?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
IB/srp: Add multichannel support
[linux/fpc-iii.git] / net / core / sock.c
blob15e0c67b1069654af22ad9afe5c993cbcaafaec5
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 *
85 *
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
90 */
91
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
99 #include <linux/in.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
120
121 #include <asm/uaccess.h>
122
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134
135 #include <linux/filter.h>
136
137 #include <trace/events/sock.h>
138
139 #ifdef CONFIG_INET
140 #include <net/tcp.h>
141 #endif
142
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 #ifdef CONFIG_MEMCG_KMEM
198 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
199 {
200 struct proto *proto;
201 int ret = 0;
202
203 mutex_lock(&proto_list_mutex);
204 list_for_each_entry(proto, &proto_list, node) {
205 if (proto->init_cgroup) {
206 ret = proto->init_cgroup(memcg, ss);
207 if (ret)
208 goto out;
209 }
210 }
211
212 mutex_unlock(&proto_list_mutex);
213 return ret;
214 out:
215 list_for_each_entry_continue_reverse(proto, &proto_list, node)
216 if (proto->destroy_cgroup)
217 proto->destroy_cgroup(memcg);
218 mutex_unlock(&proto_list_mutex);
219 return ret;
220 }
221
222 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
223 {
224 struct proto *proto;
225
226 mutex_lock(&proto_list_mutex);
227 list_for_each_entry_reverse(proto, &proto_list, node)
228 if (proto->destroy_cgroup)
229 proto->destroy_cgroup(memcg);
230 mutex_unlock(&proto_list_mutex);
231 }
232 #endif
233
234 /*
235 * Each address family might have different locking rules, so we have
236 * one slock key per address family:
237 */
238 static struct lock_class_key af_family_keys[AF_MAX];
239 static struct lock_class_key af_family_slock_keys[AF_MAX];
240
241 #if defined(CONFIG_MEMCG_KMEM)
242 struct static_key memcg_socket_limit_enabled;
243 EXPORT_SYMBOL(memcg_socket_limit_enabled);
244 #endif
245
246 /*
247 * Make lock validator output more readable. (we pre-construct these
248 * strings build-time, so that runtime initialization of socket
249 * locks is fast):
250 */
251 static const char *const af_family_key_strings[AF_MAX+1] = {
252 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
253 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
254 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
255 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
256 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
257 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
258 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
259 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
260 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
261 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
262 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
263 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
264 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
265 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
266 };
267 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
268 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
269 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
270 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
271 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
272 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
273 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
274 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
275 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
276 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
277 "slock-27" , "slock-28" , "slock-AF_CAN" ,
278 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
279 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
280 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
281 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
282 };
283 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
284 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
285 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
286 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
287 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
288 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
289 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
290 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
291 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
292 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
293 "clock-27" , "clock-28" , "clock-AF_CAN" ,
294 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
295 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
296 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
297 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
298 };
299
300 /*
301 * sk_callback_lock 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
306 /* Take into consideration the size of the struct sk_buff overhead in the
307 * determination of these values, since that is non-constant across
308 * platforms. This makes socket queueing behavior and performance
309 * not depend upon such differences.
310 */
311 #define _SK_MEM_PACKETS 256
312 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
313 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
314 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315
316 /* Run time adjustable parameters. */
317 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
318 EXPORT_SYMBOL(sysctl_wmem_max);
319 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
320 EXPORT_SYMBOL(sysctl_rmem_max);
321 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
322 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
323
324 /* Maximal space eaten by iovec or ancillary data plus some space */
325 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
326 EXPORT_SYMBOL(sysctl_optmem_max);
327
328 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
329 EXPORT_SYMBOL_GPL(memalloc_socks);
330
331 /**
332 * sk_set_memalloc - sets %SOCK_MEMALLOC
333 * @sk: socket to set it on
334 *
335 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
336 * It's the responsibility of the admin to adjust min_free_kbytes
337 * to meet the requirements
338 */
339 void sk_set_memalloc(struct sock *sk)
340 {
341 sock_set_flag(sk, SOCK_MEMALLOC);
342 sk->sk_allocation |= __GFP_MEMALLOC;
343 static_key_slow_inc(&memalloc_socks);
344 }
345 EXPORT_SYMBOL_GPL(sk_set_memalloc);
346
347 void sk_clear_memalloc(struct sock *sk)
348 {
349 sock_reset_flag(sk, SOCK_MEMALLOC);
350 sk->sk_allocation &= ~__GFP_MEMALLOC;
351 static_key_slow_dec(&memalloc_socks);
352
353 /*
354 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
355 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
356 * it has rmem allocations there is a risk that the user of the
357 * socket cannot make forward progress due to exceeding the rmem
358 * limits. By rights, sk_clear_memalloc() should only be called
359 * on sockets being torn down but warn and reset the accounting if
360 * that assumption breaks.
361 */
362 if (WARN_ON(sk->sk_forward_alloc))
363 sk_mem_reclaim(sk);
364 }
365 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
366
367 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
368 {
369 int ret;
370 unsigned long pflags = current->flags;
371
372 /* these should have been dropped before queueing */
373 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
374
375 current->flags |= PF_MEMALLOC;
376 ret = sk->sk_backlog_rcv(sk, skb);
377 tsk_restore_flags(current, pflags, PF_MEMALLOC);
378
379 return ret;
380 }
381 EXPORT_SYMBOL(__sk_backlog_rcv);
382
383 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
384 {
385 struct timeval tv;
386
387 if (optlen < sizeof(tv))
388 return -EINVAL;
389 if (copy_from_user(&tv, optval, sizeof(tv)))
390 return -EFAULT;
391 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
392 return -EDOM;
393
394 if (tv.tv_sec < 0) {
395 static int warned __read_mostly;
396
397 *timeo_p = 0;
398 if (warned < 10 && net_ratelimit()) {
399 warned++;
400 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
401 __func__, current->comm, task_pid_nr(current));
402 }
403 return 0;
404 }
405 *timeo_p = MAX_SCHEDULE_TIMEOUT;
406 if (tv.tv_sec == 0 && tv.tv_usec == 0)
407 return 0;
408 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
409 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
410 return 0;
411 }
412
413 static void sock_warn_obsolete_bsdism(const char *name)
414 {
415 static int warned;
416 static char warncomm[TASK_COMM_LEN];
417 if (strcmp(warncomm, current->comm) && warned < 5) {
418 strcpy(warncomm, current->comm);
419 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
420 warncomm, name);
421 warned++;
422 }
423 }
424
425 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
426
427 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
428 {
429 if (sk->sk_flags & flags) {
430 sk->sk_flags &= ~flags;
431 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
432 net_disable_timestamp();
433 }
434 }
435
436
437 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
438 {
439 int err;
440 unsigned long flags;
441 struct sk_buff_head *list = &sk->sk_receive_queue;
442
443 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
444 atomic_inc(&sk->sk_drops);
445 trace_sock_rcvqueue_full(sk, skb);
446 return -ENOMEM;
447 }
448
449 err = sk_filter(sk, skb);
450 if (err)
451 return err;
452
453 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
454 atomic_inc(&sk->sk_drops);
455 return -ENOBUFS;
456 }
457
458 skb->dev = NULL;
459 skb_set_owner_r(skb, sk);
460
461 /* we escape from rcu protected region, make sure we dont leak
462 * a norefcounted dst
463 */
464 skb_dst_force(skb);
465
466 spin_lock_irqsave(&list->lock, flags);
467 skb->dropcount = atomic_read(&sk->sk_drops);
468 __skb_queue_tail(list, skb);
469 spin_unlock_irqrestore(&list->lock, flags);
470
471 if (!sock_flag(sk, SOCK_DEAD))
472 sk->sk_data_ready(sk);
473 return 0;
474 }
475 EXPORT_SYMBOL(sock_queue_rcv_skb);
476
477 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
478 {
479 int rc = NET_RX_SUCCESS;
480
481 if (sk_filter(sk, skb))
482 goto discard_and_relse;
483
484 skb->dev = NULL;
485
486 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
487 atomic_inc(&sk->sk_drops);
488 goto discard_and_relse;
489 }
490 if (nested)
491 bh_lock_sock_nested(sk);
492 else
493 bh_lock_sock(sk);
494 if (!sock_owned_by_user(sk)) {
495 /*
496 * trylock + unlock semantics:
497 */
498 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
499
500 rc = sk_backlog_rcv(sk, skb);
501
502 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
503 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
504 bh_unlock_sock(sk);
505 atomic_inc(&sk->sk_drops);
506 goto discard_and_relse;
507 }
508
509 bh_unlock_sock(sk);
510 out:
511 sock_put(sk);
512 return rc;
513 discard_and_relse:
514 kfree_skb(skb);
515 goto out;
516 }
517 EXPORT_SYMBOL(sk_receive_skb);
518
519 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
520 {
521 struct dst_entry *dst = __sk_dst_get(sk);
522
523 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
524 sk_tx_queue_clear(sk);
525 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
526 dst_release(dst);
527 return NULL;
528 }
529
530 return dst;
531 }
532 EXPORT_SYMBOL(__sk_dst_check);
533
534 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
535 {
536 struct dst_entry *dst = sk_dst_get(sk);
537
538 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
539 sk_dst_reset(sk);
540 dst_release(dst);
541 return NULL;
542 }
543
544 return dst;
545 }
546 EXPORT_SYMBOL(sk_dst_check);
547
548 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
549 int optlen)
550 {
551 int ret = -ENOPROTOOPT;
552 #ifdef CONFIG_NETDEVICES
553 struct net *net = sock_net(sk);
554 char devname[IFNAMSIZ];
555 int index;
556
557 /* Sorry... */
558 ret = -EPERM;
559 if (!ns_capable(net->user_ns, CAP_NET_RAW))
560 goto out;
561
562 ret = -EINVAL;
563 if (optlen < 0)
564 goto out;
565
566 /* Bind this socket to a particular device like "eth0",
567 * as specified in the passed interface name. If the
568 * name is "" or the option length is zero the socket
569 * is not bound.
570 */
571 if (optlen > IFNAMSIZ - 1)
572 optlen = IFNAMSIZ - 1;
573 memset(devname, 0, sizeof(devname));
574
575 ret = -EFAULT;
576 if (copy_from_user(devname, optval, optlen))
577 goto out;
578
579 index = 0;
580 if (devname[0] != '\0') {
581 struct net_device *dev;
582
583 rcu_read_lock();
584 dev = dev_get_by_name_rcu(net, devname);
585 if (dev)
586 index = dev->ifindex;
587 rcu_read_unlock();
588 ret = -ENODEV;
589 if (!dev)
590 goto out;
591 }
592
593 lock_sock(sk);
594 sk->sk_bound_dev_if = index;
595 sk_dst_reset(sk);
596 release_sock(sk);
597
598 ret = 0;
599
600 out:
601 #endif
602
603 return ret;
604 }
605
606 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
607 int __user *optlen, int len)
608 {
609 int ret = -ENOPROTOOPT;
610 #ifdef CONFIG_NETDEVICES
611 struct net *net = sock_net(sk);
612 char devname[IFNAMSIZ];
613
614 if (sk->sk_bound_dev_if == 0) {
615 len = 0;
616 goto zero;
617 }
618
619 ret = -EINVAL;
620 if (len < IFNAMSIZ)
621 goto out;
622
623 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
624 if (ret)
625 goto out;
626
627 len = strlen(devname) + 1;
628
629 ret = -EFAULT;
630 if (copy_to_user(optval, devname, len))
631 goto out;
632
633 zero:
634 ret = -EFAULT;
635 if (put_user(len, optlen))
636 goto out;
637
638 ret = 0;
639
640 out:
641 #endif
642
643 return ret;
644 }
645
646 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
647 {
648 if (valbool)
649 sock_set_flag(sk, bit);
650 else
651 sock_reset_flag(sk, bit);
652 }
653
654 /*
655 * This is meant for all protocols to use and covers goings on
656 * at the socket level. Everything here is generic.
657 */
658
659 int sock_setsockopt(struct socket *sock, int level, int optname,
660 char __user *optval, unsigned int optlen)
661 {
662 struct sock *sk = sock->sk;
663 int val;
664 int valbool;
665 struct linger ling;
666 int ret = 0;
667
668 /*
669 * Options without arguments
670 */
671
672 if (optname == SO_BINDTODEVICE)
673 return sock_setbindtodevice(sk, optval, optlen);
674
675 if (optlen < sizeof(int))
676 return -EINVAL;
677
678 if (get_user(val, (int __user *)optval))
679 return -EFAULT;
680
681 valbool = val ? 1 : 0;
682
683 lock_sock(sk);
684
685 switch (optname) {
686 case SO_DEBUG:
687 if (val && !capable(CAP_NET_ADMIN))
688 ret = -EACCES;
689 else
690 sock_valbool_flag(sk, SOCK_DBG, valbool);
691 break;
692 case SO_REUSEADDR:
693 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
694 break;
695 case SO_REUSEPORT:
696 sk->sk_reuseport = valbool;
697 break;
698 case SO_TYPE:
699 case SO_PROTOCOL:
700 case SO_DOMAIN:
701 case SO_ERROR:
702 ret = -ENOPROTOOPT;
703 break;
704 case SO_DONTROUTE:
705 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
706 break;
707 case SO_BROADCAST:
708 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
709 break;
710 case SO_SNDBUF:
711 /* Don't error on this BSD doesn't and if you think
712 * about it this is right. Otherwise apps have to
713 * play 'guess the biggest size' games. RCVBUF/SNDBUF
714 * are treated in BSD as hints
715 */
716 val = min_t(u32, val, sysctl_wmem_max);
717 set_sndbuf:
718 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
719 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
720 /* Wake up sending tasks if we upped the value. */
721 sk->sk_write_space(sk);
722 break;
723
724 case SO_SNDBUFFORCE:
725 if (!capable(CAP_NET_ADMIN)) {
726 ret = -EPERM;
727 break;
728 }
729 goto set_sndbuf;
730
731 case SO_RCVBUF:
732 /* Don't error on this BSD doesn't and if you think
733 * about it this is right. Otherwise apps have to
734 * play 'guess the biggest size' games. RCVBUF/SNDBUF
735 * are treated in BSD as hints
736 */
737 val = min_t(u32, val, sysctl_rmem_max);
738 set_rcvbuf:
739 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
740 /*
741 * We double it on the way in to account for
742 * "struct sk_buff" etc. overhead. Applications
743 * assume that the SO_RCVBUF setting they make will
744 * allow that much actual data to be received on that
745 * socket.
746 *
747 * Applications are unaware that "struct sk_buff" and
748 * other overheads allocate from the receive buffer
749 * during socket buffer allocation.
750 *
751 * And after considering the possible alternatives,
752 * returning the value we actually used in getsockopt
753 * is the most desirable behavior.
754 */
755 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
756 break;
757
758 case SO_RCVBUFFORCE:
759 if (!capable(CAP_NET_ADMIN)) {
760 ret = -EPERM;
761 break;
762 }
763 goto set_rcvbuf;
764
765 case SO_KEEPALIVE:
766 #ifdef CONFIG_INET
767 if (sk->sk_protocol == IPPROTO_TCP &&
768 sk->sk_type == SOCK_STREAM)
769 tcp_set_keepalive(sk, valbool);
770 #endif
771 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
772 break;
773
774 case SO_OOBINLINE:
775 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
776 break;
777
778 case SO_NO_CHECK:
779 sk->sk_no_check_tx = valbool;
780 break;
781
782 case SO_PRIORITY:
783 if ((val >= 0 && val <= 6) ||
784 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
785 sk->sk_priority = val;
786 else
787 ret = -EPERM;
788 break;
789
790 case SO_LINGER:
791 if (optlen < sizeof(ling)) {
792 ret = -EINVAL; /* 1003.1g */
793 break;
794 }
795 if (copy_from_user(&ling, optval, sizeof(ling))) {
796 ret = -EFAULT;
797 break;
798 }
799 if (!ling.l_onoff)
800 sock_reset_flag(sk, SOCK_LINGER);
801 else {
802 #if (BITS_PER_LONG == 32)
803 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
804 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
805 else
806 #endif
807 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
808 sock_set_flag(sk, SOCK_LINGER);
809 }
810 break;
811
812 case SO_BSDCOMPAT:
813 sock_warn_obsolete_bsdism("setsockopt");
814 break;
815
816 case SO_PASSCRED:
817 if (valbool)
818 set_bit(SOCK_PASSCRED, &sock->flags);
819 else
820 clear_bit(SOCK_PASSCRED, &sock->flags);
821 break;
822
823 case SO_TIMESTAMP:
824 case SO_TIMESTAMPNS:
825 if (valbool) {
826 if (optname == SO_TIMESTAMP)
827 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
828 else
829 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
830 sock_set_flag(sk, SOCK_RCVTSTAMP);
831 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
832 } else {
833 sock_reset_flag(sk, SOCK_RCVTSTAMP);
834 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
835 }
836 break;
837
838 case SO_TIMESTAMPING:
839 if (val & ~SOF_TIMESTAMPING_MASK) {
840 ret = -EINVAL;
841 break;
842 }
843 if (val & SOF_TIMESTAMPING_OPT_ID &&
844 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
845 if (sk->sk_protocol == IPPROTO_TCP) {
846 if (sk->sk_state != TCP_ESTABLISHED) {
847 ret = -EINVAL;
848 break;
849 }
850 sk->sk_tskey = tcp_sk(sk)->snd_una;
851 } else {
852 sk->sk_tskey = 0;
853 }
854 }
855 sk->sk_tsflags = val;
856 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
857 sock_enable_timestamp(sk,
858 SOCK_TIMESTAMPING_RX_SOFTWARE);
859 else
860 sock_disable_timestamp(sk,
861 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
862 break;
863
864 case SO_RCVLOWAT:
865 if (val < 0)
866 val = INT_MAX;
867 sk->sk_rcvlowat = val ? : 1;
868 break;
869
870 case SO_RCVTIMEO:
871 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
872 break;
873
874 case SO_SNDTIMEO:
875 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
876 break;
877
878 case SO_ATTACH_FILTER:
879 ret = -EINVAL;
880 if (optlen == sizeof(struct sock_fprog)) {
881 struct sock_fprog fprog;
882
883 ret = -EFAULT;
884 if (copy_from_user(&fprog, optval, sizeof(fprog)))
885 break;
886
887 ret = sk_attach_filter(&fprog, sk);
888 }
889 break;
890
891 case SO_DETACH_FILTER:
892 ret = sk_detach_filter(sk);
893 break;
894
895 case SO_LOCK_FILTER:
896 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
897 ret = -EPERM;
898 else
899 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
900 break;
901
902 case SO_PASSSEC:
903 if (valbool)
904 set_bit(SOCK_PASSSEC, &sock->flags);
905 else
906 clear_bit(SOCK_PASSSEC, &sock->flags);
907 break;
908 case SO_MARK:
909 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
910 ret = -EPERM;
911 else
912 sk->sk_mark = val;
913 break;
914
915 /* We implement the SO_SNDLOWAT etc to
916 not be settable (1003.1g 5.3) */
917 case SO_RXQ_OVFL:
918 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
919 break;
920
921 case SO_WIFI_STATUS:
922 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
923 break;
924
925 case SO_PEEK_OFF:
926 if (sock->ops->set_peek_off)
927 ret = sock->ops->set_peek_off(sk, val);
928 else
929 ret = -EOPNOTSUPP;
930 break;
931
932 case SO_NOFCS:
933 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
934 break;
935
936 case SO_SELECT_ERR_QUEUE:
937 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
938 break;
939
940 #ifdef CONFIG_NET_RX_BUSY_POLL
941 case SO_BUSY_POLL:
942 /* allow unprivileged users to decrease the value */
943 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
944 ret = -EPERM;
945 else {
946 if (val < 0)
947 ret = -EINVAL;
948 else
949 sk->sk_ll_usec = val;
950 }
951 break;
952 #endif
953
954 case SO_MAX_PACING_RATE:
955 sk->sk_max_pacing_rate = val;
956 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
957 sk->sk_max_pacing_rate);
958 break;
959
960 default:
961 ret = -ENOPROTOOPT;
962 break;
963 }
964 release_sock(sk);
965 return ret;
966 }
967 EXPORT_SYMBOL(sock_setsockopt);
968
969
970 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
971 struct ucred *ucred)
972 {
973 ucred->pid = pid_vnr(pid);
974 ucred->uid = ucred->gid = -1;
975 if (cred) {
976 struct user_namespace *current_ns = current_user_ns();
977
978 ucred->uid = from_kuid_munged(current_ns, cred->euid);
979 ucred->gid = from_kgid_munged(current_ns, cred->egid);
980 }
981 }
982
983 int sock_getsockopt(struct socket *sock, int level, int optname,
984 char __user *optval, int __user *optlen)
985 {
986 struct sock *sk = sock->sk;
987
988 union {
989 int val;
990 struct linger ling;
991 struct timeval tm;
992 } v;
993
994 int lv = sizeof(int);
995 int len;
996
997 if (get_user(len, optlen))
998 return -EFAULT;
999 if (len < 0)
1000 return -EINVAL;
1001
1002 memset(&v, 0, sizeof(v));
1003
1004 switch (optname) {
1005 case SO_DEBUG:
1006 v.val = sock_flag(sk, SOCK_DBG);
1007 break;
1008
1009 case SO_DONTROUTE:
1010 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1011 break;
1012
1013 case SO_BROADCAST:
1014 v.val = sock_flag(sk, SOCK_BROADCAST);
1015 break;
1016
1017 case SO_SNDBUF:
1018 v.val = sk->sk_sndbuf;
1019 break;
1020
1021 case SO_RCVBUF:
1022 v.val = sk->sk_rcvbuf;
1023 break;
1024
1025 case SO_REUSEADDR:
1026 v.val = sk->sk_reuse;
1027 break;
1028
1029 case SO_REUSEPORT:
1030 v.val = sk->sk_reuseport;
1031 break;
1032
1033 case SO_KEEPALIVE:
1034 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1035 break;
1036
1037 case SO_TYPE:
1038 v.val = sk->sk_type;
1039 break;
1040
1041 case SO_PROTOCOL:
1042 v.val = sk->sk_protocol;
1043 break;
1044
1045 case SO_DOMAIN:
1046 v.val = sk->sk_family;
1047 break;
1048
1049 case SO_ERROR:
1050 v.val = -sock_error(sk);
1051 if (v.val == 0)
1052 v.val = xchg(&sk->sk_err_soft, 0);
1053 break;
1054
1055 case SO_OOBINLINE:
1056 v.val = sock_flag(sk, SOCK_URGINLINE);
1057 break;
1058
1059 case SO_NO_CHECK:
1060 v.val = sk->sk_no_check_tx;
1061 break;
1062
1063 case SO_PRIORITY:
1064 v.val = sk->sk_priority;
1065 break;
1066
1067 case SO_LINGER:
1068 lv = sizeof(v.ling);
1069 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1070 v.ling.l_linger = sk->sk_lingertime / HZ;
1071 break;
1072
1073 case SO_BSDCOMPAT:
1074 sock_warn_obsolete_bsdism("getsockopt");
1075 break;
1076
1077 case SO_TIMESTAMP:
1078 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1079 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1080 break;
1081
1082 case SO_TIMESTAMPNS:
1083 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1084 break;
1085
1086 case SO_TIMESTAMPING:
1087 v.val = sk->sk_tsflags;
1088 break;
1089
1090 case SO_RCVTIMEO:
1091 lv = sizeof(struct timeval);
1092 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1093 v.tm.tv_sec = 0;
1094 v.tm.tv_usec = 0;
1095 } else {
1096 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1097 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1098 }
1099 break;
1100
1101 case SO_SNDTIMEO:
1102 lv = sizeof(struct timeval);
1103 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1104 v.tm.tv_sec = 0;
1105 v.tm.tv_usec = 0;
1106 } else {
1107 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1108 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1109 }
1110 break;
1111
1112 case SO_RCVLOWAT:
1113 v.val = sk->sk_rcvlowat;
1114 break;
1115
1116 case SO_SNDLOWAT:
1117 v.val = 1;
1118 break;
1119
1120 case SO_PASSCRED:
1121 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1122 break;
1123
1124 case SO_PEERCRED:
1125 {
1126 struct ucred peercred;
1127 if (len > sizeof(peercred))
1128 len = sizeof(peercred);
1129 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1130 if (copy_to_user(optval, &peercred, len))
1131 return -EFAULT;
1132 goto lenout;
1133 }
1134
1135 case SO_PEERNAME:
1136 {
1137 char address[128];
1138
1139 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1140 return -ENOTCONN;
1141 if (lv < len)
1142 return -EINVAL;
1143 if (copy_to_user(optval, address, len))
1144 return -EFAULT;
1145 goto lenout;
1146 }
1147
1148 /* Dubious BSD thing... Probably nobody even uses it, but
1149 * the UNIX standard wants it for whatever reason... -DaveM
1150 */
1151 case SO_ACCEPTCONN:
1152 v.val = sk->sk_state == TCP_LISTEN;
1153 break;
1154
1155 case SO_PASSSEC:
1156 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1157 break;
1158
1159 case SO_PEERSEC:
1160 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1161
1162 case SO_MARK:
1163 v.val = sk->sk_mark;
1164 break;
1165
1166 case SO_RXQ_OVFL:
1167 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1168 break;
1169
1170 case SO_WIFI_STATUS:
1171 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1172 break;
1173
1174 case SO_PEEK_OFF:
1175 if (!sock->ops->set_peek_off)
1176 return -EOPNOTSUPP;
1177
1178 v.val = sk->sk_peek_off;
1179 break;
1180 case SO_NOFCS:
1181 v.val = sock_flag(sk, SOCK_NOFCS);
1182 break;
1183
1184 case SO_BINDTODEVICE:
1185 return sock_getbindtodevice(sk, optval, optlen, len);
1186
1187 case SO_GET_FILTER:
1188 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1189 if (len < 0)
1190 return len;
1191
1192 goto lenout;
1193
1194 case SO_LOCK_FILTER:
1195 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1196 break;
1197
1198 case SO_BPF_EXTENSIONS:
1199 v.val = bpf_tell_extensions();
1200 break;
1201
1202 case SO_SELECT_ERR_QUEUE:
1203 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1204 break;
1205
1206 #ifdef CONFIG_NET_RX_BUSY_POLL
1207 case SO_BUSY_POLL:
1208 v.val = sk->sk_ll_usec;
1209 break;
1210 #endif
1211
1212 case SO_MAX_PACING_RATE:
1213 v.val = sk->sk_max_pacing_rate;
1214 break;
1215
1216 default:
1217 return -ENOPROTOOPT;
1218 }
1219
1220 if (len > lv)
1221 len = lv;
1222 if (copy_to_user(optval, &v, len))
1223 return -EFAULT;
1224 lenout:
1225 if (put_user(len, optlen))
1226 return -EFAULT;
1227 return 0;
1228 }
1229
1230 /*
1231 * Initialize an sk_lock.
1232 *
1233 * (We also register the sk_lock with the lock validator.)
1234 */
1235 static inline void sock_lock_init(struct sock *sk)
1236 {
1237 sock_lock_init_class_and_name(sk,
1238 af_family_slock_key_strings[sk->sk_family],
1239 af_family_slock_keys + sk->sk_family,
1240 af_family_key_strings[sk->sk_family],
1241 af_family_keys + sk->sk_family);
1242 }
1243
1244 /*
1245 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1246 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1247 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1248 */
1249 static void sock_copy(struct sock *nsk, const struct sock *osk)
1250 {
1251 #ifdef CONFIG_SECURITY_NETWORK
1252 void *sptr = nsk->sk_security;
1253 #endif
1254 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1255
1256 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1257 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1258
1259 #ifdef CONFIG_SECURITY_NETWORK
1260 nsk->sk_security = sptr;
1261 security_sk_clone(osk, nsk);
1262 #endif
1263 }
1264
1265 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1266 {
1267 unsigned long nulls1, nulls2;
1268
1269 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1270 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1271 if (nulls1 > nulls2)
1272 swap(nulls1, nulls2);
1273
1274 if (nulls1 != 0)
1275 memset((char *)sk, 0, nulls1);
1276 memset((char *)sk + nulls1 + sizeof(void *), 0,
1277 nulls2 - nulls1 - sizeof(void *));
1278 memset((char *)sk + nulls2 + sizeof(void *), 0,
1279 size - nulls2 - sizeof(void *));
1280 }
1281 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1282
1283 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1284 int family)
1285 {
1286 struct sock *sk;
1287 struct kmem_cache *slab;
1288
1289 slab = prot->slab;
1290 if (slab != NULL) {
1291 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1292 if (!sk)
1293 return sk;
1294 if (priority & __GFP_ZERO) {
1295 if (prot->clear_sk)
1296 prot->clear_sk(sk, prot->obj_size);
1297 else
1298 sk_prot_clear_nulls(sk, prot->obj_size);
1299 }
1300 } else
1301 sk = kmalloc(prot->obj_size, priority);
1302
1303 if (sk != NULL) {
1304 kmemcheck_annotate_bitfield(sk, flags);
1305
1306 if (security_sk_alloc(sk, family, priority))
1307 goto out_free;
1308
1309 if (!try_module_get(prot->owner))
1310 goto out_free_sec;
1311 sk_tx_queue_clear(sk);
1312 }
1313
1314 return sk;
1315
1316 out_free_sec:
1317 security_sk_free(sk);
1318 out_free:
1319 if (slab != NULL)
1320 kmem_cache_free(slab, sk);
1321 else
1322 kfree(sk);
1323 return NULL;
1324 }
1325
1326 static void sk_prot_free(struct proto *prot, struct sock *sk)
1327 {
1328 struct kmem_cache *slab;
1329 struct module *owner;
1330
1331 owner = prot->owner;
1332 slab = prot->slab;
1333
1334 security_sk_free(sk);
1335 if (slab != NULL)
1336 kmem_cache_free(slab, sk);
1337 else
1338 kfree(sk);
1339 module_put(owner);
1340 }
1341
1342 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1343 void sock_update_netprioidx(struct sock *sk)
1344 {
1345 if (in_interrupt())
1346 return;
1347
1348 sk->sk_cgrp_prioidx = task_netprioidx(current);
1349 }
1350 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1351 #endif
1352
1353 /**
1354 * sk_alloc - All socket objects are allocated here
1355 * @net: the applicable net namespace
1356 * @family: protocol family
1357 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1358 * @prot: struct proto associated with this new sock instance
1359 */
1360 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1361 struct proto *prot)
1362 {
1363 struct sock *sk;
1364
1365 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1366 if (sk) {
1367 sk->sk_family = family;
1368 /*
1369 * See comment in struct sock definition to understand
1370 * why we need sk_prot_creator -acme
1371 */
1372 sk->sk_prot = sk->sk_prot_creator = prot;
1373 sock_lock_init(sk);
1374 sock_net_set(sk, get_net(net));
1375 atomic_set(&sk->sk_wmem_alloc, 1);
1376
1377 sock_update_classid(sk);
1378 sock_update_netprioidx(sk);
1379 }
1380
1381 return sk;
1382 }
1383 EXPORT_SYMBOL(sk_alloc);
1384
1385 static void __sk_free(struct sock *sk)
1386 {
1387 struct sk_filter *filter;
1388
1389 if (sk->sk_destruct)
1390 sk->sk_destruct(sk);
1391
1392 filter = rcu_dereference_check(sk->sk_filter,
1393 atomic_read(&sk->sk_wmem_alloc) == 0);
1394 if (filter) {
1395 sk_filter_uncharge(sk, filter);
1396 RCU_INIT_POINTER(sk->sk_filter, NULL);
1397 }
1398
1399 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1400
1401 if (atomic_read(&sk->sk_omem_alloc))
1402 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1403 __func__, atomic_read(&sk->sk_omem_alloc));
1404
1405 if (sk->sk_peer_cred)
1406 put_cred(sk->sk_peer_cred);
1407 put_pid(sk->sk_peer_pid);
1408 put_net(sock_net(sk));
1409 sk_prot_free(sk->sk_prot_creator, sk);
1410 }
1411
1412 void sk_free(struct sock *sk)
1413 {
1414 /*
1415 * We subtract one from sk_wmem_alloc and can know if
1416 * some packets are still in some tx queue.
1417 * If not null, sock_wfree() will call __sk_free(sk) later
1418 */
1419 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1420 __sk_free(sk);
1421 }
1422 EXPORT_SYMBOL(sk_free);
1423
1424 /*
1425 * Last sock_put should drop reference to sk->sk_net. It has already
1426 * been dropped in sk_change_net. Taking reference to stopping namespace
1427 * is not an option.
1428 * Take reference to a socket to remove it from hash _alive_ and after that
1429 * destroy it in the context of init_net.
1430 */
1431 void sk_release_kernel(struct sock *sk)
1432 {
1433 if (sk == NULL || sk->sk_socket == NULL)
1434 return;
1435
1436 sock_hold(sk);
1437 sock_release(sk->sk_socket);
1438 release_net(sock_net(sk));
1439 sock_net_set(sk, get_net(&init_net));
1440 sock_put(sk);
1441 }
1442 EXPORT_SYMBOL(sk_release_kernel);
1443
1444 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1445 {
1446 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1447 sock_update_memcg(newsk);
1448 }
1449
1450 /**
1451 * sk_clone_lock - clone a socket, and lock its clone
1452 * @sk: the socket to clone
1453 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1454 *
1455 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1456 */
1457 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1458 {
1459 struct sock *newsk;
1460 bool is_charged = true;
1461
1462 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1463 if (newsk != NULL) {
1464 struct sk_filter *filter;
1465
1466 sock_copy(newsk, sk);
1467
1468 /* SANITY */
1469 get_net(sock_net(newsk));
1470 sk_node_init(&newsk->sk_node);
1471 sock_lock_init(newsk);
1472 bh_lock_sock(newsk);
1473 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1474 newsk->sk_backlog.len = 0;
1475
1476 atomic_set(&newsk->sk_rmem_alloc, 0);
1477 /*
1478 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1479 */
1480 atomic_set(&newsk->sk_wmem_alloc, 1);
1481 atomic_set(&newsk->sk_omem_alloc, 0);
1482 skb_queue_head_init(&newsk->sk_receive_queue);
1483 skb_queue_head_init(&newsk->sk_write_queue);
1484
1485 spin_lock_init(&newsk->sk_dst_lock);
1486 rwlock_init(&newsk->sk_callback_lock);
1487 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1488 af_callback_keys + newsk->sk_family,
1489 af_family_clock_key_strings[newsk->sk_family]);
1490
1491 newsk->sk_dst_cache = NULL;
1492 newsk->sk_wmem_queued = 0;
1493 newsk->sk_forward_alloc = 0;
1494 newsk->sk_send_head = NULL;
1495 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1496
1497 sock_reset_flag(newsk, SOCK_DONE);
1498 skb_queue_head_init(&newsk->sk_error_queue);
1499
1500 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1501 if (filter != NULL)
1502 /* though it's an empty new sock, the charging may fail
1503 * if sysctl_optmem_max was changed between creation of
1504 * original socket and cloning
1505 */
1506 is_charged = sk_filter_charge(newsk, filter);
1507
1508 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1509 /* It is still raw copy of parent, so invalidate
1510 * destructor and make plain sk_free() */
1511 newsk->sk_destruct = NULL;
1512 bh_unlock_sock(newsk);
1513 sk_free(newsk);
1514 newsk = NULL;
1515 goto out;
1516 }
1517
1518 newsk->sk_err = 0;
1519 newsk->sk_priority = 0;
1520 /*
1521 * Before updating sk_refcnt, we must commit prior changes to memory
1522 * (Documentation/RCU/rculist_nulls.txt for details)
1523 */
1524 smp_wmb();
1525 atomic_set(&newsk->sk_refcnt, 2);
1526
1527 /*
1528 * Increment the counter in the same struct proto as the master
1529 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1530 * is the same as sk->sk_prot->socks, as this field was copied
1531 * with memcpy).
1532 *
1533 * This _changes_ the previous behaviour, where
1534 * tcp_create_openreq_child always was incrementing the
1535 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1536 * to be taken into account in all callers. -acme
1537 */
1538 sk_refcnt_debug_inc(newsk);
1539 sk_set_socket(newsk, NULL);
1540 newsk->sk_wq = NULL;
1541
1542 sk_update_clone(sk, newsk);
1543
1544 if (newsk->sk_prot->sockets_allocated)
1545 sk_sockets_allocated_inc(newsk);
1546
1547 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1548 net_enable_timestamp();
1549 }
1550 out:
1551 return newsk;
1552 }
1553 EXPORT_SYMBOL_GPL(sk_clone_lock);
1554
1555 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1556 {
1557 __sk_dst_set(sk, dst);
1558 sk->sk_route_caps = dst->dev->features;
1559 if (sk->sk_route_caps & NETIF_F_GSO)
1560 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1561 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1562 if (sk_can_gso(sk)) {
1563 if (dst->header_len) {
1564 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1565 } else {
1566 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1567 sk->sk_gso_max_size = dst->dev->gso_max_size;
1568 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1569 }
1570 }
1571 }
1572 EXPORT_SYMBOL_GPL(sk_setup_caps);
1573
1574 /*
1575 * Simple resource managers for sockets.
1576 */
1577
1578
1579 /*
1580 * Write buffer destructor automatically called from kfree_skb.
1581 */
1582 void sock_wfree(struct sk_buff *skb)
1583 {
1584 struct sock *sk = skb->sk;
1585 unsigned int len = skb->truesize;
1586
1587 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1588 /*
1589 * Keep a reference on sk_wmem_alloc, this will be released
1590 * after sk_write_space() call
1591 */
1592 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1593 sk->sk_write_space(sk);
1594 len = 1;
1595 }
1596 /*
1597 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1598 * could not do because of in-flight packets
1599 */
1600 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1601 __sk_free(sk);
1602 }
1603 EXPORT_SYMBOL(sock_wfree);
1604
1605 void skb_orphan_partial(struct sk_buff *skb)
1606 {
1607 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1608 * so we do not completely orphan skb, but transfert all
1609 * accounted bytes but one, to avoid unexpected reorders.
1610 */
1611 if (skb->destructor == sock_wfree
1612 #ifdef CONFIG_INET
1613 || skb->destructor == tcp_wfree
1614 #endif
1615 ) {
1616 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1617 skb->truesize = 1;
1618 } else {
1619 skb_orphan(skb);
1620 }
1621 }
1622 EXPORT_SYMBOL(skb_orphan_partial);
1623
1624 /*
1625 * Read buffer destructor automatically called from kfree_skb.
1626 */
1627 void sock_rfree(struct sk_buff *skb)
1628 {
1629 struct sock *sk = skb->sk;
1630 unsigned int len = skb->truesize;
1631
1632 atomic_sub(len, &sk->sk_rmem_alloc);
1633 sk_mem_uncharge(sk, len);
1634 }
1635 EXPORT_SYMBOL(sock_rfree);
1636
1637 void sock_efree(struct sk_buff *skb)
1638 {
1639 sock_put(skb->sk);
1640 }
1641 EXPORT_SYMBOL(sock_efree);
1642
1643 #ifdef CONFIG_INET
1644 void sock_edemux(struct sk_buff *skb)
1645 {
1646 struct sock *sk = skb->sk;
1647
1648 if (sk->sk_state == TCP_TIME_WAIT)
1649 inet_twsk_put(inet_twsk(sk));
1650 else
1651 sock_put(sk);
1652 }
1653 EXPORT_SYMBOL(sock_edemux);
1654 #endif
1655
1656 kuid_t sock_i_uid(struct sock *sk)
1657 {
1658 kuid_t uid;
1659
1660 read_lock_bh(&sk->sk_callback_lock);
1661 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1662 read_unlock_bh(&sk->sk_callback_lock);
1663 return uid;
1664 }
1665 EXPORT_SYMBOL(sock_i_uid);
1666
1667 unsigned long sock_i_ino(struct sock *sk)
1668 {
1669 unsigned long ino;
1670
1671 read_lock_bh(&sk->sk_callback_lock);
1672 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1673 read_unlock_bh(&sk->sk_callback_lock);
1674 return ino;
1675 }
1676 EXPORT_SYMBOL(sock_i_ino);
1677
1678 /*
1679 * Allocate a skb from the socket's send buffer.
1680 */
1681 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1682 gfp_t priority)
1683 {
1684 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1685 struct sk_buff *skb = alloc_skb(size, priority);
1686 if (skb) {
1687 skb_set_owner_w(skb, sk);
1688 return skb;
1689 }
1690 }
1691 return NULL;
1692 }
1693 EXPORT_SYMBOL(sock_wmalloc);
1694
1695 /*
1696 * Allocate a memory block from the socket's option memory buffer.
1697 */
1698 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1699 {
1700 if ((unsigned int)size <= sysctl_optmem_max &&
1701 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1702 void *mem;
1703 /* First do the add, to avoid the race if kmalloc
1704 * might sleep.
1705 */
1706 atomic_add(size, &sk->sk_omem_alloc);
1707 mem = kmalloc(size, priority);
1708 if (mem)
1709 return mem;
1710 atomic_sub(size, &sk->sk_omem_alloc);
1711 }
1712 return NULL;
1713 }
1714 EXPORT_SYMBOL(sock_kmalloc);
1715
1716 /*
1717 * Free an option memory block.
1718 */
1719 void sock_kfree_s(struct sock *sk, void *mem, int size)
1720 {
1721 if (WARN_ON_ONCE(!mem))
1722 return;
1723 kfree(mem);
1724 atomic_sub(size, &sk->sk_omem_alloc);
1725 }
1726 EXPORT_SYMBOL(sock_kfree_s);
1727
1728 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1729 I think, these locks should be removed for datagram sockets.
1730 */
1731 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1732 {
1733 DEFINE_WAIT(wait);
1734
1735 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1736 for (;;) {
1737 if (!timeo)
1738 break;
1739 if (signal_pending(current))
1740 break;
1741 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1742 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1743 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1744 break;
1745 if (sk->sk_shutdown & SEND_SHUTDOWN)
1746 break;
1747 if (sk->sk_err)
1748 break;
1749 timeo = schedule_timeout(timeo);
1750 }
1751 finish_wait(sk_sleep(sk), &wait);
1752 return timeo;
1753 }
1754
1755
1756 /*
1757 * Generic send/receive buffer handlers
1758 */
1759
1760 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1761 unsigned long data_len, int noblock,
1762 int *errcode, int max_page_order)
1763 {
1764 struct sk_buff *skb;
1765 long timeo;
1766 int err;
1767
1768 timeo = sock_sndtimeo(sk, noblock);
1769 for (;;) {
1770 err = sock_error(sk);
1771 if (err != 0)
1772 goto failure;
1773
1774 err = -EPIPE;
1775 if (sk->sk_shutdown & SEND_SHUTDOWN)
1776 goto failure;
1777
1778 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1779 break;
1780
1781 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1782 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1783 err = -EAGAIN;
1784 if (!timeo)
1785 goto failure;
1786 if (signal_pending(current))
1787 goto interrupted;
1788 timeo = sock_wait_for_wmem(sk, timeo);
1789 }
1790 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1791 errcode, sk->sk_allocation);
1792 if (skb)
1793 skb_set_owner_w(skb, sk);
1794 return skb;
1795
1796 interrupted:
1797 err = sock_intr_errno(timeo);
1798 failure:
1799 *errcode = err;
1800 return NULL;
1801 }
1802 EXPORT_SYMBOL(sock_alloc_send_pskb);
1803
1804 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1805 int noblock, int *errcode)
1806 {
1807 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1808 }
1809 EXPORT_SYMBOL(sock_alloc_send_skb);
1810
1811 /* On 32bit arches, an skb frag is limited to 2^15 */
1812 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1813
1814 /**
1815 * skb_page_frag_refill - check that a page_frag contains enough room
1816 * @sz: minimum size of the fragment we want to get
1817 * @pfrag: pointer to page_frag
1818 * @gfp: priority for memory allocation
1819 *
1820 * Note: While this allocator tries to use high order pages, there is
1821 * no guarantee that allocations succeed. Therefore, @sz MUST be
1822 * less or equal than PAGE_SIZE.
1823 */
1824 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1825 {
1826 if (pfrag->page) {
1827 if (atomic_read(&pfrag->page->_count) == 1) {
1828 pfrag->offset = 0;
1829 return true;
1830 }
1831 if (pfrag->offset + sz <= pfrag->size)
1832 return true;
1833 put_page(pfrag->page);
1834 }
1835
1836 pfrag->offset = 0;
1837 if (SKB_FRAG_PAGE_ORDER) {
1838 pfrag->page = alloc_pages(gfp | __GFP_COMP |
1839 __GFP_NOWARN | __GFP_NORETRY,
1840 SKB_FRAG_PAGE_ORDER);
1841 if (likely(pfrag->page)) {
1842 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1843 return true;
1844 }
1845 }
1846 pfrag->page = alloc_page(gfp);
1847 if (likely(pfrag->page)) {
1848 pfrag->size = PAGE_SIZE;
1849 return true;
1850 }
1851 return false;
1852 }
1853 EXPORT_SYMBOL(skb_page_frag_refill);
1854
1855 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1856 {
1857 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1858 return true;
1859
1860 sk_enter_memory_pressure(sk);
1861 sk_stream_moderate_sndbuf(sk);
1862 return false;
1863 }
1864 EXPORT_SYMBOL(sk_page_frag_refill);
1865
1866 static void __lock_sock(struct sock *sk)
1867 __releases(&sk->sk_lock.slock)
1868 __acquires(&sk->sk_lock.slock)
1869 {
1870 DEFINE_WAIT(wait);
1871
1872 for (;;) {
1873 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1874 TASK_UNINTERRUPTIBLE);
1875 spin_unlock_bh(&sk->sk_lock.slock);
1876 schedule();
1877 spin_lock_bh(&sk->sk_lock.slock);
1878 if (!sock_owned_by_user(sk))
1879 break;
1880 }
1881 finish_wait(&sk->sk_lock.wq, &wait);
1882 }
1883
1884 static void __release_sock(struct sock *sk)
1885 __releases(&sk->sk_lock.slock)
1886 __acquires(&sk->sk_lock.slock)
1887 {
1888 struct sk_buff *skb = sk->sk_backlog.head;
1889
1890 do {
1891 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1892 bh_unlock_sock(sk);
1893
1894 do {
1895 struct sk_buff *next = skb->next;
1896
1897 prefetch(next);
1898 WARN_ON_ONCE(skb_dst_is_noref(skb));
1899 skb->next = NULL;
1900 sk_backlog_rcv(sk, skb);
1901
1902 /*
1903 * We are in process context here with softirqs
1904 * disabled, use cond_resched_softirq() to preempt.
1905 * This is safe to do because we've taken the backlog
1906 * queue private:
1907 */
1908 cond_resched_softirq();
1909
1910 skb = next;
1911 } while (skb != NULL);
1912
1913 bh_lock_sock(sk);
1914 } while ((skb = sk->sk_backlog.head) != NULL);
1915
1916 /*
1917 * Doing the zeroing here guarantee we can not loop forever
1918 * while a wild producer attempts to flood us.
1919 */
1920 sk->sk_backlog.len = 0;
1921 }
1922
1923 /**
1924 * sk_wait_data - wait for data to arrive at sk_receive_queue
1925 * @sk: sock to wait on
1926 * @timeo: for how long
1927 *
1928 * Now socket state including sk->sk_err is changed only under lock,
1929 * hence we may omit checks after joining wait queue.
1930 * We check receive queue before schedule() only as optimization;
1931 * it is very likely that release_sock() added new data.
1932 */
1933 int sk_wait_data(struct sock *sk, long *timeo)
1934 {
1935 int rc;
1936 DEFINE_WAIT(wait);
1937
1938 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1939 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1940 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1941 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1942 finish_wait(sk_sleep(sk), &wait);
1943 return rc;
1944 }
1945 EXPORT_SYMBOL(sk_wait_data);
1946
1947 /**
1948 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1949 * @sk: socket
1950 * @size: memory size to allocate
1951 * @kind: allocation type
1952 *
1953 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1954 * rmem allocation. This function assumes that protocols which have
1955 * memory_pressure use sk_wmem_queued as write buffer accounting.
1956 */
1957 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1958 {
1959 struct proto *prot = sk->sk_prot;
1960 int amt = sk_mem_pages(size);
1961 long allocated;
1962 int parent_status = UNDER_LIMIT;
1963
1964 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1965
1966 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1967
1968 /* Under limit. */
1969 if (parent_status == UNDER_LIMIT &&
1970 allocated <= sk_prot_mem_limits(sk, 0)) {
1971 sk_leave_memory_pressure(sk);
1972 return 1;
1973 }
1974
1975 /* Under pressure. (we or our parents) */
1976 if ((parent_status > SOFT_LIMIT) ||
1977 allocated > sk_prot_mem_limits(sk, 1))
1978 sk_enter_memory_pressure(sk);
1979
1980 /* Over hard limit (we or our parents) */
1981 if ((parent_status == OVER_LIMIT) ||
1982 (allocated > sk_prot_mem_limits(sk, 2)))
1983 goto suppress_allocation;
1984
1985 /* guarantee minimum buffer size under pressure */
1986 if (kind == SK_MEM_RECV) {
1987 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1988 return 1;
1989
1990 } else { /* SK_MEM_SEND */
1991 if (sk->sk_type == SOCK_STREAM) {
1992 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1993 return 1;
1994 } else if (atomic_read(&sk->sk_wmem_alloc) <
1995 prot->sysctl_wmem[0])
1996 return 1;
1997 }
1998
1999 if (sk_has_memory_pressure(sk)) {
2000 int alloc;
2001
2002 if (!sk_under_memory_pressure(sk))
2003 return 1;
2004 alloc = sk_sockets_allocated_read_positive(sk);
2005 if (sk_prot_mem_limits(sk, 2) > alloc *
2006 sk_mem_pages(sk->sk_wmem_queued +
2007 atomic_read(&sk->sk_rmem_alloc) +
2008 sk->sk_forward_alloc))
2009 return 1;
2010 }
2011
2012 suppress_allocation:
2013
2014 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2015 sk_stream_moderate_sndbuf(sk);
2016
2017 /* Fail only if socket is _under_ its sndbuf.
2018 * In this case we cannot block, so that we have to fail.
2019 */
2020 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2021 return 1;
2022 }
2023
2024 trace_sock_exceed_buf_limit(sk, prot, allocated);
2025
2026 /* Alas. Undo changes. */
2027 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2028
2029 sk_memory_allocated_sub(sk, amt);
2030
2031 return 0;
2032 }
2033 EXPORT_SYMBOL(__sk_mem_schedule);
2034
2035 /**
2036 * __sk_reclaim - reclaim memory_allocated
2037 * @sk: socket
2038 */
2039 void __sk_mem_reclaim(struct sock *sk)
2040 {
2041 sk_memory_allocated_sub(sk,
2042 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2043 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2044
2045 if (sk_under_memory_pressure(sk) &&
2046 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2047 sk_leave_memory_pressure(sk);
2048 }
2049 EXPORT_SYMBOL(__sk_mem_reclaim);
2050
2051
2052 /*
2053 * Set of default routines for initialising struct proto_ops when
2054 * the protocol does not support a particular function. In certain
2055 * cases where it makes no sense for a protocol to have a "do nothing"
2056 * function, some default processing is provided.
2057 */
2058
2059 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2060 {
2061 return -EOPNOTSUPP;
2062 }
2063 EXPORT_SYMBOL(sock_no_bind);
2064
2065 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2066 int len, int flags)
2067 {
2068 return -EOPNOTSUPP;
2069 }
2070 EXPORT_SYMBOL(sock_no_connect);
2071
2072 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2073 {
2074 return -EOPNOTSUPP;
2075 }
2076 EXPORT_SYMBOL(sock_no_socketpair);
2077
2078 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2079 {
2080 return -EOPNOTSUPP;
2081 }
2082 EXPORT_SYMBOL(sock_no_accept);
2083
2084 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2085 int *len, int peer)
2086 {
2087 return -EOPNOTSUPP;
2088 }
2089 EXPORT_SYMBOL(sock_no_getname);
2090
2091 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2092 {
2093 return 0;
2094 }
2095 EXPORT_SYMBOL(sock_no_poll);
2096
2097 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2098 {
2099 return -EOPNOTSUPP;
2100 }
2101 EXPORT_SYMBOL(sock_no_ioctl);
2102
2103 int sock_no_listen(struct socket *sock, int backlog)
2104 {
2105 return -EOPNOTSUPP;
2106 }
2107 EXPORT_SYMBOL(sock_no_listen);
2108
2109 int sock_no_shutdown(struct socket *sock, int how)
2110 {
2111 return -EOPNOTSUPP;
2112 }
2113 EXPORT_SYMBOL(sock_no_shutdown);
2114
2115 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2116 char __user *optval, unsigned int optlen)
2117 {
2118 return -EOPNOTSUPP;
2119 }
2120 EXPORT_SYMBOL(sock_no_setsockopt);
2121
2122 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2123 char __user *optval, int __user *optlen)
2124 {
2125 return -EOPNOTSUPP;
2126 }
2127 EXPORT_SYMBOL(sock_no_getsockopt);
2128
2129 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2130 size_t len)
2131 {
2132 return -EOPNOTSUPP;
2133 }
2134 EXPORT_SYMBOL(sock_no_sendmsg);
2135
2136 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2137 size_t len, int flags)
2138 {
2139 return -EOPNOTSUPP;
2140 }
2141 EXPORT_SYMBOL(sock_no_recvmsg);
2142
2143 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2144 {
2145 /* Mirror missing mmap method error code */
2146 return -ENODEV;
2147 }
2148 EXPORT_SYMBOL(sock_no_mmap);
2149
2150 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2151 {
2152 ssize_t res;
2153 struct msghdr msg = {.msg_flags = flags};
2154 struct kvec iov;
2155 char *kaddr = kmap(page);
2156 iov.iov_base = kaddr + offset;
2157 iov.iov_len = size;
2158 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2159 kunmap(page);
2160 return res;
2161 }
2162 EXPORT_SYMBOL(sock_no_sendpage);
2163
2164 /*
2165 * Default Socket Callbacks
2166 */
2167
2168 static void sock_def_wakeup(struct sock *sk)
2169 {
2170 struct socket_wq *wq;
2171
2172 rcu_read_lock();
2173 wq = rcu_dereference(sk->sk_wq);
2174 if (wq_has_sleeper(wq))
2175 wake_up_interruptible_all(&wq->wait);
2176 rcu_read_unlock();
2177 }
2178
2179 static void sock_def_error_report(struct sock *sk)
2180 {
2181 struct socket_wq *wq;
2182
2183 rcu_read_lock();
2184 wq = rcu_dereference(sk->sk_wq);
2185 if (wq_has_sleeper(wq))
2186 wake_up_interruptible_poll(&wq->wait, POLLERR);
2187 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2188 rcu_read_unlock();
2189 }
2190
2191 static void sock_def_readable(struct sock *sk)
2192 {
2193 struct socket_wq *wq;
2194
2195 rcu_read_lock();
2196 wq = rcu_dereference(sk->sk_wq);
2197 if (wq_has_sleeper(wq))
2198 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2199 POLLRDNORM | POLLRDBAND);
2200 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2201 rcu_read_unlock();
2202 }
2203
2204 static void sock_def_write_space(struct sock *sk)
2205 {
2206 struct socket_wq *wq;
2207
2208 rcu_read_lock();
2209
2210 /* Do not wake up a writer until he can make "significant"
2211 * progress. --DaveM
2212 */
2213 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2214 wq = rcu_dereference(sk->sk_wq);
2215 if (wq_has_sleeper(wq))
2216 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2217 POLLWRNORM | POLLWRBAND);
2218
2219 /* Should agree with poll, otherwise some programs break */
2220 if (sock_writeable(sk))
2221 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2222 }
2223
2224 rcu_read_unlock();
2225 }
2226
2227 static void sock_def_destruct(struct sock *sk)
2228 {
2229 kfree(sk->sk_protinfo);
2230 }
2231
2232 void sk_send_sigurg(struct sock *sk)
2233 {
2234 if (sk->sk_socket && sk->sk_socket->file)
2235 if (send_sigurg(&sk->sk_socket->file->f_owner))
2236 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2237 }
2238 EXPORT_SYMBOL(sk_send_sigurg);
2239
2240 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2241 unsigned long expires)
2242 {
2243 if (!mod_timer(timer, expires))
2244 sock_hold(sk);
2245 }
2246 EXPORT_SYMBOL(sk_reset_timer);
2247
2248 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2249 {
2250 if (del_timer(timer))
2251 __sock_put(sk);
2252 }
2253 EXPORT_SYMBOL(sk_stop_timer);
2254
2255 void sock_init_data(struct socket *sock, struct sock *sk)
2256 {
2257 skb_queue_head_init(&sk->sk_receive_queue);
2258 skb_queue_head_init(&sk->sk_write_queue);
2259 skb_queue_head_init(&sk->sk_error_queue);
2260
2261 sk->sk_send_head = NULL;
2262
2263 init_timer(&sk->sk_timer);
2264
2265 sk->sk_allocation = GFP_KERNEL;
2266 sk->sk_rcvbuf = sysctl_rmem_default;
2267 sk->sk_sndbuf = sysctl_wmem_default;
2268 sk->sk_state = TCP_CLOSE;
2269 sk_set_socket(sk, sock);
2270
2271 sock_set_flag(sk, SOCK_ZAPPED);
2272
2273 if (sock) {
2274 sk->sk_type = sock->type;
2275 sk->sk_wq = sock->wq;
2276 sock->sk = sk;
2277 } else
2278 sk->sk_wq = NULL;
2279
2280 spin_lock_init(&sk->sk_dst_lock);
2281 rwlock_init(&sk->sk_callback_lock);
2282 lockdep_set_class_and_name(&sk->sk_callback_lock,
2283 af_callback_keys + sk->sk_family,
2284 af_family_clock_key_strings[sk->sk_family]);
2285
2286 sk->sk_state_change = sock_def_wakeup;
2287 sk->sk_data_ready = sock_def_readable;
2288 sk->sk_write_space = sock_def_write_space;
2289 sk->sk_error_report = sock_def_error_report;
2290 sk->sk_destruct = sock_def_destruct;
2291
2292 sk->sk_frag.page = NULL;
2293 sk->sk_frag.offset = 0;
2294 sk->sk_peek_off = -1;
2295
2296 sk->sk_peer_pid = NULL;
2297 sk->sk_peer_cred = NULL;
2298 sk->sk_write_pending = 0;
2299 sk->sk_rcvlowat = 1;
2300 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2301 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2302
2303 sk->sk_stamp = ktime_set(-1L, 0);
2304
2305 #ifdef CONFIG_NET_RX_BUSY_POLL
2306 sk->sk_napi_id = 0;
2307 sk->sk_ll_usec = sysctl_net_busy_read;
2308 #endif
2309
2310 sk->sk_max_pacing_rate = ~0U;
2311 sk->sk_pacing_rate = ~0U;
2312 /*
2313 * Before updating sk_refcnt, we must commit prior changes to memory
2314 * (Documentation/RCU/rculist_nulls.txt for details)
2315 */
2316 smp_wmb();
2317 atomic_set(&sk->sk_refcnt, 1);
2318 atomic_set(&sk->sk_drops, 0);
2319 }
2320 EXPORT_SYMBOL(sock_init_data);
2321
2322 void lock_sock_nested(struct sock *sk, int subclass)
2323 {
2324 might_sleep();
2325 spin_lock_bh(&sk->sk_lock.slock);
2326 if (sk->sk_lock.owned)
2327 __lock_sock(sk);
2328 sk->sk_lock.owned = 1;
2329 spin_unlock(&sk->sk_lock.slock);
2330 /*
2331 * The sk_lock has mutex_lock() semantics here:
2332 */
2333 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2334 local_bh_enable();
2335 }
2336 EXPORT_SYMBOL(lock_sock_nested);
2337
2338 void release_sock(struct sock *sk)
2339 {
2340 /*
2341 * The sk_lock has mutex_unlock() semantics:
2342 */
2343 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2344
2345 spin_lock_bh(&sk->sk_lock.slock);
2346 if (sk->sk_backlog.tail)
2347 __release_sock(sk);
2348
2349 /* Warning : release_cb() might need to release sk ownership,
2350 * ie call sock_release_ownership(sk) before us.
2351 */
2352 if (sk->sk_prot->release_cb)
2353 sk->sk_prot->release_cb(sk);
2354
2355 sock_release_ownership(sk);
2356 if (waitqueue_active(&sk->sk_lock.wq))
2357 wake_up(&sk->sk_lock.wq);
2358 spin_unlock_bh(&sk->sk_lock.slock);
2359 }
2360 EXPORT_SYMBOL(release_sock);
2361
2362 /**
2363 * lock_sock_fast - fast version of lock_sock
2364 * @sk: socket
2365 *
2366 * This version should be used for very small section, where process wont block
2367 * return false if fast path is taken
2368 * sk_lock.slock locked, owned = 0, BH disabled
2369 * return true if slow path is taken
2370 * sk_lock.slock unlocked, owned = 1, BH enabled
2371 */
2372 bool lock_sock_fast(struct sock *sk)
2373 {
2374 might_sleep();
2375 spin_lock_bh(&sk->sk_lock.slock);
2376
2377 if (!sk->sk_lock.owned)
2378 /*
2379 * Note : We must disable BH
2380 */
2381 return false;
2382
2383 __lock_sock(sk);
2384 sk->sk_lock.owned = 1;
2385 spin_unlock(&sk->sk_lock.slock);
2386 /*
2387 * The sk_lock has mutex_lock() semantics here:
2388 */
2389 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2390 local_bh_enable();
2391 return true;
2392 }
2393 EXPORT_SYMBOL(lock_sock_fast);
2394
2395 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2396 {
2397 struct timeval tv;
2398 if (!sock_flag(sk, SOCK_TIMESTAMP))
2399 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2400 tv = ktime_to_timeval(sk->sk_stamp);
2401 if (tv.tv_sec == -1)
2402 return -ENOENT;
2403 if (tv.tv_sec == 0) {
2404 sk->sk_stamp = ktime_get_real();
2405 tv = ktime_to_timeval(sk->sk_stamp);
2406 }
2407 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2408 }
2409 EXPORT_SYMBOL(sock_get_timestamp);
2410
2411 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2412 {
2413 struct timespec ts;
2414 if (!sock_flag(sk, SOCK_TIMESTAMP))
2415 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2416 ts = ktime_to_timespec(sk->sk_stamp);
2417 if (ts.tv_sec == -1)
2418 return -ENOENT;
2419 if (ts.tv_sec == 0) {
2420 sk->sk_stamp = ktime_get_real();
2421 ts = ktime_to_timespec(sk->sk_stamp);
2422 }
2423 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2424 }
2425 EXPORT_SYMBOL(sock_get_timestampns);
2426
2427 void sock_enable_timestamp(struct sock *sk, int flag)
2428 {
2429 if (!sock_flag(sk, flag)) {
2430 unsigned long previous_flags = sk->sk_flags;
2431
2432 sock_set_flag(sk, flag);
2433 /*
2434 * we just set one of the two flags which require net
2435 * time stamping, but time stamping might have been on
2436 * already because of the other one
2437 */
2438 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2439 net_enable_timestamp();
2440 }
2441 }
2442
2443 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2444 int level, int type)
2445 {
2446 struct sock_exterr_skb *serr;
2447 struct sk_buff *skb;
2448 int copied, err;
2449
2450 err = -EAGAIN;
2451 skb = sock_dequeue_err_skb(sk);
2452 if (skb == NULL)
2453 goto out;
2454
2455 copied = skb->len;
2456 if (copied > len) {
2457 msg->msg_flags |= MSG_TRUNC;
2458 copied = len;
2459 }
2460 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2461 if (err)
2462 goto out_free_skb;
2463
2464 sock_recv_timestamp(msg, sk, skb);
2465
2466 serr = SKB_EXT_ERR(skb);
2467 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2468
2469 msg->msg_flags |= MSG_ERRQUEUE;
2470 err = copied;
2471
2472 out_free_skb:
2473 kfree_skb(skb);
2474 out:
2475 return err;
2476 }
2477 EXPORT_SYMBOL(sock_recv_errqueue);
2478
2479 /*
2480 * Get a socket option on an socket.
2481 *
2482 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2483 * asynchronous errors should be reported by getsockopt. We assume
2484 * this means if you specify SO_ERROR (otherwise whats the point of it).
2485 */
2486 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2487 char __user *optval, int __user *optlen)
2488 {
2489 struct sock *sk = sock->sk;
2490
2491 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2492 }
2493 EXPORT_SYMBOL(sock_common_getsockopt);
2494
2495 #ifdef CONFIG_COMPAT
2496 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2497 char __user *optval, int __user *optlen)
2498 {
2499 struct sock *sk = sock->sk;
2500
2501 if (sk->sk_prot->compat_getsockopt != NULL)
2502 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2503 optval, optlen);
2504 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2505 }
2506 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2507 #endif
2508
2509 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2510 struct msghdr *msg, size_t size, int flags)
2511 {
2512 struct sock *sk = sock->sk;
2513 int addr_len = 0;
2514 int err;
2515
2516 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2517 flags & ~MSG_DONTWAIT, &addr_len);
2518 if (err >= 0)
2519 msg->msg_namelen = addr_len;
2520 return err;
2521 }
2522 EXPORT_SYMBOL(sock_common_recvmsg);
2523
2524 /*
2525 * Set socket options on an inet socket.
2526 */
2527 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2528 char __user *optval, unsigned int optlen)
2529 {
2530 struct sock *sk = sock->sk;
2531
2532 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2533 }
2534 EXPORT_SYMBOL(sock_common_setsockopt);
2535
2536 #ifdef CONFIG_COMPAT
2537 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2538 char __user *optval, unsigned int optlen)
2539 {
2540 struct sock *sk = sock->sk;
2541
2542 if (sk->sk_prot->compat_setsockopt != NULL)
2543 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2544 optval, optlen);
2545 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2546 }
2547 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2548 #endif
2549
2550 void sk_common_release(struct sock *sk)
2551 {
2552 if (sk->sk_prot->destroy)
2553 sk->sk_prot->destroy(sk);
2554
2555 /*
2556 * Observation: when sock_common_release is called, processes have
2557 * no access to socket. But net still has.
2558 * Step one, detach it from networking:
2559 *
2560 * A. Remove from hash tables.
2561 */
2562
2563 sk->sk_prot->unhash(sk);
2564
2565 /*
2566 * In this point socket cannot receive new packets, but it is possible
2567 * that some packets are in flight because some CPU runs receiver and
2568 * did hash table lookup before we unhashed socket. They will achieve
2569 * receive queue and will be purged by socket destructor.
2570 *
2571 * Also we still have packets pending on receive queue and probably,
2572 * our own packets waiting in device queues. sock_destroy will drain
2573 * receive queue, but transmitted packets will delay socket destruction
2574 * until the last reference will be released.
2575 */
2576
2577 sock_orphan(sk);
2578
2579 xfrm_sk_free_policy(sk);
2580
2581 sk_refcnt_debug_release(sk);
2582
2583 if (sk->sk_frag.page) {
2584 put_page(sk->sk_frag.page);
2585 sk->sk_frag.page = NULL;
2586 }
2587
2588 sock_put(sk);
2589 }
2590 EXPORT_SYMBOL(sk_common_release);
2591
2592 #ifdef CONFIG_PROC_FS
2593 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2594 struct prot_inuse {
2595 int val[PROTO_INUSE_NR];
2596 };
2597
2598 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2599
2600 #ifdef CONFIG_NET_NS
2601 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2602 {
2603 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2604 }
2605 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2606
2607 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2608 {
2609 int cpu, idx = prot->inuse_idx;
2610 int res = 0;
2611
2612 for_each_possible_cpu(cpu)
2613 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2614
2615 return res >= 0 ? res : 0;
2616 }
2617 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2618
2619 static int __net_init sock_inuse_init_net(struct net *net)
2620 {
2621 net->core.inuse = alloc_percpu(struct prot_inuse);
2622 return net->core.inuse ? 0 : -ENOMEM;
2623 }
2624
2625 static void __net_exit sock_inuse_exit_net(struct net *net)
2626 {
2627 free_percpu(net->core.inuse);
2628 }
2629
2630 static struct pernet_operations net_inuse_ops = {
2631 .init = sock_inuse_init_net,
2632 .exit = sock_inuse_exit_net,
2633 };
2634
2635 static __init int net_inuse_init(void)
2636 {
2637 if (register_pernet_subsys(&net_inuse_ops))
2638 panic("Cannot initialize net inuse counters");
2639
2640 return 0;
2641 }
2642
2643 core_initcall(net_inuse_init);
2644 #else
2645 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2646
2647 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2648 {
2649 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2650 }
2651 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2652
2653 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2654 {
2655 int cpu, idx = prot->inuse_idx;
2656 int res = 0;
2657
2658 for_each_possible_cpu(cpu)
2659 res += per_cpu(prot_inuse, cpu).val[idx];
2660
2661 return res >= 0 ? res : 0;
2662 }
2663 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2664 #endif
2665
2666 static void assign_proto_idx(struct proto *prot)
2667 {
2668 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2669
2670 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2671 pr_err("PROTO_INUSE_NR exhausted\n");
2672 return;
2673 }
2674
2675 set_bit(prot->inuse_idx, proto_inuse_idx);
2676 }
2677
2678 static void release_proto_idx(struct proto *prot)
2679 {
2680 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2681 clear_bit(prot->inuse_idx, proto_inuse_idx);
2682 }
2683 #else
2684 static inline void assign_proto_idx(struct proto *prot)
2685 {
2686 }
2687
2688 static inline void release_proto_idx(struct proto *prot)
2689 {
2690 }
2691 #endif
2692
2693 int proto_register(struct proto *prot, int alloc_slab)
2694 {
2695 if (alloc_slab) {
2696 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2697 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2698 NULL);
2699
2700 if (prot->slab == NULL) {
2701 pr_crit("%s: Can't create sock SLAB cache!\n",
2702 prot->name);
2703 goto out;
2704 }
2705
2706 if (prot->rsk_prot != NULL) {
2707 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2708 if (prot->rsk_prot->slab_name == NULL)
2709 goto out_free_sock_slab;
2710
2711 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2712 prot->rsk_prot->obj_size, 0,
2713 SLAB_HWCACHE_ALIGN, NULL);
2714
2715 if (prot->rsk_prot->slab == NULL) {
2716 pr_crit("%s: Can't create request sock SLAB cache!\n",
2717 prot->name);
2718 goto out_free_request_sock_slab_name;
2719 }
2720 }
2721
2722 if (prot->twsk_prot != NULL) {
2723 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2724
2725 if (prot->twsk_prot->twsk_slab_name == NULL)
2726 goto out_free_request_sock_slab;
2727
2728 prot->twsk_prot->twsk_slab =
2729 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2730 prot->twsk_prot->twsk_obj_size,
2731 0,
2732 SLAB_HWCACHE_ALIGN |
2733 prot->slab_flags,
2734 NULL);
2735 if (prot->twsk_prot->twsk_slab == NULL)
2736 goto out_free_timewait_sock_slab_name;
2737 }
2738 }
2739
2740 mutex_lock(&proto_list_mutex);
2741 list_add(&prot->node, &proto_list);
2742 assign_proto_idx(prot);
2743 mutex_unlock(&proto_list_mutex);
2744 return 0;
2745
2746 out_free_timewait_sock_slab_name:
2747 kfree(prot->twsk_prot->twsk_slab_name);
2748 out_free_request_sock_slab:
2749 if (prot->rsk_prot && prot->rsk_prot->slab) {
2750 kmem_cache_destroy(prot->rsk_prot->slab);
2751 prot->rsk_prot->slab = NULL;
2752 }
2753 out_free_request_sock_slab_name:
2754 if (prot->rsk_prot)
2755 kfree(prot->rsk_prot->slab_name);
2756 out_free_sock_slab:
2757 kmem_cache_destroy(prot->slab);
2758 prot->slab = NULL;
2759 out:
2760 return -ENOBUFS;
2761 }
2762 EXPORT_SYMBOL(proto_register);
2763
2764 void proto_unregister(struct proto *prot)
2765 {
2766 mutex_lock(&proto_list_mutex);
2767 release_proto_idx(prot);
2768 list_del(&prot->node);
2769 mutex_unlock(&proto_list_mutex);
2770
2771 if (prot->slab != NULL) {
2772 kmem_cache_destroy(prot->slab);
2773 prot->slab = NULL;
2774 }
2775
2776 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2777 kmem_cache_destroy(prot->rsk_prot->slab);
2778 kfree(prot->rsk_prot->slab_name);
2779 prot->rsk_prot->slab = NULL;
2780 }
2781
2782 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2783 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2784 kfree(prot->twsk_prot->twsk_slab_name);
2785 prot->twsk_prot->twsk_slab = NULL;
2786 }
2787 }
2788 EXPORT_SYMBOL(proto_unregister);
2789
2790 #ifdef CONFIG_PROC_FS
2791 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2792 __acquires(proto_list_mutex)
2793 {
2794 mutex_lock(&proto_list_mutex);
2795 return seq_list_start_head(&proto_list, *pos);
2796 }
2797
2798 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2799 {
2800 return seq_list_next(v, &proto_list, pos);
2801 }
2802
2803 static void proto_seq_stop(struct seq_file *seq, void *v)
2804 __releases(proto_list_mutex)
2805 {
2806 mutex_unlock(&proto_list_mutex);
2807 }
2808
2809 static char proto_method_implemented(const void *method)
2810 {
2811 return method == NULL ? 'n' : 'y';
2812 }
2813 static long sock_prot_memory_allocated(struct proto *proto)
2814 {
2815 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2816 }
2817
2818 static char *sock_prot_memory_pressure(struct proto *proto)
2819 {
2820 return proto->memory_pressure != NULL ?
2821 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2822 }
2823
2824 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2825 {
2826
2827 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2828 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2829 proto->name,
2830 proto->obj_size,
2831 sock_prot_inuse_get(seq_file_net(seq), proto),
2832 sock_prot_memory_allocated(proto),
2833 sock_prot_memory_pressure(proto),
2834 proto->max_header,
2835 proto->slab == NULL ? "no" : "yes",
2836 module_name(proto->owner),
2837 proto_method_implemented(proto->close),
2838 proto_method_implemented(proto->connect),
2839 proto_method_implemented(proto->disconnect),
2840 proto_method_implemented(proto->accept),
2841 proto_method_implemented(proto->ioctl),
2842 proto_method_implemented(proto->init),
2843 proto_method_implemented(proto->destroy),
2844 proto_method_implemented(proto->shutdown),
2845 proto_method_implemented(proto->setsockopt),
2846 proto_method_implemented(proto->getsockopt),
2847 proto_method_implemented(proto->sendmsg),
2848 proto_method_implemented(proto->recvmsg),
2849 proto_method_implemented(proto->sendpage),
2850 proto_method_implemented(proto->bind),
2851 proto_method_implemented(proto->backlog_rcv),
2852 proto_method_implemented(proto->hash),
2853 proto_method_implemented(proto->unhash),
2854 proto_method_implemented(proto->get_port),
2855 proto_method_implemented(proto->enter_memory_pressure));
2856 }
2857
2858 static int proto_seq_show(struct seq_file *seq, void *v)
2859 {
2860 if (v == &proto_list)
2861 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2862 "protocol",
2863 "size",
2864 "sockets",
2865 "memory",
2866 "press",
2867 "maxhdr",
2868 "slab",
2869 "module",
2870 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2871 else
2872 proto_seq_printf(seq, list_entry(v, struct proto, node));
2873 return 0;
2874 }
2875
2876 static const struct seq_operations proto_seq_ops = {
2877 .start = proto_seq_start,
2878 .next = proto_seq_next,
2879 .stop = proto_seq_stop,
2880 .show = proto_seq_show,
2881 };
2882
2883 static int proto_seq_open(struct inode *inode, struct file *file)
2884 {
2885 return seq_open_net(inode, file, &proto_seq_ops,
2886 sizeof(struct seq_net_private));
2887 }
2888
2889 static const struct file_operations proto_seq_fops = {
2890 .owner = THIS_MODULE,
2891 .open = proto_seq_open,
2892 .read = seq_read,
2893 .llseek = seq_lseek,
2894 .release = seq_release_net,
2895 };
2896
2897 static __net_init int proto_init_net(struct net *net)
2898 {
2899 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2900 return -ENOMEM;
2901
2902 return 0;
2903 }
2904
2905 static __net_exit void proto_exit_net(struct net *net)
2906 {
2907 remove_proc_entry("protocols", net->proc_net);
2908 }
2909
2910
2911 static __net_initdata struct pernet_operations proto_net_ops = {
2912 .init = proto_init_net,
2913 .exit = proto_exit_net,
2914 };
2915
2916 static int __init proto_init(void)
2917 {
2918 return register_pernet_subsys(&proto_net_ops);
2919 }
2920
2921 subsys_initcall(proto_init);
2922
2923 #endif /* PROC_FS */
Linux with added FPC-III support