?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mm/zsmalloc: allocate exactly size of struct zs_pool
[linux/fpc-iii.git] / net / core / sock.c
blob9a56b2000c3f374fb95aedada3327447816a9512
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_ATTACH_BPF:
892 ret = -EINVAL;
893 if (optlen == sizeof(u32)) {
894 u32 ufd;
895
896 ret = -EFAULT;
897 if (copy_from_user(&ufd, optval, sizeof(ufd)))
898 break;
899
900 ret = sk_attach_bpf(ufd, sk);
901 }
902 break;
903
904 case SO_DETACH_FILTER:
905 ret = sk_detach_filter(sk);
906 break;
907
908 case SO_LOCK_FILTER:
909 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
910 ret = -EPERM;
911 else
912 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
913 break;
914
915 case SO_PASSSEC:
916 if (valbool)
917 set_bit(SOCK_PASSSEC, &sock->flags);
918 else
919 clear_bit(SOCK_PASSSEC, &sock->flags);
920 break;
921 case SO_MARK:
922 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
923 ret = -EPERM;
924 else
925 sk->sk_mark = val;
926 break;
927
928 /* We implement the SO_SNDLOWAT etc to
929 not be settable (1003.1g 5.3) */
930 case SO_RXQ_OVFL:
931 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
932 break;
933
934 case SO_WIFI_STATUS:
935 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
936 break;
937
938 case SO_PEEK_OFF:
939 if (sock->ops->set_peek_off)
940 ret = sock->ops->set_peek_off(sk, val);
941 else
942 ret = -EOPNOTSUPP;
943 break;
944
945 case SO_NOFCS:
946 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
947 break;
948
949 case SO_SELECT_ERR_QUEUE:
950 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
951 break;
952
953 #ifdef CONFIG_NET_RX_BUSY_POLL
954 case SO_BUSY_POLL:
955 /* allow unprivileged users to decrease the value */
956 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
957 ret = -EPERM;
958 else {
959 if (val < 0)
960 ret = -EINVAL;
961 else
962 sk->sk_ll_usec = val;
963 }
964 break;
965 #endif
966
967 case SO_MAX_PACING_RATE:
968 sk->sk_max_pacing_rate = val;
969 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
970 sk->sk_max_pacing_rate);
971 break;
972
973 default:
974 ret = -ENOPROTOOPT;
975 break;
976 }
977 release_sock(sk);
978 return ret;
979 }
980 EXPORT_SYMBOL(sock_setsockopt);
981
982
983 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
984 struct ucred *ucred)
985 {
986 ucred->pid = pid_vnr(pid);
987 ucred->uid = ucred->gid = -1;
988 if (cred) {
989 struct user_namespace *current_ns = current_user_ns();
990
991 ucred->uid = from_kuid_munged(current_ns, cred->euid);
992 ucred->gid = from_kgid_munged(current_ns, cred->egid);
993 }
994 }
995
996 int sock_getsockopt(struct socket *sock, int level, int optname,
997 char __user *optval, int __user *optlen)
998 {
999 struct sock *sk = sock->sk;
1000
1001 union {
1002 int val;
1003 struct linger ling;
1004 struct timeval tm;
1005 } v;
1006
1007 int lv = sizeof(int);
1008 int len;
1009
1010 if (get_user(len, optlen))
1011 return -EFAULT;
1012 if (len < 0)
1013 return -EINVAL;
1014
1015 memset(&v, 0, sizeof(v));
1016
1017 switch (optname) {
1018 case SO_DEBUG:
1019 v.val = sock_flag(sk, SOCK_DBG);
1020 break;
1021
1022 case SO_DONTROUTE:
1023 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1024 break;
1025
1026 case SO_BROADCAST:
1027 v.val = sock_flag(sk, SOCK_BROADCAST);
1028 break;
1029
1030 case SO_SNDBUF:
1031 v.val = sk->sk_sndbuf;
1032 break;
1033
1034 case SO_RCVBUF:
1035 v.val = sk->sk_rcvbuf;
1036 break;
1037
1038 case SO_REUSEADDR:
1039 v.val = sk->sk_reuse;
1040 break;
1041
1042 case SO_REUSEPORT:
1043 v.val = sk->sk_reuseport;
1044 break;
1045
1046 case SO_KEEPALIVE:
1047 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1048 break;
1049
1050 case SO_TYPE:
1051 v.val = sk->sk_type;
1052 break;
1053
1054 case SO_PROTOCOL:
1055 v.val = sk->sk_protocol;
1056 break;
1057
1058 case SO_DOMAIN:
1059 v.val = sk->sk_family;
1060 break;
1061
1062 case SO_ERROR:
1063 v.val = -sock_error(sk);
1064 if (v.val == 0)
1065 v.val = xchg(&sk->sk_err_soft, 0);
1066 break;
1067
1068 case SO_OOBINLINE:
1069 v.val = sock_flag(sk, SOCK_URGINLINE);
1070 break;
1071
1072 case SO_NO_CHECK:
1073 v.val = sk->sk_no_check_tx;
1074 break;
1075
1076 case SO_PRIORITY:
1077 v.val = sk->sk_priority;
1078 break;
1079
1080 case SO_LINGER:
1081 lv = sizeof(v.ling);
1082 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1083 v.ling.l_linger = sk->sk_lingertime / HZ;
1084 break;
1085
1086 case SO_BSDCOMPAT:
1087 sock_warn_obsolete_bsdism("getsockopt");
1088 break;
1089
1090 case SO_TIMESTAMP:
1091 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1092 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1093 break;
1094
1095 case SO_TIMESTAMPNS:
1096 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1097 break;
1098
1099 case SO_TIMESTAMPING:
1100 v.val = sk->sk_tsflags;
1101 break;
1102
1103 case SO_RCVTIMEO:
1104 lv = sizeof(struct timeval);
1105 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1106 v.tm.tv_sec = 0;
1107 v.tm.tv_usec = 0;
1108 } else {
1109 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1110 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1111 }
1112 break;
1113
1114 case SO_SNDTIMEO:
1115 lv = sizeof(struct timeval);
1116 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1117 v.tm.tv_sec = 0;
1118 v.tm.tv_usec = 0;
1119 } else {
1120 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1121 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1122 }
1123 break;
1124
1125 case SO_RCVLOWAT:
1126 v.val = sk->sk_rcvlowat;
1127 break;
1128
1129 case SO_SNDLOWAT:
1130 v.val = 1;
1131 break;
1132
1133 case SO_PASSCRED:
1134 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1135 break;
1136
1137 case SO_PEERCRED:
1138 {
1139 struct ucred peercred;
1140 if (len > sizeof(peercred))
1141 len = sizeof(peercred);
1142 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1143 if (copy_to_user(optval, &peercred, len))
1144 return -EFAULT;
1145 goto lenout;
1146 }
1147
1148 case SO_PEERNAME:
1149 {
1150 char address[128];
1151
1152 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1153 return -ENOTCONN;
1154 if (lv < len)
1155 return -EINVAL;
1156 if (copy_to_user(optval, address, len))
1157 return -EFAULT;
1158 goto lenout;
1159 }
1160
1161 /* Dubious BSD thing... Probably nobody even uses it, but
1162 * the UNIX standard wants it for whatever reason... -DaveM
1163 */
1164 case SO_ACCEPTCONN:
1165 v.val = sk->sk_state == TCP_LISTEN;
1166 break;
1167
1168 case SO_PASSSEC:
1169 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1170 break;
1171
1172 case SO_PEERSEC:
1173 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1174
1175 case SO_MARK:
1176 v.val = sk->sk_mark;
1177 break;
1178
1179 case SO_RXQ_OVFL:
1180 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1181 break;
1182
1183 case SO_WIFI_STATUS:
1184 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1185 break;
1186
1187 case SO_PEEK_OFF:
1188 if (!sock->ops->set_peek_off)
1189 return -EOPNOTSUPP;
1190
1191 v.val = sk->sk_peek_off;
1192 break;
1193 case SO_NOFCS:
1194 v.val = sock_flag(sk, SOCK_NOFCS);
1195 break;
1196
1197 case SO_BINDTODEVICE:
1198 return sock_getbindtodevice(sk, optval, optlen, len);
1199
1200 case SO_GET_FILTER:
1201 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1202 if (len < 0)
1203 return len;
1204
1205 goto lenout;
1206
1207 case SO_LOCK_FILTER:
1208 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1209 break;
1210
1211 case SO_BPF_EXTENSIONS:
1212 v.val = bpf_tell_extensions();
1213 break;
1214
1215 case SO_SELECT_ERR_QUEUE:
1216 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1217 break;
1218
1219 #ifdef CONFIG_NET_RX_BUSY_POLL
1220 case SO_BUSY_POLL:
1221 v.val = sk->sk_ll_usec;
1222 break;
1223 #endif
1224
1225 case SO_MAX_PACING_RATE:
1226 v.val = sk->sk_max_pacing_rate;
1227 break;
1228
1229 case SO_INCOMING_CPU:
1230 v.val = sk->sk_incoming_cpu;
1231 break;
1232
1233 default:
1234 return -ENOPROTOOPT;
1235 }
1236
1237 if (len > lv)
1238 len = lv;
1239 if (copy_to_user(optval, &v, len))
1240 return -EFAULT;
1241 lenout:
1242 if (put_user(len, optlen))
1243 return -EFAULT;
1244 return 0;
1245 }
1246
1247 /*
1248 * Initialize an sk_lock.
1249 *
1250 * (We also register the sk_lock with the lock validator.)
1251 */
1252 static inline void sock_lock_init(struct sock *sk)
1253 {
1254 sock_lock_init_class_and_name(sk,
1255 af_family_slock_key_strings[sk->sk_family],
1256 af_family_slock_keys + sk->sk_family,
1257 af_family_key_strings[sk->sk_family],
1258 af_family_keys + sk->sk_family);
1259 }
1260
1261 /*
1262 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1263 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1264 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1265 */
1266 static void sock_copy(struct sock *nsk, const struct sock *osk)
1267 {
1268 #ifdef CONFIG_SECURITY_NETWORK
1269 void *sptr = nsk->sk_security;
1270 #endif
1271 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1272
1273 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1274 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1275
1276 #ifdef CONFIG_SECURITY_NETWORK
1277 nsk->sk_security = sptr;
1278 security_sk_clone(osk, nsk);
1279 #endif
1280 }
1281
1282 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1283 {
1284 unsigned long nulls1, nulls2;
1285
1286 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1287 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1288 if (nulls1 > nulls2)
1289 swap(nulls1, nulls2);
1290
1291 if (nulls1 != 0)
1292 memset((char *)sk, 0, nulls1);
1293 memset((char *)sk + nulls1 + sizeof(void *), 0,
1294 nulls2 - nulls1 - sizeof(void *));
1295 memset((char *)sk + nulls2 + sizeof(void *), 0,
1296 size - nulls2 - sizeof(void *));
1297 }
1298 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1299
1300 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1301 int family)
1302 {
1303 struct sock *sk;
1304 struct kmem_cache *slab;
1305
1306 slab = prot->slab;
1307 if (slab != NULL) {
1308 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1309 if (!sk)
1310 return sk;
1311 if (priority & __GFP_ZERO) {
1312 if (prot->clear_sk)
1313 prot->clear_sk(sk, prot->obj_size);
1314 else
1315 sk_prot_clear_nulls(sk, prot->obj_size);
1316 }
1317 } else
1318 sk = kmalloc(prot->obj_size, priority);
1319
1320 if (sk != NULL) {
1321 kmemcheck_annotate_bitfield(sk, flags);
1322
1323 if (security_sk_alloc(sk, family, priority))
1324 goto out_free;
1325
1326 if (!try_module_get(prot->owner))
1327 goto out_free_sec;
1328 sk_tx_queue_clear(sk);
1329 }
1330
1331 return sk;
1332
1333 out_free_sec:
1334 security_sk_free(sk);
1335 out_free:
1336 if (slab != NULL)
1337 kmem_cache_free(slab, sk);
1338 else
1339 kfree(sk);
1340 return NULL;
1341 }
1342
1343 static void sk_prot_free(struct proto *prot, struct sock *sk)
1344 {
1345 struct kmem_cache *slab;
1346 struct module *owner;
1347
1348 owner = prot->owner;
1349 slab = prot->slab;
1350
1351 security_sk_free(sk);
1352 if (slab != NULL)
1353 kmem_cache_free(slab, sk);
1354 else
1355 kfree(sk);
1356 module_put(owner);
1357 }
1358
1359 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1360 void sock_update_netprioidx(struct sock *sk)
1361 {
1362 if (in_interrupt())
1363 return;
1364
1365 sk->sk_cgrp_prioidx = task_netprioidx(current);
1366 }
1367 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1368 #endif
1369
1370 /**
1371 * sk_alloc - All socket objects are allocated here
1372 * @net: the applicable net namespace
1373 * @family: protocol family
1374 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1375 * @prot: struct proto associated with this new sock instance
1376 */
1377 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1378 struct proto *prot)
1379 {
1380 struct sock *sk;
1381
1382 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1383 if (sk) {
1384 sk->sk_family = family;
1385 /*
1386 * See comment in struct sock definition to understand
1387 * why we need sk_prot_creator -acme
1388 */
1389 sk->sk_prot = sk->sk_prot_creator = prot;
1390 sock_lock_init(sk);
1391 sock_net_set(sk, get_net(net));
1392 atomic_set(&sk->sk_wmem_alloc, 1);
1393
1394 sock_update_classid(sk);
1395 sock_update_netprioidx(sk);
1396 }
1397
1398 return sk;
1399 }
1400 EXPORT_SYMBOL(sk_alloc);
1401
1402 static void __sk_free(struct sock *sk)
1403 {
1404 struct sk_filter *filter;
1405
1406 if (sk->sk_destruct)
1407 sk->sk_destruct(sk);
1408
1409 filter = rcu_dereference_check(sk->sk_filter,
1410 atomic_read(&sk->sk_wmem_alloc) == 0);
1411 if (filter) {
1412 sk_filter_uncharge(sk, filter);
1413 RCU_INIT_POINTER(sk->sk_filter, NULL);
1414 }
1415
1416 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1417
1418 if (atomic_read(&sk->sk_omem_alloc))
1419 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1420 __func__, atomic_read(&sk->sk_omem_alloc));
1421
1422 if (sk->sk_peer_cred)
1423 put_cred(sk->sk_peer_cred);
1424 put_pid(sk->sk_peer_pid);
1425 put_net(sock_net(sk));
1426 sk_prot_free(sk->sk_prot_creator, sk);
1427 }
1428
1429 void sk_free(struct sock *sk)
1430 {
1431 /*
1432 * We subtract one from sk_wmem_alloc and can know if
1433 * some packets are still in some tx queue.
1434 * If not null, sock_wfree() will call __sk_free(sk) later
1435 */
1436 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1437 __sk_free(sk);
1438 }
1439 EXPORT_SYMBOL(sk_free);
1440
1441 /*
1442 * Last sock_put should drop reference to sk->sk_net. It has already
1443 * been dropped in sk_change_net. Taking reference to stopping namespace
1444 * is not an option.
1445 * Take reference to a socket to remove it from hash _alive_ and after that
1446 * destroy it in the context of init_net.
1447 */
1448 void sk_release_kernel(struct sock *sk)
1449 {
1450 if (sk == NULL || sk->sk_socket == NULL)
1451 return;
1452
1453 sock_hold(sk);
1454 sock_release(sk->sk_socket);
1455 release_net(sock_net(sk));
1456 sock_net_set(sk, get_net(&init_net));
1457 sock_put(sk);
1458 }
1459 EXPORT_SYMBOL(sk_release_kernel);
1460
1461 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1462 {
1463 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1464 sock_update_memcg(newsk);
1465 }
1466
1467 /**
1468 * sk_clone_lock - clone a socket, and lock its clone
1469 * @sk: the socket to clone
1470 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1471 *
1472 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1473 */
1474 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1475 {
1476 struct sock *newsk;
1477 bool is_charged = true;
1478
1479 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1480 if (newsk != NULL) {
1481 struct sk_filter *filter;
1482
1483 sock_copy(newsk, sk);
1484
1485 /* SANITY */
1486 get_net(sock_net(newsk));
1487 sk_node_init(&newsk->sk_node);
1488 sock_lock_init(newsk);
1489 bh_lock_sock(newsk);
1490 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1491 newsk->sk_backlog.len = 0;
1492
1493 atomic_set(&newsk->sk_rmem_alloc, 0);
1494 /*
1495 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1496 */
1497 atomic_set(&newsk->sk_wmem_alloc, 1);
1498 atomic_set(&newsk->sk_omem_alloc, 0);
1499 skb_queue_head_init(&newsk->sk_receive_queue);
1500 skb_queue_head_init(&newsk->sk_write_queue);
1501
1502 spin_lock_init(&newsk->sk_dst_lock);
1503 rwlock_init(&newsk->sk_callback_lock);
1504 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1505 af_callback_keys + newsk->sk_family,
1506 af_family_clock_key_strings[newsk->sk_family]);
1507
1508 newsk->sk_dst_cache = NULL;
1509 newsk->sk_wmem_queued = 0;
1510 newsk->sk_forward_alloc = 0;
1511 newsk->sk_send_head = NULL;
1512 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1513
1514 sock_reset_flag(newsk, SOCK_DONE);
1515 skb_queue_head_init(&newsk->sk_error_queue);
1516
1517 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1518 if (filter != NULL)
1519 /* though it's an empty new sock, the charging may fail
1520 * if sysctl_optmem_max was changed between creation of
1521 * original socket and cloning
1522 */
1523 is_charged = sk_filter_charge(newsk, filter);
1524
1525 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1526 /* It is still raw copy of parent, so invalidate
1527 * destructor and make plain sk_free() */
1528 newsk->sk_destruct = NULL;
1529 bh_unlock_sock(newsk);
1530 sk_free(newsk);
1531 newsk = NULL;
1532 goto out;
1533 }
1534
1535 newsk->sk_err = 0;
1536 newsk->sk_priority = 0;
1537 newsk->sk_incoming_cpu = raw_smp_processor_id();
1538 /*
1539 * Before updating sk_refcnt, we must commit prior changes to memory
1540 * (Documentation/RCU/rculist_nulls.txt for details)
1541 */
1542 smp_wmb();
1543 atomic_set(&newsk->sk_refcnt, 2);
1544
1545 /*
1546 * Increment the counter in the same struct proto as the master
1547 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1548 * is the same as sk->sk_prot->socks, as this field was copied
1549 * with memcpy).
1550 *
1551 * This _changes_ the previous behaviour, where
1552 * tcp_create_openreq_child always was incrementing the
1553 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1554 * to be taken into account in all callers. -acme
1555 */
1556 sk_refcnt_debug_inc(newsk);
1557 sk_set_socket(newsk, NULL);
1558 newsk->sk_wq = NULL;
1559
1560 sk_update_clone(sk, newsk);
1561
1562 if (newsk->sk_prot->sockets_allocated)
1563 sk_sockets_allocated_inc(newsk);
1564
1565 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1566 net_enable_timestamp();
1567 }
1568 out:
1569 return newsk;
1570 }
1571 EXPORT_SYMBOL_GPL(sk_clone_lock);
1572
1573 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1574 {
1575 __sk_dst_set(sk, dst);
1576 sk->sk_route_caps = dst->dev->features;
1577 if (sk->sk_route_caps & NETIF_F_GSO)
1578 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1579 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1580 if (sk_can_gso(sk)) {
1581 if (dst->header_len) {
1582 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1583 } else {
1584 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1585 sk->sk_gso_max_size = dst->dev->gso_max_size;
1586 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1587 }
1588 }
1589 }
1590 EXPORT_SYMBOL_GPL(sk_setup_caps);
1591
1592 /*
1593 * Simple resource managers for sockets.
1594 */
1595
1596
1597 /*
1598 * Write buffer destructor automatically called from kfree_skb.
1599 */
1600 void sock_wfree(struct sk_buff *skb)
1601 {
1602 struct sock *sk = skb->sk;
1603 unsigned int len = skb->truesize;
1604
1605 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1606 /*
1607 * Keep a reference on sk_wmem_alloc, this will be released
1608 * after sk_write_space() call
1609 */
1610 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1611 sk->sk_write_space(sk);
1612 len = 1;
1613 }
1614 /*
1615 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1616 * could not do because of in-flight packets
1617 */
1618 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1619 __sk_free(sk);
1620 }
1621 EXPORT_SYMBOL(sock_wfree);
1622
1623 void skb_orphan_partial(struct sk_buff *skb)
1624 {
1625 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1626 * so we do not completely orphan skb, but transfert all
1627 * accounted bytes but one, to avoid unexpected reorders.
1628 */
1629 if (skb->destructor == sock_wfree
1630 #ifdef CONFIG_INET
1631 || skb->destructor == tcp_wfree
1632 #endif
1633 ) {
1634 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1635 skb->truesize = 1;
1636 } else {
1637 skb_orphan(skb);
1638 }
1639 }
1640 EXPORT_SYMBOL(skb_orphan_partial);
1641
1642 /*
1643 * Read buffer destructor automatically called from kfree_skb.
1644 */
1645 void sock_rfree(struct sk_buff *skb)
1646 {
1647 struct sock *sk = skb->sk;
1648 unsigned int len = skb->truesize;
1649
1650 atomic_sub(len, &sk->sk_rmem_alloc);
1651 sk_mem_uncharge(sk, len);
1652 }
1653 EXPORT_SYMBOL(sock_rfree);
1654
1655 void sock_efree(struct sk_buff *skb)
1656 {
1657 sock_put(skb->sk);
1658 }
1659 EXPORT_SYMBOL(sock_efree);
1660
1661 #ifdef CONFIG_INET
1662 void sock_edemux(struct sk_buff *skb)
1663 {
1664 struct sock *sk = skb->sk;
1665
1666 if (sk->sk_state == TCP_TIME_WAIT)
1667 inet_twsk_put(inet_twsk(sk));
1668 else
1669 sock_put(sk);
1670 }
1671 EXPORT_SYMBOL(sock_edemux);
1672 #endif
1673
1674 kuid_t sock_i_uid(struct sock *sk)
1675 {
1676 kuid_t uid;
1677
1678 read_lock_bh(&sk->sk_callback_lock);
1679 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1680 read_unlock_bh(&sk->sk_callback_lock);
1681 return uid;
1682 }
1683 EXPORT_SYMBOL(sock_i_uid);
1684
1685 unsigned long sock_i_ino(struct sock *sk)
1686 {
1687 unsigned long ino;
1688
1689 read_lock_bh(&sk->sk_callback_lock);
1690 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1691 read_unlock_bh(&sk->sk_callback_lock);
1692 return ino;
1693 }
1694 EXPORT_SYMBOL(sock_i_ino);
1695
1696 /*
1697 * Allocate a skb from the socket's send buffer.
1698 */
1699 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1700 gfp_t priority)
1701 {
1702 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1703 struct sk_buff *skb = alloc_skb(size, priority);
1704 if (skb) {
1705 skb_set_owner_w(skb, sk);
1706 return skb;
1707 }
1708 }
1709 return NULL;
1710 }
1711 EXPORT_SYMBOL(sock_wmalloc);
1712
1713 /*
1714 * Allocate a memory block from the socket's option memory buffer.
1715 */
1716 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1717 {
1718 if ((unsigned int)size <= sysctl_optmem_max &&
1719 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1720 void *mem;
1721 /* First do the add, to avoid the race if kmalloc
1722 * might sleep.
1723 */
1724 atomic_add(size, &sk->sk_omem_alloc);
1725 mem = kmalloc(size, priority);
1726 if (mem)
1727 return mem;
1728 atomic_sub(size, &sk->sk_omem_alloc);
1729 }
1730 return NULL;
1731 }
1732 EXPORT_SYMBOL(sock_kmalloc);
1733
1734 /*
1735 * Free an option memory block.
1736 */
1737 void sock_kfree_s(struct sock *sk, void *mem, int size)
1738 {
1739 if (WARN_ON_ONCE(!mem))
1740 return;
1741 kfree(mem);
1742 atomic_sub(size, &sk->sk_omem_alloc);
1743 }
1744 EXPORT_SYMBOL(sock_kfree_s);
1745
1746 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1747 I think, these locks should be removed for datagram sockets.
1748 */
1749 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1750 {
1751 DEFINE_WAIT(wait);
1752
1753 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1754 for (;;) {
1755 if (!timeo)
1756 break;
1757 if (signal_pending(current))
1758 break;
1759 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1760 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1761 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1762 break;
1763 if (sk->sk_shutdown & SEND_SHUTDOWN)
1764 break;
1765 if (sk->sk_err)
1766 break;
1767 timeo = schedule_timeout(timeo);
1768 }
1769 finish_wait(sk_sleep(sk), &wait);
1770 return timeo;
1771 }
1772
1773
1774 /*
1775 * Generic send/receive buffer handlers
1776 */
1777
1778 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1779 unsigned long data_len, int noblock,
1780 int *errcode, int max_page_order)
1781 {
1782 struct sk_buff *skb;
1783 long timeo;
1784 int err;
1785
1786 timeo = sock_sndtimeo(sk, noblock);
1787 for (;;) {
1788 err = sock_error(sk);
1789 if (err != 0)
1790 goto failure;
1791
1792 err = -EPIPE;
1793 if (sk->sk_shutdown & SEND_SHUTDOWN)
1794 goto failure;
1795
1796 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1797 break;
1798
1799 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1800 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1801 err = -EAGAIN;
1802 if (!timeo)
1803 goto failure;
1804 if (signal_pending(current))
1805 goto interrupted;
1806 timeo = sock_wait_for_wmem(sk, timeo);
1807 }
1808 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1809 errcode, sk->sk_allocation);
1810 if (skb)
1811 skb_set_owner_w(skb, sk);
1812 return skb;
1813
1814 interrupted:
1815 err = sock_intr_errno(timeo);
1816 failure:
1817 *errcode = err;
1818 return NULL;
1819 }
1820 EXPORT_SYMBOL(sock_alloc_send_pskb);
1821
1822 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1823 int noblock, int *errcode)
1824 {
1825 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1826 }
1827 EXPORT_SYMBOL(sock_alloc_send_skb);
1828
1829 /* On 32bit arches, an skb frag is limited to 2^15 */
1830 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1831
1832 /**
1833 * skb_page_frag_refill - check that a page_frag contains enough room
1834 * @sz: minimum size of the fragment we want to get
1835 * @pfrag: pointer to page_frag
1836 * @gfp: priority for memory allocation
1837 *
1838 * Note: While this allocator tries to use high order pages, there is
1839 * no guarantee that allocations succeed. Therefore, @sz MUST be
1840 * less or equal than PAGE_SIZE.
1841 */
1842 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1843 {
1844 if (pfrag->page) {
1845 if (atomic_read(&pfrag->page->_count) == 1) {
1846 pfrag->offset = 0;
1847 return true;
1848 }
1849 if (pfrag->offset + sz <= pfrag->size)
1850 return true;
1851 put_page(pfrag->page);
1852 }
1853
1854 pfrag->offset = 0;
1855 if (SKB_FRAG_PAGE_ORDER) {
1856 pfrag->page = alloc_pages(gfp | __GFP_COMP |
1857 __GFP_NOWARN | __GFP_NORETRY,
1858 SKB_FRAG_PAGE_ORDER);
1859 if (likely(pfrag->page)) {
1860 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1861 return true;
1862 }
1863 }
1864 pfrag->page = alloc_page(gfp);
1865 if (likely(pfrag->page)) {
1866 pfrag->size = PAGE_SIZE;
1867 return true;
1868 }
1869 return false;
1870 }
1871 EXPORT_SYMBOL(skb_page_frag_refill);
1872
1873 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1874 {
1875 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1876 return true;
1877
1878 sk_enter_memory_pressure(sk);
1879 sk_stream_moderate_sndbuf(sk);
1880 return false;
1881 }
1882 EXPORT_SYMBOL(sk_page_frag_refill);
1883
1884 static void __lock_sock(struct sock *sk)
1885 __releases(&sk->sk_lock.slock)
1886 __acquires(&sk->sk_lock.slock)
1887 {
1888 DEFINE_WAIT(wait);
1889
1890 for (;;) {
1891 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1892 TASK_UNINTERRUPTIBLE);
1893 spin_unlock_bh(&sk->sk_lock.slock);
1894 schedule();
1895 spin_lock_bh(&sk->sk_lock.slock);
1896 if (!sock_owned_by_user(sk))
1897 break;
1898 }
1899 finish_wait(&sk->sk_lock.wq, &wait);
1900 }
1901
1902 static void __release_sock(struct sock *sk)
1903 __releases(&sk->sk_lock.slock)
1904 __acquires(&sk->sk_lock.slock)
1905 {
1906 struct sk_buff *skb = sk->sk_backlog.head;
1907
1908 do {
1909 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1910 bh_unlock_sock(sk);
1911
1912 do {
1913 struct sk_buff *next = skb->next;
1914
1915 prefetch(next);
1916 WARN_ON_ONCE(skb_dst_is_noref(skb));
1917 skb->next = NULL;
1918 sk_backlog_rcv(sk, skb);
1919
1920 /*
1921 * We are in process context here with softirqs
1922 * disabled, use cond_resched_softirq() to preempt.
1923 * This is safe to do because we've taken the backlog
1924 * queue private:
1925 */
1926 cond_resched_softirq();
1927
1928 skb = next;
1929 } while (skb != NULL);
1930
1931 bh_lock_sock(sk);
1932 } while ((skb = sk->sk_backlog.head) != NULL);
1933
1934 /*
1935 * Doing the zeroing here guarantee we can not loop forever
1936 * while a wild producer attempts to flood us.
1937 */
1938 sk->sk_backlog.len = 0;
1939 }
1940
1941 /**
1942 * sk_wait_data - wait for data to arrive at sk_receive_queue
1943 * @sk: sock to wait on
1944 * @timeo: for how long
1945 *
1946 * Now socket state including sk->sk_err is changed only under lock,
1947 * hence we may omit checks after joining wait queue.
1948 * We check receive queue before schedule() only as optimization;
1949 * it is very likely that release_sock() added new data.
1950 */
1951 int sk_wait_data(struct sock *sk, long *timeo)
1952 {
1953 int rc;
1954 DEFINE_WAIT(wait);
1955
1956 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1957 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1958 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1959 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1960 finish_wait(sk_sleep(sk), &wait);
1961 return rc;
1962 }
1963 EXPORT_SYMBOL(sk_wait_data);
1964
1965 /**
1966 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1967 * @sk: socket
1968 * @size: memory size to allocate
1969 * @kind: allocation type
1970 *
1971 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1972 * rmem allocation. This function assumes that protocols which have
1973 * memory_pressure use sk_wmem_queued as write buffer accounting.
1974 */
1975 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1976 {
1977 struct proto *prot = sk->sk_prot;
1978 int amt = sk_mem_pages(size);
1979 long allocated;
1980 int parent_status = UNDER_LIMIT;
1981
1982 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1983
1984 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1985
1986 /* Under limit. */
1987 if (parent_status == UNDER_LIMIT &&
1988 allocated <= sk_prot_mem_limits(sk, 0)) {
1989 sk_leave_memory_pressure(sk);
1990 return 1;
1991 }
1992
1993 /* Under pressure. (we or our parents) */
1994 if ((parent_status > SOFT_LIMIT) ||
1995 allocated > sk_prot_mem_limits(sk, 1))
1996 sk_enter_memory_pressure(sk);
1997
1998 /* Over hard limit (we or our parents) */
1999 if ((parent_status == OVER_LIMIT) ||
2000 (allocated > sk_prot_mem_limits(sk, 2)))
2001 goto suppress_allocation;
2002
2003 /* guarantee minimum buffer size under pressure */
2004 if (kind == SK_MEM_RECV) {
2005 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2006 return 1;
2007
2008 } else { /* SK_MEM_SEND */
2009 if (sk->sk_type == SOCK_STREAM) {
2010 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2011 return 1;
2012 } else if (atomic_read(&sk->sk_wmem_alloc) <
2013 prot->sysctl_wmem[0])
2014 return 1;
2015 }
2016
2017 if (sk_has_memory_pressure(sk)) {
2018 int alloc;
2019
2020 if (!sk_under_memory_pressure(sk))
2021 return 1;
2022 alloc = sk_sockets_allocated_read_positive(sk);
2023 if (sk_prot_mem_limits(sk, 2) > alloc *
2024 sk_mem_pages(sk->sk_wmem_queued +
2025 atomic_read(&sk->sk_rmem_alloc) +
2026 sk->sk_forward_alloc))
2027 return 1;
2028 }
2029
2030 suppress_allocation:
2031
2032 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2033 sk_stream_moderate_sndbuf(sk);
2034
2035 /* Fail only if socket is _under_ its sndbuf.
2036 * In this case we cannot block, so that we have to fail.
2037 */
2038 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2039 return 1;
2040 }
2041
2042 trace_sock_exceed_buf_limit(sk, prot, allocated);
2043
2044 /* Alas. Undo changes. */
2045 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2046
2047 sk_memory_allocated_sub(sk, amt);
2048
2049 return 0;
2050 }
2051 EXPORT_SYMBOL(__sk_mem_schedule);
2052
2053 /**
2054 * __sk_reclaim - reclaim memory_allocated
2055 * @sk: socket
2056 */
2057 void __sk_mem_reclaim(struct sock *sk)
2058 {
2059 sk_memory_allocated_sub(sk,
2060 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2061 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2062
2063 if (sk_under_memory_pressure(sk) &&
2064 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2065 sk_leave_memory_pressure(sk);
2066 }
2067 EXPORT_SYMBOL(__sk_mem_reclaim);
2068
2069
2070 /*
2071 * Set of default routines for initialising struct proto_ops when
2072 * the protocol does not support a particular function. In certain
2073 * cases where it makes no sense for a protocol to have a "do nothing"
2074 * function, some default processing is provided.
2075 */
2076
2077 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2078 {
2079 return -EOPNOTSUPP;
2080 }
2081 EXPORT_SYMBOL(sock_no_bind);
2082
2083 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2084 int len, int flags)
2085 {
2086 return -EOPNOTSUPP;
2087 }
2088 EXPORT_SYMBOL(sock_no_connect);
2089
2090 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2091 {
2092 return -EOPNOTSUPP;
2093 }
2094 EXPORT_SYMBOL(sock_no_socketpair);
2095
2096 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2097 {
2098 return -EOPNOTSUPP;
2099 }
2100 EXPORT_SYMBOL(sock_no_accept);
2101
2102 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2103 int *len, int peer)
2104 {
2105 return -EOPNOTSUPP;
2106 }
2107 EXPORT_SYMBOL(sock_no_getname);
2108
2109 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2110 {
2111 return 0;
2112 }
2113 EXPORT_SYMBOL(sock_no_poll);
2114
2115 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2116 {
2117 return -EOPNOTSUPP;
2118 }
2119 EXPORT_SYMBOL(sock_no_ioctl);
2120
2121 int sock_no_listen(struct socket *sock, int backlog)
2122 {
2123 return -EOPNOTSUPP;
2124 }
2125 EXPORT_SYMBOL(sock_no_listen);
2126
2127 int sock_no_shutdown(struct socket *sock, int how)
2128 {
2129 return -EOPNOTSUPP;
2130 }
2131 EXPORT_SYMBOL(sock_no_shutdown);
2132
2133 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2134 char __user *optval, unsigned int optlen)
2135 {
2136 return -EOPNOTSUPP;
2137 }
2138 EXPORT_SYMBOL(sock_no_setsockopt);
2139
2140 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2141 char __user *optval, int __user *optlen)
2142 {
2143 return -EOPNOTSUPP;
2144 }
2145 EXPORT_SYMBOL(sock_no_getsockopt);
2146
2147 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2148 size_t len)
2149 {
2150 return -EOPNOTSUPP;
2151 }
2152 EXPORT_SYMBOL(sock_no_sendmsg);
2153
2154 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2155 size_t len, int flags)
2156 {
2157 return -EOPNOTSUPP;
2158 }
2159 EXPORT_SYMBOL(sock_no_recvmsg);
2160
2161 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2162 {
2163 /* Mirror missing mmap method error code */
2164 return -ENODEV;
2165 }
2166 EXPORT_SYMBOL(sock_no_mmap);
2167
2168 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2169 {
2170 ssize_t res;
2171 struct msghdr msg = {.msg_flags = flags};
2172 struct kvec iov;
2173 char *kaddr = kmap(page);
2174 iov.iov_base = kaddr + offset;
2175 iov.iov_len = size;
2176 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2177 kunmap(page);
2178 return res;
2179 }
2180 EXPORT_SYMBOL(sock_no_sendpage);
2181
2182 /*
2183 * Default Socket Callbacks
2184 */
2185
2186 static void sock_def_wakeup(struct sock *sk)
2187 {
2188 struct socket_wq *wq;
2189
2190 rcu_read_lock();
2191 wq = rcu_dereference(sk->sk_wq);
2192 if (wq_has_sleeper(wq))
2193 wake_up_interruptible_all(&wq->wait);
2194 rcu_read_unlock();
2195 }
2196
2197 static void sock_def_error_report(struct sock *sk)
2198 {
2199 struct socket_wq *wq;
2200
2201 rcu_read_lock();
2202 wq = rcu_dereference(sk->sk_wq);
2203 if (wq_has_sleeper(wq))
2204 wake_up_interruptible_poll(&wq->wait, POLLERR);
2205 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2206 rcu_read_unlock();
2207 }
2208
2209 static void sock_def_readable(struct sock *sk)
2210 {
2211 struct socket_wq *wq;
2212
2213 rcu_read_lock();
2214 wq = rcu_dereference(sk->sk_wq);
2215 if (wq_has_sleeper(wq))
2216 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2217 POLLRDNORM | POLLRDBAND);
2218 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2219 rcu_read_unlock();
2220 }
2221
2222 static void sock_def_write_space(struct sock *sk)
2223 {
2224 struct socket_wq *wq;
2225
2226 rcu_read_lock();
2227
2228 /* Do not wake up a writer until he can make "significant"
2229 * progress. --DaveM
2230 */
2231 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2232 wq = rcu_dereference(sk->sk_wq);
2233 if (wq_has_sleeper(wq))
2234 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2235 POLLWRNORM | POLLWRBAND);
2236
2237 /* Should agree with poll, otherwise some programs break */
2238 if (sock_writeable(sk))
2239 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2240 }
2241
2242 rcu_read_unlock();
2243 }
2244
2245 static void sock_def_destruct(struct sock *sk)
2246 {
2247 kfree(sk->sk_protinfo);
2248 }
2249
2250 void sk_send_sigurg(struct sock *sk)
2251 {
2252 if (sk->sk_socket && sk->sk_socket->file)
2253 if (send_sigurg(&sk->sk_socket->file->f_owner))
2254 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2255 }
2256 EXPORT_SYMBOL(sk_send_sigurg);
2257
2258 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2259 unsigned long expires)
2260 {
2261 if (!mod_timer(timer, expires))
2262 sock_hold(sk);
2263 }
2264 EXPORT_SYMBOL(sk_reset_timer);
2265
2266 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2267 {
2268 if (del_timer(timer))
2269 __sock_put(sk);
2270 }
2271 EXPORT_SYMBOL(sk_stop_timer);
2272
2273 void sock_init_data(struct socket *sock, struct sock *sk)
2274 {
2275 skb_queue_head_init(&sk->sk_receive_queue);
2276 skb_queue_head_init(&sk->sk_write_queue);
2277 skb_queue_head_init(&sk->sk_error_queue);
2278
2279 sk->sk_send_head = NULL;
2280
2281 init_timer(&sk->sk_timer);
2282
2283 sk->sk_allocation = GFP_KERNEL;
2284 sk->sk_rcvbuf = sysctl_rmem_default;
2285 sk->sk_sndbuf = sysctl_wmem_default;
2286 sk->sk_state = TCP_CLOSE;
2287 sk_set_socket(sk, sock);
2288
2289 sock_set_flag(sk, SOCK_ZAPPED);
2290
2291 if (sock) {
2292 sk->sk_type = sock->type;
2293 sk->sk_wq = sock->wq;
2294 sock->sk = sk;
2295 } else
2296 sk->sk_wq = NULL;
2297
2298 spin_lock_init(&sk->sk_dst_lock);
2299 rwlock_init(&sk->sk_callback_lock);
2300 lockdep_set_class_and_name(&sk->sk_callback_lock,
2301 af_callback_keys + sk->sk_family,
2302 af_family_clock_key_strings[sk->sk_family]);
2303
2304 sk->sk_state_change = sock_def_wakeup;
2305 sk->sk_data_ready = sock_def_readable;
2306 sk->sk_write_space = sock_def_write_space;
2307 sk->sk_error_report = sock_def_error_report;
2308 sk->sk_destruct = sock_def_destruct;
2309
2310 sk->sk_frag.page = NULL;
2311 sk->sk_frag.offset = 0;
2312 sk->sk_peek_off = -1;
2313
2314 sk->sk_peer_pid = NULL;
2315 sk->sk_peer_cred = NULL;
2316 sk->sk_write_pending = 0;
2317 sk->sk_rcvlowat = 1;
2318 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2319 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2320
2321 sk->sk_stamp = ktime_set(-1L, 0);
2322
2323 #ifdef CONFIG_NET_RX_BUSY_POLL
2324 sk->sk_napi_id = 0;
2325 sk->sk_ll_usec = sysctl_net_busy_read;
2326 #endif
2327
2328 sk->sk_max_pacing_rate = ~0U;
2329 sk->sk_pacing_rate = ~0U;
2330 /*
2331 * Before updating sk_refcnt, we must commit prior changes to memory
2332 * (Documentation/RCU/rculist_nulls.txt for details)
2333 */
2334 smp_wmb();
2335 atomic_set(&sk->sk_refcnt, 1);
2336 atomic_set(&sk->sk_drops, 0);
2337 }
2338 EXPORT_SYMBOL(sock_init_data);
2339
2340 void lock_sock_nested(struct sock *sk, int subclass)
2341 {
2342 might_sleep();
2343 spin_lock_bh(&sk->sk_lock.slock);
2344 if (sk->sk_lock.owned)
2345 __lock_sock(sk);
2346 sk->sk_lock.owned = 1;
2347 spin_unlock(&sk->sk_lock.slock);
2348 /*
2349 * The sk_lock has mutex_lock() semantics here:
2350 */
2351 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2352 local_bh_enable();
2353 }
2354 EXPORT_SYMBOL(lock_sock_nested);
2355
2356 void release_sock(struct sock *sk)
2357 {
2358 /*
2359 * The sk_lock has mutex_unlock() semantics:
2360 */
2361 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2362
2363 spin_lock_bh(&sk->sk_lock.slock);
2364 if (sk->sk_backlog.tail)
2365 __release_sock(sk);
2366
2367 /* Warning : release_cb() might need to release sk ownership,
2368 * ie call sock_release_ownership(sk) before us.
2369 */
2370 if (sk->sk_prot->release_cb)
2371 sk->sk_prot->release_cb(sk);
2372
2373 sock_release_ownership(sk);
2374 if (waitqueue_active(&sk->sk_lock.wq))
2375 wake_up(&sk->sk_lock.wq);
2376 spin_unlock_bh(&sk->sk_lock.slock);
2377 }
2378 EXPORT_SYMBOL(release_sock);
2379
2380 /**
2381 * lock_sock_fast - fast version of lock_sock
2382 * @sk: socket
2383 *
2384 * This version should be used for very small section, where process wont block
2385 * return false if fast path is taken
2386 * sk_lock.slock locked, owned = 0, BH disabled
2387 * return true if slow path is taken
2388 * sk_lock.slock unlocked, owned = 1, BH enabled
2389 */
2390 bool lock_sock_fast(struct sock *sk)
2391 {
2392 might_sleep();
2393 spin_lock_bh(&sk->sk_lock.slock);
2394
2395 if (!sk->sk_lock.owned)
2396 /*
2397 * Note : We must disable BH
2398 */
2399 return false;
2400
2401 __lock_sock(sk);
2402 sk->sk_lock.owned = 1;
2403 spin_unlock(&sk->sk_lock.slock);
2404 /*
2405 * The sk_lock has mutex_lock() semantics here:
2406 */
2407 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2408 local_bh_enable();
2409 return true;
2410 }
2411 EXPORT_SYMBOL(lock_sock_fast);
2412
2413 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2414 {
2415 struct timeval tv;
2416 if (!sock_flag(sk, SOCK_TIMESTAMP))
2417 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2418 tv = ktime_to_timeval(sk->sk_stamp);
2419 if (tv.tv_sec == -1)
2420 return -ENOENT;
2421 if (tv.tv_sec == 0) {
2422 sk->sk_stamp = ktime_get_real();
2423 tv = ktime_to_timeval(sk->sk_stamp);
2424 }
2425 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2426 }
2427 EXPORT_SYMBOL(sock_get_timestamp);
2428
2429 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2430 {
2431 struct timespec ts;
2432 if (!sock_flag(sk, SOCK_TIMESTAMP))
2433 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2434 ts = ktime_to_timespec(sk->sk_stamp);
2435 if (ts.tv_sec == -1)
2436 return -ENOENT;
2437 if (ts.tv_sec == 0) {
2438 sk->sk_stamp = ktime_get_real();
2439 ts = ktime_to_timespec(sk->sk_stamp);
2440 }
2441 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2442 }
2443 EXPORT_SYMBOL(sock_get_timestampns);
2444
2445 void sock_enable_timestamp(struct sock *sk, int flag)
2446 {
2447 if (!sock_flag(sk, flag)) {
2448 unsigned long previous_flags = sk->sk_flags;
2449
2450 sock_set_flag(sk, flag);
2451 /*
2452 * we just set one of the two flags which require net
2453 * time stamping, but time stamping might have been on
2454 * already because of the other one
2455 */
2456 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2457 net_enable_timestamp();
2458 }
2459 }
2460
2461 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2462 int level, int type)
2463 {
2464 struct sock_exterr_skb *serr;
2465 struct sk_buff *skb;
2466 int copied, err;
2467
2468 err = -EAGAIN;
2469 skb = sock_dequeue_err_skb(sk);
2470 if (skb == NULL)
2471 goto out;
2472
2473 copied = skb->len;
2474 if (copied > len) {
2475 msg->msg_flags |= MSG_TRUNC;
2476 copied = len;
2477 }
2478 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2479 if (err)
2480 goto out_free_skb;
2481
2482 sock_recv_timestamp(msg, sk, skb);
2483
2484 serr = SKB_EXT_ERR(skb);
2485 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2486
2487 msg->msg_flags |= MSG_ERRQUEUE;
2488 err = copied;
2489
2490 out_free_skb:
2491 kfree_skb(skb);
2492 out:
2493 return err;
2494 }
2495 EXPORT_SYMBOL(sock_recv_errqueue);
2496
2497 /*
2498 * Get a socket option on an socket.
2499 *
2500 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2501 * asynchronous errors should be reported by getsockopt. We assume
2502 * this means if you specify SO_ERROR (otherwise whats the point of it).
2503 */
2504 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2505 char __user *optval, int __user *optlen)
2506 {
2507 struct sock *sk = sock->sk;
2508
2509 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2510 }
2511 EXPORT_SYMBOL(sock_common_getsockopt);
2512
2513 #ifdef CONFIG_COMPAT
2514 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2515 char __user *optval, int __user *optlen)
2516 {
2517 struct sock *sk = sock->sk;
2518
2519 if (sk->sk_prot->compat_getsockopt != NULL)
2520 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2521 optval, optlen);
2522 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2523 }
2524 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2525 #endif
2526
2527 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2528 struct msghdr *msg, size_t size, int flags)
2529 {
2530 struct sock *sk = sock->sk;
2531 int addr_len = 0;
2532 int err;
2533
2534 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2535 flags & ~MSG_DONTWAIT, &addr_len);
2536 if (err >= 0)
2537 msg->msg_namelen = addr_len;
2538 return err;
2539 }
2540 EXPORT_SYMBOL(sock_common_recvmsg);
2541
2542 /*
2543 * Set socket options on an inet socket.
2544 */
2545 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2546 char __user *optval, unsigned int optlen)
2547 {
2548 struct sock *sk = sock->sk;
2549
2550 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2551 }
2552 EXPORT_SYMBOL(sock_common_setsockopt);
2553
2554 #ifdef CONFIG_COMPAT
2555 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2556 char __user *optval, unsigned int optlen)
2557 {
2558 struct sock *sk = sock->sk;
2559
2560 if (sk->sk_prot->compat_setsockopt != NULL)
2561 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2562 optval, optlen);
2563 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2564 }
2565 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2566 #endif
2567
2568 void sk_common_release(struct sock *sk)
2569 {
2570 if (sk->sk_prot->destroy)
2571 sk->sk_prot->destroy(sk);
2572
2573 /*
2574 * Observation: when sock_common_release is called, processes have
2575 * no access to socket. But net still has.
2576 * Step one, detach it from networking:
2577 *
2578 * A. Remove from hash tables.
2579 */
2580
2581 sk->sk_prot->unhash(sk);
2582
2583 /*
2584 * In this point socket cannot receive new packets, but it is possible
2585 * that some packets are in flight because some CPU runs receiver and
2586 * did hash table lookup before we unhashed socket. They will achieve
2587 * receive queue and will be purged by socket destructor.
2588 *
2589 * Also we still have packets pending on receive queue and probably,
2590 * our own packets waiting in device queues. sock_destroy will drain
2591 * receive queue, but transmitted packets will delay socket destruction
2592 * until the last reference will be released.
2593 */
2594
2595 sock_orphan(sk);
2596
2597 xfrm_sk_free_policy(sk);
2598
2599 sk_refcnt_debug_release(sk);
2600
2601 if (sk->sk_frag.page) {
2602 put_page(sk->sk_frag.page);
2603 sk->sk_frag.page = NULL;
2604 }
2605
2606 sock_put(sk);
2607 }
2608 EXPORT_SYMBOL(sk_common_release);
2609
2610 #ifdef CONFIG_PROC_FS
2611 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2612 struct prot_inuse {
2613 int val[PROTO_INUSE_NR];
2614 };
2615
2616 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2617
2618 #ifdef CONFIG_NET_NS
2619 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2620 {
2621 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2622 }
2623 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2624
2625 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2626 {
2627 int cpu, idx = prot->inuse_idx;
2628 int res = 0;
2629
2630 for_each_possible_cpu(cpu)
2631 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2632
2633 return res >= 0 ? res : 0;
2634 }
2635 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2636
2637 static int __net_init sock_inuse_init_net(struct net *net)
2638 {
2639 net->core.inuse = alloc_percpu(struct prot_inuse);
2640 return net->core.inuse ? 0 : -ENOMEM;
2641 }
2642
2643 static void __net_exit sock_inuse_exit_net(struct net *net)
2644 {
2645 free_percpu(net->core.inuse);
2646 }
2647
2648 static struct pernet_operations net_inuse_ops = {
2649 .init = sock_inuse_init_net,
2650 .exit = sock_inuse_exit_net,
2651 };
2652
2653 static __init int net_inuse_init(void)
2654 {
2655 if (register_pernet_subsys(&net_inuse_ops))
2656 panic("Cannot initialize net inuse counters");
2657
2658 return 0;
2659 }
2660
2661 core_initcall(net_inuse_init);
2662 #else
2663 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2664
2665 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2666 {
2667 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2668 }
2669 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2670
2671 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2672 {
2673 int cpu, idx = prot->inuse_idx;
2674 int res = 0;
2675
2676 for_each_possible_cpu(cpu)
2677 res += per_cpu(prot_inuse, cpu).val[idx];
2678
2679 return res >= 0 ? res : 0;
2680 }
2681 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2682 #endif
2683
2684 static void assign_proto_idx(struct proto *prot)
2685 {
2686 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2687
2688 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2689 pr_err("PROTO_INUSE_NR exhausted\n");
2690 return;
2691 }
2692
2693 set_bit(prot->inuse_idx, proto_inuse_idx);
2694 }
2695
2696 static void release_proto_idx(struct proto *prot)
2697 {
2698 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2699 clear_bit(prot->inuse_idx, proto_inuse_idx);
2700 }
2701 #else
2702 static inline void assign_proto_idx(struct proto *prot)
2703 {
2704 }
2705
2706 static inline void release_proto_idx(struct proto *prot)
2707 {
2708 }
2709 #endif
2710
2711 int proto_register(struct proto *prot, int alloc_slab)
2712 {
2713 if (alloc_slab) {
2714 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2715 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2716 NULL);
2717
2718 if (prot->slab == NULL) {
2719 pr_crit("%s: Can't create sock SLAB cache!\n",
2720 prot->name);
2721 goto out;
2722 }
2723
2724 if (prot->rsk_prot != NULL) {
2725 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2726 if (prot->rsk_prot->slab_name == NULL)
2727 goto out_free_sock_slab;
2728
2729 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2730 prot->rsk_prot->obj_size, 0,
2731 SLAB_HWCACHE_ALIGN, NULL);
2732
2733 if (prot->rsk_prot->slab == NULL) {
2734 pr_crit("%s: Can't create request sock SLAB cache!\n",
2735 prot->name);
2736 goto out_free_request_sock_slab_name;
2737 }
2738 }
2739
2740 if (prot->twsk_prot != NULL) {
2741 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2742
2743 if (prot->twsk_prot->twsk_slab_name == NULL)
2744 goto out_free_request_sock_slab;
2745
2746 prot->twsk_prot->twsk_slab =
2747 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2748 prot->twsk_prot->twsk_obj_size,
2749 0,
2750 SLAB_HWCACHE_ALIGN |
2751 prot->slab_flags,
2752 NULL);
2753 if (prot->twsk_prot->twsk_slab == NULL)
2754 goto out_free_timewait_sock_slab_name;
2755 }
2756 }
2757
2758 mutex_lock(&proto_list_mutex);
2759 list_add(&prot->node, &proto_list);
2760 assign_proto_idx(prot);
2761 mutex_unlock(&proto_list_mutex);
2762 return 0;
2763
2764 out_free_timewait_sock_slab_name:
2765 kfree(prot->twsk_prot->twsk_slab_name);
2766 out_free_request_sock_slab:
2767 if (prot->rsk_prot && prot->rsk_prot->slab) {
2768 kmem_cache_destroy(prot->rsk_prot->slab);
2769 prot->rsk_prot->slab = NULL;
2770 }
2771 out_free_request_sock_slab_name:
2772 if (prot->rsk_prot)
2773 kfree(prot->rsk_prot->slab_name);
2774 out_free_sock_slab:
2775 kmem_cache_destroy(prot->slab);
2776 prot->slab = NULL;
2777 out:
2778 return -ENOBUFS;
2779 }
2780 EXPORT_SYMBOL(proto_register);
2781
2782 void proto_unregister(struct proto *prot)
2783 {
2784 mutex_lock(&proto_list_mutex);
2785 release_proto_idx(prot);
2786 list_del(&prot->node);
2787 mutex_unlock(&proto_list_mutex);
2788
2789 if (prot->slab != NULL) {
2790 kmem_cache_destroy(prot->slab);
2791 prot->slab = NULL;
2792 }
2793
2794 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2795 kmem_cache_destroy(prot->rsk_prot->slab);
2796 kfree(prot->rsk_prot->slab_name);
2797 prot->rsk_prot->slab = NULL;
2798 }
2799
2800 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2801 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2802 kfree(prot->twsk_prot->twsk_slab_name);
2803 prot->twsk_prot->twsk_slab = NULL;
2804 }
2805 }
2806 EXPORT_SYMBOL(proto_unregister);
2807
2808 #ifdef CONFIG_PROC_FS
2809 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2810 __acquires(proto_list_mutex)
2811 {
2812 mutex_lock(&proto_list_mutex);
2813 return seq_list_start_head(&proto_list, *pos);
2814 }
2815
2816 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2817 {
2818 return seq_list_next(v, &proto_list, pos);
2819 }
2820
2821 static void proto_seq_stop(struct seq_file *seq, void *v)
2822 __releases(proto_list_mutex)
2823 {
2824 mutex_unlock(&proto_list_mutex);
2825 }
2826
2827 static char proto_method_implemented(const void *method)
2828 {
2829 return method == NULL ? 'n' : 'y';
2830 }
2831 static long sock_prot_memory_allocated(struct proto *proto)
2832 {
2833 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2834 }
2835
2836 static char *sock_prot_memory_pressure(struct proto *proto)
2837 {
2838 return proto->memory_pressure != NULL ?
2839 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2840 }
2841
2842 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2843 {
2844
2845 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2846 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2847 proto->name,
2848 proto->obj_size,
2849 sock_prot_inuse_get(seq_file_net(seq), proto),
2850 sock_prot_memory_allocated(proto),
2851 sock_prot_memory_pressure(proto),
2852 proto->max_header,
2853 proto->slab == NULL ? "no" : "yes",
2854 module_name(proto->owner),
2855 proto_method_implemented(proto->close),
2856 proto_method_implemented(proto->connect),
2857 proto_method_implemented(proto->disconnect),
2858 proto_method_implemented(proto->accept),
2859 proto_method_implemented(proto->ioctl),
2860 proto_method_implemented(proto->init),
2861 proto_method_implemented(proto->destroy),
2862 proto_method_implemented(proto->shutdown),
2863 proto_method_implemented(proto->setsockopt),
2864 proto_method_implemented(proto->getsockopt),
2865 proto_method_implemented(proto->sendmsg),
2866 proto_method_implemented(proto->recvmsg),
2867 proto_method_implemented(proto->sendpage),
2868 proto_method_implemented(proto->bind),
2869 proto_method_implemented(proto->backlog_rcv),
2870 proto_method_implemented(proto->hash),
2871 proto_method_implemented(proto->unhash),
2872 proto_method_implemented(proto->get_port),
2873 proto_method_implemented(proto->enter_memory_pressure));
2874 }
2875
2876 static int proto_seq_show(struct seq_file *seq, void *v)
2877 {
2878 if (v == &proto_list)
2879 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2880 "protocol",
2881 "size",
2882 "sockets",
2883 "memory",
2884 "press",
2885 "maxhdr",
2886 "slab",
2887 "module",
2888 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2889 else
2890 proto_seq_printf(seq, list_entry(v, struct proto, node));
2891 return 0;
2892 }
2893
2894 static const struct seq_operations proto_seq_ops = {
2895 .start = proto_seq_start,
2896 .next = proto_seq_next,
2897 .stop = proto_seq_stop,
2898 .show = proto_seq_show,
2899 };
2900
2901 static int proto_seq_open(struct inode *inode, struct file *file)
2902 {
2903 return seq_open_net(inode, file, &proto_seq_ops,
2904 sizeof(struct seq_net_private));
2905 }
2906
2907 static const struct file_operations proto_seq_fops = {
2908 .owner = THIS_MODULE,
2909 .open = proto_seq_open,
2910 .read = seq_read,
2911 .llseek = seq_lseek,
2912 .release = seq_release_net,
2913 };
2914
2915 static __net_init int proto_init_net(struct net *net)
2916 {
2917 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2918 return -ENOMEM;
2919
2920 return 0;
2921 }
2922
2923 static __net_exit void proto_exit_net(struct net *net)
2924 {
2925 remove_proc_entry("protocols", net->proc_net);
2926 }
2927
2928
2929 static __net_initdata struct pernet_operations proto_net_ops = {
2930 .init = proto_init_net,
2931 .exit = proto_exit_net,
2932 };
2933
2934 static int __init proto_init(void)
2935 {
2936 return register_pernet_subsys(&proto_net_ops);
2937 }
2938
2939 subsys_initcall(proto_init);
2940
2941 #endif /* PROC_FS */
Linux with added FPC-III support