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 /**
149 * sk_ns_capable - General socket capability test
150 * @sk: Socket to use a capability on or through
151 * @user_ns: The user namespace of the capability to use
152 * @cap: The capability to use
153 *
154 * Test to see if the opener of the socket had when the socket was
155 * created and the current process has the capability @cap in the user
156 * namespace @user_ns.
157 */
158 bool sk_ns_capable(const struct sock *sk,
159 struct user_namespace *user_ns, int cap)
160 {
161 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
162 ns_capable(user_ns, cap);
163 }
164 EXPORT_SYMBOL(sk_ns_capable);
165
166 /**
167 * sk_capable - Socket global capability test
168 * @sk: Socket to use a capability on or through
169 * @cap: The global capability to use
170 *
171 * Test to see if the opener of the socket had when the socket was
172 * created and the current process has the capability @cap in all user
173 * namespaces.
174 */
175 bool sk_capable(const struct sock *sk, int cap)
176 {
177 return sk_ns_capable(sk, &init_user_ns, cap);
178 }
179 EXPORT_SYMBOL(sk_capable);
180
181 /**
182 * sk_net_capable - Network namespace socket capability test
183 * @sk: Socket to use a capability on or through
184 * @cap: The capability to use
185 *
186 * Test to see if the opener of the socket had when the socket was created
187 * and the current process has the capability @cap over the network namespace
188 * the socket is a member of.
189 */
190 bool sk_net_capable(const struct sock *sk, int cap)
191 {
192 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193 }
194 EXPORT_SYMBOL(sk_net_capable);
195
196 /*
197 * Each address family might have different locking rules, so we have
198 * one slock key per address family and separate keys for internal and
199 * userspace sockets.
200 */
201 static struct lock_class_key af_family_keys[AF_MAX];
202 static struct lock_class_key af_family_kern_keys[AF_MAX];
203 static struct lock_class_key af_family_slock_keys[AF_MAX];
204 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
205
206 /*
207 * Make lock validator output more readable. (we pre-construct these
208 * strings build-time, so that runtime initialization of socket
209 * locks is fast):
210 */
211
212 #define _sock_locks(x) \
213 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
214 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
215 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
216 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
217 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
218 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
219 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
220 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
221 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
222 x "27" , x "28" , x "AF_CAN" , \
223 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
224 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
225 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
226 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
227 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX"
228
229 static const char *const af_family_key_strings[AF_MAX+1] = {
230 _sock_locks("sk_lock-")
231 };
232 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
233 _sock_locks("slock-")
234 };
235 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
236 _sock_locks("clock-")
237 };
238
239 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
240 _sock_locks("k-sk_lock-")
241 };
242 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("k-slock-")
244 };
245 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-clock-")
247 };
248 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
249 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" ,
250 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK",
251 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" ,
252 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" ,
253 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" ,
254 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" ,
255 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" ,
256 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" ,
257 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" ,
258 "rlock-27" , "rlock-28" , "rlock-AF_CAN" ,
259 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" ,
260 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" ,
261 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" ,
262 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" ,
263 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX"
264 };
265 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
266 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" ,
267 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK",
268 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" ,
269 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" ,
270 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" ,
271 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" ,
272 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" ,
273 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" ,
274 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" ,
275 "wlock-27" , "wlock-28" , "wlock-AF_CAN" ,
276 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" ,
277 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" ,
278 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" ,
279 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" ,
280 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX"
281 };
282 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
283 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" ,
284 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK",
285 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" ,
286 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" ,
287 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" ,
288 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" ,
289 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" ,
290 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" ,
291 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" ,
292 "elock-27" , "elock-28" , "elock-AF_CAN" ,
293 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" ,
294 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" ,
295 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" ,
296 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" ,
297 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX"
298 };
299
300 /*
301 * sk_callback_lock and sk queues locking rules are per-address-family,
302 * so split the lock classes by using a per-AF key:
303 */
304 static struct lock_class_key af_callback_keys[AF_MAX];
305 static struct lock_class_key af_rlock_keys[AF_MAX];
306 static struct lock_class_key af_wlock_keys[AF_MAX];
307 static struct lock_class_key af_elock_keys[AF_MAX];
308 static struct lock_class_key af_kern_callback_keys[AF_MAX];
309
310 /* Run time adjustable parameters. */
311 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
312 EXPORT_SYMBOL(sysctl_wmem_max);
313 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
314 EXPORT_SYMBOL(sysctl_rmem_max);
315 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
316 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
317
318 /* Maximal space eaten by iovec or ancillary data plus some space */
319 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
320 EXPORT_SYMBOL(sysctl_optmem_max);
321
322 int sysctl_tstamp_allow_data __read_mostly = 1;
323
324 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
325 EXPORT_SYMBOL_GPL(memalloc_socks);
326
327 /**
328 * sk_set_memalloc - sets %SOCK_MEMALLOC
329 * @sk: socket to set it on
330 *
331 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
332 * It's the responsibility of the admin to adjust min_free_kbytes
333 * to meet the requirements
334 */
335 void sk_set_memalloc(struct sock *sk)
336 {
337 sock_set_flag(sk, SOCK_MEMALLOC);
338 sk->sk_allocation |= __GFP_MEMALLOC;
339 static_key_slow_inc(&memalloc_socks);
340 }
341 EXPORT_SYMBOL_GPL(sk_set_memalloc);
342
343 void sk_clear_memalloc(struct sock *sk)
344 {
345 sock_reset_flag(sk, SOCK_MEMALLOC);
346 sk->sk_allocation &= ~__GFP_MEMALLOC;
347 static_key_slow_dec(&memalloc_socks);
348
349 /*
350 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
351 * progress of swapping. SOCK_MEMALLOC may be cleared while
352 * it has rmem allocations due to the last swapfile being deactivated
353 * but there is a risk that the socket is unusable due to exceeding
354 * the rmem limits. Reclaim the reserves and obey rmem limits again.
355 */
356 sk_mem_reclaim(sk);
357 }
358 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
359
360 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
361 {
362 int ret;
363 unsigned int noreclaim_flag;
364
365 /* these should have been dropped before queueing */
366 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
367
368 noreclaim_flag = memalloc_noreclaim_save();
369 ret = sk->sk_backlog_rcv(sk, skb);
370 memalloc_noreclaim_restore(noreclaim_flag);
371
372 return ret;
373 }
374 EXPORT_SYMBOL(__sk_backlog_rcv);
375
376 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
377 {
378 struct timeval tv;
379
380 if (optlen < sizeof(tv))
381 return -EINVAL;
382 if (copy_from_user(&tv, optval, sizeof(tv)))
383 return -EFAULT;
384 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
385 return -EDOM;
386
387 if (tv.tv_sec < 0) {
388 static int warned __read_mostly;
389
390 *timeo_p = 0;
391 if (warned < 10 && net_ratelimit()) {
392 warned++;
393 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
394 __func__, current->comm, task_pid_nr(current));
395 }
396 return 0;
397 }
398 *timeo_p = MAX_SCHEDULE_TIMEOUT;
399 if (tv.tv_sec == 0 && tv.tv_usec == 0)
400 return 0;
401 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
402 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
403 return 0;
404 }
405
406 static void sock_warn_obsolete_bsdism(const char *name)
407 {
408 static int warned;
409 static char warncomm[TASK_COMM_LEN];
410 if (strcmp(warncomm, current->comm) && warned < 5) {
411 strcpy(warncomm, current->comm);
412 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
413 warncomm, name);
414 warned++;
415 }
416 }
417
418 static bool sock_needs_netstamp(const struct sock *sk)
419 {
420 switch (sk->sk_family) {
421 case AF_UNSPEC:
422 case AF_UNIX:
423 return false;
424 default:
425 return true;
426 }
427 }
428
429 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
430 {
431 if (sk->sk_flags & flags) {
432 sk->sk_flags &= ~flags;
433 if (sock_needs_netstamp(sk) &&
434 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
435 net_disable_timestamp();
436 }
437 }
438
439
440 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
441 {
442 unsigned long flags;
443 struct sk_buff_head *list = &sk->sk_receive_queue;
444
445 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
446 atomic_inc(&sk->sk_drops);
447 trace_sock_rcvqueue_full(sk, skb);
448 return -ENOMEM;
449 }
450
451 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
452 atomic_inc(&sk->sk_drops);
453 return -ENOBUFS;
454 }
455
456 skb->dev = NULL;
457 skb_set_owner_r(skb, sk);
458
459 /* we escape from rcu protected region, make sure we dont leak
460 * a norefcounted dst
461 */
462 skb_dst_force(skb);
463
464 spin_lock_irqsave(&list->lock, flags);
465 sock_skb_set_dropcount(sk, skb);
466 __skb_queue_tail(list, skb);
467 spin_unlock_irqrestore(&list->lock, flags);
468
469 if (!sock_flag(sk, SOCK_DEAD))
470 sk->sk_data_ready(sk);
471 return 0;
472 }
473 EXPORT_SYMBOL(__sock_queue_rcv_skb);
474
475 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
476 {
477 int err;
478
479 err = sk_filter(sk, skb);
480 if (err)
481 return err;
482
483 return __sock_queue_rcv_skb(sk, skb);
484 }
485 EXPORT_SYMBOL(sock_queue_rcv_skb);
486
487 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
488 const int nested, unsigned int trim_cap, bool refcounted)
489 {
490 int rc = NET_RX_SUCCESS;
491
492 if (sk_filter_trim_cap(sk, skb, trim_cap))
493 goto discard_and_relse;
494
495 skb->dev = NULL;
496
497 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
498 atomic_inc(&sk->sk_drops);
499 goto discard_and_relse;
500 }
501 if (nested)
502 bh_lock_sock_nested(sk);
503 else
504 bh_lock_sock(sk);
505 if (!sock_owned_by_user(sk)) {
506 /*
507 * trylock + unlock semantics:
508 */
509 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
510
511 rc = sk_backlog_rcv(sk, skb);
512
513 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
514 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
515 bh_unlock_sock(sk);
516 atomic_inc(&sk->sk_drops);
517 goto discard_and_relse;
518 }
519
520 bh_unlock_sock(sk);
521 out:
522 if (refcounted)
523 sock_put(sk);
524 return rc;
525 discard_and_relse:
526 kfree_skb(skb);
527 goto out;
528 }
529 EXPORT_SYMBOL(__sk_receive_skb);
530
531 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
532 {
533 struct dst_entry *dst = __sk_dst_get(sk);
534
535 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
536 sk_tx_queue_clear(sk);
537 sk->sk_dst_pending_confirm = 0;
538 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
539 dst_release(dst);
540 return NULL;
541 }
542
543 return dst;
544 }
545 EXPORT_SYMBOL(__sk_dst_check);
546
547 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
548 {
549 struct dst_entry *dst = sk_dst_get(sk);
550
551 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
552 sk_dst_reset(sk);
553 dst_release(dst);
554 return NULL;
555 }
556
557 return dst;
558 }
559 EXPORT_SYMBOL(sk_dst_check);
560
561 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
562 int optlen)
563 {
564 int ret = -ENOPROTOOPT;
565 #ifdef CONFIG_NETDEVICES
566 struct net *net = sock_net(sk);
567 char devname[IFNAMSIZ];
568 int index;
569
570 /* Sorry... */
571 ret = -EPERM;
572 if (!ns_capable(net->user_ns, CAP_NET_RAW))
573 goto out;
574
575 ret = -EINVAL;
576 if (optlen < 0)
577 goto out;
578
579 /* Bind this socket to a particular device like "eth0",
580 * as specified in the passed interface name. If the
581 * name is "" or the option length is zero the socket
582 * is not bound.
583 */
584 if (optlen > IFNAMSIZ - 1)
585 optlen = IFNAMSIZ - 1;
586 memset(devname, 0, sizeof(devname));
587
588 ret = -EFAULT;
589 if (copy_from_user(devname, optval, optlen))
590 goto out;
591
592 index = 0;
593 if (devname[0] != '\0') {
594 struct net_device *dev;
595
596 rcu_read_lock();
597 dev = dev_get_by_name_rcu(net, devname);
598 if (dev)
599 index = dev->ifindex;
600 rcu_read_unlock();
601 ret = -ENODEV;
602 if (!dev)
603 goto out;
604 }
605
606 lock_sock(sk);
607 sk->sk_bound_dev_if = index;
608 sk_dst_reset(sk);
609 release_sock(sk);
610
611 ret = 0;
612
613 out:
614 #endif
615
616 return ret;
617 }
618
619 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
620 int __user *optlen, int len)
621 {
622 int ret = -ENOPROTOOPT;
623 #ifdef CONFIG_NETDEVICES
624 struct net *net = sock_net(sk);
625 char devname[IFNAMSIZ];
626
627 if (sk->sk_bound_dev_if == 0) {
628 len = 0;
629 goto zero;
630 }
631
632 ret = -EINVAL;
633 if (len < IFNAMSIZ)
634 goto out;
635
636 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
637 if (ret)
638 goto out;
639
640 len = strlen(devname) + 1;
641
642 ret = -EFAULT;
643 if (copy_to_user(optval, devname, len))
644 goto out;
645
646 zero:
647 ret = -EFAULT;
648 if (put_user(len, optlen))
649 goto out;
650
651 ret = 0;
652
653 out:
654 #endif
655
656 return ret;
657 }
658
659 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
660 {
661 if (valbool)
662 sock_set_flag(sk, bit);
663 else
664 sock_reset_flag(sk, bit);
665 }
666
667 bool sk_mc_loop(struct sock *sk)
668 {
669 if (dev_recursion_level())
670 return false;
671 if (!sk)
672 return true;
673 switch (sk->sk_family) {
674 case AF_INET:
675 return inet_sk(sk)->mc_loop;
676 #if IS_ENABLED(CONFIG_IPV6)
677 case AF_INET6:
678 return inet6_sk(sk)->mc_loop;
679 #endif
680 }
681 WARN_ON(1);
682 return true;
683 }
684 EXPORT_SYMBOL(sk_mc_loop);
685
686 /*
687 * This is meant for all protocols to use and covers goings on
688 * at the socket level. Everything here is generic.
689 */
690
691 int sock_setsockopt(struct socket *sock, int level, int optname,
692 char __user *optval, unsigned int optlen)
693 {
694 struct sock *sk = sock->sk;
695 int val;
696 int valbool;
697 struct linger ling;
698 int ret = 0;
699
700 /*
701 * Options without arguments
702 */
703
704 if (optname == SO_BINDTODEVICE)
705 return sock_setbindtodevice(sk, optval, optlen);
706
707 if (optlen < sizeof(int))
708 return -EINVAL;
709
710 if (get_user(val, (int __user *)optval))
711 return -EFAULT;
712
713 valbool = val ? 1 : 0;
714
715 lock_sock(sk);
716
717 switch (optname) {
718 case SO_DEBUG:
719 if (val && !capable(CAP_NET_ADMIN))
720 ret = -EACCES;
721 else
722 sock_valbool_flag(sk, SOCK_DBG, valbool);
723 break;
724 case SO_REUSEADDR:
725 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
726 break;
727 case SO_REUSEPORT:
728 sk->sk_reuseport = valbool;
729 break;
730 case SO_TYPE:
731 case SO_PROTOCOL:
732 case SO_DOMAIN:
733 case SO_ERROR:
734 ret = -ENOPROTOOPT;
735 break;
736 case SO_DONTROUTE:
737 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
738 break;
739 case SO_BROADCAST:
740 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
741 break;
742 case SO_SNDBUF:
743 /* Don't error on this BSD doesn't and if you think
744 * about it this is right. Otherwise apps have to
745 * play 'guess the biggest size' games. RCVBUF/SNDBUF
746 * are treated in BSD as hints
747 */
748 val = min_t(u32, val, sysctl_wmem_max);
749 set_sndbuf:
750 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
751 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
752 /* Wake up sending tasks if we upped the value. */
753 sk->sk_write_space(sk);
754 break;
755
756 case SO_SNDBUFFORCE:
757 if (!capable(CAP_NET_ADMIN)) {
758 ret = -EPERM;
759 break;
760 }
761 goto set_sndbuf;
762
763 case SO_RCVBUF:
764 /* Don't error on this BSD doesn't and if you think
765 * about it this is right. Otherwise apps have to
766 * play 'guess the biggest size' games. RCVBUF/SNDBUF
767 * are treated in BSD as hints
768 */
769 val = min_t(u32, val, sysctl_rmem_max);
770 set_rcvbuf:
771 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
772 /*
773 * We double it on the way in to account for
774 * "struct sk_buff" etc. overhead. Applications
775 * assume that the SO_RCVBUF setting they make will
776 * allow that much actual data to be received on that
777 * socket.
778 *
779 * Applications are unaware that "struct sk_buff" and
780 * other overheads allocate from the receive buffer
781 * during socket buffer allocation.
782 *
783 * And after considering the possible alternatives,
784 * returning the value we actually used in getsockopt
785 * is the most desirable behavior.
786 */
787 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
788 break;
789
790 case SO_RCVBUFFORCE:
791 if (!capable(CAP_NET_ADMIN)) {
792 ret = -EPERM;
793 break;
794 }
795 goto set_rcvbuf;
796
797 case SO_KEEPALIVE:
798 if (sk->sk_prot->keepalive)
799 sk->sk_prot->keepalive(sk, valbool);
800 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
801 break;
802
803 case SO_OOBINLINE:
804 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
805 break;
806
807 case SO_NO_CHECK:
808 sk->sk_no_check_tx = valbool;
809 break;
810
811 case SO_PRIORITY:
812 if ((val >= 0 && val <= 6) ||
813 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
814 sk->sk_priority = val;
815 else
816 ret = -EPERM;
817 break;
818
819 case SO_LINGER:
820 if (optlen < sizeof(ling)) {
821 ret = -EINVAL; /* 1003.1g */
822 break;
823 }
824 if (copy_from_user(&ling, optval, sizeof(ling))) {
825 ret = -EFAULT;
826 break;
827 }
828 if (!ling.l_onoff)
829 sock_reset_flag(sk, SOCK_LINGER);
830 else {
831 #if (BITS_PER_LONG == 32)
832 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
833 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
834 else
835 #endif
836 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
837 sock_set_flag(sk, SOCK_LINGER);
838 }
839 break;
840
841 case SO_BSDCOMPAT:
842 sock_warn_obsolete_bsdism("setsockopt");
843 break;
844
845 case SO_PASSCRED:
846 if (valbool)
847 set_bit(SOCK_PASSCRED, &sock->flags);
848 else
849 clear_bit(SOCK_PASSCRED, &sock->flags);
850 break;
851
852 case SO_TIMESTAMP:
853 case SO_TIMESTAMPNS:
854 if (valbool) {
855 if (optname == SO_TIMESTAMP)
856 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
857 else
858 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
859 sock_set_flag(sk, SOCK_RCVTSTAMP);
860 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
861 } else {
862 sock_reset_flag(sk, SOCK_RCVTSTAMP);
863 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
864 }
865 break;
866
867 case SO_TIMESTAMPING:
868 if (val & ~SOF_TIMESTAMPING_MASK) {
869 ret = -EINVAL;
870 break;
871 }
872
873 if (val & SOF_TIMESTAMPING_OPT_ID &&
874 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
875 if (sk->sk_protocol == IPPROTO_TCP &&
876 sk->sk_type == SOCK_STREAM) {
877 if ((1 << sk->sk_state) &
878 (TCPF_CLOSE | TCPF_LISTEN)) {
879 ret = -EINVAL;
880 break;
881 }
882 sk->sk_tskey = tcp_sk(sk)->snd_una;
883 } else {
884 sk->sk_tskey = 0;
885 }
886 }
887
888 if (val & SOF_TIMESTAMPING_OPT_STATS &&
889 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
890 ret = -EINVAL;
891 break;
892 }
893
894 sk->sk_tsflags = val;
895 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
896 sock_enable_timestamp(sk,
897 SOCK_TIMESTAMPING_RX_SOFTWARE);
898 else
899 sock_disable_timestamp(sk,
900 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
901 break;
902
903 case SO_RCVLOWAT:
904 if (val < 0)
905 val = INT_MAX;
906 sk->sk_rcvlowat = val ? : 1;
907 break;
908
909 case SO_RCVTIMEO:
910 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
911 break;
912
913 case SO_SNDTIMEO:
914 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
915 break;
916
917 case SO_ATTACH_FILTER:
918 ret = -EINVAL;
919 if (optlen == sizeof(struct sock_fprog)) {
920 struct sock_fprog fprog;
921
922 ret = -EFAULT;
923 if (copy_from_user(&fprog, optval, sizeof(fprog)))
924 break;
925
926 ret = sk_attach_filter(&fprog, sk);
927 }
928 break;
929
930 case SO_ATTACH_BPF:
931 ret = -EINVAL;
932 if (optlen == sizeof(u32)) {
933 u32 ufd;
934
935 ret = -EFAULT;
936 if (copy_from_user(&ufd, optval, sizeof(ufd)))
937 break;
938
939 ret = sk_attach_bpf(ufd, sk);
940 }
941 break;
942
943 case SO_ATTACH_REUSEPORT_CBPF:
944 ret = -EINVAL;
945 if (optlen == sizeof(struct sock_fprog)) {
946 struct sock_fprog fprog;
947
948 ret = -EFAULT;
949 if (copy_from_user(&fprog, optval, sizeof(fprog)))
950 break;
951
952 ret = sk_reuseport_attach_filter(&fprog, sk);
953 }
954 break;
955
956 case SO_ATTACH_REUSEPORT_EBPF:
957 ret = -EINVAL;
958 if (optlen == sizeof(u32)) {
959 u32 ufd;
960
961 ret = -EFAULT;
962 if (copy_from_user(&ufd, optval, sizeof(ufd)))
963 break;
964
965 ret = sk_reuseport_attach_bpf(ufd, sk);
966 }
967 break;
968
969 case SO_DETACH_FILTER:
970 ret = sk_detach_filter(sk);
971 break;
972
973 case SO_LOCK_FILTER:
974 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
975 ret = -EPERM;
976 else
977 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
978 break;
979
980 case SO_PASSSEC:
981 if (valbool)
982 set_bit(SOCK_PASSSEC, &sock->flags);
983 else
984 clear_bit(SOCK_PASSSEC, &sock->flags);
985 break;
986 case SO_MARK:
987 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
988 ret = -EPERM;
989 else
990 sk->sk_mark = val;
991 break;
992
993 case SO_RXQ_OVFL:
994 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
995 break;
996
997 case SO_WIFI_STATUS:
998 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
999 break;
1000
1001 case SO_PEEK_OFF:
1002 if (sock->ops->set_peek_off)
1003 ret = sock->ops->set_peek_off(sk, val);
1004 else
1005 ret = -EOPNOTSUPP;
1006 break;
1007
1008 case SO_NOFCS:
1009 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1010 break;
1011
1012 case SO_SELECT_ERR_QUEUE:
1013 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1014 break;
1015
1016 #ifdef CONFIG_NET_RX_BUSY_POLL
1017 case SO_BUSY_POLL:
1018 /* allow unprivileged users to decrease the value */
1019 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1020 ret = -EPERM;
1021 else {
1022 if (val < 0)
1023 ret = -EINVAL;
1024 else
1025 sk->sk_ll_usec = val;
1026 }
1027 break;
1028 #endif
1029
1030 case SO_MAX_PACING_RATE:
1031 if (val != ~0U)
1032 cmpxchg(&sk->sk_pacing_status,
1033 SK_PACING_NONE,
1034 SK_PACING_NEEDED);
1035 sk->sk_max_pacing_rate = val;
1036 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1037 sk->sk_max_pacing_rate);
1038 break;
1039
1040 case SO_INCOMING_CPU:
1041 sk->sk_incoming_cpu = val;
1042 break;
1043
1044 case SO_CNX_ADVICE:
1045 if (val == 1)
1046 dst_negative_advice(sk);
1047 break;
1048
1049 case SO_ZEROCOPY:
1050 if (sk->sk_family != PF_INET && sk->sk_family != PF_INET6)
1051 ret = -ENOTSUPP;
1052 else if (sk->sk_protocol != IPPROTO_TCP)
1053 ret = -ENOTSUPP;
1054 else if (sk->sk_state != TCP_CLOSE)
1055 ret = -EBUSY;
1056 else if (val < 0 || val > 1)
1057 ret = -EINVAL;
1058 else
1059 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1060 break;
1061
1062 default:
1063 ret = -ENOPROTOOPT;
1064 break;
1065 }
1066 release_sock(sk);
1067 return ret;
1068 }
1069 EXPORT_SYMBOL(sock_setsockopt);
1070
1071
1072 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1073 struct ucred *ucred)
1074 {
1075 ucred->pid = pid_vnr(pid);
1076 ucred->uid = ucred->gid = -1;
1077 if (cred) {
1078 struct user_namespace *current_ns = current_user_ns();
1079
1080 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1081 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1082 }
1083 }
1084
1085 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1086 {
1087 struct user_namespace *user_ns = current_user_ns();
1088 int i;
1089
1090 for (i = 0; i < src->ngroups; i++)
1091 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1092 return -EFAULT;
1093
1094 return 0;
1095 }
1096
1097 int sock_getsockopt(struct socket *sock, int level, int optname,
1098 char __user *optval, int __user *optlen)
1099 {
1100 struct sock *sk = sock->sk;
1101
1102 union {
1103 int val;
1104 u64 val64;
1105 struct linger ling;
1106 struct timeval tm;
1107 } v;
1108
1109 int lv = sizeof(int);
1110 int len;
1111
1112 if (get_user(len, optlen))
1113 return -EFAULT;
1114 if (len < 0)
1115 return -EINVAL;
1116
1117 memset(&v, 0, sizeof(v));
1118
1119 switch (optname) {
1120 case SO_DEBUG:
1121 v.val = sock_flag(sk, SOCK_DBG);
1122 break;
1123
1124 case SO_DONTROUTE:
1125 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1126 break;
1127
1128 case SO_BROADCAST:
1129 v.val = sock_flag(sk, SOCK_BROADCAST);
1130 break;
1131
1132 case SO_SNDBUF:
1133 v.val = sk->sk_sndbuf;
1134 break;
1135
1136 case SO_RCVBUF:
1137 v.val = sk->sk_rcvbuf;
1138 break;
1139
1140 case SO_REUSEADDR:
1141 v.val = sk->sk_reuse;
1142 break;
1143
1144 case SO_REUSEPORT:
1145 v.val = sk->sk_reuseport;
1146 break;
1147
1148 case SO_KEEPALIVE:
1149 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1150 break;
1151
1152 case SO_TYPE:
1153 v.val = sk->sk_type;
1154 break;
1155
1156 case SO_PROTOCOL:
1157 v.val = sk->sk_protocol;
1158 break;
1159
1160 case SO_DOMAIN:
1161 v.val = sk->sk_family;
1162 break;
1163
1164 case SO_ERROR:
1165 v.val = -sock_error(sk);
1166 if (v.val == 0)
1167 v.val = xchg(&sk->sk_err_soft, 0);
1168 break;
1169
1170 case SO_OOBINLINE:
1171 v.val = sock_flag(sk, SOCK_URGINLINE);
1172 break;
1173
1174 case SO_NO_CHECK:
1175 v.val = sk->sk_no_check_tx;
1176 break;
1177
1178 case SO_PRIORITY:
1179 v.val = sk->sk_priority;
1180 break;
1181
1182 case SO_LINGER:
1183 lv = sizeof(v.ling);
1184 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1185 v.ling.l_linger = sk->sk_lingertime / HZ;
1186 break;
1187
1188 case SO_BSDCOMPAT:
1189 sock_warn_obsolete_bsdism("getsockopt");
1190 break;
1191
1192 case SO_TIMESTAMP:
1193 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1194 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1195 break;
1196
1197 case SO_TIMESTAMPNS:
1198 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1199 break;
1200
1201 case SO_TIMESTAMPING:
1202 v.val = sk->sk_tsflags;
1203 break;
1204
1205 case SO_RCVTIMEO:
1206 lv = sizeof(struct timeval);
1207 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1208 v.tm.tv_sec = 0;
1209 v.tm.tv_usec = 0;
1210 } else {
1211 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1212 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1213 }
1214 break;
1215
1216 case SO_SNDTIMEO:
1217 lv = sizeof(struct timeval);
1218 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1219 v.tm.tv_sec = 0;
1220 v.tm.tv_usec = 0;
1221 } else {
1222 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1223 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1224 }
1225 break;
1226
1227 case SO_RCVLOWAT:
1228 v.val = sk->sk_rcvlowat;
1229 break;
1230
1231 case SO_SNDLOWAT:
1232 v.val = 1;
1233 break;
1234
1235 case SO_PASSCRED:
1236 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1237 break;
1238
1239 case SO_PEERCRED:
1240 {
1241 struct ucred peercred;
1242 if (len > sizeof(peercred))
1243 len = sizeof(peercred);
1244 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1245 if (copy_to_user(optval, &peercred, len))
1246 return -EFAULT;
1247 goto lenout;
1248 }
1249
1250 case SO_PEERGROUPS:
1251 {
1252 int ret, n;
1253
1254 if (!sk->sk_peer_cred)
1255 return -ENODATA;
1256
1257 n = sk->sk_peer_cred->group_info->ngroups;
1258 if (len < n * sizeof(gid_t)) {
1259 len = n * sizeof(gid_t);
1260 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1261 }
1262 len = n * sizeof(gid_t);
1263
1264 ret = groups_to_user((gid_t __user *)optval,
1265 sk->sk_peer_cred->group_info);
1266 if (ret)
1267 return ret;
1268 goto lenout;
1269 }
1270
1271 case SO_PEERNAME:
1272 {
1273 char address[128];
1274
1275 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1276 return -ENOTCONN;
1277 if (lv < len)
1278 return -EINVAL;
1279 if (copy_to_user(optval, address, len))
1280 return -EFAULT;
1281 goto lenout;
1282 }
1283
1284 /* Dubious BSD thing... Probably nobody even uses it, but
1285 * the UNIX standard wants it for whatever reason... -DaveM
1286 */
1287 case SO_ACCEPTCONN:
1288 v.val = sk->sk_state == TCP_LISTEN;
1289 break;
1290
1291 case SO_PASSSEC:
1292 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1293 break;
1294
1295 case SO_PEERSEC:
1296 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1297
1298 case SO_MARK:
1299 v.val = sk->sk_mark;
1300 break;
1301
1302 case SO_RXQ_OVFL:
1303 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1304 break;
1305
1306 case SO_WIFI_STATUS:
1307 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1308 break;
1309
1310 case SO_PEEK_OFF:
1311 if (!sock->ops->set_peek_off)
1312 return -EOPNOTSUPP;
1313
1314 v.val = sk->sk_peek_off;
1315 break;
1316 case SO_NOFCS:
1317 v.val = sock_flag(sk, SOCK_NOFCS);
1318 break;
1319
1320 case SO_BINDTODEVICE:
1321 return sock_getbindtodevice(sk, optval, optlen, len);
1322
1323 case SO_GET_FILTER:
1324 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1325 if (len < 0)
1326 return len;
1327
1328 goto lenout;
1329
1330 case SO_LOCK_FILTER:
1331 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1332 break;
1333
1334 case SO_BPF_EXTENSIONS:
1335 v.val = bpf_tell_extensions();
1336 break;
1337
1338 case SO_SELECT_ERR_QUEUE:
1339 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1340 break;
1341
1342 #ifdef CONFIG_NET_RX_BUSY_POLL
1343 case SO_BUSY_POLL:
1344 v.val = sk->sk_ll_usec;
1345 break;
1346 #endif
1347
1348 case SO_MAX_PACING_RATE:
1349 v.val = sk->sk_max_pacing_rate;
1350 break;
1351
1352 case SO_INCOMING_CPU:
1353 v.val = sk->sk_incoming_cpu;
1354 break;
1355
1356 case SO_MEMINFO:
1357 {
1358 u32 meminfo[SK_MEMINFO_VARS];
1359
1360 if (get_user(len, optlen))
1361 return -EFAULT;
1362
1363 sk_get_meminfo(sk, meminfo);
1364
1365 len = min_t(unsigned int, len, sizeof(meminfo));
1366 if (copy_to_user(optval, &meminfo, len))
1367 return -EFAULT;
1368
1369 goto lenout;
1370 }
1371
1372 #ifdef CONFIG_NET_RX_BUSY_POLL
1373 case SO_INCOMING_NAPI_ID:
1374 v.val = READ_ONCE(sk->sk_napi_id);
1375
1376 /* aggregate non-NAPI IDs down to 0 */
1377 if (v.val < MIN_NAPI_ID)
1378 v.val = 0;
1379
1380 break;
1381 #endif
1382
1383 case SO_COOKIE:
1384 lv = sizeof(u64);
1385 if (len < lv)
1386 return -EINVAL;
1387 v.val64 = sock_gen_cookie(sk);
1388 break;
1389
1390 case SO_ZEROCOPY:
1391 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1392 break;
1393
1394 default:
1395 /* We implement the SO_SNDLOWAT etc to not be settable
1396 * (1003.1g 7).
1397 */
1398 return -ENOPROTOOPT;
1399 }
1400
1401 if (len > lv)
1402 len = lv;
1403 if (copy_to_user(optval, &v, len))
1404 return -EFAULT;
1405 lenout:
1406 if (put_user(len, optlen))
1407 return -EFAULT;
1408 return 0;
1409 }
1410
1411 /*
1412 * Initialize an sk_lock.
1413 *
1414 * (We also register the sk_lock with the lock validator.)
1415 */
1416 static inline void sock_lock_init(struct sock *sk)
1417 {
1418 if (sk->sk_kern_sock)
1419 sock_lock_init_class_and_name(
1420 sk,
1421 af_family_kern_slock_key_strings[sk->sk_family],
1422 af_family_kern_slock_keys + sk->sk_family,
1423 af_family_kern_key_strings[sk->sk_family],
1424 af_family_kern_keys + sk->sk_family);
1425 else
1426 sock_lock_init_class_and_name(
1427 sk,
1428 af_family_slock_key_strings[sk->sk_family],
1429 af_family_slock_keys + sk->sk_family,
1430 af_family_key_strings[sk->sk_family],
1431 af_family_keys + sk->sk_family);
1432 }
1433
1434 /*
1435 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1436 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1437 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1438 */
1439 static void sock_copy(struct sock *nsk, const struct sock *osk)
1440 {
1441 #ifdef CONFIG_SECURITY_NETWORK
1442 void *sptr = nsk->sk_security;
1443 #endif
1444 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1445
1446 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1447 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1448
1449 #ifdef CONFIG_SECURITY_NETWORK
1450 nsk->sk_security = sptr;
1451 security_sk_clone(osk, nsk);
1452 #endif
1453 }
1454
1455 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1456 int family)
1457 {
1458 struct sock *sk;
1459 struct kmem_cache *slab;
1460
1461 slab = prot->slab;
1462 if (slab != NULL) {
1463 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1464 if (!sk)
1465 return sk;
1466 if (priority & __GFP_ZERO)
1467 sk_prot_clear_nulls(sk, prot->obj_size);
1468 } else
1469 sk = kmalloc(prot->obj_size, priority);
1470
1471 if (sk != NULL) {
1472 if (security_sk_alloc(sk, family, priority))
1473 goto out_free;
1474
1475 if (!try_module_get(prot->owner))
1476 goto out_free_sec;
1477 sk_tx_queue_clear(sk);
1478 }
1479
1480 return sk;
1481
1482 out_free_sec:
1483 security_sk_free(sk);
1484 out_free:
1485 if (slab != NULL)
1486 kmem_cache_free(slab, sk);
1487 else
1488 kfree(sk);
1489 return NULL;
1490 }
1491
1492 static void sk_prot_free(struct proto *prot, struct sock *sk)
1493 {
1494 struct kmem_cache *slab;
1495 struct module *owner;
1496
1497 owner = prot->owner;
1498 slab = prot->slab;
1499
1500 cgroup_sk_free(&sk->sk_cgrp_data);
1501 mem_cgroup_sk_free(sk);
1502 security_sk_free(sk);
1503 if (slab != NULL)
1504 kmem_cache_free(slab, sk);
1505 else
1506 kfree(sk);
1507 module_put(owner);
1508 }
1509
1510 /**
1511 * sk_alloc - All socket objects are allocated here
1512 * @net: the applicable net namespace
1513 * @family: protocol family
1514 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1515 * @prot: struct proto associated with this new sock instance
1516 * @kern: is this to be a kernel socket?
1517 */
1518 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1519 struct proto *prot, int kern)
1520 {
1521 struct sock *sk;
1522
1523 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1524 if (sk) {
1525 sk->sk_family = family;
1526 /*
1527 * See comment in struct sock definition to understand
1528 * why we need sk_prot_creator -acme
1529 */
1530 sk->sk_prot = sk->sk_prot_creator = prot;
1531 sk->sk_kern_sock = kern;
1532 sock_lock_init(sk);
1533 sk->sk_net_refcnt = kern ? 0 : 1;
1534 if (likely(sk->sk_net_refcnt))
1535 get_net(net);
1536 sock_net_set(sk, net);
1537 refcount_set(&sk->sk_wmem_alloc, 1);
1538
1539 mem_cgroup_sk_alloc(sk);
1540 cgroup_sk_alloc(&sk->sk_cgrp_data);
1541 sock_update_classid(&sk->sk_cgrp_data);
1542 sock_update_netprioidx(&sk->sk_cgrp_data);
1543 }
1544
1545 return sk;
1546 }
1547 EXPORT_SYMBOL(sk_alloc);
1548
1549 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1550 * grace period. This is the case for UDP sockets and TCP listeners.
1551 */
1552 static void __sk_destruct(struct rcu_head *head)
1553 {
1554 struct sock *sk = container_of(head, struct sock, sk_rcu);
1555 struct sk_filter *filter;
1556
1557 if (sk->sk_destruct)
1558 sk->sk_destruct(sk);
1559
1560 filter = rcu_dereference_check(sk->sk_filter,
1561 refcount_read(&sk->sk_wmem_alloc) == 0);
1562 if (filter) {
1563 sk_filter_uncharge(sk, filter);
1564 RCU_INIT_POINTER(sk->sk_filter, NULL);
1565 }
1566 if (rcu_access_pointer(sk->sk_reuseport_cb))
1567 reuseport_detach_sock(sk);
1568
1569 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1570
1571 if (atomic_read(&sk->sk_omem_alloc))
1572 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1573 __func__, atomic_read(&sk->sk_omem_alloc));
1574
1575 if (sk->sk_frag.page) {
1576 put_page(sk->sk_frag.page);
1577 sk->sk_frag.page = NULL;
1578 }
1579
1580 if (sk->sk_peer_cred)
1581 put_cred(sk->sk_peer_cred);
1582 put_pid(sk->sk_peer_pid);
1583 if (likely(sk->sk_net_refcnt))
1584 put_net(sock_net(sk));
1585 sk_prot_free(sk->sk_prot_creator, sk);
1586 }
1587
1588 void sk_destruct(struct sock *sk)
1589 {
1590 if (sock_flag(sk, SOCK_RCU_FREE))
1591 call_rcu(&sk->sk_rcu, __sk_destruct);
1592 else
1593 __sk_destruct(&sk->sk_rcu);
1594 }
1595
1596 static void __sk_free(struct sock *sk)
1597 {
1598 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1599 sock_diag_broadcast_destroy(sk);
1600 else
1601 sk_destruct(sk);
1602 }
1603
1604 void sk_free(struct sock *sk)
1605 {
1606 /*
1607 * We subtract one from sk_wmem_alloc and can know if
1608 * some packets are still in some tx queue.
1609 * If not null, sock_wfree() will call __sk_free(sk) later
1610 */
1611 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1612 __sk_free(sk);
1613 }
1614 EXPORT_SYMBOL(sk_free);
1615
1616 static void sk_init_common(struct sock *sk)
1617 {
1618 skb_queue_head_init(&sk->sk_receive_queue);
1619 skb_queue_head_init(&sk->sk_write_queue);
1620 skb_queue_head_init(&sk->sk_error_queue);
1621
1622 rwlock_init(&sk->sk_callback_lock);
1623 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1624 af_rlock_keys + sk->sk_family,
1625 af_family_rlock_key_strings[sk->sk_family]);
1626 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1627 af_wlock_keys + sk->sk_family,
1628 af_family_wlock_key_strings[sk->sk_family]);
1629 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1630 af_elock_keys + sk->sk_family,
1631 af_family_elock_key_strings[sk->sk_family]);
1632 lockdep_set_class_and_name(&sk->sk_callback_lock,
1633 af_callback_keys + sk->sk_family,
1634 af_family_clock_key_strings[sk->sk_family]);
1635 }
1636
1637 /**
1638 * sk_clone_lock - clone a socket, and lock its clone
1639 * @sk: the socket to clone
1640 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1641 *
1642 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1643 */
1644 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1645 {
1646 struct sock *newsk;
1647 bool is_charged = true;
1648
1649 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1650 if (newsk != NULL) {
1651 struct sk_filter *filter;
1652
1653 sock_copy(newsk, sk);
1654
1655 newsk->sk_prot_creator = sk->sk_prot;
1656
1657 /* SANITY */
1658 if (likely(newsk->sk_net_refcnt))
1659 get_net(sock_net(newsk));
1660 sk_node_init(&newsk->sk_node);
1661 sock_lock_init(newsk);
1662 bh_lock_sock(newsk);
1663 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1664 newsk->sk_backlog.len = 0;
1665
1666 atomic_set(&newsk->sk_rmem_alloc, 0);
1667 /*
1668 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1669 */
1670 refcount_set(&newsk->sk_wmem_alloc, 1);
1671 atomic_set(&newsk->sk_omem_alloc, 0);
1672 sk_init_common(newsk);
1673
1674 newsk->sk_dst_cache = NULL;
1675 newsk->sk_dst_pending_confirm = 0;
1676 newsk->sk_wmem_queued = 0;
1677 newsk->sk_forward_alloc = 0;
1678 atomic_set(&newsk->sk_drops, 0);
1679 newsk->sk_send_head = NULL;
1680 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1681 atomic_set(&newsk->sk_zckey, 0);
1682
1683 sock_reset_flag(newsk, SOCK_DONE);
1684 mem_cgroup_sk_alloc(newsk);
1685 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1686
1687 rcu_read_lock();
1688 filter = rcu_dereference(sk->sk_filter);
1689 if (filter != NULL)
1690 /* though it's an empty new sock, the charging may fail
1691 * if sysctl_optmem_max was changed between creation of
1692 * original socket and cloning
1693 */
1694 is_charged = sk_filter_charge(newsk, filter);
1695 RCU_INIT_POINTER(newsk->sk_filter, filter);
1696 rcu_read_unlock();
1697
1698 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1699 /* We need to make sure that we don't uncharge the new
1700 * socket if we couldn't charge it in the first place
1701 * as otherwise we uncharge the parent's filter.
1702 */
1703 if (!is_charged)
1704 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1705 sk_free_unlock_clone(newsk);
1706 newsk = NULL;
1707 goto out;
1708 }
1709 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1710
1711 newsk->sk_err = 0;
1712 newsk->sk_err_soft = 0;
1713 newsk->sk_priority = 0;
1714 newsk->sk_incoming_cpu = raw_smp_processor_id();
1715 atomic64_set(&newsk->sk_cookie, 0);
1716
1717 /*
1718 * Before updating sk_refcnt, we must commit prior changes to memory
1719 * (Documentation/RCU/rculist_nulls.txt for details)
1720 */
1721 smp_wmb();
1722 refcount_set(&newsk->sk_refcnt, 2);
1723
1724 /*
1725 * Increment the counter in the same struct proto as the master
1726 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1727 * is the same as sk->sk_prot->socks, as this field was copied
1728 * with memcpy).
1729 *
1730 * This _changes_ the previous behaviour, where
1731 * tcp_create_openreq_child always was incrementing the
1732 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1733 * to be taken into account in all callers. -acme
1734 */
1735 sk_refcnt_debug_inc(newsk);
1736 sk_set_socket(newsk, NULL);
1737 newsk->sk_wq = NULL;
1738
1739 if (newsk->sk_prot->sockets_allocated)
1740 sk_sockets_allocated_inc(newsk);
1741
1742 if (sock_needs_netstamp(sk) &&
1743 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1744 net_enable_timestamp();
1745 }
1746 out:
1747 return newsk;
1748 }
1749 EXPORT_SYMBOL_GPL(sk_clone_lock);
1750
1751 void sk_free_unlock_clone(struct sock *sk)
1752 {
1753 /* It is still raw copy of parent, so invalidate
1754 * destructor and make plain sk_free() */
1755 sk->sk_destruct = NULL;
1756 bh_unlock_sock(sk);
1757 sk_free(sk);
1758 }
1759 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1760
1761 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1762 {
1763 u32 max_segs = 1;
1764
1765 sk_dst_set(sk, dst);
1766 sk->sk_route_caps = dst->dev->features;
1767 if (sk->sk_route_caps & NETIF_F_GSO)
1768 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1769 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1770 if (sk_can_gso(sk)) {
1771 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1772 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1773 } else {
1774 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1775 sk->sk_gso_max_size = dst->dev->gso_max_size;
1776 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1777 }
1778 }
1779 sk->sk_gso_max_segs = max_segs;
1780 }
1781 EXPORT_SYMBOL_GPL(sk_setup_caps);
1782
1783 /*
1784 * Simple resource managers for sockets.
1785 */
1786
1787
1788 /*
1789 * Write buffer destructor automatically called from kfree_skb.
1790 */
1791 void sock_wfree(struct sk_buff *skb)
1792 {
1793 struct sock *sk = skb->sk;
1794 unsigned int len = skb->truesize;
1795
1796 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1797 /*
1798 * Keep a reference on sk_wmem_alloc, this will be released
1799 * after sk_write_space() call
1800 */
1801 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1802 sk->sk_write_space(sk);
1803 len = 1;
1804 }
1805 /*
1806 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1807 * could not do because of in-flight packets
1808 */
1809 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1810 __sk_free(sk);
1811 }
1812 EXPORT_SYMBOL(sock_wfree);
1813
1814 /* This variant of sock_wfree() is used by TCP,
1815 * since it sets SOCK_USE_WRITE_QUEUE.
1816 */
1817 void __sock_wfree(struct sk_buff *skb)
1818 {
1819 struct sock *sk = skb->sk;
1820
1821 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1822 __sk_free(sk);
1823 }
1824
1825 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1826 {
1827 skb_orphan(skb);
1828 skb->sk = sk;
1829 #ifdef CONFIG_INET
1830 if (unlikely(!sk_fullsock(sk))) {
1831 skb->destructor = sock_edemux;
1832 sock_hold(sk);
1833 return;
1834 }
1835 #endif
1836 skb->destructor = sock_wfree;
1837 skb_set_hash_from_sk(skb, sk);
1838 /*
1839 * We used to take a refcount on sk, but following operation
1840 * is enough to guarantee sk_free() wont free this sock until
1841 * all in-flight packets are completed
1842 */
1843 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1844 }
1845 EXPORT_SYMBOL(skb_set_owner_w);
1846
1847 /* This helper is used by netem, as it can hold packets in its
1848 * delay queue. We want to allow the owner socket to send more
1849 * packets, as if they were already TX completed by a typical driver.
1850 * But we also want to keep skb->sk set because some packet schedulers
1851 * rely on it (sch_fq for example).
1852 */
1853 void skb_orphan_partial(struct sk_buff *skb)
1854 {
1855 if (skb_is_tcp_pure_ack(skb))
1856 return;
1857
1858 if (skb->destructor == sock_wfree
1859 #ifdef CONFIG_INET
1860 || skb->destructor == tcp_wfree
1861 #endif
1862 ) {
1863 struct sock *sk = skb->sk;
1864
1865 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1866 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
1867 skb->destructor = sock_efree;
1868 }
1869 } else {
1870 skb_orphan(skb);
1871 }
1872 }
1873 EXPORT_SYMBOL(skb_orphan_partial);
1874
1875 /*
1876 * Read buffer destructor automatically called from kfree_skb.
1877 */
1878 void sock_rfree(struct sk_buff *skb)
1879 {
1880 struct sock *sk = skb->sk;
1881 unsigned int len = skb->truesize;
1882
1883 atomic_sub(len, &sk->sk_rmem_alloc);
1884 sk_mem_uncharge(sk, len);
1885 }
1886 EXPORT_SYMBOL(sock_rfree);
1887
1888 /*
1889 * Buffer destructor for skbs that are not used directly in read or write
1890 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1891 */
1892 void sock_efree(struct sk_buff *skb)
1893 {
1894 sock_put(skb->sk);
1895 }
1896 EXPORT_SYMBOL(sock_efree);
1897
1898 kuid_t sock_i_uid(struct sock *sk)
1899 {
1900 kuid_t uid;
1901
1902 read_lock_bh(&sk->sk_callback_lock);
1903 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1904 read_unlock_bh(&sk->sk_callback_lock);
1905 return uid;
1906 }
1907 EXPORT_SYMBOL(sock_i_uid);
1908
1909 unsigned long sock_i_ino(struct sock *sk)
1910 {
1911 unsigned long ino;
1912
1913 read_lock_bh(&sk->sk_callback_lock);
1914 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1915 read_unlock_bh(&sk->sk_callback_lock);
1916 return ino;
1917 }
1918 EXPORT_SYMBOL(sock_i_ino);
1919
1920 /*
1921 * Allocate a skb from the socket's send buffer.
1922 */
1923 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1924 gfp_t priority)
1925 {
1926 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1927 struct sk_buff *skb = alloc_skb(size, priority);
1928 if (skb) {
1929 skb_set_owner_w(skb, sk);
1930 return skb;
1931 }
1932 }
1933 return NULL;
1934 }
1935 EXPORT_SYMBOL(sock_wmalloc);
1936
1937 static void sock_ofree(struct sk_buff *skb)
1938 {
1939 struct sock *sk = skb->sk;
1940
1941 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
1942 }
1943
1944 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1945 gfp_t priority)
1946 {
1947 struct sk_buff *skb;
1948
1949 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
1950 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
1951 sysctl_optmem_max)
1952 return NULL;
1953
1954 skb = alloc_skb(size, priority);
1955 if (!skb)
1956 return NULL;
1957
1958 atomic_add(skb->truesize, &sk->sk_omem_alloc);
1959 skb->sk = sk;
1960 skb->destructor = sock_ofree;
1961 return skb;
1962 }
1963
1964 /*
1965 * Allocate a memory block from the socket's option memory buffer.
1966 */
1967 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1968 {
1969 if ((unsigned int)size <= sysctl_optmem_max &&
1970 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1971 void *mem;
1972 /* First do the add, to avoid the race if kmalloc
1973 * might sleep.
1974 */
1975 atomic_add(size, &sk->sk_omem_alloc);
1976 mem = kmalloc(size, priority);
1977 if (mem)
1978 return mem;
1979 atomic_sub(size, &sk->sk_omem_alloc);
1980 }
1981 return NULL;
1982 }
1983 EXPORT_SYMBOL(sock_kmalloc);
1984
1985 /* Free an option memory block. Note, we actually want the inline
1986 * here as this allows gcc to detect the nullify and fold away the
1987 * condition entirely.
1988 */
1989 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1990 const bool nullify)
1991 {
1992 if (WARN_ON_ONCE(!mem))
1993 return;
1994 if (nullify)
1995 kzfree(mem);
1996 else
1997 kfree(mem);
1998 atomic_sub(size, &sk->sk_omem_alloc);
1999 }
2000
2001 void sock_kfree_s(struct sock *sk, void *mem, int size)
2002 {
2003 __sock_kfree_s(sk, mem, size, false);
2004 }
2005 EXPORT_SYMBOL(sock_kfree_s);
2006
2007 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2008 {
2009 __sock_kfree_s(sk, mem, size, true);
2010 }
2011 EXPORT_SYMBOL(sock_kzfree_s);
2012
2013 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2014 I think, these locks should be removed for datagram sockets.
2015 */
2016 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2017 {
2018 DEFINE_WAIT(wait);
2019
2020 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2021 for (;;) {
2022 if (!timeo)
2023 break;
2024 if (signal_pending(current))
2025 break;
2026 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2027 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2028 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2029 break;
2030 if (sk->sk_shutdown & SEND_SHUTDOWN)
2031 break;
2032 if (sk->sk_err)
2033 break;
2034 timeo = schedule_timeout(timeo);
2035 }
2036 finish_wait(sk_sleep(sk), &wait);
2037 return timeo;
2038 }
2039
2040
2041 /*
2042 * Generic send/receive buffer handlers
2043 */
2044
2045 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2046 unsigned long data_len, int noblock,
2047 int *errcode, int max_page_order)
2048 {
2049 struct sk_buff *skb;
2050 long timeo;
2051 int err;
2052
2053 timeo = sock_sndtimeo(sk, noblock);
2054 for (;;) {
2055 err = sock_error(sk);
2056 if (err != 0)
2057 goto failure;
2058
2059 err = -EPIPE;
2060 if (sk->sk_shutdown & SEND_SHUTDOWN)
2061 goto failure;
2062
2063 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2064 break;
2065
2066 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2067 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2068 err = -EAGAIN;
2069 if (!timeo)
2070 goto failure;
2071 if (signal_pending(current))
2072 goto interrupted;
2073 timeo = sock_wait_for_wmem(sk, timeo);
2074 }
2075 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2076 errcode, sk->sk_allocation);
2077 if (skb)
2078 skb_set_owner_w(skb, sk);
2079 return skb;
2080
2081 interrupted:
2082 err = sock_intr_errno(timeo);
2083 failure:
2084 *errcode = err;
2085 return NULL;
2086 }
2087 EXPORT_SYMBOL(sock_alloc_send_pskb);
2088
2089 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2090 int noblock, int *errcode)
2091 {
2092 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2093 }
2094 EXPORT_SYMBOL(sock_alloc_send_skb);
2095
2096 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2097 struct sockcm_cookie *sockc)
2098 {
2099 u32 tsflags;
2100
2101 switch (cmsg->cmsg_type) {
2102 case SO_MARK:
2103 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2104 return -EPERM;
2105 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2106 return -EINVAL;
2107 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2108 break;
2109 case SO_TIMESTAMPING:
2110 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2111 return -EINVAL;
2112
2113 tsflags = *(u32 *)CMSG_DATA(cmsg);
2114 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2115 return -EINVAL;
2116
2117 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2118 sockc->tsflags |= tsflags;
2119 break;
2120 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2121 case SCM_RIGHTS:
2122 case SCM_CREDENTIALS:
2123 break;
2124 default:
2125 return -EINVAL;
2126 }
2127 return 0;
2128 }
2129 EXPORT_SYMBOL(__sock_cmsg_send);
2130
2131 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2132 struct sockcm_cookie *sockc)
2133 {
2134 struct cmsghdr *cmsg;
2135 int ret;
2136
2137 for_each_cmsghdr(cmsg, msg) {
2138 if (!CMSG_OK(msg, cmsg))
2139 return -EINVAL;
2140 if (cmsg->cmsg_level != SOL_SOCKET)
2141 continue;
2142 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2143 if (ret)
2144 return ret;
2145 }
2146 return 0;
2147 }
2148 EXPORT_SYMBOL(sock_cmsg_send);
2149
2150 static void sk_enter_memory_pressure(struct sock *sk)
2151 {
2152 if (!sk->sk_prot->enter_memory_pressure)
2153 return;
2154
2155 sk->sk_prot->enter_memory_pressure(sk);
2156 }
2157
2158 static void sk_leave_memory_pressure(struct sock *sk)
2159 {
2160 if (sk->sk_prot->leave_memory_pressure) {
2161 sk->sk_prot->leave_memory_pressure(sk);
2162 } else {
2163 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2164
2165 if (memory_pressure && *memory_pressure)
2166 *memory_pressure = 0;
2167 }
2168 }
2169
2170 /* On 32bit arches, an skb frag is limited to 2^15 */
2171 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2172
2173 /**
2174 * skb_page_frag_refill - check that a page_frag contains enough room
2175 * @sz: minimum size of the fragment we want to get
2176 * @pfrag: pointer to page_frag
2177 * @gfp: priority for memory allocation
2178 *
2179 * Note: While this allocator tries to use high order pages, there is
2180 * no guarantee that allocations succeed. Therefore, @sz MUST be
2181 * less or equal than PAGE_SIZE.
2182 */
2183 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2184 {
2185 if (pfrag->page) {
2186 if (page_ref_count(pfrag->page) == 1) {
2187 pfrag->offset = 0;
2188 return true;
2189 }
2190 if (pfrag->offset + sz <= pfrag->size)
2191 return true;
2192 put_page(pfrag->page);
2193 }
2194
2195 pfrag->offset = 0;
2196 if (SKB_FRAG_PAGE_ORDER) {
2197 /* Avoid direct reclaim but allow kswapd to wake */
2198 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2199 __GFP_COMP | __GFP_NOWARN |
2200 __GFP_NORETRY,
2201 SKB_FRAG_PAGE_ORDER);
2202 if (likely(pfrag->page)) {
2203 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2204 return true;
2205 }
2206 }
2207 pfrag->page = alloc_page(gfp);
2208 if (likely(pfrag->page)) {
2209 pfrag->size = PAGE_SIZE;
2210 return true;
2211 }
2212 return false;
2213 }
2214 EXPORT_SYMBOL(skb_page_frag_refill);
2215
2216 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2217 {
2218 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2219 return true;
2220
2221 sk_enter_memory_pressure(sk);
2222 sk_stream_moderate_sndbuf(sk);
2223 return false;
2224 }
2225 EXPORT_SYMBOL(sk_page_frag_refill);
2226
2227 static void __lock_sock(struct sock *sk)
2228 __releases(&sk->sk_lock.slock)
2229 __acquires(&sk->sk_lock.slock)
2230 {
2231 DEFINE_WAIT(wait);
2232
2233 for (;;) {
2234 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2235 TASK_UNINTERRUPTIBLE);
2236 spin_unlock_bh(&sk->sk_lock.slock);
2237 schedule();
2238 spin_lock_bh(&sk->sk_lock.slock);
2239 if (!sock_owned_by_user(sk))
2240 break;
2241 }
2242 finish_wait(&sk->sk_lock.wq, &wait);
2243 }
2244
2245 static void __release_sock(struct sock *sk)
2246 __releases(&sk->sk_lock.slock)
2247 __acquires(&sk->sk_lock.slock)
2248 {
2249 struct sk_buff *skb, *next;
2250
2251 while ((skb = sk->sk_backlog.head) != NULL) {
2252 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2253
2254 spin_unlock_bh(&sk->sk_lock.slock);
2255
2256 do {
2257 next = skb->next;
2258 prefetch(next);
2259 WARN_ON_ONCE(skb_dst_is_noref(skb));
2260 skb->next = NULL;
2261 sk_backlog_rcv(sk, skb);
2262
2263 cond_resched();
2264
2265 skb = next;
2266 } while (skb != NULL);
2267
2268 spin_lock_bh(&sk->sk_lock.slock);
2269 }
2270
2271 /*
2272 * Doing the zeroing here guarantee we can not loop forever
2273 * while a wild producer attempts to flood us.
2274 */
2275 sk->sk_backlog.len = 0;
2276 }
2277
2278 void __sk_flush_backlog(struct sock *sk)
2279 {
2280 spin_lock_bh(&sk->sk_lock.slock);
2281 __release_sock(sk);
2282 spin_unlock_bh(&sk->sk_lock.slock);
2283 }
2284
2285 /**
2286 * sk_wait_data - wait for data to arrive at sk_receive_queue
2287 * @sk: sock to wait on
2288 * @timeo: for how long
2289 * @skb: last skb seen on sk_receive_queue
2290 *
2291 * Now socket state including sk->sk_err is changed only under lock,
2292 * hence we may omit checks after joining wait queue.
2293 * We check receive queue before schedule() only as optimization;
2294 * it is very likely that release_sock() added new data.
2295 */
2296 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2297 {
2298 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2299 int rc;
2300
2301 add_wait_queue(sk_sleep(sk), &wait);
2302 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2303 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2304 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2305 remove_wait_queue(sk_sleep(sk), &wait);
2306 return rc;
2307 }
2308 EXPORT_SYMBOL(sk_wait_data);
2309
2310 /**
2311 * __sk_mem_raise_allocated - increase memory_allocated
2312 * @sk: socket
2313 * @size: memory size to allocate
2314 * @amt: pages to allocate
2315 * @kind: allocation type
2316 *
2317 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2318 */
2319 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2320 {
2321 struct proto *prot = sk->sk_prot;
2322 long allocated = sk_memory_allocated_add(sk, amt);
2323
2324 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2325 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2326 goto suppress_allocation;
2327
2328 /* Under limit. */
2329 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2330 sk_leave_memory_pressure(sk);
2331 return 1;
2332 }
2333
2334 /* Under pressure. */
2335 if (allocated > sk_prot_mem_limits(sk, 1))
2336 sk_enter_memory_pressure(sk);
2337
2338 /* Over hard limit. */
2339 if (allocated > sk_prot_mem_limits(sk, 2))
2340 goto suppress_allocation;
2341
2342 /* guarantee minimum buffer size under pressure */
2343 if (kind == SK_MEM_RECV) {
2344 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2345 return 1;
2346
2347 } else { /* SK_MEM_SEND */
2348 int wmem0 = sk_get_wmem0(sk, prot);
2349
2350 if (sk->sk_type == SOCK_STREAM) {
2351 if (sk->sk_wmem_queued < wmem0)
2352 return 1;
2353 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2354 return 1;
2355 }
2356 }
2357
2358 if (sk_has_memory_pressure(sk)) {
2359 int alloc;
2360
2361 if (!sk_under_memory_pressure(sk))
2362 return 1;
2363 alloc = sk_sockets_allocated_read_positive(sk);
2364 if (sk_prot_mem_limits(sk, 2) > alloc *
2365 sk_mem_pages(sk->sk_wmem_queued +
2366 atomic_read(&sk->sk_rmem_alloc) +
2367 sk->sk_forward_alloc))
2368 return 1;
2369 }
2370
2371 suppress_allocation:
2372
2373 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2374 sk_stream_moderate_sndbuf(sk);
2375
2376 /* Fail only if socket is _under_ its sndbuf.
2377 * In this case we cannot block, so that we have to fail.
2378 */
2379 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2380 return 1;
2381 }
2382
2383 trace_sock_exceed_buf_limit(sk, prot, allocated);
2384
2385 sk_memory_allocated_sub(sk, amt);
2386
2387 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2388 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2389
2390 return 0;
2391 }
2392 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2393
2394 /**
2395 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2396 * @sk: socket
2397 * @size: memory size to allocate
2398 * @kind: allocation type
2399 *
2400 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2401 * rmem allocation. This function assumes that protocols which have
2402 * memory_pressure use sk_wmem_queued as write buffer accounting.
2403 */
2404 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2405 {
2406 int ret, amt = sk_mem_pages(size);
2407
2408 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2409 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2410 if (!ret)
2411 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2412 return ret;
2413 }
2414 EXPORT_SYMBOL(__sk_mem_schedule);
2415
2416 /**
2417 * __sk_mem_reduce_allocated - reclaim memory_allocated
2418 * @sk: socket
2419 * @amount: number of quanta
2420 *
2421 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2422 */
2423 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2424 {
2425 sk_memory_allocated_sub(sk, amount);
2426
2427 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2428 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2429
2430 if (sk_under_memory_pressure(sk) &&
2431 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2432 sk_leave_memory_pressure(sk);
2433 }
2434 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2435
2436 /**
2437 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2438 * @sk: socket
2439 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2440 */
2441 void __sk_mem_reclaim(struct sock *sk, int amount)
2442 {
2443 amount >>= SK_MEM_QUANTUM_SHIFT;
2444 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2445 __sk_mem_reduce_allocated(sk, amount);
2446 }
2447 EXPORT_SYMBOL(__sk_mem_reclaim);
2448
2449 int sk_set_peek_off(struct sock *sk, int val)
2450 {
2451 sk->sk_peek_off = val;
2452 return 0;
2453 }
2454 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2455
2456 /*
2457 * Set of default routines for initialising struct proto_ops when
2458 * the protocol does not support a particular function. In certain
2459 * cases where it makes no sense for a protocol to have a "do nothing"
2460 * function, some default processing is provided.
2461 */
2462
2463 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2464 {
2465 return -EOPNOTSUPP;
2466 }
2467 EXPORT_SYMBOL(sock_no_bind);
2468
2469 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2470 int len, int flags)
2471 {
2472 return -EOPNOTSUPP;
2473 }
2474 EXPORT_SYMBOL(sock_no_connect);
2475
2476 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2477 {
2478 return -EOPNOTSUPP;
2479 }
2480 EXPORT_SYMBOL(sock_no_socketpair);
2481
2482 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2483 bool kern)
2484 {
2485 return -EOPNOTSUPP;
2486 }
2487 EXPORT_SYMBOL(sock_no_accept);
2488
2489 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2490 int *len, int peer)
2491 {
2492 return -EOPNOTSUPP;
2493 }
2494 EXPORT_SYMBOL(sock_no_getname);
2495
2496 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2497 {
2498 return 0;
2499 }
2500 EXPORT_SYMBOL(sock_no_poll);
2501
2502 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2503 {
2504 return -EOPNOTSUPP;
2505 }
2506 EXPORT_SYMBOL(sock_no_ioctl);
2507
2508 int sock_no_listen(struct socket *sock, int backlog)
2509 {
2510 return -EOPNOTSUPP;
2511 }
2512 EXPORT_SYMBOL(sock_no_listen);
2513
2514 int sock_no_shutdown(struct socket *sock, int how)
2515 {
2516 return -EOPNOTSUPP;
2517 }
2518 EXPORT_SYMBOL(sock_no_shutdown);
2519
2520 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2521 char __user *optval, unsigned int optlen)
2522 {
2523 return -EOPNOTSUPP;
2524 }
2525 EXPORT_SYMBOL(sock_no_setsockopt);
2526
2527 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2528 char __user *optval, int __user *optlen)
2529 {
2530 return -EOPNOTSUPP;
2531 }
2532 EXPORT_SYMBOL(sock_no_getsockopt);
2533
2534 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2535 {
2536 return -EOPNOTSUPP;
2537 }
2538 EXPORT_SYMBOL(sock_no_sendmsg);
2539
2540 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2541 {
2542 return -EOPNOTSUPP;
2543 }
2544 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2545
2546 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2547 int flags)
2548 {
2549 return -EOPNOTSUPP;
2550 }
2551 EXPORT_SYMBOL(sock_no_recvmsg);
2552
2553 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2554 {
2555 /* Mirror missing mmap method error code */
2556 return -ENODEV;
2557 }
2558 EXPORT_SYMBOL(sock_no_mmap);
2559
2560 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2561 {
2562 ssize_t res;
2563 struct msghdr msg = {.msg_flags = flags};
2564 struct kvec iov;
2565 char *kaddr = kmap(page);
2566 iov.iov_base = kaddr + offset;
2567 iov.iov_len = size;
2568 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2569 kunmap(page);
2570 return res;
2571 }
2572 EXPORT_SYMBOL(sock_no_sendpage);
2573
2574 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2575 int offset, size_t size, int flags)
2576 {
2577 ssize_t res;
2578 struct msghdr msg = {.msg_flags = flags};
2579 struct kvec iov;
2580 char *kaddr = kmap(page);
2581
2582 iov.iov_base = kaddr + offset;
2583 iov.iov_len = size;
2584 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2585 kunmap(page);
2586 return res;
2587 }
2588 EXPORT_SYMBOL(sock_no_sendpage_locked);
2589
2590 /*
2591 * Default Socket Callbacks
2592 */
2593
2594 static void sock_def_wakeup(struct sock *sk)
2595 {
2596 struct socket_wq *wq;
2597
2598 rcu_read_lock();
2599 wq = rcu_dereference(sk->sk_wq);
2600 if (skwq_has_sleeper(wq))
2601 wake_up_interruptible_all(&wq->wait);
2602 rcu_read_unlock();
2603 }
2604
2605 static void sock_def_error_report(struct sock *sk)
2606 {
2607 struct socket_wq *wq;
2608
2609 rcu_read_lock();
2610 wq = rcu_dereference(sk->sk_wq);
2611 if (skwq_has_sleeper(wq))
2612 wake_up_interruptible_poll(&wq->wait, POLLERR);
2613 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2614 rcu_read_unlock();
2615 }
2616
2617 static void sock_def_readable(struct sock *sk)
2618 {
2619 struct socket_wq *wq;
2620
2621 rcu_read_lock();
2622 wq = rcu_dereference(sk->sk_wq);
2623 if (skwq_has_sleeper(wq))
2624 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2625 POLLRDNORM | POLLRDBAND);
2626 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2627 rcu_read_unlock();
2628 }
2629
2630 static void sock_def_write_space(struct sock *sk)
2631 {
2632 struct socket_wq *wq;
2633
2634 rcu_read_lock();
2635
2636 /* Do not wake up a writer until he can make "significant"
2637 * progress. --DaveM
2638 */
2639 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2640 wq = rcu_dereference(sk->sk_wq);
2641 if (skwq_has_sleeper(wq))
2642 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2643 POLLWRNORM | POLLWRBAND);
2644
2645 /* Should agree with poll, otherwise some programs break */
2646 if (sock_writeable(sk))
2647 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2648 }
2649
2650 rcu_read_unlock();
2651 }
2652
2653 static void sock_def_destruct(struct sock *sk)
2654 {
2655 }
2656
2657 void sk_send_sigurg(struct sock *sk)
2658 {
2659 if (sk->sk_socket && sk->sk_socket->file)
2660 if (send_sigurg(&sk->sk_socket->file->f_owner))
2661 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2662 }
2663 EXPORT_SYMBOL(sk_send_sigurg);
2664
2665 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2666 unsigned long expires)
2667 {
2668 if (!mod_timer(timer, expires))
2669 sock_hold(sk);
2670 }
2671 EXPORT_SYMBOL(sk_reset_timer);
2672
2673 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2674 {
2675 if (del_timer(timer))
2676 __sock_put(sk);
2677 }
2678 EXPORT_SYMBOL(sk_stop_timer);
2679
2680 void sock_init_data(struct socket *sock, struct sock *sk)
2681 {
2682 sk_init_common(sk);
2683 sk->sk_send_head = NULL;
2684
2685 timer_setup(&sk->sk_timer, NULL, 0);
2686
2687 sk->sk_allocation = GFP_KERNEL;
2688 sk->sk_rcvbuf = sysctl_rmem_default;
2689 sk->sk_sndbuf = sysctl_wmem_default;
2690 sk->sk_state = TCP_CLOSE;
2691 sk_set_socket(sk, sock);
2692
2693 sock_set_flag(sk, SOCK_ZAPPED);
2694
2695 if (sock) {
2696 sk->sk_type = sock->type;
2697 sk->sk_wq = sock->wq;
2698 sock->sk = sk;
2699 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2700 } else {
2701 sk->sk_wq = NULL;
2702 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2703 }
2704
2705 rwlock_init(&sk->sk_callback_lock);
2706 if (sk->sk_kern_sock)
2707 lockdep_set_class_and_name(
2708 &sk->sk_callback_lock,
2709 af_kern_callback_keys + sk->sk_family,
2710 af_family_kern_clock_key_strings[sk->sk_family]);
2711 else
2712 lockdep_set_class_and_name(
2713 &sk->sk_callback_lock,
2714 af_callback_keys + sk->sk_family,
2715 af_family_clock_key_strings[sk->sk_family]);
2716
2717 sk->sk_state_change = sock_def_wakeup;
2718 sk->sk_data_ready = sock_def_readable;
2719 sk->sk_write_space = sock_def_write_space;
2720 sk->sk_error_report = sock_def_error_report;
2721 sk->sk_destruct = sock_def_destruct;
2722
2723 sk->sk_frag.page = NULL;
2724 sk->sk_frag.offset = 0;
2725 sk->sk_peek_off = -1;
2726
2727 sk->sk_peer_pid = NULL;
2728 sk->sk_peer_cred = NULL;
2729 sk->sk_write_pending = 0;
2730 sk->sk_rcvlowat = 1;
2731 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2732 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2733
2734 sk->sk_stamp = SK_DEFAULT_STAMP;
2735 atomic_set(&sk->sk_zckey, 0);
2736
2737 #ifdef CONFIG_NET_RX_BUSY_POLL
2738 sk->sk_napi_id = 0;
2739 sk->sk_ll_usec = sysctl_net_busy_read;
2740 #endif
2741
2742 sk->sk_max_pacing_rate = ~0U;
2743 sk->sk_pacing_rate = ~0U;
2744 sk->sk_pacing_shift = 10;
2745 sk->sk_incoming_cpu = -1;
2746 /*
2747 * Before updating sk_refcnt, we must commit prior changes to memory
2748 * (Documentation/RCU/rculist_nulls.txt for details)
2749 */
2750 smp_wmb();
2751 refcount_set(&sk->sk_refcnt, 1);
2752 atomic_set(&sk->sk_drops, 0);
2753 }
2754 EXPORT_SYMBOL(sock_init_data);
2755
2756 void lock_sock_nested(struct sock *sk, int subclass)
2757 {
2758 might_sleep();
2759 spin_lock_bh(&sk->sk_lock.slock);
2760 if (sk->sk_lock.owned)
2761 __lock_sock(sk);
2762 sk->sk_lock.owned = 1;
2763 spin_unlock(&sk->sk_lock.slock);
2764 /*
2765 * The sk_lock has mutex_lock() semantics here:
2766 */
2767 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2768 local_bh_enable();
2769 }
2770 EXPORT_SYMBOL(lock_sock_nested);
2771
2772 void release_sock(struct sock *sk)
2773 {
2774 spin_lock_bh(&sk->sk_lock.slock);
2775 if (sk->sk_backlog.tail)
2776 __release_sock(sk);
2777
2778 /* Warning : release_cb() might need to release sk ownership,
2779 * ie call sock_release_ownership(sk) before us.
2780 */
2781 if (sk->sk_prot->release_cb)
2782 sk->sk_prot->release_cb(sk);
2783
2784 sock_release_ownership(sk);
2785 if (waitqueue_active(&sk->sk_lock.wq))
2786 wake_up(&sk->sk_lock.wq);
2787 spin_unlock_bh(&sk->sk_lock.slock);
2788 }
2789 EXPORT_SYMBOL(release_sock);
2790
2791 /**
2792 * lock_sock_fast - fast version of lock_sock
2793 * @sk: socket
2794 *
2795 * This version should be used for very small section, where process wont block
2796 * return false if fast path is taken:
2797 *
2798 * sk_lock.slock locked, owned = 0, BH disabled
2799 *
2800 * return true if slow path is taken:
2801 *
2802 * sk_lock.slock unlocked, owned = 1, BH enabled
2803 */
2804 bool lock_sock_fast(struct sock *sk)
2805 {
2806 might_sleep();
2807 spin_lock_bh(&sk->sk_lock.slock);
2808
2809 if (!sk->sk_lock.owned)
2810 /*
2811 * Note : We must disable BH
2812 */
2813 return false;
2814
2815 __lock_sock(sk);
2816 sk->sk_lock.owned = 1;
2817 spin_unlock(&sk->sk_lock.slock);
2818 /*
2819 * The sk_lock has mutex_lock() semantics here:
2820 */
2821 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2822 local_bh_enable();
2823 return true;
2824 }
2825 EXPORT_SYMBOL(lock_sock_fast);
2826
2827 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2828 {
2829 struct timeval tv;
2830 if (!sock_flag(sk, SOCK_TIMESTAMP))
2831 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2832 tv = ktime_to_timeval(sk->sk_stamp);
2833 if (tv.tv_sec == -1)
2834 return -ENOENT;
2835 if (tv.tv_sec == 0) {
2836 sk->sk_stamp = ktime_get_real();
2837 tv = ktime_to_timeval(sk->sk_stamp);
2838 }
2839 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2840 }
2841 EXPORT_SYMBOL(sock_get_timestamp);
2842
2843 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2844 {
2845 struct timespec ts;
2846 if (!sock_flag(sk, SOCK_TIMESTAMP))
2847 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2848 ts = ktime_to_timespec(sk->sk_stamp);
2849 if (ts.tv_sec == -1)
2850 return -ENOENT;
2851 if (ts.tv_sec == 0) {
2852 sk->sk_stamp = ktime_get_real();
2853 ts = ktime_to_timespec(sk->sk_stamp);
2854 }
2855 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2856 }
2857 EXPORT_SYMBOL(sock_get_timestampns);
2858
2859 void sock_enable_timestamp(struct sock *sk, int flag)
2860 {
2861 if (!sock_flag(sk, flag)) {
2862 unsigned long previous_flags = sk->sk_flags;
2863
2864 sock_set_flag(sk, flag);
2865 /*
2866 * we just set one of the two flags which require net
2867 * time stamping, but time stamping might have been on
2868 * already because of the other one
2869 */
2870 if (sock_needs_netstamp(sk) &&
2871 !(previous_flags & SK_FLAGS_TIMESTAMP))
2872 net_enable_timestamp();
2873 }
2874 }
2875
2876 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2877 int level, int type)
2878 {
2879 struct sock_exterr_skb *serr;
2880 struct sk_buff *skb;
2881 int copied, err;
2882
2883 err = -EAGAIN;
2884 skb = sock_dequeue_err_skb(sk);
2885 if (skb == NULL)
2886 goto out;
2887
2888 copied = skb->len;
2889 if (copied > len) {
2890 msg->msg_flags |= MSG_TRUNC;
2891 copied = len;
2892 }
2893 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2894 if (err)
2895 goto out_free_skb;
2896
2897 sock_recv_timestamp(msg, sk, skb);
2898
2899 serr = SKB_EXT_ERR(skb);
2900 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2901
2902 msg->msg_flags |= MSG_ERRQUEUE;
2903 err = copied;
2904
2905 out_free_skb:
2906 kfree_skb(skb);
2907 out:
2908 return err;
2909 }
2910 EXPORT_SYMBOL(sock_recv_errqueue);
2911
2912 /*
2913 * Get a socket option on an socket.
2914 *
2915 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2916 * asynchronous errors should be reported by getsockopt. We assume
2917 * this means if you specify SO_ERROR (otherwise whats the point of it).
2918 */
2919 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2920 char __user *optval, int __user *optlen)
2921 {
2922 struct sock *sk = sock->sk;
2923
2924 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2925 }
2926 EXPORT_SYMBOL(sock_common_getsockopt);
2927
2928 #ifdef CONFIG_COMPAT
2929 int compat_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 if (sk->sk_prot->compat_getsockopt != NULL)
2935 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2936 optval, optlen);
2937 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2938 }
2939 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2940 #endif
2941
2942 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2943 int flags)
2944 {
2945 struct sock *sk = sock->sk;
2946 int addr_len = 0;
2947 int err;
2948
2949 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2950 flags & ~MSG_DONTWAIT, &addr_len);
2951 if (err >= 0)
2952 msg->msg_namelen = addr_len;
2953 return err;
2954 }
2955 EXPORT_SYMBOL(sock_common_recvmsg);
2956
2957 /*
2958 * Set socket options on an inet socket.
2959 */
2960 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2961 char __user *optval, unsigned int optlen)
2962 {
2963 struct sock *sk = sock->sk;
2964
2965 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2966 }
2967 EXPORT_SYMBOL(sock_common_setsockopt);
2968
2969 #ifdef CONFIG_COMPAT
2970 int compat_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 if (sk->sk_prot->compat_setsockopt != NULL)
2976 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2977 optval, optlen);
2978 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2979 }
2980 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2981 #endif
2982
2983 void sk_common_release(struct sock *sk)
2984 {
2985 if (sk->sk_prot->destroy)
2986 sk->sk_prot->destroy(sk);
2987
2988 /*
2989 * Observation: when sock_common_release is called, processes have
2990 * no access to socket. But net still has.
2991 * Step one, detach it from networking:
2992 *
2993 * A. Remove from hash tables.
2994 */
2995
2996 sk->sk_prot->unhash(sk);
2997
2998 /*
2999 * In this point socket cannot receive new packets, but it is possible
3000 * that some packets are in flight because some CPU runs receiver and
3001 * did hash table lookup before we unhashed socket. They will achieve
3002 * receive queue and will be purged by socket destructor.
3003 *
3004 * Also we still have packets pending on receive queue and probably,
3005 * our own packets waiting in device queues. sock_destroy will drain
3006 * receive queue, but transmitted packets will delay socket destruction
3007 * until the last reference will be released.
3008 */
3009
3010 sock_orphan(sk);
3011
3012 xfrm_sk_free_policy(sk);
3013
3014 sk_refcnt_debug_release(sk);
3015
3016 sock_put(sk);
3017 }
3018 EXPORT_SYMBOL(sk_common_release);
3019
3020 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3021 {
3022 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3023
3024 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3025 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3026 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3027 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3028 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3029 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3030 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3031 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3032 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3033 }
3034
3035 #ifdef CONFIG_PROC_FS
3036 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3037 struct prot_inuse {
3038 int val[PROTO_INUSE_NR];
3039 };
3040
3041 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3042
3043 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3044 {
3045 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
3046 }
3047 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3048
3049 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3050 {
3051 int cpu, idx = prot->inuse_idx;
3052 int res = 0;
3053
3054 for_each_possible_cpu(cpu)
3055 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
3056
3057 return res >= 0 ? res : 0;
3058 }
3059 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3060
3061 static int __net_init sock_inuse_init_net(struct net *net)
3062 {
3063 net->core.inuse = alloc_percpu(struct prot_inuse);
3064 return net->core.inuse ? 0 : -ENOMEM;
3065 }
3066
3067 static void __net_exit sock_inuse_exit_net(struct net *net)
3068 {
3069 free_percpu(net->core.inuse);
3070 }
3071
3072 static struct pernet_operations net_inuse_ops = {
3073 .init = sock_inuse_init_net,
3074 .exit = sock_inuse_exit_net,
3075 };
3076
3077 static __init int net_inuse_init(void)
3078 {
3079 if (register_pernet_subsys(&net_inuse_ops))
3080 panic("Cannot initialize net inuse counters");
3081
3082 return 0;
3083 }
3084
3085 core_initcall(net_inuse_init);
3086
3087 static void assign_proto_idx(struct proto *prot)
3088 {
3089 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3090
3091 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3092 pr_err("PROTO_INUSE_NR exhausted\n");
3093 return;
3094 }
3095
3096 set_bit(prot->inuse_idx, proto_inuse_idx);
3097 }
3098
3099 static void release_proto_idx(struct proto *prot)
3100 {
3101 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3102 clear_bit(prot->inuse_idx, proto_inuse_idx);
3103 }
3104 #else
3105 static inline void assign_proto_idx(struct proto *prot)
3106 {
3107 }
3108
3109 static inline void release_proto_idx(struct proto *prot)
3110 {
3111 }
3112 #endif
3113
3114 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3115 {
3116 if (!rsk_prot)
3117 return;
3118 kfree(rsk_prot->slab_name);
3119 rsk_prot->slab_name = NULL;
3120 kmem_cache_destroy(rsk_prot->slab);
3121 rsk_prot->slab = NULL;
3122 }
3123
3124 static int req_prot_init(const struct proto *prot)
3125 {
3126 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3127
3128 if (!rsk_prot)
3129 return 0;
3130
3131 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3132 prot->name);
3133 if (!rsk_prot->slab_name)
3134 return -ENOMEM;
3135
3136 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3137 rsk_prot->obj_size, 0,
3138 prot->slab_flags, NULL);
3139
3140 if (!rsk_prot->slab) {
3141 pr_crit("%s: Can't create request sock SLAB cache!\n",
3142 prot->name);
3143 return -ENOMEM;
3144 }
3145 return 0;
3146 }
3147
3148 int proto_register(struct proto *prot, int alloc_slab)
3149 {
3150 if (alloc_slab) {
3151 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
3152 SLAB_HWCACHE_ALIGN | prot->slab_flags,
3153 NULL);
3154
3155 if (prot->slab == NULL) {
3156 pr_crit("%s: Can't create sock SLAB cache!\n",
3157 prot->name);
3158 goto out;
3159 }
3160
3161 if (req_prot_init(prot))
3162 goto out_free_request_sock_slab;
3163
3164 if (prot->twsk_prot != NULL) {
3165 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3166
3167 if (prot->twsk_prot->twsk_slab_name == NULL)
3168 goto out_free_request_sock_slab;
3169
3170 prot->twsk_prot->twsk_slab =
3171 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3172 prot->twsk_prot->twsk_obj_size,
3173 0,
3174 prot->slab_flags,
3175 NULL);
3176 if (prot->twsk_prot->twsk_slab == NULL)
3177 goto out_free_timewait_sock_slab_name;
3178 }
3179 }
3180
3181 mutex_lock(&proto_list_mutex);
3182 list_add(&prot->node, &proto_list);
3183 assign_proto_idx(prot);
3184 mutex_unlock(&proto_list_mutex);
3185 return 0;
3186
3187 out_free_timewait_sock_slab_name:
3188 kfree(prot->twsk_prot->twsk_slab_name);
3189 out_free_request_sock_slab:
3190 req_prot_cleanup(prot->rsk_prot);
3191
3192 kmem_cache_destroy(prot->slab);
3193 prot->slab = NULL;
3194 out:
3195 return -ENOBUFS;
3196 }
3197 EXPORT_SYMBOL(proto_register);
3198
3199 void proto_unregister(struct proto *prot)
3200 {
3201 mutex_lock(&proto_list_mutex);
3202 release_proto_idx(prot);
3203 list_del(&prot->node);
3204 mutex_unlock(&proto_list_mutex);
3205
3206 kmem_cache_destroy(prot->slab);
3207 prot->slab = NULL;
3208
3209 req_prot_cleanup(prot->rsk_prot);
3210
3211 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3212 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3213 kfree(prot->twsk_prot->twsk_slab_name);
3214 prot->twsk_prot->twsk_slab = NULL;
3215 }
3216 }
3217 EXPORT_SYMBOL(proto_unregister);
3218
3219 #ifdef CONFIG_PROC_FS
3220 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3221 __acquires(proto_list_mutex)
3222 {
3223 mutex_lock(&proto_list_mutex);
3224 return seq_list_start_head(&proto_list, *pos);
3225 }
3226
3227 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3228 {
3229 return seq_list_next(v, &proto_list, pos);
3230 }
3231
3232 static void proto_seq_stop(struct seq_file *seq, void *v)
3233 __releases(proto_list_mutex)
3234 {
3235 mutex_unlock(&proto_list_mutex);
3236 }
3237
3238 static char proto_method_implemented(const void *method)
3239 {
3240 return method == NULL ? 'n' : 'y';
3241 }
3242 static long sock_prot_memory_allocated(struct proto *proto)
3243 {
3244 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3245 }
3246
3247 static char *sock_prot_memory_pressure(struct proto *proto)
3248 {
3249 return proto->memory_pressure != NULL ?
3250 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3251 }
3252
3253 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3254 {
3255
3256 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3257 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3258 proto->name,
3259 proto->obj_size,
3260 sock_prot_inuse_get(seq_file_net(seq), proto),
3261 sock_prot_memory_allocated(proto),
3262 sock_prot_memory_pressure(proto),
3263 proto->max_header,
3264 proto->slab == NULL ? "no" : "yes",
3265 module_name(proto->owner),
3266 proto_method_implemented(proto->close),
3267 proto_method_implemented(proto->connect),
3268 proto_method_implemented(proto->disconnect),
3269 proto_method_implemented(proto->accept),
3270 proto_method_implemented(proto->ioctl),
3271 proto_method_implemented(proto->init),
3272 proto_method_implemented(proto->destroy),
3273 proto_method_implemented(proto->shutdown),
3274 proto_method_implemented(proto->setsockopt),
3275 proto_method_implemented(proto->getsockopt),
3276 proto_method_implemented(proto->sendmsg),
3277 proto_method_implemented(proto->recvmsg),
3278 proto_method_implemented(proto->sendpage),
3279 proto_method_implemented(proto->bind),
3280 proto_method_implemented(proto->backlog_rcv),
3281 proto_method_implemented(proto->hash),
3282 proto_method_implemented(proto->unhash),
3283 proto_method_implemented(proto->get_port),
3284 proto_method_implemented(proto->enter_memory_pressure));
3285 }
3286
3287 static int proto_seq_show(struct seq_file *seq, void *v)
3288 {
3289 if (v == &proto_list)
3290 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3291 "protocol",
3292 "size",
3293 "sockets",
3294 "memory",
3295 "press",
3296 "maxhdr",
3297 "slab",
3298 "module",
3299 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3300 else
3301 proto_seq_printf(seq, list_entry(v, struct proto, node));
3302 return 0;
3303 }
3304
3305 static const struct seq_operations proto_seq_ops = {
3306 .start = proto_seq_start,
3307 .next = proto_seq_next,
3308 .stop = proto_seq_stop,
3309 .show = proto_seq_show,
3310 };
3311
3312 static int proto_seq_open(struct inode *inode, struct file *file)
3313 {
3314 return seq_open_net(inode, file, &proto_seq_ops,
3315 sizeof(struct seq_net_private));
3316 }
3317
3318 static const struct file_operations proto_seq_fops = {
3319 .owner = THIS_MODULE,
3320 .open = proto_seq_open,
3321 .read = seq_read,
3322 .llseek = seq_lseek,
3323 .release = seq_release_net,
3324 };
3325
3326 static __net_init int proto_init_net(struct net *net)
3327 {
3328 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3329 return -ENOMEM;
3330
3331 return 0;
3332 }
3333
3334 static __net_exit void proto_exit_net(struct net *net)
3335 {
3336 remove_proc_entry("protocols", net->proc_net);
3337 }
3338
3339
3340 static __net_initdata struct pernet_operations proto_net_ops = {
3341 .init = proto_init_net,
3342 .exit = proto_exit_net,
3343 };
3344
3345 static int __init proto_init(void)
3346 {
3347 return register_pernet_subsys(&proto_net_ops);
3348 }
3349
3350 subsys_initcall(proto_init);
3351
3352 #endif /* PROC_FS */
3353
3354 #ifdef CONFIG_NET_RX_BUSY_POLL
3355 bool sk_busy_loop_end(void *p, unsigned long start_time)
3356 {
3357 struct sock *sk = p;
3358
3359 return !skb_queue_empty(&sk->sk_receive_queue) ||
3360 sk_busy_loop_timeout(sk, start_time);
3361 }
3362 EXPORT_SYMBOL(sk_busy_loop_end);
3363 #endif /* CONFIG_NET_RX_BUSY_POLL */
3364
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.