!capable(CAP_NET_ADMIN)) {
794 ret = -EPERM;
795 break;
796 }
797 goto set_rcvbuf;
798
799 case SO_KEEPALIVE:
800 if (sk->sk_prot->keepalive)
801 sk->sk_prot->keepalive(sk, valbool);
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 sk->sk_type == SOCK_STREAM) {
879 if ((1 << sk->sk_state) &
880 (TCPF_CLOSE | TCPF_LISTEN)) {
881 ret = -EINVAL;
882 break;
883 }
884 sk->sk_tskey = tcp_sk(sk)->snd_una;
885 } else {
886 sk->sk_tskey = 0;
887 }
888 }
889
890 if (val & SOF_TIMESTAMPING_OPT_STATS &&
891 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
892 ret = -EINVAL;
893 break;
894 }
895
896 sk->sk_tsflags = val;
897 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
898 sock_enable_timestamp(sk,
899 SOCK_TIMESTAMPING_RX_SOFTWARE);
900 else
901 sock_disable_timestamp(sk,
902 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
903 break;
904
905 case SO_RCVLOWAT:
906 if (val < 0)
907 val = INT_MAX;
908 sk->sk_rcvlowat = val ? : 1;
909 break;
910
911 case SO_RCVTIMEO:
912 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
913 break;
914
915 case SO_SNDTIMEO:
916 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
917 break;
918
919 case SO_ATTACH_FILTER:
920 ret = -EINVAL;
921 if (optlen == sizeof(struct sock_fprog)) {
922 struct sock_fprog fprog;
923
924 ret = -EFAULT;
925 if (copy_from_user(&fprog, optval, sizeof(fprog)))
926 break;
927
928 ret = sk_attach_filter(&fprog, sk);
929 }
930 break;
931
932 case SO_ATTACH_BPF:
933 ret = -EINVAL;
934 if (optlen == sizeof(u32)) {
935 u32 ufd;
936
937 ret = -EFAULT;
938 if (copy_from_user(&ufd, optval, sizeof(ufd)))
939 break;
940
941 ret = sk_attach_bpf(ufd, sk);
942 }
943 break;
944
945 case SO_ATTACH_REUSEPORT_CBPF:
946 ret = -EINVAL;
947 if (optlen == sizeof(struct sock_fprog)) {
948 struct sock_fprog fprog;
949
950 ret = -EFAULT;
951 if (copy_from_user(&fprog, optval, sizeof(fprog)))
952 break;
953
954 ret = sk_reuseport_attach_filter(&fprog, sk);
955 }
956 break;
957
958 case SO_ATTACH_REUSEPORT_EBPF:
959 ret = -EINVAL;
960 if (optlen == sizeof(u32)) {
961 u32 ufd;
962
963 ret = -EFAULT;
964 if (copy_from_user(&ufd, optval, sizeof(ufd)))
965 break;
966
967 ret = sk_reuseport_attach_bpf(ufd, sk);
968 }
969 break;
970
971 case SO_DETACH_FILTER:
972 ret = sk_detach_filter(sk);
973 break;
974
975 case SO_LOCK_FILTER:
976 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
977 ret = -EPERM;
978 else
979 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
980 break;
981
982 case SO_PASSSEC:
983 if (valbool)
984 set_bit(SOCK_PASSSEC, &sock->flags);
985 else
986 clear_bit(SOCK_PASSSEC, &sock->flags);
987 break;
988 case SO_MARK:
989 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
990 ret = -EPERM;
991 else
992 sk->sk_mark = val;
993 break;
994
995 case SO_RXQ_OVFL:
996 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
997 break;
998
999 case SO_WIFI_STATUS:
1000 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1001 break;
1002
1003 case SO_PEEK_OFF:
1004 if (sock->ops->set_peek_off)
1005 ret = sock->ops->set_peek_off(sk, val);
1006 else
1007 ret = -EOPNOTSUPP;
1008 break;
1009
1010 case SO_NOFCS:
1011 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1012 break;
1013
1014 case SO_SELECT_ERR_QUEUE:
1015 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1016 break;
1017
1018 #ifdef CONFIG_NET_RX_BUSY_POLL
1019 case SO_BUSY_POLL:
1020 /* allow unprivileged users to decrease the value */
1021 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1022 ret = -EPERM;
1023 else {
1024 if (val < 0)
1025 ret = -EINVAL;
1026 else
1027 sk->sk_ll_usec = val;
1028 }
1029 break;
1030 #endif
1031
1032 case SO_MAX_PACING_RATE:
1033 if (val != ~0U)
1034 cmpxchg(&sk->sk_pacing_status,
1035 SK_PACING_NONE,
1036 SK_PACING_NEEDED);
1037 sk->sk_max_pacing_rate = val;
1038 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1039 sk->sk_max_pacing_rate);
1040 break;
1041
1042 case SO_INCOMING_CPU:
1043 sk->sk_incoming_cpu = val;
1044 break;
1045
1046 case SO_CNX_ADVICE:
1047 if (val == 1)
1048 dst_negative_advice(sk);
1049 break;
1050
1051 case SO_ZEROCOPY:
1052 if (sk->sk_family != PF_INET && sk->sk_family != PF_INET6)
1053 ret = -ENOTSUPP;
1054 else if (sk->sk_protocol != IPPROTO_TCP)
1055 ret = -ENOTSUPP;
1056 else if (sk->sk_state != TCP_CLOSE)
1057 ret = -EBUSY;
1058 else if (val < 0 || val > 1)
1059 ret = -EINVAL;
1060 else
1061 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1062 break;
1063
1064 default:
1065 ret = -ENOPROTOOPT;
1066 break;
1067 }
1068 release_sock(sk);
1069 return ret;
1070 }
1071 EXPORT_SYMBOL(sock_setsockopt);
1072
1073
1074 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1075 struct ucred *ucred)
1076 {
1077 ucred->pid = pid_vnr(pid);
1078 ucred->uid = ucred->gid = -1;
1079 if (cred) {
1080 struct user_namespace *current_ns = current_user_ns();
1081
1082 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1083 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1084 }
1085 }
1086
1087 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1088 {
1089 struct user_namespace *user_ns = current_user_ns();
1090 int i;
1091
1092 for (i = 0; i < src->ngroups; i++)
1093 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1094 return -EFAULT;
1095
1096 return 0;
1097 }
1098
1099 int sock_getsockopt(struct socket *sock, int level, int optname,
1100 char __user *optval, int __user *optlen)
1101 {
1102 struct sock *sk = sock->sk;
1103
1104 union {
1105 int val;
1106 u64 val64;
1107 struct linger ling;
1108 struct timeval tm;
1109 } v;
1110
1111 int lv = sizeof(int);
1112 int len;
1113
1114 if (get_user(len, optlen))
1115 return -EFAULT;
1116 if (len < 0)
1117 return -EINVAL;
1118
1119 memset(&v, 0, sizeof(v));
1120
1121 switch (optname) {
1122 case SO_DEBUG:
1123 v.val = sock_flag(sk, SOCK_DBG);
1124 break;
1125
1126 case SO_DONTROUTE:
1127 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1128 break;
1129
1130 case SO_BROADCAST:
1131 v.val = sock_flag(sk, SOCK_BROADCAST);
1132 break;
1133
1134 case SO_SNDBUF:
1135 v.val = sk->sk_sndbuf;
1136 break;
1137
1138 case SO_RCVBUF:
1139 v.val = sk->sk_rcvbuf;
1140 break;
1141
1142 case SO_REUSEADDR:
1143 v.val = sk->sk_reuse;
1144 break;
1145
1146 case SO_REUSEPORT:
1147 v.val = sk->sk_reuseport;
1148 break;
1149
1150 case SO_KEEPALIVE:
1151 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1152 break;
1153
1154 case SO_TYPE:
1155 v.val = sk->sk_type;
1156 break;
1157
1158 case SO_PROTOCOL:
1159 v.val = sk->sk_protocol;
1160 break;
1161
1162 case SO_DOMAIN:
1163 v.val = sk->sk_family;
1164 break;
1165
1166 case SO_ERROR:
1167 v.val = -sock_error(sk);
1168 if (v.val == 0)
1169 v.val = xchg(&sk->sk_err_soft, 0);
1170 break;
1171
1172 case SO_OOBINLINE:
1173 v.val = sock_flag(sk, SOCK_URGINLINE);
1174 break;
1175
1176 case SO_NO_CHECK:
1177 v.val = sk->sk_no_check_tx;
1178 break;
1179
1180 case SO_PRIORITY:
1181 v.val = sk->sk_priority;
1182 break;
1183
1184 case SO_LINGER:
1185 lv = sizeof(v.ling);
1186 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1187 v.ling.l_linger = sk->sk_lingertime / HZ;
1188 break;
1189
1190 case SO_BSDCOMPAT:
1191 sock_warn_obsolete_bsdism("getsockopt");
1192 break;
1193
1194 case SO_TIMESTAMP:
1195 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1196 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1197 break;
1198
1199 case SO_TIMESTAMPNS:
1200 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1201 break;
1202
1203 case SO_TIMESTAMPING:
1204 v.val = sk->sk_tsflags;
1205 break;
1206
1207 case SO_RCVTIMEO:
1208 lv = sizeof(struct timeval);
1209 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1210 v.tm.tv_sec = 0;
1211 v.tm.tv_usec = 0;
1212 } else {
1213 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1214 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1215 }
1216 break;
1217
1218 case SO_SNDTIMEO:
1219 lv = sizeof(struct timeval);
1220 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1221 v.tm.tv_sec = 0;
1222 v.tm.tv_usec = 0;
1223 } else {
1224 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1225 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1226 }
1227 break;
1228
1229 case SO_RCVLOWAT:
1230 v.val = sk->sk_rcvlowat;
1231 break;
1232
1233 case SO_SNDLOWAT:
1234 v.val = 1;
1235 break;
1236
1237 case SO_PASSCRED:
1238 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1239 break;
1240
1241 case SO_PEERCRED:
1242 {
1243 struct ucred peercred;
1244 if (len > sizeof(peercred))
1245 len = sizeof(peercred);
1246 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1247 if (copy_to_user(optval, &peercred, len))
1248 return -EFAULT;
1249 goto lenout;
1250 }
1251
1252 case SO_PEERGROUPS:
1253 {
1254 int ret, n;
1255
1256 if (!sk->sk_peer_cred)
1257 return -ENODATA;
1258
1259 n = sk->sk_peer_cred->group_info->ngroups;
1260 if (len < n * sizeof(gid_t)) {
1261 len = n * sizeof(gid_t);
1262 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1263 }
1264 len = n * sizeof(gid_t);
1265
1266 ret = groups_to_user((gid_t __user *)optval,
1267 sk->sk_peer_cred->group_info);
1268 if (ret)
1269 return ret;
1270 goto lenout;
1271 }
1272
1273 case SO_PEERNAME:
1274 {
1275 char address[128];
1276
1277 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1278 return -ENOTCONN;
1279 if (lv < len)
1280 return -EINVAL;
1281 if (copy_to_user(optval, address, len))
1282 return -EFAULT;
1283 goto lenout;
1284 }
1285
1286 /* Dubious BSD thing... Probably nobody even uses it, but
1287 * the UNIX standard wants it for whatever reason... -DaveM
1288 */
1289 case SO_ACCEPTCONN:
1290 v.val = sk->sk_state == TCP_LISTEN;
1291 break;
1292
1293 case SO_PASSSEC:
1294 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1295 break;
1296
1297 case SO_PEERSEC:
1298 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1299
1300 case SO_MARK:
1301 v.val = sk->sk_mark;
1302 break;
1303
1304 case SO_RXQ_OVFL:
1305 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1306 break;
1307
1308 case SO_WIFI_STATUS:
1309 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1310 break;
1311
1312 case SO_PEEK_OFF:
1313 if (!sock->ops->set_peek_off)
1314 return -EOPNOTSUPP;
1315
1316 v.val = sk->sk_peek_off;
1317 break;
1318 case SO_NOFCS:
1319 v.val = sock_flag(sk, SOCK_NOFCS);
1320 break;
1321
1322 case SO_BINDTODEVICE:
1323 return sock_getbindtodevice(sk, optval, optlen, len);
1324
1325 case SO_GET_FILTER:
1326 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1327 if (len < 0)
1328 return len;
1329
1330 goto lenout;
1331
1332 case SO_LOCK_FILTER:
1333 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1334 break;
1335
1336 case SO_BPF_EXTENSIONS:
1337 v.val = bpf_tell_extensions();
1338 break;
1339
1340 case SO_SELECT_ERR_QUEUE:
1341 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1342 break;
1343
1344 #ifdef CONFIG_NET_RX_BUSY_POLL
1345 case SO_BUSY_POLL:
1346 v.val = sk->sk_ll_usec;
1347 break;
1348 #endif
1349
1350 case SO_MAX_PACING_RATE:
1351 v.val = sk->sk_max_pacing_rate;
1352 break;
1353
1354 case SO_INCOMING_CPU:
1355 v.val = sk->sk_incoming_cpu;
1356 break;
1357
1358 case SO_MEMINFO:
1359 {
1360 u32 meminfo[SK_MEMINFO_VARS];
1361
1362 if (get_user(len, optlen))
1363 return -EFAULT;
1364
1365 sk_get_meminfo(sk, meminfo);
1366
1367 len = min_t(unsigned int, len, sizeof(meminfo));
1368 if (copy_to_user(optval, &meminfo, len))
1369 return -EFAULT;
1370
1371 goto lenout;
1372 }
1373
1374 #ifdef CONFIG_NET_RX_BUSY_POLL
1375 case SO_INCOMING_NAPI_ID:
1376 v.val = READ_ONCE(sk->sk_napi_id);
1377
1378 /* aggregate non-NAPI IDs down to 0 */
1379 if (v.val < MIN_NAPI_ID)
1380 v.val = 0;
1381
1382 break;
1383 #endif
1384
1385 case SO_COOKIE:
1386 lv = sizeof(u64);
1387 if (len < lv)
1388 return -EINVAL;
1389 v.val64 = sock_gen_cookie(sk);
1390 break;
1391
1392 case SO_ZEROCOPY:
1393 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1394 break;
1395
1396 default:
1397 /* We implement the SO_SNDLOWAT etc to not be settable
1398 * (1003.1g 7).
1399 */
1400 return -ENOPROTOOPT;
1401 }
1402
1403 if (len > lv)
1404 len = lv;
1405 if (copy_to_user(optval, &v, len))
1406 return -EFAULT;
1407 lenout:
1408 if (put_user(len, optlen))
1409 return -EFAULT;
1410 return 0;
1411 }
1412
1413 /*
1414 * Initialize an sk_lock.
1415 *
1416 * (We also register the sk_lock with the lock validator.)
1417 */
1418 static inline void sock_lock_init(struct sock *sk)
1419 {
1420 if (sk->sk_kern_sock)
1421 sock_lock_init_class_and_name(
1422 sk,
1423 af_family_kern_slock_key_strings[sk->sk_family],
1424 af_family_kern_slock_keys + sk->sk_family,
1425 af_family_kern_key_strings[sk->sk_family],
1426 af_family_kern_keys + sk->sk_family);
1427 else
1428 sock_lock_init_class_and_name(
1429 sk,
1430 af_family_slock_key_strings[sk->sk_family],
1431 af_family_slock_keys + sk->sk_family,
1432 af_family_key_strings[sk->sk_family],
1433 af_family_keys + sk->sk_family);
1434 }
1435
1436 /*
1437 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1438 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1439 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1440 */
1441 static void sock_copy(struct sock *nsk, const struct sock *osk)
1442 {
1443 #ifdef CONFIG_SECURITY_NETWORK
1444 void *sptr = nsk->sk_security;
1445 #endif
1446 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1447
1448 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1449 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1450
1451 #ifdef CONFIG_SECURITY_NETWORK
1452 nsk->sk_security = sptr;
1453 security_sk_clone(osk, nsk);
1454 #endif
1455 }
1456
1457 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1458 int family)
1459 {
1460 struct sock *sk;
1461 struct kmem_cache *slab;
1462
1463 slab = prot->slab;
1464 if (slab != NULL) {
1465 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1466 if (!sk)
1467 return sk;
1468 if (priority & __GFP_ZERO)
1469 sk_prot_clear_nulls(sk, prot->obj_size);
1470 } else
1471 sk = kmalloc(prot->obj_size, priority);
1472
1473 if (sk != NULL) {
1474 if (security_sk_alloc(sk, family, priority))
1475 goto out_free;
1476
1477 if (!try_module_get(prot->owner))
1478 goto out_free_sec;
1479 sk_tx_queue_clear(sk);
1480 }
1481
1482 return sk;
1483
1484 out_free_sec:
1485 security_sk_free(sk);
1486 out_free:
1487 if (slab != NULL)
1488 kmem_cache_free(slab, sk);
1489 else
1490 kfree(sk);
1491 return NULL;
1492 }
1493
1494 static void sk_prot_free(struct proto *prot, struct sock *sk)
1495 {
1496 struct kmem_cache *slab;
1497 struct module *owner;
1498
1499 owner = prot->owner;
1500 slab = prot->slab;
1501
1502 cgroup_sk_free(&sk->sk_cgrp_data);
1503 mem_cgroup_sk_free(sk);
1504 security_sk_free(sk);
1505 if (slab != NULL)
1506 kmem_cache_free(slab, sk);
1507 else
1508 kfree(sk);
1509 module_put(owner);
1510 }
1511
1512 /**
1513 * sk_alloc - All socket objects are allocated here
1514 * @net: the applicable net namespace
1515 * @family: protocol family
1516 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1517 * @prot: struct proto associated with this new sock instance
1518 * @kern: is this to be a kernel socket?
1519 */
1520 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1521 struct proto *prot, int kern)
1522 {
1523 struct sock *sk;
1524
1525 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1526 if (sk) {
1527 sk->sk_family = family;
1528 /*
1529 * See comment in struct sock definition to understand
1530 * why we need sk_prot_creator -acme
1531 */
1532 sk->sk_prot = sk->sk_prot_creator = prot;
1533 sk->sk_kern_sock = kern;
1534 sock_lock_init(sk);
1535 sk->sk_net_refcnt = kern ? 0 : 1;
1536 if (likely(sk->sk_net_refcnt)) {
1537 get_net(net);
1538 sock_inuse_add(net, 1);
1539 }
1540
1541 sock_net_set(sk, net);
1542 refcount_set(&sk->sk_wmem_alloc, 1);
1543
1544 mem_cgroup_sk_alloc(sk);
1545 cgroup_sk_alloc(&sk->sk_cgrp_data);
1546 sock_update_classid(&sk->sk_cgrp_data);
1547 sock_update_netprioidx(&sk->sk_cgrp_data);
1548 }
1549
1550 return sk;
1551 }
1552 EXPORT_SYMBOL(sk_alloc);
1553
1554 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1555 * grace period. This is the case for UDP sockets and TCP listeners.
1556 */
1557 static void __sk_destruct(struct rcu_head *head)
1558 {
1559 struct sock *sk = container_of(head, struct sock, sk_rcu);
1560 struct sk_filter *filter;
1561
1562 if (sk->sk_destruct)
1563 sk->sk_destruct(sk);
1564
1565 filter = rcu_dereference_check(sk->sk_filter,
1566 refcount_read(&sk->sk_wmem_alloc) == 0);
1567 if (filter) {
1568 sk_filter_uncharge(sk, filter);
1569 RCU_INIT_POINTER(sk->sk_filter, NULL);
1570 }
1571 if (rcu_access_pointer(sk->sk_reuseport_cb))
1572 reuseport_detach_sock(sk);
1573
1574 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1575
1576 if (atomic_read(&sk->sk_omem_alloc))
1577 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1578 __func__, atomic_read(&sk->sk_omem_alloc));
1579
1580 if (sk->sk_frag.page) {
1581 put_page(sk->sk_frag.page);
1582 sk->sk_frag.page = NULL;
1583 }
1584
1585 if (sk->sk_peer_cred)
1586 put_cred(sk->sk_peer_cred);
1587 put_pid(sk->sk_peer_pid);
1588 if (likely(sk->sk_net_refcnt))
1589 put_net(sock_net(sk));
1590 sk_prot_free(sk->sk_prot_creator, sk);
1591 }
1592
1593 void sk_destruct(struct sock *sk)
1594 {
1595 if (sock_flag(sk, SOCK_RCU_FREE))
1596 call_rcu(&sk->sk_rcu, __sk_destruct);
1597 else
1598 __sk_destruct(&sk->sk_rcu);
1599 }
1600
1601 static void __sk_free(struct sock *sk)
1602 {
1603 if (likely(sk->sk_net_refcnt))
1604 sock_inuse_add(sock_net(sk), -1);
1605
1606 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1607 sock_diag_broadcast_destroy(sk);
1608 else
1609 sk_destruct(sk);
1610 }
1611
1612 void sk_free(struct sock *sk)
1613 {
1614 /*
1615 * We subtract one from sk_wmem_alloc and can know if
1616 * some packets are still in some tx queue.
1617 * If not null, sock_wfree() will call __sk_free(sk) later
1618 */
1619 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1620 __sk_free(sk);
1621 }
1622 EXPORT_SYMBOL(sk_free);
1623
1624 static void sk_init_common(struct sock *sk)
1625 {
1626 skb_queue_head_init(&sk->sk_receive_queue);
1627 skb_queue_head_init(&sk->sk_write_queue);
1628 skb_queue_head_init(&sk->sk_error_queue);
1629
1630 rwlock_init(&sk->sk_callback_lock);
1631 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1632 af_rlock_keys + sk->sk_family,
1633 af_family_rlock_key_strings[sk->sk_family]);
1634 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1635 af_wlock_keys + sk->sk_family,
1636 af_family_wlock_key_strings[sk->sk_family]);
1637 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1638 af_elock_keys + sk->sk_family,
1639 af_family_elock_key_strings[sk->sk_family]);
1640 lockdep_set_class_and_name(&sk->sk_callback_lock,
1641 af_callback_keys + sk->sk_family,
1642 af_family_clock_key_strings[sk->sk_family]);
1643 }
1644
1645 /**
1646 * sk_clone_lock - clone a socket, and lock its clone
1647 * @sk: the socket to clone
1648 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1649 *
1650 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1651 */
1652 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1653 {
1654 struct sock *newsk;
1655 bool is_charged = true;
1656
1657 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1658 if (newsk != NULL) {
1659 struct sk_filter *filter;
1660
1661 sock_copy(newsk, sk);
1662
1663 newsk->sk_prot_creator = sk->sk_prot;
1664
1665 /* SANITY */
1666 if (likely(newsk->sk_net_refcnt))
1667 get_net(sock_net(newsk));
1668 sk_node_init(&newsk->sk_node);
1669 sock_lock_init(newsk);
1670 bh_lock_sock(newsk);
1671 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1672 newsk->sk_backlog.len = 0;
1673
1674 atomic_set(&newsk->sk_rmem_alloc, 0);
1675 /*
1676 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1677 */
1678 refcount_set(&newsk->sk_wmem_alloc, 1);
1679 atomic_set(&newsk->sk_omem_alloc, 0);
1680 sk_init_common(newsk);
1681
1682 newsk->sk_dst_cache = NULL;
1683 newsk->sk_dst_pending_confirm = 0;
1684 newsk->sk_wmem_queued = 0;
1685 newsk->sk_forward_alloc = 0;
1686 atomic_set(&newsk->sk_drops, 0);
1687 newsk->sk_send_head = NULL;
1688 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1689 atomic_set(&newsk->sk_zckey, 0);
1690
1691 sock_reset_flag(newsk, SOCK_DONE);
1692 mem_cgroup_sk_alloc(newsk);
1693 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1694
1695 rcu_read_lock();
1696 filter = rcu_dereference(sk->sk_filter);
1697 if (filter != NULL)
1698 /* though it's an empty new sock, the charging may fail
1699 * if sysctl_optmem_max was changed between creation of
1700 * original socket and cloning
1701 */
1702 is_charged = sk_filter_charge(newsk, filter);
1703 RCU_INIT_POINTER(newsk->sk_filter, filter);
1704 rcu_read_unlock();
1705
1706 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1707 /* We need to make sure that we don't uncharge the new
1708 * socket if we couldn't charge it in the first place
1709 * as otherwise we uncharge the parent's filter.
1710 */
1711 if (!is_charged)
1712 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1713 sk_free_unlock_clone(newsk);
1714 newsk = NULL;
1715 goto out;
1716 }
1717 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1718
1719 newsk->sk_err = 0;
1720 newsk->sk_err_soft = 0;
1721 newsk->sk_priority = 0;
1722 newsk->sk_incoming_cpu = raw_smp_processor_id();
1723 atomic64_set(&newsk->sk_cookie, 0);
1724 if (likely(newsk->sk_net_refcnt))
1725 sock_inuse_add(sock_net(newsk), 1);
1726
1727 /*
1728 * Before updating sk_refcnt, we must commit prior changes to memory
1729 * (Documentation/RCU/rculist_nulls.txt for details)
1730 */
1731 smp_wmb();
1732 refcount_set(&newsk->sk_refcnt, 2);
1733
1734 /*
1735 * Increment the counter in the same struct proto as the master
1736 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1737 * is the same as sk->sk_prot->socks, as this field was copied
1738 * with memcpy).
1739 *
1740 * This _changes_ the previous behaviour, where
1741 * tcp_create_openreq_child always was incrementing the
1742 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1743 * to be taken into account in all callers. -acme
1744 */
1745 sk_refcnt_debug_inc(newsk);
1746 sk_set_socket(newsk, NULL);
1747 newsk->sk_wq = NULL;
1748
1749 if (newsk->sk_prot->sockets_allocated)
1750 sk_sockets_allocated_inc(newsk);
1751
1752 if (sock_needs_netstamp(sk) &&
1753 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1754 net_enable_timestamp();
1755 }
1756 out:
1757 return newsk;
1758 }
1759 EXPORT_SYMBOL_GPL(sk_clone_lock);
1760
1761 void sk_free_unlock_clone(struct sock *sk)
1762 {
1763 /* It is still raw copy of parent, so invalidate
1764 * destructor and make plain sk_free() */
1765 sk->sk_destruct = NULL;
1766 bh_unlock_sock(sk);
1767 sk_free(sk);
1768 }
1769 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1770
1771 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1772 {
1773 u32 max_segs = 1;
1774
1775 sk_dst_set(sk, dst);
1776 sk->sk_route_caps = dst->dev->features;
1777 if (sk->sk_route_caps & NETIF_F_GSO)
1778 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1779 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1780 if (sk_can_gso(sk)) {
1781 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1782 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1783 } else {
1784 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1785 sk->sk_gso_max_size = dst->dev->gso_max_size;
1786 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1787 }
1788 }
1789 sk->sk_gso_max_segs = max_segs;
1790 }
1791 EXPORT_SYMBOL_GPL(sk_setup_caps);
1792
1793 /*
1794 * Simple resource managers for sockets.
1795 */
1796
1797
1798 /*
1799 * Write buffer destructor automatically called from kfree_skb.
1800 */
1801 void sock_wfree(struct sk_buff *skb)
1802 {
1803 struct sock *sk = skb->sk;
1804 unsigned int len = skb->truesize;
1805
1806 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1807 /*
1808 * Keep a reference on sk_wmem_alloc, this will be released
1809 * after sk_write_space() call
1810 */
1811 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1812 sk->sk_write_space(sk);
1813 len = 1;
1814 }
1815 /*
1816 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1817 * could not do because of in-flight packets
1818 */
1819 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1820 __sk_free(sk);
1821 }
1822 EXPORT_SYMBOL(sock_wfree);
1823
1824 /* This variant of sock_wfree() is used by TCP,
1825 * since it sets SOCK_USE_WRITE_QUEUE.
1826 */
1827 void __sock_wfree(struct sk_buff *skb)
1828 {
1829 struct sock *sk = skb->sk;
1830
1831 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1832 __sk_free(sk);
1833 }
1834
1835 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1836 {
1837 skb_orphan(skb);
1838 skb->sk = sk;
1839 #ifdef CONFIG_INET
1840 if (unlikely(!sk_fullsock(sk))) {
1841 skb->destructor = sock_edemux;
1842 sock_hold(sk);
1843 return;
1844 }
1845 #endif
1846 skb->destructor = sock_wfree;
1847 skb_set_hash_from_sk(skb, sk);
1848 /*
1849 * We used to take a refcount on sk, but following operation
1850 * is enough to guarantee sk_free() wont free this sock until
1851 * all in-flight packets are completed
1852 */
1853 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1854 }
1855 EXPORT_SYMBOL(skb_set_owner_w);
1856
1857 /* This helper is used by netem, as it can hold packets in its
1858 * delay queue. We want to allow the owner socket to send more
1859 * packets, as if they were already TX completed by a typical driver.
1860 * But we also want to keep skb->sk set because some packet schedulers
1861 * rely on it (sch_fq for example).
1862 */
1863 void skb_orphan_partial(struct sk_buff *skb)
1864 {
1865 if (skb_is_tcp_pure_ack(skb))
1866 return;
1867
1868 if (skb->destructor == sock_wfree
1869 #ifdef CONFIG_INET
1870 || skb->destructor == tcp_wfree
1871 #endif
1872 ) {
1873 struct sock *sk = skb->sk;
1874
1875 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1876 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
1877 skb->destructor = sock_efree;
1878 }
1879 } else {
1880 skb_orphan(skb);
1881 }
1882 }
1883 EXPORT_SYMBOL(skb_orphan_partial);
1884
1885 /*
1886 * Read buffer destructor automatically called from kfree_skb.
1887 */
1888 void sock_rfree(struct sk_buff *skb)
1889 {
1890 struct sock *sk = skb->sk;
1891 unsigned int len = skb->truesize;
1892
1893 atomic_sub(len, &sk->sk_rmem_alloc);
1894 sk_mem_uncharge(sk, len);
1895 }
1896 EXPORT_SYMBOL(sock_rfree);
1897
1898 /*
1899 * Buffer destructor for skbs that are not used directly in read or write
1900 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1901 */
1902 void sock_efree(struct sk_buff *skb)
1903 {
1904 sock_put(skb->sk);
1905 }
1906 EXPORT_SYMBOL(sock_efree);
1907
1908 kuid_t sock_i_uid(struct sock *sk)
1909 {
1910 kuid_t uid;
1911
1912 read_lock_bh(&sk->sk_callback_lock);
1913 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1914 read_unlock_bh(&sk->sk_callback_lock);
1915 return uid;
1916 }
1917 EXPORT_SYMBOL(sock_i_uid);
1918
1919 unsigned long sock_i_ino(struct sock *sk)
1920 {
1921 unsigned long ino;
1922
1923 read_lock_bh(&sk->sk_callback_lock);
1924 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1925 read_unlock_bh(&sk->sk_callback_lock);
1926 return ino;
1927 }
1928 EXPORT_SYMBOL(sock_i_ino);
1929
1930 /*
1931 * Allocate a skb from the socket's send buffer.
1932 */
1933 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1934 gfp_t priority)
1935 {
1936 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1937 struct sk_buff *skb = alloc_skb(size, priority);
1938 if (skb) {
1939 skb_set_owner_w(skb, sk);
1940 return skb;
1941 }
1942 }
1943 return NULL;
1944 }
1945 EXPORT_SYMBOL(sock_wmalloc);
1946
1947 static void sock_ofree(struct sk_buff *skb)
1948 {
1949 struct sock *sk = skb->sk;
1950
1951 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
1952 }
1953
1954 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1955 gfp_t priority)
1956 {
1957 struct sk_buff *skb;
1958
1959 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
1960 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
1961 sysctl_optmem_max)
1962 return NULL;
1963
1964 skb = alloc_skb(size, priority);
1965 if (!skb)
1966 return NULL;
1967
1968 atomic_add(skb->truesize, &sk->sk_omem_alloc);
1969 skb->sk = sk;
1970 skb->destructor = sock_ofree;
1971 return skb;
1972 }
1973
1974 /*
1975 * Allocate a memory block from the socket's option memory buffer.
1976 */
1977 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1978 {
1979 if ((unsigned int)size <= sysctl_optmem_max &&
1980 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1981 void *mem;
1982 /* First do the add, to avoid the race if kmalloc
1983 * might sleep.
1984 */
1985 atomic_add(size, &sk->sk_omem_alloc);
1986 mem = kmalloc(size, priority);
1987 if (mem)
1988 return mem;
1989 atomic_sub(size, &sk->sk_omem_alloc);
1990 }
1991 return NULL;
1992 }
1993 EXPORT_SYMBOL(sock_kmalloc);
1994
1995 /* Free an option memory block. Note, we actually want the inline
1996 * here as this allows gcc to detect the nullify and fold away the
1997 * condition entirely.
1998 */
1999 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2000 const bool nullify)
2001 {
2002 if (WARN_ON_ONCE(!mem))
2003 return;
2004 if (nullify)
2005 kzfree(mem);
2006 else
2007 kfree(mem);
2008 atomic_sub(size, &sk->sk_omem_alloc);
2009 }
2010
2011 void sock_kfree_s(struct sock *sk, void *mem, int size)
2012 {
2013 __sock_kfree_s(sk, mem, size, false);
2014 }
2015 EXPORT_SYMBOL(sock_kfree_s);
2016
2017 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2018 {
2019 __sock_kfree_s(sk, mem, size, true);
2020 }
2021 EXPORT_SYMBOL(sock_kzfree_s);
2022
2023 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2024 I think, these locks should be removed for datagram sockets.
2025 */
2026 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2027 {
2028 DEFINE_WAIT(wait);
2029
2030 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2031 for (;;) {
2032 if (!timeo)
2033 break;
2034 if (signal_pending(current))
2035 break;
2036 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2037 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2038 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2039 break;
2040 if (sk->sk_shutdown & SEND_SHUTDOWN)
2041 break;
2042 if (sk->sk_err)
2043 break;
2044 timeo = schedule_timeout(timeo);
2045 }
2046 finish_wait(sk_sleep(sk), &wait);
2047 return timeo;
2048 }
2049
2050
2051 /*
2052 * Generic send/receive buffer handlers
2053 */
2054
2055 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2056 unsigned long data_len, int noblock,
2057 int *errcode, int max_page_order)
2058 {
2059 struct sk_buff *skb;
2060 long timeo;
2061 int err;
2062
2063 timeo = sock_sndtimeo(sk, noblock);
2064 for (;;) {
2065 err = sock_error(sk);
2066 if (err != 0)
2067 goto failure;
2068
2069 err = -EPIPE;
2070 if (sk->sk_shutdown & SEND_SHUTDOWN)
2071 goto failure;
2072
2073 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2074 break;
2075
2076 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2077 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2078 err = -EAGAIN;
2079 if (!timeo)
2080 goto failure;
2081 if (signal_pending(current))
2082 goto interrupted;
2083 timeo = sock_wait_for_wmem(sk, timeo);
2084 }
2085 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2086 errcode, sk->sk_allocation);
2087 if (skb)
2088 skb_set_owner_w(skb, sk);
2089 return skb;
2090
2091 interrupted:
2092 err = sock_intr_errno(timeo);
2093 failure:
2094 *errcode = err;
2095 return NULL;
2096 }
2097 EXPORT_SYMBOL(sock_alloc_send_pskb);
2098
2099 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2100 int noblock, int *errcode)
2101 {
2102 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2103 }
2104 EXPORT_SYMBOL(sock_alloc_send_skb);
2105
2106 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2107 struct sockcm_cookie *sockc)
2108 {
2109 u32 tsflags;
2110
2111 switch (cmsg->cmsg_type) {
2112 case SO_MARK:
2113 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2114 return -EPERM;
2115 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2116 return -EINVAL;
2117 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2118 break;
2119 case SO_TIMESTAMPING:
2120 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2121 return -EINVAL;
2122
2123 tsflags = *(u32 *)CMSG_DATA(cmsg);
2124 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2125 return -EINVAL;
2126
2127 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2128 sockc->tsflags |= tsflags;
2129 break;
2130 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2131 case SCM_RIGHTS:
2132 case SCM_CREDENTIALS:
2133 break;
2134 default:
2135 return -EINVAL;
2136 }
2137 return 0;
2138 }
2139 EXPORT_SYMBOL(__sock_cmsg_send);
2140
2141 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2142 struct sockcm_cookie *sockc)
2143 {
2144 struct cmsghdr *cmsg;
2145 int ret;
2146
2147 for_each_cmsghdr(cmsg, msg) {
2148 if (!CMSG_OK(msg, cmsg))
2149 return -EINVAL;
2150 if (cmsg->cmsg_level != SOL_SOCKET)
2151 continue;
2152 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2153 if (ret)
2154 return ret;
2155 }
2156 return 0;
2157 }
2158 EXPORT_SYMBOL(sock_cmsg_send);
2159
2160 static void sk_enter_memory_pressure(struct sock *sk)
2161 {
2162 if (!sk->sk_prot->enter_memory_pressure)
2163 return;
2164
2165 sk->sk_prot->enter_memory_pressure(sk);
2166 }
2167
2168 static void sk_leave_memory_pressure(struct sock *sk)
2169 {
2170 if (sk->sk_prot->leave_memory_pressure) {
2171 sk->sk_prot->leave_memory_pressure(sk);
2172 } else {
2173 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2174
2175 if (memory_pressure && *memory_pressure)
2176 *memory_pressure = 0;
2177 }
2178 }
2179
2180 /* On 32bit arches, an skb frag is limited to 2^15 */
2181 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2182
2183 /**
2184 * skb_page_frag_refill - check that a page_frag contains enough room
2185 * @sz: minimum size of the fragment we want to get
2186 * @pfrag: pointer to page_frag
2187 * @gfp: priority for memory allocation
2188 *
2189 * Note: While this allocator tries to use high order pages, there is
2190 * no guarantee that allocations succeed. Therefore, @sz MUST be
2191 * less or equal than PAGE_SIZE.
2192 */
2193 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2194 {
2195 if (pfrag->page) {
2196 if (page_ref_count(pfrag->page) == 1) {
2197 pfrag->offset = 0;
2198 return true;
2199 }
2200 if (pfrag->offset + sz <= pfrag->size)
2201 return true;
2202 put_page(pfrag->page);
2203 }
2204
2205 pfrag->offset = 0;
2206 if (SKB_FRAG_PAGE_ORDER) {
2207 /* Avoid direct reclaim but allow kswapd to wake */
2208 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2209 __GFP_COMP | __GFP_NOWARN |
2210 __GFP_NORETRY,
2211 SKB_FRAG_PAGE_ORDER);
2212 if (likely(pfrag->page)) {
2213 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2214 return true;
2215 }
2216 }
2217 pfrag->page = alloc_page(gfp);
2218 if (likely(pfrag->page)) {
2219 pfrag->size = PAGE_SIZE;
2220 return true;
2221 }
2222 return false;
2223 }
2224 EXPORT_SYMBOL(skb_page_frag_refill);
2225
2226 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2227 {
2228 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2229 return true;
2230
2231 sk_enter_memory_pressure(sk);
2232 sk_stream_moderate_sndbuf(sk);
2233 return false;
2234 }
2235 EXPORT_SYMBOL(sk_page_frag_refill);
2236
2237 static void __lock_sock(struct sock *sk)
2238 __releases(&sk->sk_lock.slock)
2239 __acquires(&sk->sk_lock.slock)
2240 {
2241 DEFINE_WAIT(wait);
2242
2243 for (;;) {
2244 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2245 TASK_UNINTERRUPTIBLE);
2246 spin_unlock_bh(&sk->sk_lock.slock);
2247 schedule();
2248 spin_lock_bh(&sk->sk_lock.slock);
2249 if (!sock_owned_by_user(sk))
2250 break;
2251 }
2252 finish_wait(&sk->sk_lock.wq, &wait);
2253 }
2254
2255 static void __release_sock(struct sock *sk)
2256 __releases(&sk->sk_lock.slock)
2257 __acquires(&sk->sk_lock.slock)
2258 {
2259 struct sk_buff *skb, *next;
2260
2261 while ((skb = sk->sk_backlog.head) != NULL) {
2262 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2263
2264 spin_unlock_bh(&sk->sk_lock.slock);
2265
2266 do {
2267 next = skb->next;
2268 prefetch(next);
2269 WARN_ON_ONCE(skb_dst_is_noref(skb));
2270 skb->next = NULL;
2271 sk_backlog_rcv(sk, skb);
2272
2273 cond_resched();
2274
2275 skb = next;
2276 } while (skb != NULL);
2277
2278 spin_lock_bh(&sk->sk_lock.slock);
2279 }
2280
2281 /*
2282 * Doing the zeroing here guarantee we can not loop forever
2283 * while a wild producer attempts to flood us.
2284 */
2285 sk->sk_backlog.len = 0;
2286 }
2287
2288 void __sk_flush_backlog(struct sock *sk)
2289 {
2290 spin_lock_bh(&sk->sk_lock.slock);
2291 __release_sock(sk);
2292 spin_unlock_bh(&sk->sk_lock.slock);
2293 }
2294
2295 /**
2296 * sk_wait_data - wait for data to arrive at sk_receive_queue
2297 * @sk: sock to wait on
2298 * @timeo: for how long
2299 * @skb: last skb seen on sk_receive_queue
2300 *
2301 * Now socket state including sk->sk_err is changed only under lock,
2302 * hence we may omit checks after joining wait queue.
2303 * We check receive queue before schedule() only as optimization;
2304 * it is very likely that release_sock() added new data.
2305 */
2306 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2307 {
2308 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2309 int rc;
2310
2311 add_wait_queue(sk_sleep(sk), &wait);
2312 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2313 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2314 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2315 remove_wait_queue(sk_sleep(sk), &wait);
2316 return rc;
2317 }
2318 EXPORT_SYMBOL(sk_wait_data);
2319
2320 /**
2321 * __sk_mem_raise_allocated - increase memory_allocated
2322 * @sk: socket
2323 * @size: memory size to allocate
2324 * @amt: pages to allocate
2325 * @kind: allocation type
2326 *
2327 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2328 */
2329 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2330 {
2331 struct proto *prot = sk->sk_prot;
2332 long allocated = sk_memory_allocated_add(sk, amt);
2333
2334 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2335 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2336 goto suppress_allocation;
2337
2338 /* Under limit. */
2339 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2340 sk_leave_memory_pressure(sk);
2341 return 1;
2342 }
2343
2344 /* Under pressure. */
2345 if (allocated > sk_prot_mem_limits(sk, 1))
2346 sk_enter_memory_pressure(sk);
2347
2348 /* Over hard limit. */
2349 if (allocated > sk_prot_mem_limits(sk, 2))
2350 goto suppress_allocation;
2351
2352 /* guarantee minimum buffer size under pressure */
2353 if (kind == SK_MEM_RECV) {
2354 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2355 return 1;
2356
2357 } else { /* SK_MEM_SEND */
2358 int wmem0 = sk_get_wmem0(sk, prot);
2359
2360 if (sk->sk_type == SOCK_STREAM) {
2361 if (sk->sk_wmem_queued < wmem0)
2362 return 1;
2363 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2364 return 1;
2365 }
2366 }
2367
2368 if (sk_has_memory_pressure(sk)) {
2369 int alloc;
2370
2371 if (!sk_under_memory_pressure(sk))
2372 return 1;
2373 alloc = sk_sockets_allocated_read_positive(sk);
2374 if (sk_prot_mem_limits(sk, 2) > alloc *
2375 sk_mem_pages(sk->sk_wmem_queued +
2376 atomic_read(&sk->sk_rmem_alloc) +
2377 sk->sk_forward_alloc))
2378 return 1;
2379 }
2380
2381 suppress_allocation:
2382
2383 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2384 sk_stream_moderate_sndbuf(sk);
2385
2386 /* Fail only if socket is _under_ its sndbuf.
2387 * In this case we cannot block, so that we have to fail.
2388 */
2389 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2390 return 1;
2391 }
2392
2393 trace_sock_exceed_buf_limit(sk, prot, allocated);
2394
2395 sk_memory_allocated_sub(sk, amt);
2396
2397 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2398 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2399
2400 return 0;
2401 }
2402 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2403
2404 /**
2405 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2406 * @sk: socket
2407 * @size: memory size to allocate
2408 * @kind: allocation type
2409 *
2410 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2411 * rmem allocation. This function assumes that protocols which have
2412 * memory_pressure use sk_wmem_queued as write buffer accounting.
2413 */
2414 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2415 {
2416 int ret, amt = sk_mem_pages(size);
2417
2418 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2419 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2420 if (!ret)
2421 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2422 return ret;
2423 }
2424 EXPORT_SYMBOL(__sk_mem_schedule);
2425
2426 /**
2427 * __sk_mem_reduce_allocated - reclaim memory_allocated
2428 * @sk: socket
2429 * @amount: number of quanta
2430 *
2431 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2432 */
2433 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2434 {
2435 sk_memory_allocated_sub(sk, amount);
2436
2437 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2438 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2439
2440 if (sk_under_memory_pressure(sk) &&
2441 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2442 sk_leave_memory_pressure(sk);
2443 }
2444 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2445
2446 /**
2447 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2448 * @sk: socket
2449 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2450 */
2451 void __sk_mem_reclaim(struct sock *sk, int amount)
2452 {
2453 amount >>= SK_MEM_QUANTUM_SHIFT;
2454 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2455 __sk_mem_reduce_allocated(sk, amount);
2456 }
2457 EXPORT_SYMBOL(__sk_mem_reclaim);
2458
2459 int sk_set_peek_off(struct sock *sk, int val)
2460 {
2461 sk->sk_peek_off = val;
2462 return 0;
2463 }
2464 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2465
2466 /*
2467 * Set of default routines for initialising struct proto_ops when
2468 * the protocol does not support a particular function. In certain
2469 * cases where it makes no sense for a protocol to have a "do nothing"
2470 * function, some default processing is provided.
2471 */
2472
2473 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2474 {
2475 return -EOPNOTSUPP;
2476 }
2477 EXPORT_SYMBOL(sock_no_bind);
2478
2479 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2480 int len, int flags)
2481 {
2482 return -EOPNOTSUPP;
2483 }
2484 EXPORT_SYMBOL(sock_no_connect);
2485
2486 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2487 {
2488 return -EOPNOTSUPP;
2489 }
2490 EXPORT_SYMBOL(sock_no_socketpair);
2491
2492 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2493 bool kern)
2494 {
2495 return -EOPNOTSUPP;
2496 }
2497 EXPORT_SYMBOL(sock_no_accept);
2498
2499 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2500 int *len, int peer)
2501 {
2502 return -EOPNOTSUPP;
2503 }
2504 EXPORT_SYMBOL(sock_no_getname);
2505
2506 __poll_t sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2507 {
2508 return 0;
2509 }
2510 EXPORT_SYMBOL(sock_no_poll);
2511
2512 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2513 {
2514 return -EOPNOTSUPP;
2515 }
2516 EXPORT_SYMBOL(sock_no_ioctl);
2517
2518 int sock_no_listen(struct socket *sock, int backlog)
2519 {
2520 return -EOPNOTSUPP;
2521 }
2522 EXPORT_SYMBOL(sock_no_listen);
2523
2524 int sock_no_shutdown(struct socket *sock, int how)
2525 {
2526 return -EOPNOTSUPP;
2527 }
2528 EXPORT_SYMBOL(sock_no_shutdown);
2529
2530 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2531 char __user *optval, unsigned int optlen)
2532 {
2533 return -EOPNOTSUPP;
2534 }
2535 EXPORT_SYMBOL(sock_no_setsockopt);
2536
2537 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2538 char __user *optval, int __user *optlen)
2539 {
2540 return -EOPNOTSUPP;
2541 }
2542 EXPORT_SYMBOL(sock_no_getsockopt);
2543
2544 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2545 {
2546 return -EOPNOTSUPP;
2547 }
2548 EXPORT_SYMBOL(sock_no_sendmsg);
2549
2550 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2551 {
2552 return -EOPNOTSUPP;
2553 }
2554 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2555
2556 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2557 int flags)
2558 {
2559 return -EOPNOTSUPP;
2560 }
2561 EXPORT_SYMBOL(sock_no_recvmsg);
2562
2563 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2564 {
2565 /* Mirror missing mmap method error code */
2566 return -ENODEV;
2567 }
2568 EXPORT_SYMBOL(sock_no_mmap);
2569
2570 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2571 {
2572 ssize_t res;
2573 struct msghdr msg = {.msg_flags = flags};
2574 struct kvec iov;
2575 char *kaddr = kmap(page);
2576 iov.iov_base = kaddr + offset;
2577 iov.iov_len = size;
2578 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2579 kunmap(page);
2580 return res;
2581 }
2582 EXPORT_SYMBOL(sock_no_sendpage);
2583
2584 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2585 int offset, size_t size, int flags)
2586 {
2587 ssize_t res;
2588 struct msghdr msg = {.msg_flags = flags};
2589 struct kvec iov;
2590 char *kaddr = kmap(page);
2591
2592 iov.iov_base = kaddr + offset;
2593 iov.iov_len = size;
2594 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2595 kunmap(page);
2596 return res;
2597 }
2598 EXPORT_SYMBOL(sock_no_sendpage_locked);
2599
2600 /*
2601 * Default Socket Callbacks
2602 */
2603
2604 static void sock_def_wakeup(struct sock *sk)
2605 {
2606 struct socket_wq *wq;
2607
2608 rcu_read_lock();
2609 wq = rcu_dereference(sk->sk_wq);
2610 if (skwq_has_sleeper(wq))
2611 wake_up_interruptible_all(&wq->wait);
2612 rcu_read_unlock();
2613 }
2614
2615 static void sock_def_error_report(struct sock *sk)
2616 {
2617 struct socket_wq *wq;
2618
2619 rcu_read_lock();
2620 wq = rcu_dereference(sk->sk_wq);
2621 if (skwq_has_sleeper(wq))
2622 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2623 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2624 rcu_read_unlock();
2625 }
2626
2627 static void sock_def_readable(struct sock *sk)
2628 {
2629 struct socket_wq *wq;
2630
2631 rcu_read_lock();
2632 wq = rcu_dereference(sk->sk_wq);
2633 if (skwq_has_sleeper(wq))
2634 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2635 EPOLLRDNORM | EPOLLRDBAND);
2636 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2637 rcu_read_unlock();
2638 }
2639
2640 static void sock_def_write_space(struct sock *sk)
2641 {
2642 struct socket_wq *wq;
2643
2644 rcu_read_lock();
2645
2646 /* Do not wake up a writer until he can make "significant"
2647 * progress. --DaveM
2648 */
2649 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2650 wq = rcu_dereference(sk->sk_wq);
2651 if (skwq_has_sleeper(wq))
2652 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2653 EPOLLWRNORM | EPOLLWRBAND);
2654
2655 /* Should agree with poll, otherwise some programs break */
2656 if (sock_writeable(sk))
2657 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2658 }
2659
2660 rcu_read_unlock();
2661 }
2662
2663 static void sock_def_destruct(struct sock *sk)
2664 {
2665 }
2666
2667 void sk_send_sigurg(struct sock *sk)
2668 {
2669 if (sk->sk_socket && sk->sk_socket->file)
2670 if (send_sigurg(&sk->sk_socket->file->f_owner))
2671 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2672 }
2673 EXPORT_SYMBOL(sk_send_sigurg);
2674
2675 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2676 unsigned long expires)
2677 {
2678 if (!mod_timer(timer, expires))
2679 sock_hold(sk);
2680 }
2681 EXPORT_SYMBOL(sk_reset_timer);
2682
2683 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2684 {
2685 if (del_timer(timer))
2686 __sock_put(sk);
2687 }
2688 EXPORT_SYMBOL(sk_stop_timer);
2689
2690 void sock_init_data(struct socket *sock, struct sock *sk)
2691 {
2692 sk_init_common(sk);
2693 sk->sk_send_head = NULL;
2694
2695 timer_setup(&sk->sk_timer, NULL, 0);
2696
2697 sk->sk_allocation = GFP_KERNEL;
2698 sk->sk_rcvbuf = sysctl_rmem_default;
2699 sk->sk_sndbuf = sysctl_wmem_default;
2700 sk->sk_state = TCP_CLOSE;
2701 sk_set_socket(sk, sock);
2702
2703 sock_set_flag(sk, SOCK_ZAPPED);
2704
2705 if (sock) {
2706 sk->sk_type = sock->type;
2707 sk->sk_wq = sock->wq;
2708 sock->sk = sk;
2709 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2710 } else {
2711 sk->sk_wq = NULL;
2712 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2713 }
2714
2715 rwlock_init(&sk->sk_callback_lock);
2716 if (sk->sk_kern_sock)
2717 lockdep_set_class_and_name(
2718 &sk->sk_callback_lock,
2719 af_kern_callback_keys + sk->sk_family,
2720 af_family_kern_clock_key_strings[sk->sk_family]);
2721 else
2722 lockdep_set_class_and_name(
2723 &sk->sk_callback_lock,
2724 af_callback_keys + sk->sk_family,
2725 af_family_clock_key_strings[sk->sk_family]);
2726
2727 sk->sk_state_change = sock_def_wakeup;
2728 sk->sk_data_ready = sock_def_readable;
2729 sk->sk_write_space = sock_def_write_space;
2730 sk->sk_error_report = sock_def_error_report;
2731 sk->sk_destruct = sock_def_destruct;
2732
2733 sk->sk_frag.page = NULL;
2734 sk->sk_frag.offset = 0;
2735 sk->sk_peek_off = -1;
2736
2737 sk->sk_peer_pid = NULL;
2738 sk->sk_peer_cred = NULL;
2739 sk->sk_write_pending = 0;
2740 sk->sk_rcvlowat = 1;
2741 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2742 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2743
2744 sk->sk_stamp = SK_DEFAULT_STAMP;
2745 atomic_set(&sk->sk_zckey, 0);
2746
2747 #ifdef CONFIG_NET_RX_BUSY_POLL
2748 sk->sk_napi_id = 0;
2749 sk->sk_ll_usec = sysctl_net_busy_read;
2750 #endif
2751
2752 sk->sk_max_pacing_rate = ~0U;
2753 sk->sk_pacing_rate = ~0U;
2754 sk->sk_pacing_shift = 10;
2755 sk->sk_incoming_cpu = -1;
2756 /*
2757 * Before updating sk_refcnt, we must commit prior changes to memory
2758 * (Documentation/RCU/rculist_nulls.txt for details)
2759 */
2760 smp_wmb();
2761 refcount_set(&sk->sk_refcnt, 1);
2762 atomic_set(&sk->sk_drops, 0);
2763 }
2764 EXPORT_SYMBOL(sock_init_data);
2765
2766 void lock_sock_nested(struct sock *sk, int subclass)
2767 {
2768 might_sleep();
2769 spin_lock_bh(&sk->sk_lock.slock);
2770 if (sk->sk_lock.owned)
2771 __lock_sock(sk);
2772 sk->sk_lock.owned = 1;
2773 spin_unlock(&sk->sk_lock.slock);
2774 /*
2775 * The sk_lock has mutex_lock() semantics here:
2776 */
2777 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2778 local_bh_enable();
2779 }
2780 EXPORT_SYMBOL(lock_sock_nested);
2781
2782 void release_sock(struct sock *sk)
2783 {
2784 spin_lock_bh(&sk->sk_lock.slock);
2785 if (sk->sk_backlog.tail)
2786 __release_sock(sk);
2787
2788 /* Warning : release_cb() might need to release sk ownership,
2789 * ie call sock_release_ownership(sk) before us.
2790 */
2791 if (sk->sk_prot->release_cb)
2792 sk->sk_prot->release_cb(sk);
2793
2794 sock_release_ownership(sk);
2795 if (waitqueue_active(&sk->sk_lock.wq))
2796 wake_up(&sk->sk_lock.wq);
2797 spin_unlock_bh(&sk->sk_lock.slock);
2798 }
2799 EXPORT_SYMBOL(release_sock);
2800
2801 /**
2802 * lock_sock_fast - fast version of lock_sock
2803 * @sk: socket
2804 *
2805 * This version should be used for very small section, where process wont block
2806 * return false if fast path is taken:
2807 *
2808 * sk_lock.slock locked, owned = 0, BH disabled
2809 *
2810 * return true if slow path is taken:
2811 *
2812 * sk_lock.slock unlocked, owned = 1, BH enabled
2813 */
2814 bool lock_sock_fast(struct sock *sk)
2815 {
2816 might_sleep();
2817 spin_lock_bh(&sk->sk_lock.slock);
2818
2819 if (!sk->sk_lock.owned)
2820 /*
2821 * Note : We must disable BH
2822 */
2823 return false;
2824
2825 __lock_sock(sk);
2826 sk->sk_lock.owned = 1;
2827 spin_unlock(&sk->sk_lock.slock);
2828 /*
2829 * The sk_lock has mutex_lock() semantics here:
2830 */
2831 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2832 local_bh_enable();
2833 return true;
2834 }
2835 EXPORT_SYMBOL(lock_sock_fast);
2836
2837 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2838 {
2839 struct timeval tv;
2840 if (!sock_flag(sk, SOCK_TIMESTAMP))
2841 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2842 tv = ktime_to_timeval(sk->sk_stamp);
2843 if (tv.tv_sec == -1)
2844 return -ENOENT;
2845 if (tv.tv_sec == 0) {
2846 sk->sk_stamp = ktime_get_real();
2847 tv = ktime_to_timeval(sk->sk_stamp);
2848 }
2849 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2850 }
2851 EXPORT_SYMBOL(sock_get_timestamp);
2852
2853 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2854 {
2855 struct timespec ts;
2856 if (!sock_flag(sk, SOCK_TIMESTAMP))
2857 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2858 ts = ktime_to_timespec(sk->sk_stamp);
2859 if (ts.tv_sec == -1)
2860 return -ENOENT;
2861 if (ts.tv_sec == 0) {
2862 sk->sk_stamp = ktime_get_real();
2863 ts = ktime_to_timespec(sk->sk_stamp);
2864 }
2865 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2866 }
2867 EXPORT_SYMBOL(sock_get_timestampns);
2868
2869 void sock_enable_timestamp(struct sock *sk, int flag)
2870 {
2871 if (!sock_flag(sk, flag)) {
2872 unsigned long previous_flags = sk->sk_flags;
2873
2874 sock_set_flag(sk, flag);
2875 /*
2876 * we just set one of the two flags which require net
2877 * time stamping, but time stamping might have been on
2878 * already because of the other one
2879 */
2880 if (sock_needs_netstamp(sk) &&
2881 !(previous_flags & SK_FLAGS_TIMESTAMP))
2882 net_enable_timestamp();
2883 }
2884 }
2885
2886 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2887 int level, int type)
2888 {
2889 struct sock_exterr_skb *serr;
2890 struct sk_buff *skb;
2891 int copied, err;
2892
2893 err = -EAGAIN;
2894 skb = sock_dequeue_err_skb(sk);
2895 if (skb == NULL)
2896 goto out;
2897
2898 copied = skb->len;
2899 if (copied > len) {
2900 msg->msg_flags |= MSG_TRUNC;
2901 copied = len;
2902 }
2903 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2904 if (err)
2905 goto out_free_skb;
2906
2907 sock_recv_timestamp(msg, sk, skb);
2908
2909 serr = SKB_EXT_ERR(skb);
2910 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2911
2912 msg->msg_flags |= MSG_ERRQUEUE;
2913 err = copied;
2914
2915 out_free_skb:
2916 kfree_skb(skb);
2917 out:
2918 return err;
2919 }
2920 EXPORT_SYMBOL(sock_recv_errqueue);
2921
2922 /*
2923 * Get a socket option on an socket.
2924 *
2925 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2926 * asynchronous errors should be reported by getsockopt. We assume
2927 * this means if you specify SO_ERROR (otherwise whats the point of it).
2928 */
2929 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2930 char __user *optval, int __user *optlen)
2931 {
2932 struct sock *sk = sock->sk;
2933
2934 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2935 }
2936 EXPORT_SYMBOL(sock_common_getsockopt);
2937
2938 #ifdef CONFIG_COMPAT
2939 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2940 char __user *optval, int __user *optlen)
2941 {
2942 struct sock *sk = sock->sk;
2943
2944 if (sk->sk_prot->compat_getsockopt != NULL)
2945 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2946 optval, optlen);
2947 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2948 }
2949 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2950 #endif
2951
2952 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2953 int flags)
2954 {
2955 struct sock *sk = sock->sk;
2956 int addr_len = 0;
2957 int err;
2958
2959 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2960 flags & ~MSG_DONTWAIT, &addr_len);
2961 if (err >= 0)
2962 msg->msg_namelen = addr_len;
2963 return err;
2964 }
2965 EXPORT_SYMBOL(sock_common_recvmsg);
2966
2967 /*
2968 * Set socket options on an inet socket.
2969 */
2970 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2971 char __user *optval, unsigned int optlen)
2972 {
2973 struct sock *sk = sock->sk;
2974
2975 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2976 }
2977 EXPORT_SYMBOL(sock_common_setsockopt);
2978
2979 #ifdef CONFIG_COMPAT
2980 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2981 char __user *optval, unsigned int optlen)
2982 {
2983 struct sock *sk = sock->sk;
2984
2985 if (sk->sk_prot->compat_setsockopt != NULL)
2986 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2987 optval, optlen);
2988 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2989 }
2990 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2991 #endif
2992
2993 void sk_common_release(struct sock *sk)
2994 {
2995 if (sk->sk_prot->destroy)
2996 sk->sk_prot->destroy(sk);
2997
2998 /*
2999 * Observation: when sock_common_release is called, processes have
3000 * no access to socket. But net still has.
3001 * Step one, detach it from networking:
3002 *
3003 * A. Remove from hash tables.
3004 */
3005
3006 sk->sk_prot->unhash(sk);
3007
3008 /*
3009 * In this point socket cannot receive new packets, but it is possible
3010 * that some packets are in flight because some CPU runs receiver and
3011 * did hash table lookup before we unhashed socket. They will achieve
3012 * receive queue and will be purged by socket destructor.
3013 *
3014 * Also we still have packets pending on receive queue and probably,
3015 * our own packets waiting in device queues. sock_destroy will drain
3016 * receive queue, but transmitted packets will delay socket destruction
3017 * until the last reference will be released.
3018 */
3019
3020 sock_orphan(sk);
3021
3022 xfrm_sk_free_policy(sk);
3023
3024 sk_refcnt_debug_release(sk);
3025
3026 sock_put(sk);
3027 }
3028 EXPORT_SYMBOL(sk_common_release);
3029
3030 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3031 {
3032 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3033
3034 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3035 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3036 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3037 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3038 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3039 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3040 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3041 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3042 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3043 }
3044
3045 #ifdef CONFIG_PROC_FS
3046 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3047 struct prot_inuse {
3048 int val[PROTO_INUSE_NR];
3049 };
3050
3051 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3052
3053 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3054 {
3055 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3056 }
3057 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3058
3059 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3060 {
3061 int cpu, idx = prot->inuse_idx;
3062 int res = 0;
3063
3064 for_each_possible_cpu(cpu)
3065 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3066
3067 return res >= 0 ? res : 0;
3068 }
3069 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3070
3071 static void sock_inuse_add(struct net *net, int val)
3072 {
3073 this_cpu_add(*net->core.sock_inuse, val);
3074 }
3075
3076 int sock_inuse_get(struct net *net)
3077 {
3078 int cpu, res = 0;
3079
3080 for_each_possible_cpu(cpu)
3081 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3082
3083 return res;
3084 }
3085
3086 EXPORT_SYMBOL_GPL(sock_inuse_get);
3087
3088 static int __net_init sock_inuse_init_net(struct net *net)
3089 {
3090 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3091 if (net->core.prot_inuse == NULL)
3092 return -ENOMEM;
3093
3094 net->core.sock_inuse = alloc_percpu(int);
3095 if (net->core.sock_inuse == NULL)
3096 goto out;
3097
3098 return 0;
3099
3100 out:
3101 free_percpu(net->core.prot_inuse);
3102 return -ENOMEM;
3103 }
3104
3105 static void __net_exit sock_inuse_exit_net(struct net *net)
3106 {
3107 free_percpu(net->core.prot_inuse);
3108 free_percpu(net->core.sock_inuse);
3109 }
3110
3111 static struct pernet_operations net_inuse_ops = {
3112 .init = sock_inuse_init_net,
3113 .exit = sock_inuse_exit_net,
3114 };
3115
3116 static __init int net_inuse_init(void)
3117 {
3118 if (register_pernet_subsys(&net_inuse_ops))
3119 panic("Cannot initialize net inuse counters");
3120
3121 return 0;
3122 }
3123
3124 core_initcall(net_inuse_init);
3125
3126 static void assign_proto_idx(struct proto *prot)
3127 {
3128 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3129
3130 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3131 pr_err("PROTO_INUSE_NR exhausted\n");
3132 return;
3133 }
3134
3135 set_bit(prot->inuse_idx, proto_inuse_idx);
3136 }
3137
3138 static void release_proto_idx(struct proto *prot)
3139 {
3140 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3141 clear_bit(prot->inuse_idx, proto_inuse_idx);
3142 }
3143 #else
3144 static inline void assign_proto_idx(struct proto *prot)
3145 {
3146 }
3147
3148 static inline void release_proto_idx(struct proto *prot)
3149 {
3150 }
3151
3152 static void sock_inuse_add(struct net *net, int val)
3153 {
3154 }
3155 #endif
3156
3157 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3158 {
3159 if (!rsk_prot)
3160 return;
3161 kfree(rsk_prot->slab_name);
3162 rsk_prot->slab_name = NULL;
3163 kmem_cache_destroy(rsk_prot->slab);
3164 rsk_prot->slab = NULL;
3165 }
3166
3167 static int req_prot_init(const struct proto *prot)
3168 {
3169 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3170
3171 if (!rsk_prot)
3172 return 0;
3173
3174 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3175 prot->name);
3176 if (!rsk_prot->slab_name)
3177 return -ENOMEM;
3178
3179 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3180 rsk_prot->obj_size, 0,
3181 prot->slab_flags, NULL);
3182
3183 if (!rsk_prot->slab) {
3184 pr_crit("%s: Can't create request sock SLAB cache!\n",
3185 prot->name);
3186 return -ENOMEM;
3187 }
3188 return 0;
3189 }
3190
3191 int proto_register(struct proto *prot, int alloc_slab)
3192 {
3193 if (alloc_slab) {
3194 prot->slab = kmem_cache_create_usercopy(prot->name,
3195 prot->obj_size, 0,
3196 SLAB_HWCACHE_ALIGN | prot->slab_flags,
3197 prot->useroffset, prot->usersize,
3198 NULL);
3199
3200 if (prot->slab == NULL) {
3201 pr_crit("%s: Can't create sock SLAB cache!\n",
3202 prot->name);
3203 goto out;
3204 }
3205
3206 if (req_prot_init(prot))
3207 goto out_free_request_sock_slab;
3208
3209 if (prot->twsk_prot != NULL) {
3210 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3211
3212 if (prot->twsk_prot->twsk_slab_name == NULL)
3213 goto out_free_request_sock_slab;
3214
3215 prot->twsk_prot->twsk_slab =
3216 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3217 prot->twsk_prot->twsk_obj_size,
3218 0,
3219 prot->slab_flags,
3220 NULL);
3221 if (prot->twsk_prot->twsk_slab == NULL)
3222 goto out_free_timewait_sock_slab_name;
3223 }
3224 }
3225
3226 mutex_lock(&proto_list_mutex);
3227 list_add(&prot->node, &proto_list);
3228 assign_proto_idx(prot);
3229 mutex_unlock(&proto_list_mutex);
3230 return 0;
3231
3232 out_free_timewait_sock_slab_name:
3233 kfree(prot->twsk_prot->twsk_slab_name);
3234 out_free_request_sock_slab:
3235 req_prot_cleanup(prot->rsk_prot);
3236
3237 kmem_cache_destroy(prot->slab);
3238 prot->slab = NULL;
3239 out:
3240 return -ENOBUFS;
3241 }
3242 EXPORT_SYMBOL(proto_register);
3243
3244 void proto_unregister(struct proto *prot)
3245 {
3246 mutex_lock(&proto_list_mutex);
3247 release_proto_idx(prot);
3248 list_del(&prot->node);
3249 mutex_unlock(&proto_list_mutex);
3250
3251 kmem_cache_destroy(prot->slab);
3252 prot->slab = NULL;
3253
3254 req_prot_cleanup(prot->rsk_prot);
3255
3256 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3257 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3258 kfree(prot->twsk_prot->twsk_slab_name);
3259 prot->twsk_prot->twsk_slab = NULL;
3260 }
3261 }
3262 EXPORT_SYMBOL(proto_unregister);
3263
3264 #ifdef CONFIG_PROC_FS
3265 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3266 __acquires(proto_list_mutex)
3267 {
3268 mutex_lock(&proto_list_mutex);
3269 return seq_list_start_head(&proto_list, *pos);
3270 }
3271
3272 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3273 {
3274 return seq_list_next(v, &proto_list, pos);
3275 }
3276
3277 static void proto_seq_stop(struct seq_file *seq, void *v)
3278 __releases(proto_list_mutex)
3279 {
3280 mutex_unlock(&proto_list_mutex);
3281 }
3282
3283 static char proto_method_implemented(const void *method)
3284 {
3285 return method == NULL ? 'n' : 'y';
3286 }
3287 static long sock_prot_memory_allocated(struct proto *proto)
3288 {
3289 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3290 }
3291
3292 static char *sock_prot_memory_pressure(struct proto *proto)
3293 {
3294 return proto->memory_pressure != NULL ?
3295 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3296 }
3297
3298 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3299 {
3300
3301 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3302 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3303 proto->name,
3304 proto->obj_size,
3305 sock_prot_inuse_get(seq_file_net(seq), proto),
3306 sock_prot_memory_allocated(proto),
3307 sock_prot_memory_pressure(proto),
3308 proto->max_header,
3309 proto->slab == NULL ? "no" : "yes",
3310 module_name(proto->owner),
3311 proto_method_implemented(proto->close),
3312 proto_method_implemented(proto->connect),
3313 proto_method_implemented(proto->disconnect),
3314 proto_method_implemented(proto->accept),
3315 proto_method_implemented(proto->ioctl),
3316 proto_method_implemented(proto->init),
3317 proto_method_implemented(proto->destroy),
3318 proto_method_implemented(proto->shutdown),
3319 proto_method_implemented(proto->setsockopt),
3320 proto_method_implemented(proto->getsockopt),
3321 proto_method_implemented(proto->sendmsg),
3322 proto_method_implemented(proto->recvmsg),
3323 proto_method_implemented(proto->sendpage),
3324 proto_method_implemented(proto->bind),
3325 proto_method_implemented(proto->backlog_rcv),
3326 proto_method_implemented(proto->hash),
3327 proto_method_implemented(proto->unhash),
3328 proto_method_implemented(proto->get_port),
3329 proto_method_implemented(proto->enter_memory_pressure));
3330 }
3331
3332 static int proto_seq_show(struct seq_file *seq, void *v)
3333 {
3334 if (v == &proto_list)
3335 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3336 "protocol",
3337 "size",
3338 "sockets",
3339 "memory",
3340 "press",
3341 "maxhdr",
3342 "slab",
3343 "module",
3344 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3345 else
3346 proto_seq_printf(seq, list_entry(v, struct proto, node));
3347 return 0;
3348 }
3349
3350 static const struct seq_operations proto_seq_ops = {
3351 .start = proto_seq_start,
3352 .next = proto_seq_next,
3353 .stop = proto_seq_stop,
3354 .show = proto_seq_show,
3355 };
3356
3357 static int proto_seq_open(struct inode *inode, struct file *file)
3358 {
3359 return seq_open_net(inode, file, &proto_seq_ops,
3360 sizeof(struct seq_net_private));
3361 }
3362
3363 static const struct file_operations proto_seq_fops = {
3364 .open = proto_seq_open,
3365 .read = seq_read,
3366 .llseek = seq_lseek,
3367 .release = seq_release_net,
3368 };
3369
3370 static __net_init int proto_init_net(struct net *net)
3371 {
3372 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3373 return -ENOMEM;
3374
3375 return 0;
3376 }
3377
3378 static __net_exit void proto_exit_net(struct net *net)
3379 {
3380 remove_proc_entry("protocols", net->proc_net);
3381 }
3382
3383
3384 static __net_initdata struct pernet_operations proto_net_ops = {
3385 .init = proto_init_net,
3386 .exit = proto_exit_net,
3387 };
3388
3389 static int __init proto_init(void)
3390 {
3391 return register_pernet_subsys(&proto_net_ops);
3392 }
3393
3394 subsys_initcall(proto_init);
3395
3396 #endif /* PROC_FS */
3397
3398 #ifdef CONFIG_NET_RX_BUSY_POLL
3399 bool sk_busy_loop_end(void *p, unsigned long start_time)
3400 {
3401 struct sock *sk = p;
3402
3403 return !skb_queue_empty(&sk->sk_receive_queue) ||
3404 sk_busy_loop_timeout(sk, start_time);
3405 }
3406 EXPORT_SYMBOL(sk_busy_loop_end);
3407 #endif /* CONFIG_NET_RX_BUSY_POLL */
CRIS linux kernel tree