projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
powerpc: Speed up clear_page by unrolling it
[deliverable/linux.git] / net / core / sock.c
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 *
85 *
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
90 */
91
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
99 #include <linux/in.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
120
121 #include <asm/uaccess.h>
122
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134
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 int skb_len;
441 unsigned long flags;
442 struct sk_buff_head *list = &sk->sk_receive_queue;
443
444 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
445 atomic_inc(&sk->sk_drops);
446 trace_sock_rcvqueue_full(sk, skb);
447 return -ENOMEM;
448 }
449
450 err = sk_filter(sk, skb);
451 if (err)
452 return err;
453
454 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
455 atomic_inc(&sk->sk_drops);
456 return -ENOBUFS;
457 }
458
459 skb->dev = NULL;
460 skb_set_owner_r(skb, sk);
461
462 /* Cache the SKB length before we tack it onto the receive
463 * queue. Once it is added it no longer belongs to us and
464 * may be freed by other threads of control pulling packets
465 * from the queue.
466 */
467 skb_len = skb->len;
468
469 /* we escape from rcu protected region, make sure we dont leak
470 * a norefcounted dst
471 */
472 skb_dst_force(skb);
473
474 spin_lock_irqsave(&list->lock, flags);
475 skb->dropcount = atomic_read(&sk->sk_drops);
476 __skb_queue_tail(list, skb);
477 spin_unlock_irqrestore(&list->lock, flags);
478
479 if (!sock_flag(sk, SOCK_DEAD))
480 sk->sk_data_ready(sk);
481 return 0;
482 }
483 EXPORT_SYMBOL(sock_queue_rcv_skb);
484
485 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
486 {
487 int rc = NET_RX_SUCCESS;
488
489 if (sk_filter(sk, skb))
490 goto discard_and_relse;
491
492 skb->dev = NULL;
493
494 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
495 atomic_inc(&sk->sk_drops);
496 goto discard_and_relse;
497 }
498 if (nested)
499 bh_lock_sock_nested(sk);
500 else
501 bh_lock_sock(sk);
502 if (!sock_owned_by_user(sk)) {
503 /*
504 * trylock + unlock semantics:
505 */
506 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
507
508 rc = sk_backlog_rcv(sk, skb);
509
510 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
511 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
512 bh_unlock_sock(sk);
513 atomic_inc(&sk->sk_drops);
514 goto discard_and_relse;
515 }
516
517 bh_unlock_sock(sk);
518 out:
519 sock_put(sk);
520 return rc;
521 discard_and_relse:
522 kfree_skb(skb);
523 goto out;
524 }
525 EXPORT_SYMBOL(sk_receive_skb);
526
527 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
528 {
529 struct dst_entry *dst = __sk_dst_get(sk);
530
531 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
532 sk_tx_queue_clear(sk);
533 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
534 dst_release(dst);
535 return NULL;
536 }
537
538 return dst;
539 }
540 EXPORT_SYMBOL(__sk_dst_check);
541
542 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
543 {
544 struct dst_entry *dst = sk_dst_get(sk);
545
546 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
547 sk_dst_reset(sk);
548 dst_release(dst);
549 return NULL;
550 }
551
552 return dst;
553 }
554 EXPORT_SYMBOL(sk_dst_check);
555
556 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
557 int optlen)
558 {
559 int ret = -ENOPROTOOPT;
560 #ifdef CONFIG_NETDEVICES
561 struct net *net = sock_net(sk);
562 char devname[IFNAMSIZ];
563 int index;
564
565 /* Sorry... */
566 ret = -EPERM;
567 if (!ns_capable(net->user_ns, CAP_NET_RAW))
568 goto out;
569
570 ret = -EINVAL;
571 if (optlen < 0)
572 goto out;
573
574 /* Bind this socket to a particular device like "eth0",
575 * as specified in the passed interface name. If the
576 * name is "" or the option length is zero the socket
577 * is not bound.
578 */
579 if (optlen > IFNAMSIZ - 1)
580 optlen = IFNAMSIZ - 1;
581 memset(devname, 0, sizeof(devname));
582
583 ret = -EFAULT;
584 if (copy_from_user(devname, optval, optlen))
585 goto out;
586
587 index = 0;
588 if (devname[0] != '\0') {
589 struct net_device *dev;
590
591 rcu_read_lock();
592 dev = dev_get_by_name_rcu(net, devname);
593 if (dev)
594 index = dev->ifindex;
595 rcu_read_unlock();
596 ret = -ENODEV;
597 if (!dev)
598 goto out;
599 }
600
601 lock_sock(sk);
602 sk->sk_bound_dev_if = index;
603 sk_dst_reset(sk);
604 release_sock(sk);
605
606 ret = 0;
607
608 out:
609 #endif
610
611 return ret;
612 }
613
614 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
615 int __user *optlen, int len)
616 {
617 int ret = -ENOPROTOOPT;
618 #ifdef CONFIG_NETDEVICES
619 struct net *net = sock_net(sk);
620 char devname[IFNAMSIZ];
621
622 if (sk->sk_bound_dev_if == 0) {
623 len = 0;
624 goto zero;
625 }
626
627 ret = -EINVAL;
628 if (len < IFNAMSIZ)
629 goto out;
630
631 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
632 if (ret)
633 goto out;
634
635 len = strlen(devname) + 1;
636
637 ret = -EFAULT;
638 if (copy_to_user(optval, devname, len))
639 goto out;
640
641 zero:
642 ret = -EFAULT;
643 if (put_user(len, optlen))
644 goto out;
645
646 ret = 0;
647
648 out:
649 #endif
650
651 return ret;
652 }
653
654 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
655 {
656 if (valbool)
657 sock_set_flag(sk, bit);
658 else
659 sock_reset_flag(sk, bit);
660 }
661
662 /*
663 * This is meant for all protocols to use and covers goings on
664 * at the socket level. Everything here is generic.
665 */
666
667 int sock_setsockopt(struct socket *sock, int level, int optname,
668 char __user *optval, unsigned int optlen)
669 {
670 struct sock *sk = sock->sk;
671 int val;
672 int valbool;
673 struct linger ling;
674 int ret = 0;
675
676 /*
677 * Options without arguments
678 */
679
680 if (optname == SO_BINDTODEVICE)
681 return sock_setbindtodevice(sk, optval, optlen);
682
683 if (optlen < sizeof(int))
684 return -EINVAL;
685
686 if (get_user(val, (int __user *)optval))
687 return -EFAULT;
688
689 valbool = val ? 1 : 0;
690
691 lock_sock(sk);
692
693 switch (optname) {
694 case SO_DEBUG:
695 if (val && !capable(CAP_NET_ADMIN))
696 ret = -EACCES;
697 else
698 sock_valbool_flag(sk, SOCK_DBG, valbool);
699 break;
700 case SO_REUSEADDR:
701 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
702 break;
703 case SO_REUSEPORT:
704 sk->sk_reuseport = valbool;
705 break;
706 case SO_TYPE:
707 case SO_PROTOCOL:
708 case SO_DOMAIN:
709 case SO_ERROR:
710 ret = -ENOPROTOOPT;
711 break;
712 case SO_DONTROUTE:
713 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
714 break;
715 case SO_BROADCAST:
716 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
717 break;
718 case SO_SNDBUF:
719 /* Don't error on this BSD doesn't and if you think
720 * about it this is right. Otherwise apps have to
721 * play 'guess the biggest size' games. RCVBUF/SNDBUF
722 * are treated in BSD as hints
723 */
724 val = min_t(u32, val, sysctl_wmem_max);
725 set_sndbuf:
726 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
727 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
728 /* Wake up sending tasks if we upped the value. */
729 sk->sk_write_space(sk);
730 break;
731
732 case SO_SNDBUFFORCE:
733 if (!capable(CAP_NET_ADMIN)) {
734 ret = -EPERM;
735 break;
736 }
737 goto set_sndbuf;
738
739 case SO_RCVBUF:
740 /* Don't error on this BSD doesn't and if you think
741 * about it this is right. Otherwise apps have to
742 * play 'guess the biggest size' games. RCVBUF/SNDBUF
743 * are treated in BSD as hints
744 */
745 val = min_t(u32, val, sysctl_rmem_max);
746 set_rcvbuf:
747 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
748 /*
749 * We double it on the way in to account for
750 * "struct sk_buff" etc. overhead. Applications
751 * assume that the SO_RCVBUF setting they make will
752 * allow that much actual data to be received on that
753 * socket.
754 *
755 * Applications are unaware that "struct sk_buff" and
756 * other overheads allocate from the receive buffer
757 * during socket buffer allocation.
758 *
759 * And after considering the possible alternatives,
760 * returning the value we actually used in getsockopt
761 * is the most desirable behavior.
762 */
763 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
764 break;
765
766 case SO_RCVBUFFORCE:
767 if (!capable(CAP_NET_ADMIN)) {
768 ret = -EPERM;
769 break;
770 }
771 goto set_rcvbuf;
772
773 case SO_KEEPALIVE:
774 #ifdef CONFIG_INET
775 if (sk->sk_protocol == IPPROTO_TCP &&
776 sk->sk_type == SOCK_STREAM)
777 tcp_set_keepalive(sk, valbool);
778 #endif
779 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
780 break;
781
782 case SO_OOBINLINE:
783 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
784 break;
785
786 case SO_NO_CHECK:
787 sk->sk_no_check_tx = valbool;
788 break;
789
790 case SO_PRIORITY:
791 if ((val >= 0 && val <= 6) ||
792 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
793 sk->sk_priority = val;
794 else
795 ret = -EPERM;
796 break;
797
798 case SO_LINGER:
799 if (optlen < sizeof(ling)) {
800 ret = -EINVAL; /* 1003.1g */
801 break;
802 }
803 if (copy_from_user(&ling, optval, sizeof(ling))) {
804 ret = -EFAULT;
805 break;
806 }
807 if (!ling.l_onoff)
808 sock_reset_flag(sk, SOCK_LINGER);
809 else {
810 #if (BITS_PER_LONG == 32)
811 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
812 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
813 else
814 #endif
815 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
816 sock_set_flag(sk, SOCK_LINGER);
817 }
818 break;
819
820 case SO_BSDCOMPAT:
821 sock_warn_obsolete_bsdism("setsockopt");
822 break;
823
824 case SO_PASSCRED:
825 if (valbool)
826 set_bit(SOCK_PASSCRED, &sock->flags);
827 else
828 clear_bit(SOCK_PASSCRED, &sock->flags);
829 break;
830
831 case SO_TIMESTAMP:
832 case SO_TIMESTAMPNS:
833 if (valbool) {
834 if (optname == SO_TIMESTAMP)
835 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
836 else
837 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
838 sock_set_flag(sk, SOCK_RCVTSTAMP);
839 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
840 } else {
841 sock_reset_flag(sk, SOCK_RCVTSTAMP);
842 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
843 }
844 break;
845
846 case SO_TIMESTAMPING:
847 if (val & ~SOF_TIMESTAMPING_MASK) {
848 ret = -EINVAL;
849 break;
850 }
851 if (val & SOF_TIMESTAMPING_OPT_ID &&
852 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
853 if (sk->sk_protocol == IPPROTO_TCP) {
854 if (sk->sk_state != TCP_ESTABLISHED) {
855 ret = -EINVAL;
856 break;
857 }
858 sk->sk_tskey = tcp_sk(sk)->snd_una;
859 } else {
860 sk->sk_tskey = 0;
861 }
862 }
863 sk->sk_tsflags = val;
864 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
865 sock_enable_timestamp(sk,
866 SOCK_TIMESTAMPING_RX_SOFTWARE);
867 else
868 sock_disable_timestamp(sk,
869 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
870 break;
871
872 case SO_RCVLOWAT:
873 if (val < 0)
874 val = INT_MAX;
875 sk->sk_rcvlowat = val ? : 1;
876 break;
877
878 case SO_RCVTIMEO:
879 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
880 break;
881
882 case SO_SNDTIMEO:
883 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
884 break;
885
886 case SO_ATTACH_FILTER:
887 ret = -EINVAL;
888 if (optlen == sizeof(struct sock_fprog)) {
889 struct sock_fprog fprog;
890
891 ret = -EFAULT;
892 if (copy_from_user(&fprog, optval, sizeof(fprog)))
893 break;
894
895 ret = sk_attach_filter(&fprog, sk);
896 }
897 break;
898
899 case SO_DETACH_FILTER:
900 ret = sk_detach_filter(sk);
901 break;
902
903 case SO_LOCK_FILTER:
904 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
905 ret = -EPERM;
906 else
907 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
908 break;
909
910 case SO_PASSSEC:
911 if (valbool)
912 set_bit(SOCK_PASSSEC, &sock->flags);
913 else
914 clear_bit(SOCK_PASSSEC, &sock->flags);
915 break;
916 case SO_MARK:
917 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
918 ret = -EPERM;
919 else
920 sk->sk_mark = val;
921 break;
922
923 /* We implement the SO_SNDLOWAT etc to
924 not be settable (1003.1g 5.3) */
925 case SO_RXQ_OVFL:
926 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
927 break;
928
929 case SO_WIFI_STATUS:
930 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
931 break;
932
933 case SO_PEEK_OFF:
934 if (sock->ops->set_peek_off)
935 ret = sock->ops->set_peek_off(sk, val);
936 else
937 ret = -EOPNOTSUPP;
938 break;
939
940 case SO_NOFCS:
941 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
942 break;
943
944 case SO_SELECT_ERR_QUEUE:
945 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
946 break;
947
948 #ifdef CONFIG_NET_RX_BUSY_POLL
949 case SO_BUSY_POLL:
950 /* allow unprivileged users to decrease the value */
951 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
952 ret = -EPERM;
953 else {
954 if (val < 0)
955 ret = -EINVAL;
956 else
957 sk->sk_ll_usec = val;
958 }
959 break;
960 #endif
961
962 case SO_MAX_PACING_RATE:
963 sk->sk_max_pacing_rate = val;
964 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
965 sk->sk_max_pacing_rate);
966 break;
967
968 default:
969 ret = -ENOPROTOOPT;
970 break;
971 }
972 release_sock(sk);
973 return ret;
974 }
975 EXPORT_SYMBOL(sock_setsockopt);
976
977
978 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
979 struct ucred *ucred)
980 {
981 ucred->pid = pid_vnr(pid);
982 ucred->uid = ucred->gid = -1;
983 if (cred) {
984 struct user_namespace *current_ns = current_user_ns();
985
986 ucred->uid = from_kuid_munged(current_ns, cred->euid);
987 ucred->gid = from_kgid_munged(current_ns, cred->egid);
988 }
989 }
990
991 int sock_getsockopt(struct socket *sock, int level, int optname,
992 char __user *optval, int __user *optlen)
993 {
994 struct sock *sk = sock->sk;
995
996 union {
997 int val;
998 struct linger ling;
999 struct timeval tm;
1000 } v;
1001
1002 int lv = sizeof(int);
1003 int len;
1004
1005 if (get_user(len, optlen))
1006 return -EFAULT;
1007 if (len < 0)
1008 return -EINVAL;
1009
1010 memset(&v, 0, sizeof(v));
1011
1012 switch (optname) {
1013 case SO_DEBUG:
1014 v.val = sock_flag(sk, SOCK_DBG);
1015 break;
1016
1017 case SO_DONTROUTE:
1018 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1019 break;
1020
1021 case SO_BROADCAST:
1022 v.val = sock_flag(sk, SOCK_BROADCAST);
1023 break;
1024
1025 case SO_SNDBUF:
1026 v.val = sk->sk_sndbuf;
1027 break;
1028
1029 case SO_RCVBUF:
1030 v.val = sk->sk_rcvbuf;
1031 break;
1032
1033 case SO_REUSEADDR:
1034 v.val = sk->sk_reuse;
1035 break;
1036
1037 case SO_REUSEPORT:
1038 v.val = sk->sk_reuseport;
1039 break;
1040
1041 case SO_KEEPALIVE:
1042 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1043 break;
1044
1045 case SO_TYPE:
1046 v.val = sk->sk_type;
1047 break;
1048
1049 case SO_PROTOCOL:
1050 v.val = sk->sk_protocol;
1051 break;
1052
1053 case SO_DOMAIN:
1054 v.val = sk->sk_family;
1055 break;
1056
1057 case SO_ERROR:
1058 v.val = -sock_error(sk);
1059 if (v.val == 0)
1060 v.val = xchg(&sk->sk_err_soft, 0);
1061 break;
1062
1063 case SO_OOBINLINE:
1064 v.val = sock_flag(sk, SOCK_URGINLINE);
1065 break;
1066
1067 case SO_NO_CHECK:
1068 v.val = sk->sk_no_check_tx;
1069 break;
1070
1071 case SO_PRIORITY:
1072 v.val = sk->sk_priority;
1073 break;
1074
1075 case SO_LINGER:
1076 lv = sizeof(v.ling);
1077 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1078 v.ling.l_linger = sk->sk_lingertime / HZ;
1079 break;
1080
1081 case SO_BSDCOMPAT:
1082 sock_warn_obsolete_bsdism("getsockopt");
1083 break;
1084
1085 case SO_TIMESTAMP:
1086 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1087 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1088 break;
1089
1090 case SO_TIMESTAMPNS:
1091 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1092 break;
1093
1094 case SO_TIMESTAMPING:
1095 v.val = sk->sk_tsflags;
1096 break;
1097
1098 case SO_RCVTIMEO:
1099 lv = sizeof(struct timeval);
1100 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1101 v.tm.tv_sec = 0;
1102 v.tm.tv_usec = 0;
1103 } else {
1104 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1105 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1106 }
1107 break;
1108
1109 case SO_SNDTIMEO:
1110 lv = sizeof(struct timeval);
1111 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1112 v.tm.tv_sec = 0;
1113 v.tm.tv_usec = 0;
1114 } else {
1115 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1116 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1117 }
1118 break;
1119
1120 case SO_RCVLOWAT:
1121 v.val = sk->sk_rcvlowat;
1122 break;
1123
1124 case SO_SNDLOWAT:
1125 v.val = 1;
1126 break;
1127
1128 case SO_PASSCRED:
1129 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1130 break;
1131
1132 case SO_PEERCRED:
1133 {
1134 struct ucred peercred;
1135 if (len > sizeof(peercred))
1136 len = sizeof(peercred);
1137 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1138 if (copy_to_user(optval, &peercred, len))
1139 return -EFAULT;
1140 goto lenout;
1141 }
1142
1143 case SO_PEERNAME:
1144 {
1145 char address[128];
1146
1147 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1148 return -ENOTCONN;
1149 if (lv < len)
1150 return -EINVAL;
1151 if (copy_to_user(optval, address, len))
1152 return -EFAULT;
1153 goto lenout;
1154 }
1155
1156 /* Dubious BSD thing... Probably nobody even uses it, but
1157 * the UNIX standard wants it for whatever reason... -DaveM
1158 */
1159 case SO_ACCEPTCONN:
1160 v.val = sk->sk_state == TCP_LISTEN;
1161 break;
1162
1163 case SO_PASSSEC:
1164 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1165 break;
1166
1167 case SO_PEERSEC:
1168 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1169
1170 case SO_MARK:
1171 v.val = sk->sk_mark;
1172 break;
1173
1174 case SO_RXQ_OVFL:
1175 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1176 break;
1177
1178 case SO_WIFI_STATUS:
1179 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1180 break;
1181
1182 case SO_PEEK_OFF:
1183 if (!sock->ops->set_peek_off)
1184 return -EOPNOTSUPP;
1185
1186 v.val = sk->sk_peek_off;
1187 break;
1188 case SO_NOFCS:
1189 v.val = sock_flag(sk, SOCK_NOFCS);
1190 break;
1191
1192 case SO_BINDTODEVICE:
1193 return sock_getbindtodevice(sk, optval, optlen, len);
1194
1195 case SO_GET_FILTER:
1196 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1197 if (len < 0)
1198 return len;
1199
1200 goto lenout;
1201
1202 case SO_LOCK_FILTER:
1203 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1204 break;
1205
1206 case SO_BPF_EXTENSIONS:
1207 v.val = bpf_tell_extensions();
1208 break;
1209
1210 case SO_SELECT_ERR_QUEUE:
1211 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1212 break;
1213
1214 #ifdef CONFIG_NET_RX_BUSY_POLL
1215 case SO_BUSY_POLL:
1216 v.val = sk->sk_ll_usec;
1217 break;
1218 #endif
1219
1220 case SO_MAX_PACING_RATE:
1221 v.val = sk->sk_max_pacing_rate;
1222 break;
1223
1224 default:
1225 return -ENOPROTOOPT;
1226 }
1227
1228 if (len > lv)
1229 len = lv;
1230 if (copy_to_user(optval, &v, len))
1231 return -EFAULT;
1232 lenout:
1233 if (put_user(len, optlen))
1234 return -EFAULT;
1235 return 0;
1236 }
1237
1238 /*
1239 * Initialize an sk_lock.
1240 *
1241 * (We also register the sk_lock with the lock validator.)
1242 */
1243 static inline void sock_lock_init(struct sock *sk)
1244 {
1245 sock_lock_init_class_and_name(sk,
1246 af_family_slock_key_strings[sk->sk_family],
1247 af_family_slock_keys + sk->sk_family,
1248 af_family_key_strings[sk->sk_family],
1249 af_family_keys + sk->sk_family);
1250 }
1251
1252 /*
1253 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1254 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1255 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1256 */
1257 static void sock_copy(struct sock *nsk, const struct sock *osk)
1258 {
1259 #ifdef CONFIG_SECURITY_NETWORK
1260 void *sptr = nsk->sk_security;
1261 #endif
1262 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1263
1264 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1265 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1266
1267 #ifdef CONFIG_SECURITY_NETWORK
1268 nsk->sk_security = sptr;
1269 security_sk_clone(osk, nsk);
1270 #endif
1271 }
1272
1273 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1274 {
1275 unsigned long nulls1, nulls2;
1276
1277 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1278 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1279 if (nulls1 > nulls2)
1280 swap(nulls1, nulls2);
1281
1282 if (nulls1 != 0)
1283 memset((char *)sk, 0, nulls1);
1284 memset((char *)sk + nulls1 + sizeof(void *), 0,
1285 nulls2 - nulls1 - sizeof(void *));
1286 memset((char *)sk + nulls2 + sizeof(void *), 0,
1287 size - nulls2 - sizeof(void *));
1288 }
1289 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1290
1291 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1292 int family)
1293 {
1294 struct sock *sk;
1295 struct kmem_cache *slab;
1296
1297 slab = prot->slab;
1298 if (slab != NULL) {
1299 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1300 if (!sk)
1301 return sk;
1302 if (priority & __GFP_ZERO) {
1303 if (prot->clear_sk)
1304 prot->clear_sk(sk, prot->obj_size);
1305 else
1306 sk_prot_clear_nulls(sk, prot->obj_size);
1307 }
1308 } else
1309 sk = kmalloc(prot->obj_size, priority);
1310
1311 if (sk != NULL) {
1312 kmemcheck_annotate_bitfield(sk, flags);
1313
1314 if (security_sk_alloc(sk, family, priority))
1315 goto out_free;
1316
1317 if (!try_module_get(prot->owner))
1318 goto out_free_sec;
1319 sk_tx_queue_clear(sk);
1320 }
1321
1322 return sk;
1323
1324 out_free_sec:
1325 security_sk_free(sk);
1326 out_free:
1327 if (slab != NULL)
1328 kmem_cache_free(slab, sk);
1329 else
1330 kfree(sk);
1331 return NULL;
1332 }
1333
1334 static void sk_prot_free(struct proto *prot, struct sock *sk)
1335 {
1336 struct kmem_cache *slab;
1337 struct module *owner;
1338
1339 owner = prot->owner;
1340 slab = prot->slab;
1341
1342 security_sk_free(sk);
1343 if (slab != NULL)
1344 kmem_cache_free(slab, sk);
1345 else
1346 kfree(sk);
1347 module_put(owner);
1348 }
1349
1350 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1351 void sock_update_netprioidx(struct sock *sk)
1352 {
1353 if (in_interrupt())
1354 return;
1355
1356 sk->sk_cgrp_prioidx = task_netprioidx(current);
1357 }
1358 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1359 #endif
1360
1361 /**
1362 * sk_alloc - All socket objects are allocated here
1363 * @net: the applicable net namespace
1364 * @family: protocol family
1365 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1366 * @prot: struct proto associated with this new sock instance
1367 */
1368 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1369 struct proto *prot)
1370 {
1371 struct sock *sk;
1372
1373 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1374 if (sk) {
1375 sk->sk_family = family;
1376 /*
1377 * See comment in struct sock definition to understand
1378 * why we need sk_prot_creator -acme
1379 */
1380 sk->sk_prot = sk->sk_prot_creator = prot;
1381 sock_lock_init(sk);
1382 sock_net_set(sk, get_net(net));
1383 atomic_set(&sk->sk_wmem_alloc, 1);
1384
1385 sock_update_classid(sk);
1386 sock_update_netprioidx(sk);
1387 }
1388
1389 return sk;
1390 }
1391 EXPORT_SYMBOL(sk_alloc);
1392
1393 static void __sk_free(struct sock *sk)
1394 {
1395 struct sk_filter *filter;
1396
1397 if (sk->sk_destruct)
1398 sk->sk_destruct(sk);
1399
1400 filter = rcu_dereference_check(sk->sk_filter,
1401 atomic_read(&sk->sk_wmem_alloc) == 0);
1402 if (filter) {
1403 sk_filter_uncharge(sk, filter);
1404 RCU_INIT_POINTER(sk->sk_filter, NULL);
1405 }
1406
1407 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1408
1409 if (atomic_read(&sk->sk_omem_alloc))
1410 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1411 __func__, atomic_read(&sk->sk_omem_alloc));
1412
1413 if (sk->sk_peer_cred)
1414 put_cred(sk->sk_peer_cred);
1415 put_pid(sk->sk_peer_pid);
1416 put_net(sock_net(sk));
1417 sk_prot_free(sk->sk_prot_creator, sk);
1418 }
1419
1420 void sk_free(struct sock *sk)
1421 {
1422 /*
1423 * We subtract one from sk_wmem_alloc and can know if
1424 * some packets are still in some tx queue.
1425 * If not null, sock_wfree() will call __sk_free(sk) later
1426 */
1427 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1428 __sk_free(sk);
1429 }
1430 EXPORT_SYMBOL(sk_free);
1431
1432 /*
1433 * Last sock_put should drop reference to sk->sk_net. It has already
1434 * been dropped in sk_change_net. Taking reference to stopping namespace
1435 * is not an option.
1436 * Take reference to a socket to remove it from hash _alive_ and after that
1437 * destroy it in the context of init_net.
1438 */
1439 void sk_release_kernel(struct sock *sk)
1440 {
1441 if (sk == NULL || sk->sk_socket == NULL)
1442 return;
1443
1444 sock_hold(sk);
1445 sock_release(sk->sk_socket);
1446 release_net(sock_net(sk));
1447 sock_net_set(sk, get_net(&init_net));
1448 sock_put(sk);
1449 }
1450 EXPORT_SYMBOL(sk_release_kernel);
1451
1452 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1453 {
1454 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1455 sock_update_memcg(newsk);
1456 }
1457
1458 /**
1459 * sk_clone_lock - clone a socket, and lock its clone
1460 * @sk: the socket to clone
1461 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1462 *
1463 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1464 */
1465 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1466 {
1467 struct sock *newsk;
1468 bool is_charged = true;
1469
1470 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1471 if (newsk != NULL) {
1472 struct sk_filter *filter;
1473
1474 sock_copy(newsk, sk);
1475
1476 /* SANITY */
1477 get_net(sock_net(newsk));
1478 sk_node_init(&newsk->sk_node);
1479 sock_lock_init(newsk);
1480 bh_lock_sock(newsk);
1481 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1482 newsk->sk_backlog.len = 0;
1483
1484 atomic_set(&newsk->sk_rmem_alloc, 0);
1485 /*
1486 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1487 */
1488 atomic_set(&newsk->sk_wmem_alloc, 1);
1489 atomic_set(&newsk->sk_omem_alloc, 0);
1490 skb_queue_head_init(&newsk->sk_receive_queue);
1491 skb_queue_head_init(&newsk->sk_write_queue);
1492 #ifdef CONFIG_NET_DMA
1493 skb_queue_head_init(&newsk->sk_async_wait_queue);
1494 #endif
1495
1496 spin_lock_init(&newsk->sk_dst_lock);
1497 rwlock_init(&newsk->sk_callback_lock);
1498 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1499 af_callback_keys + newsk->sk_family,
1500 af_family_clock_key_strings[newsk->sk_family]);
1501
1502 newsk->sk_dst_cache = NULL;
1503 newsk->sk_wmem_queued = 0;
1504 newsk->sk_forward_alloc = 0;
1505 newsk->sk_send_head = NULL;
1506 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1507
1508 sock_reset_flag(newsk, SOCK_DONE);
1509 skb_queue_head_init(&newsk->sk_error_queue);
1510
1511 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1512 if (filter != NULL)
1513 /* though it's an empty new sock, the charging may fail
1514 * if sysctl_optmem_max was changed between creation of
1515 * original socket and cloning
1516 */
1517 is_charged = sk_filter_charge(newsk, filter);
1518
1519 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1520 /* It is still raw copy of parent, so invalidate
1521 * destructor and make plain sk_free() */
1522 newsk->sk_destruct = NULL;
1523 bh_unlock_sock(newsk);
1524 sk_free(newsk);
1525 newsk = NULL;
1526 goto out;
1527 }
1528
1529 newsk->sk_err = 0;
1530 newsk->sk_priority = 0;
1531 /*
1532 * Before updating sk_refcnt, we must commit prior changes to memory
1533 * (Documentation/RCU/rculist_nulls.txt for details)
1534 */
1535 smp_wmb();
1536 atomic_set(&newsk->sk_refcnt, 2);
1537
1538 /*
1539 * Increment the counter in the same struct proto as the master
1540 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1541 * is the same as sk->sk_prot->socks, as this field was copied
1542 * with memcpy).
1543 *
1544 * This _changes_ the previous behaviour, where
1545 * tcp_create_openreq_child always was incrementing the
1546 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1547 * to be taken into account in all callers. -acme
1548 */
1549 sk_refcnt_debug_inc(newsk);
1550 sk_set_socket(newsk, NULL);
1551 newsk->sk_wq = NULL;
1552
1553 sk_update_clone(sk, newsk);
1554
1555 if (newsk->sk_prot->sockets_allocated)
1556 sk_sockets_allocated_inc(newsk);
1557
1558 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1559 net_enable_timestamp();
1560 }
1561 out:
1562 return newsk;
1563 }
1564 EXPORT_SYMBOL_GPL(sk_clone_lock);
1565
1566 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1567 {
1568 __sk_dst_set(sk, dst);
1569 sk->sk_route_caps = dst->dev->features;
1570 if (sk->sk_route_caps & NETIF_F_GSO)
1571 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1572 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1573 if (sk_can_gso(sk)) {
1574 if (dst->header_len) {
1575 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1576 } else {
1577 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1578 sk->sk_gso_max_size = dst->dev->gso_max_size;
1579 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1580 }
1581 }
1582 }
1583 EXPORT_SYMBOL_GPL(sk_setup_caps);
1584
1585 /*
1586 * Simple resource managers for sockets.
1587 */
1588
1589
1590 /*
1591 * Write buffer destructor automatically called from kfree_skb.
1592 */
1593 void sock_wfree(struct sk_buff *skb)
1594 {
1595 struct sock *sk = skb->sk;
1596 unsigned int len = skb->truesize;
1597
1598 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1599 /*
1600 * Keep a reference on sk_wmem_alloc, this will be released
1601 * after sk_write_space() call
1602 */
1603 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1604 sk->sk_write_space(sk);
1605 len = 1;
1606 }
1607 /*
1608 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1609 * could not do because of in-flight packets
1610 */
1611 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1612 __sk_free(sk);
1613 }
1614 EXPORT_SYMBOL(sock_wfree);
1615
1616 void skb_orphan_partial(struct sk_buff *skb)
1617 {
1618 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1619 * so we do not completely orphan skb, but transfert all
1620 * accounted bytes but one, to avoid unexpected reorders.
1621 */
1622 if (skb->destructor == sock_wfree
1623 #ifdef CONFIG_INET
1624 || skb->destructor == tcp_wfree
1625 #endif
1626 ) {
1627 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1628 skb->truesize = 1;
1629 } else {
1630 skb_orphan(skb);
1631 }
1632 }
1633 EXPORT_SYMBOL(skb_orphan_partial);
1634
1635 /*
1636 * Read buffer destructor automatically called from kfree_skb.
1637 */
1638 void sock_rfree(struct sk_buff *skb)
1639 {
1640 struct sock *sk = skb->sk;
1641 unsigned int len = skb->truesize;
1642
1643 atomic_sub(len, &sk->sk_rmem_alloc);
1644 sk_mem_uncharge(sk, len);
1645 }
1646 EXPORT_SYMBOL(sock_rfree);
1647
1648 void sock_edemux(struct sk_buff *skb)
1649 {
1650 struct sock *sk = skb->sk;
1651
1652 #ifdef CONFIG_INET
1653 if (sk->sk_state == TCP_TIME_WAIT)
1654 inet_twsk_put(inet_twsk(sk));
1655 else
1656 #endif
1657 sock_put(sk);
1658 }
1659 EXPORT_SYMBOL(sock_edemux);
1660
1661 kuid_t sock_i_uid(struct sock *sk)
1662 {
1663 kuid_t uid;
1664
1665 read_lock_bh(&sk->sk_callback_lock);
1666 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1667 read_unlock_bh(&sk->sk_callback_lock);
1668 return uid;
1669 }
1670 EXPORT_SYMBOL(sock_i_uid);
1671
1672 unsigned long sock_i_ino(struct sock *sk)
1673 {
1674 unsigned long ino;
1675
1676 read_lock_bh(&sk->sk_callback_lock);
1677 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1678 read_unlock_bh(&sk->sk_callback_lock);
1679 return ino;
1680 }
1681 EXPORT_SYMBOL(sock_i_ino);
1682
1683 /*
1684 * Allocate a skb from the socket's send buffer.
1685 */
1686 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1687 gfp_t priority)
1688 {
1689 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1690 struct sk_buff *skb = alloc_skb(size, priority);
1691 if (skb) {
1692 skb_set_owner_w(skb, sk);
1693 return skb;
1694 }
1695 }
1696 return NULL;
1697 }
1698 EXPORT_SYMBOL(sock_wmalloc);
1699
1700 /*
1701 * Allocate a memory block from the socket's option memory buffer.
1702 */
1703 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1704 {
1705 if ((unsigned int)size <= sysctl_optmem_max &&
1706 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1707 void *mem;
1708 /* First do the add, to avoid the race if kmalloc
1709 * might sleep.
1710 */
1711 atomic_add(size, &sk->sk_omem_alloc);
1712 mem = kmalloc(size, priority);
1713 if (mem)
1714 return mem;
1715 atomic_sub(size, &sk->sk_omem_alloc);
1716 }
1717 return NULL;
1718 }
1719 EXPORT_SYMBOL(sock_kmalloc);
1720
1721 /*
1722 * Free an option memory block.
1723 */
1724 void sock_kfree_s(struct sock *sk, void *mem, int size)
1725 {
1726 kfree(mem);
1727 atomic_sub(size, &sk->sk_omem_alloc);
1728 }
1729 EXPORT_SYMBOL(sock_kfree_s);
1730
1731 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1732 I think, these locks should be removed for datagram sockets.
1733 */
1734 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1735 {
1736 DEFINE_WAIT(wait);
1737
1738 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1739 for (;;) {
1740 if (!timeo)
1741 break;
1742 if (signal_pending(current))
1743 break;
1744 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1745 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1746 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1747 break;
1748 if (sk->sk_shutdown & SEND_SHUTDOWN)
1749 break;
1750 if (sk->sk_err)
1751 break;
1752 timeo = schedule_timeout(timeo);
1753 }
1754 finish_wait(sk_sleep(sk), &wait);
1755 return timeo;
1756 }
1757
1758
1759 /*
1760 * Generic send/receive buffer handlers
1761 */
1762
1763 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1764 unsigned long data_len, int noblock,
1765 int *errcode, int max_page_order)
1766 {
1767 struct sk_buff *skb = NULL;
1768 unsigned long chunk;
1769 gfp_t gfp_mask;
1770 long timeo;
1771 int err;
1772 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1773 struct page *page;
1774 int i;
1775
1776 err = -EMSGSIZE;
1777 if (npages > MAX_SKB_FRAGS)
1778 goto failure;
1779
1780 timeo = sock_sndtimeo(sk, noblock);
1781 while (!skb) {
1782 err = sock_error(sk);
1783 if (err != 0)
1784 goto failure;
1785
1786 err = -EPIPE;
1787 if (sk->sk_shutdown & SEND_SHUTDOWN)
1788 goto failure;
1789
1790 if (atomic_read(&sk->sk_wmem_alloc) >= sk->sk_sndbuf) {
1791 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1792 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1793 err = -EAGAIN;
1794 if (!timeo)
1795 goto failure;
1796 if (signal_pending(current))
1797 goto interrupted;
1798 timeo = sock_wait_for_wmem(sk, timeo);
1799 continue;
1800 }
1801
1802 err = -ENOBUFS;
1803 gfp_mask = sk->sk_allocation;
1804 if (gfp_mask & __GFP_WAIT)
1805 gfp_mask |= __GFP_REPEAT;
1806
1807 skb = alloc_skb(header_len, gfp_mask);
1808 if (!skb)
1809 goto failure;
1810
1811 skb->truesize += data_len;
1812
1813 for (i = 0; npages > 0; i++) {
1814 int order = max_page_order;
1815
1816 while (order) {
1817 if (npages >= 1 << order) {
1818 page = alloc_pages(sk->sk_allocation |
1819 __GFP_COMP |
1820 __GFP_NOWARN |
1821 __GFP_NORETRY,
1822 order);
1823 if (page)
1824 goto fill_page;
1825 /* Do not retry other high order allocations */
1826 order = 1;
1827 max_page_order = 0;
1828 }
1829 order--;
1830 }
1831 page = alloc_page(sk->sk_allocation);
1832 if (!page)
1833 goto failure;
1834 fill_page:
1835 chunk = min_t(unsigned long, data_len,
1836 PAGE_SIZE << order);
1837 skb_fill_page_desc(skb, i, page, 0, chunk);
1838 data_len -= chunk;
1839 npages -= 1 << order;
1840 }
1841 }
1842
1843 skb_set_owner_w(skb, sk);
1844 return skb;
1845
1846 interrupted:
1847 err = sock_intr_errno(timeo);
1848 failure:
1849 kfree_skb(skb);
1850 *errcode = err;
1851 return NULL;
1852 }
1853 EXPORT_SYMBOL(sock_alloc_send_pskb);
1854
1855 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1856 int noblock, int *errcode)
1857 {
1858 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1859 }
1860 EXPORT_SYMBOL(sock_alloc_send_skb);
1861
1862 /* On 32bit arches, an skb frag is limited to 2^15 */
1863 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1864
1865 /**
1866 * skb_page_frag_refill - check that a page_frag contains enough room
1867 * @sz: minimum size of the fragment we want to get
1868 * @pfrag: pointer to page_frag
1869 * @prio: priority for memory allocation
1870 *
1871 * Note: While this allocator tries to use high order pages, there is
1872 * no guarantee that allocations succeed. Therefore, @sz MUST be
1873 * less or equal than PAGE_SIZE.
1874 */
1875 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1876 {
1877 if (pfrag->page) {
1878 if (atomic_read(&pfrag->page->_count) == 1) {
1879 pfrag->offset = 0;
1880 return true;
1881 }
1882 if (pfrag->offset + sz <= pfrag->size)
1883 return true;
1884 put_page(pfrag->page);
1885 }
1886
1887 pfrag->offset = 0;
1888 if (SKB_FRAG_PAGE_ORDER) {
1889 pfrag->page = alloc_pages(gfp | __GFP_COMP |
1890 __GFP_NOWARN | __GFP_NORETRY,
1891 SKB_FRAG_PAGE_ORDER);
1892 if (likely(pfrag->page)) {
1893 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1894 return true;
1895 }
1896 }
1897 pfrag->page = alloc_page(gfp);
1898 if (likely(pfrag->page)) {
1899 pfrag->size = PAGE_SIZE;
1900 return true;
1901 }
1902 return false;
1903 }
1904 EXPORT_SYMBOL(skb_page_frag_refill);
1905
1906 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1907 {
1908 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1909 return true;
1910
1911 sk_enter_memory_pressure(sk);
1912 sk_stream_moderate_sndbuf(sk);
1913 return false;
1914 }
1915 EXPORT_SYMBOL(sk_page_frag_refill);
1916
1917 static void __lock_sock(struct sock *sk)
1918 __releases(&sk->sk_lock.slock)
1919 __acquires(&sk->sk_lock.slock)
1920 {
1921 DEFINE_WAIT(wait);
1922
1923 for (;;) {
1924 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1925 TASK_UNINTERRUPTIBLE);
1926 spin_unlock_bh(&sk->sk_lock.slock);
1927 schedule();
1928 spin_lock_bh(&sk->sk_lock.slock);
1929 if (!sock_owned_by_user(sk))
1930 break;
1931 }
1932 finish_wait(&sk->sk_lock.wq, &wait);
1933 }
1934
1935 static void __release_sock(struct sock *sk)
1936 __releases(&sk->sk_lock.slock)
1937 __acquires(&sk->sk_lock.slock)
1938 {
1939 struct sk_buff *skb = sk->sk_backlog.head;
1940
1941 do {
1942 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1943 bh_unlock_sock(sk);
1944
1945 do {
1946 struct sk_buff *next = skb->next;
1947
1948 prefetch(next);
1949 WARN_ON_ONCE(skb_dst_is_noref(skb));
1950 skb->next = NULL;
1951 sk_backlog_rcv(sk, skb);
1952
1953 /*
1954 * We are in process context here with softirqs
1955 * disabled, use cond_resched_softirq() to preempt.
1956 * This is safe to do because we've taken the backlog
1957 * queue private:
1958 */
1959 cond_resched_softirq();
1960
1961 skb = next;
1962 } while (skb != NULL);
1963
1964 bh_lock_sock(sk);
1965 } while ((skb = sk->sk_backlog.head) != NULL);
1966
1967 /*
1968 * Doing the zeroing here guarantee we can not loop forever
1969 * while a wild producer attempts to flood us.
1970 */
1971 sk->sk_backlog.len = 0;
1972 }
1973
1974 /**
1975 * sk_wait_data - wait for data to arrive at sk_receive_queue
1976 * @sk: sock to wait on
1977 * @timeo: for how long
1978 *
1979 * Now socket state including sk->sk_err is changed only under lock,
1980 * hence we may omit checks after joining wait queue.
1981 * We check receive queue before schedule() only as optimization;
1982 * it is very likely that release_sock() added new data.
1983 */
1984 int sk_wait_data(struct sock *sk, long *timeo)
1985 {
1986 int rc;
1987 DEFINE_WAIT(wait);
1988
1989 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1990 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1991 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1992 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1993 finish_wait(sk_sleep(sk), &wait);
1994 return rc;
1995 }
1996 EXPORT_SYMBOL(sk_wait_data);
1997
1998 /**
1999 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2000 * @sk: socket
2001 * @size: memory size to allocate
2002 * @kind: allocation type
2003 *
2004 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2005 * rmem allocation. This function assumes that protocols which have
2006 * memory_pressure use sk_wmem_queued as write buffer accounting.
2007 */
2008 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2009 {
2010 struct proto *prot = sk->sk_prot;
2011 int amt = sk_mem_pages(size);
2012 long allocated;
2013 int parent_status = UNDER_LIMIT;
2014
2015 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2016
2017 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2018
2019 /* Under limit. */
2020 if (parent_status == UNDER_LIMIT &&
2021 allocated <= sk_prot_mem_limits(sk, 0)) {
2022 sk_leave_memory_pressure(sk);
2023 return 1;
2024 }
2025
2026 /* Under pressure. (we or our parents) */
2027 if ((parent_status > SOFT_LIMIT) ||
2028 allocated > sk_prot_mem_limits(sk, 1))
2029 sk_enter_memory_pressure(sk);
2030
2031 /* Over hard limit (we or our parents) */
2032 if ((parent_status == OVER_LIMIT) ||
2033 (allocated > sk_prot_mem_limits(sk, 2)))
2034 goto suppress_allocation;
2035
2036 /* guarantee minimum buffer size under pressure */
2037 if (kind == SK_MEM_RECV) {
2038 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2039 return 1;
2040
2041 } else { /* SK_MEM_SEND */
2042 if (sk->sk_type == SOCK_STREAM) {
2043 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2044 return 1;
2045 } else if (atomic_read(&sk->sk_wmem_alloc) <
2046 prot->sysctl_wmem[0])
2047 return 1;
2048 }
2049
2050 if (sk_has_memory_pressure(sk)) {
2051 int alloc;
2052
2053 if (!sk_under_memory_pressure(sk))
2054 return 1;
2055 alloc = sk_sockets_allocated_read_positive(sk);
2056 if (sk_prot_mem_limits(sk, 2) > alloc *
2057 sk_mem_pages(sk->sk_wmem_queued +
2058 atomic_read(&sk->sk_rmem_alloc) +
2059 sk->sk_forward_alloc))
2060 return 1;
2061 }
2062
2063 suppress_allocation:
2064
2065 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2066 sk_stream_moderate_sndbuf(sk);
2067
2068 /* Fail only if socket is _under_ its sndbuf.
2069 * In this case we cannot block, so that we have to fail.
2070 */
2071 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2072 return 1;
2073 }
2074
2075 trace_sock_exceed_buf_limit(sk, prot, allocated);
2076
2077 /* Alas. Undo changes. */
2078 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2079
2080 sk_memory_allocated_sub(sk, amt);
2081
2082 return 0;
2083 }
2084 EXPORT_SYMBOL(__sk_mem_schedule);
2085
2086 /**
2087 * __sk_reclaim - reclaim memory_allocated
2088 * @sk: socket
2089 */
2090 void __sk_mem_reclaim(struct sock *sk)
2091 {
2092 sk_memory_allocated_sub(sk,
2093 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2094 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2095
2096 if (sk_under_memory_pressure(sk) &&
2097 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2098 sk_leave_memory_pressure(sk);
2099 }
2100 EXPORT_SYMBOL(__sk_mem_reclaim);
2101
2102
2103 /*
2104 * Set of default routines for initialising struct proto_ops when
2105 * the protocol does not support a particular function. In certain
2106 * cases where it makes no sense for a protocol to have a "do nothing"
2107 * function, some default processing is provided.
2108 */
2109
2110 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2111 {
2112 return -EOPNOTSUPP;
2113 }
2114 EXPORT_SYMBOL(sock_no_bind);
2115
2116 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2117 int len, int flags)
2118 {
2119 return -EOPNOTSUPP;
2120 }
2121 EXPORT_SYMBOL(sock_no_connect);
2122
2123 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2124 {
2125 return -EOPNOTSUPP;
2126 }
2127 EXPORT_SYMBOL(sock_no_socketpair);
2128
2129 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2130 {
2131 return -EOPNOTSUPP;
2132 }
2133 EXPORT_SYMBOL(sock_no_accept);
2134
2135 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2136 int *len, int peer)
2137 {
2138 return -EOPNOTSUPP;
2139 }
2140 EXPORT_SYMBOL(sock_no_getname);
2141
2142 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2143 {
2144 return 0;
2145 }
2146 EXPORT_SYMBOL(sock_no_poll);
2147
2148 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2149 {
2150 return -EOPNOTSUPP;
2151 }
2152 EXPORT_SYMBOL(sock_no_ioctl);
2153
2154 int sock_no_listen(struct socket *sock, int backlog)
2155 {
2156 return -EOPNOTSUPP;
2157 }
2158 EXPORT_SYMBOL(sock_no_listen);
2159
2160 int sock_no_shutdown(struct socket *sock, int how)
2161 {
2162 return -EOPNOTSUPP;
2163 }
2164 EXPORT_SYMBOL(sock_no_shutdown);
2165
2166 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2167 char __user *optval, unsigned int optlen)
2168 {
2169 return -EOPNOTSUPP;
2170 }
2171 EXPORT_SYMBOL(sock_no_setsockopt);
2172
2173 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2174 char __user *optval, int __user *optlen)
2175 {
2176 return -EOPNOTSUPP;
2177 }
2178 EXPORT_SYMBOL(sock_no_getsockopt);
2179
2180 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2181 size_t len)
2182 {
2183 return -EOPNOTSUPP;
2184 }
2185 EXPORT_SYMBOL(sock_no_sendmsg);
2186
2187 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2188 size_t len, int flags)
2189 {
2190 return -EOPNOTSUPP;
2191 }
2192 EXPORT_SYMBOL(sock_no_recvmsg);
2193
2194 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2195 {
2196 /* Mirror missing mmap method error code */
2197 return -ENODEV;
2198 }
2199 EXPORT_SYMBOL(sock_no_mmap);
2200
2201 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2202 {
2203 ssize_t res;
2204 struct msghdr msg = {.msg_flags = flags};
2205 struct kvec iov;
2206 char *kaddr = kmap(page);
2207 iov.iov_base = kaddr + offset;
2208 iov.iov_len = size;
2209 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2210 kunmap(page);
2211 return res;
2212 }
2213 EXPORT_SYMBOL(sock_no_sendpage);
2214
2215 /*
2216 * Default Socket Callbacks
2217 */
2218
2219 static void sock_def_wakeup(struct sock *sk)
2220 {
2221 struct socket_wq *wq;
2222
2223 rcu_read_lock();
2224 wq = rcu_dereference(sk->sk_wq);
2225 if (wq_has_sleeper(wq))
2226 wake_up_interruptible_all(&wq->wait);
2227 rcu_read_unlock();
2228 }
2229
2230 static void sock_def_error_report(struct sock *sk)
2231 {
2232 struct socket_wq *wq;
2233
2234 rcu_read_lock();
2235 wq = rcu_dereference(sk->sk_wq);
2236 if (wq_has_sleeper(wq))
2237 wake_up_interruptible_poll(&wq->wait, POLLERR);
2238 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2239 rcu_read_unlock();
2240 }
2241
2242 static void sock_def_readable(struct sock *sk)
2243 {
2244 struct socket_wq *wq;
2245
2246 rcu_read_lock();
2247 wq = rcu_dereference(sk->sk_wq);
2248 if (wq_has_sleeper(wq))
2249 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2250 POLLRDNORM | POLLRDBAND);
2251 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2252 rcu_read_unlock();
2253 }
2254
2255 static void sock_def_write_space(struct sock *sk)
2256 {
2257 struct socket_wq *wq;
2258
2259 rcu_read_lock();
2260
2261 /* Do not wake up a writer until he can make "significant"
2262 * progress. --DaveM
2263 */
2264 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2265 wq = rcu_dereference(sk->sk_wq);
2266 if (wq_has_sleeper(wq))
2267 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2268 POLLWRNORM | POLLWRBAND);
2269
2270 /* Should agree with poll, otherwise some programs break */
2271 if (sock_writeable(sk))
2272 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2273 }
2274
2275 rcu_read_unlock();
2276 }
2277
2278 static void sock_def_destruct(struct sock *sk)
2279 {
2280 kfree(sk->sk_protinfo);
2281 }
2282
2283 void sk_send_sigurg(struct sock *sk)
2284 {
2285 if (sk->sk_socket && sk->sk_socket->file)
2286 if (send_sigurg(&sk->sk_socket->file->f_owner))
2287 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2288 }
2289 EXPORT_SYMBOL(sk_send_sigurg);
2290
2291 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2292 unsigned long expires)
2293 {
2294 if (!mod_timer(timer, expires))
2295 sock_hold(sk);
2296 }
2297 EXPORT_SYMBOL(sk_reset_timer);
2298
2299 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2300 {
2301 if (del_timer(timer))
2302 __sock_put(sk);
2303 }
2304 EXPORT_SYMBOL(sk_stop_timer);
2305
2306 void sock_init_data(struct socket *sock, struct sock *sk)
2307 {
2308 skb_queue_head_init(&sk->sk_receive_queue);
2309 skb_queue_head_init(&sk->sk_write_queue);
2310 skb_queue_head_init(&sk->sk_error_queue);
2311 #ifdef CONFIG_NET_DMA
2312 skb_queue_head_init(&sk->sk_async_wait_queue);
2313 #endif
2314
2315 sk->sk_send_head = NULL;
2316
2317 init_timer(&sk->sk_timer);
2318
2319 sk->sk_allocation = GFP_KERNEL;
2320 sk->sk_rcvbuf = sysctl_rmem_default;
2321 sk->sk_sndbuf = sysctl_wmem_default;
2322 sk->sk_state = TCP_CLOSE;
2323 sk_set_socket(sk, sock);
2324
2325 sock_set_flag(sk, SOCK_ZAPPED);
2326
2327 if (sock) {
2328 sk->sk_type = sock->type;
2329 sk->sk_wq = sock->wq;
2330 sock->sk = sk;
2331 } else
2332 sk->sk_wq = NULL;
2333
2334 spin_lock_init(&sk->sk_dst_lock);
2335 rwlock_init(&sk->sk_callback_lock);
2336 lockdep_set_class_and_name(&sk->sk_callback_lock,
2337 af_callback_keys + sk->sk_family,
2338 af_family_clock_key_strings[sk->sk_family]);
2339
2340 sk->sk_state_change = sock_def_wakeup;
2341 sk->sk_data_ready = sock_def_readable;
2342 sk->sk_write_space = sock_def_write_space;
2343 sk->sk_error_report = sock_def_error_report;
2344 sk->sk_destruct = sock_def_destruct;
2345
2346 sk->sk_frag.page = NULL;
2347 sk->sk_frag.offset = 0;
2348 sk->sk_peek_off = -1;
2349
2350 sk->sk_peer_pid = NULL;
2351 sk->sk_peer_cred = NULL;
2352 sk->sk_write_pending = 0;
2353 sk->sk_rcvlowat = 1;
2354 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2355 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2356
2357 sk->sk_stamp = ktime_set(-1L, 0);
2358
2359 #ifdef CONFIG_NET_RX_BUSY_POLL
2360 sk->sk_napi_id = 0;
2361 sk->sk_ll_usec = sysctl_net_busy_read;
2362 #endif
2363
2364 sk->sk_max_pacing_rate = ~0U;
2365 sk->sk_pacing_rate = ~0U;
2366 /*
2367 * Before updating sk_refcnt, we must commit prior changes to memory
2368 * (Documentation/RCU/rculist_nulls.txt for details)
2369 */
2370 smp_wmb();
2371 atomic_set(&sk->sk_refcnt, 1);
2372 atomic_set(&sk->sk_drops, 0);
2373 }
2374 EXPORT_SYMBOL(sock_init_data);
2375
2376 void lock_sock_nested(struct sock *sk, int subclass)
2377 {
2378 might_sleep();
2379 spin_lock_bh(&sk->sk_lock.slock);
2380 if (sk->sk_lock.owned)
2381 __lock_sock(sk);
2382 sk->sk_lock.owned = 1;
2383 spin_unlock(&sk->sk_lock.slock);
2384 /*
2385 * The sk_lock has mutex_lock() semantics here:
2386 */
2387 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2388 local_bh_enable();
2389 }
2390 EXPORT_SYMBOL(lock_sock_nested);
2391
2392 void release_sock(struct sock *sk)
2393 {
2394 /*
2395 * The sk_lock has mutex_unlock() semantics:
2396 */
2397 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2398
2399 spin_lock_bh(&sk->sk_lock.slock);
2400 if (sk->sk_backlog.tail)
2401 __release_sock(sk);
2402
2403 /* Warning : release_cb() might need to release sk ownership,
2404 * ie call sock_release_ownership(sk) before us.
2405 */
2406 if (sk->sk_prot->release_cb)
2407 sk->sk_prot->release_cb(sk);
2408
2409 sock_release_ownership(sk);
2410 if (waitqueue_active(&sk->sk_lock.wq))
2411 wake_up(&sk->sk_lock.wq);
2412 spin_unlock_bh(&sk->sk_lock.slock);
2413 }
2414 EXPORT_SYMBOL(release_sock);
2415
2416 /**
2417 * lock_sock_fast - fast version of lock_sock
2418 * @sk: socket
2419 *
2420 * This version should be used for very small section, where process wont block
2421 * return false if fast path is taken
2422 * sk_lock.slock locked, owned = 0, BH disabled
2423 * return true if slow path is taken
2424 * sk_lock.slock unlocked, owned = 1, BH enabled
2425 */
2426 bool lock_sock_fast(struct sock *sk)
2427 {
2428 might_sleep();
2429 spin_lock_bh(&sk->sk_lock.slock);
2430
2431 if (!sk->sk_lock.owned)
2432 /*
2433 * Note : We must disable BH
2434 */
2435 return false;
2436
2437 __lock_sock(sk);
2438 sk->sk_lock.owned = 1;
2439 spin_unlock(&sk->sk_lock.slock);
2440 /*
2441 * The sk_lock has mutex_lock() semantics here:
2442 */
2443 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2444 local_bh_enable();
2445 return true;
2446 }
2447 EXPORT_SYMBOL(lock_sock_fast);
2448
2449 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2450 {
2451 struct timeval tv;
2452 if (!sock_flag(sk, SOCK_TIMESTAMP))
2453 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2454 tv = ktime_to_timeval(sk->sk_stamp);
2455 if (tv.tv_sec == -1)
2456 return -ENOENT;
2457 if (tv.tv_sec == 0) {
2458 sk->sk_stamp = ktime_get_real();
2459 tv = ktime_to_timeval(sk->sk_stamp);
2460 }
2461 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2462 }
2463 EXPORT_SYMBOL(sock_get_timestamp);
2464
2465 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2466 {
2467 struct timespec ts;
2468 if (!sock_flag(sk, SOCK_TIMESTAMP))
2469 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2470 ts = ktime_to_timespec(sk->sk_stamp);
2471 if (ts.tv_sec == -1)
2472 return -ENOENT;
2473 if (ts.tv_sec == 0) {
2474 sk->sk_stamp = ktime_get_real();
2475 ts = ktime_to_timespec(sk->sk_stamp);
2476 }
2477 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2478 }
2479 EXPORT_SYMBOL(sock_get_timestampns);
2480
2481 void sock_enable_timestamp(struct sock *sk, int flag)
2482 {
2483 if (!sock_flag(sk, flag)) {
2484 unsigned long previous_flags = sk->sk_flags;
2485
2486 sock_set_flag(sk, flag);
2487 /*
2488 * we just set one of the two flags which require net
2489 * time stamping, but time stamping might have been on
2490 * already because of the other one
2491 */
2492 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2493 net_enable_timestamp();
2494 }
2495 }
2496
2497 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2498 int level, int type)
2499 {
2500 struct sock_exterr_skb *serr;
2501 struct sk_buff *skb, *skb2;
2502 int copied, err;
2503
2504 err = -EAGAIN;
2505 skb = skb_dequeue(&sk->sk_error_queue);
2506 if (skb == NULL)
2507 goto out;
2508
2509 copied = skb->len;
2510 if (copied > len) {
2511 msg->msg_flags |= MSG_TRUNC;
2512 copied = len;
2513 }
2514 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2515 if (err)
2516 goto out_free_skb;
2517
2518 sock_recv_timestamp(msg, sk, skb);
2519
2520 serr = SKB_EXT_ERR(skb);
2521 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2522
2523 msg->msg_flags |= MSG_ERRQUEUE;
2524 err = copied;
2525
2526 /* Reset and regenerate socket error */
2527 spin_lock_bh(&sk->sk_error_queue.lock);
2528 sk->sk_err = 0;
2529 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2530 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2531 spin_unlock_bh(&sk->sk_error_queue.lock);
2532 sk->sk_error_report(sk);
2533 } else
2534 spin_unlock_bh(&sk->sk_error_queue.lock);
2535
2536 out_free_skb:
2537 kfree_skb(skb);
2538 out:
2539 return err;
2540 }
2541 EXPORT_SYMBOL(sock_recv_errqueue);
2542
2543 /*
2544 * Get a socket option on an socket.
2545 *
2546 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2547 * asynchronous errors should be reported by getsockopt. We assume
2548 * this means if you specify SO_ERROR (otherwise whats the point of it).
2549 */
2550 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2551 char __user *optval, int __user *optlen)
2552 {
2553 struct sock *sk = sock->sk;
2554
2555 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2556 }
2557 EXPORT_SYMBOL(sock_common_getsockopt);
2558
2559 #ifdef CONFIG_COMPAT
2560 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2561 char __user *optval, int __user *optlen)
2562 {
2563 struct sock *sk = sock->sk;
2564
2565 if (sk->sk_prot->compat_getsockopt != NULL)
2566 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2567 optval, optlen);
2568 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2569 }
2570 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2571 #endif
2572
2573 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2574 struct msghdr *msg, size_t size, int flags)
2575 {
2576 struct sock *sk = sock->sk;
2577 int addr_len = 0;
2578 int err;
2579
2580 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2581 flags & ~MSG_DONTWAIT, &addr_len);
2582 if (err >= 0)
2583 msg->msg_namelen = addr_len;
2584 return err;
2585 }
2586 EXPORT_SYMBOL(sock_common_recvmsg);
2587
2588 /*
2589 * Set socket options on an inet socket.
2590 */
2591 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2592 char __user *optval, unsigned int optlen)
2593 {
2594 struct sock *sk = sock->sk;
2595
2596 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2597 }
2598 EXPORT_SYMBOL(sock_common_setsockopt);
2599
2600 #ifdef CONFIG_COMPAT
2601 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2602 char __user *optval, unsigned int optlen)
2603 {
2604 struct sock *sk = sock->sk;
2605
2606 if (sk->sk_prot->compat_setsockopt != NULL)
2607 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2608 optval, optlen);
2609 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2610 }
2611 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2612 #endif
2613
2614 void sk_common_release(struct sock *sk)
2615 {
2616 if (sk->sk_prot->destroy)
2617 sk->sk_prot->destroy(sk);
2618
2619 /*
2620 * Observation: when sock_common_release is called, processes have
2621 * no access to socket. But net still has.
2622 * Step one, detach it from networking:
2623 *
2624 * A. Remove from hash tables.
2625 */
2626
2627 sk->sk_prot->unhash(sk);
2628
2629 /*
2630 * In this point socket cannot receive new packets, but it is possible
2631 * that some packets are in flight because some CPU runs receiver and
2632 * did hash table lookup before we unhashed socket. They will achieve
2633 * receive queue and will be purged by socket destructor.
2634 *
2635 * Also we still have packets pending on receive queue and probably,
2636 * our own packets waiting in device queues. sock_destroy will drain
2637 * receive queue, but transmitted packets will delay socket destruction
2638 * until the last reference will be released.
2639 */
2640
2641 sock_orphan(sk);
2642
2643 xfrm_sk_free_policy(sk);
2644
2645 sk_refcnt_debug_release(sk);
2646
2647 if (sk->sk_frag.page) {
2648 put_page(sk->sk_frag.page);
2649 sk->sk_frag.page = NULL;
2650 }
2651
2652 sock_put(sk);
2653 }
2654 EXPORT_SYMBOL(sk_common_release);
2655
2656 #ifdef CONFIG_PROC_FS
2657 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2658 struct prot_inuse {
2659 int val[PROTO_INUSE_NR];
2660 };
2661
2662 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2663
2664 #ifdef CONFIG_NET_NS
2665 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2666 {
2667 __this_cpu_add(net->core.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_ptr(net->core.inuse, cpu)->val[idx];
2678
2679 return res >= 0 ? res : 0;
2680 }
2681 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2682
2683 static int __net_init sock_inuse_init_net(struct net *net)
2684 {
2685 net->core.inuse = alloc_percpu(struct prot_inuse);
2686 return net->core.inuse ? 0 : -ENOMEM;
2687 }
2688
2689 static void __net_exit sock_inuse_exit_net(struct net *net)
2690 {
2691 free_percpu(net->core.inuse);
2692 }
2693
2694 static struct pernet_operations net_inuse_ops = {
2695 .init = sock_inuse_init_net,
2696 .exit = sock_inuse_exit_net,
2697 };
2698
2699 static __init int net_inuse_init(void)
2700 {
2701 if (register_pernet_subsys(&net_inuse_ops))
2702 panic("Cannot initialize net inuse counters");
2703
2704 return 0;
2705 }
2706
2707 core_initcall(net_inuse_init);
2708 #else
2709 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2710
2711 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2712 {
2713 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2714 }
2715 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2716
2717 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2718 {
2719 int cpu, idx = prot->inuse_idx;
2720 int res = 0;
2721
2722 for_each_possible_cpu(cpu)
2723 res += per_cpu(prot_inuse, cpu).val[idx];
2724
2725 return res >= 0 ? res : 0;
2726 }
2727 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2728 #endif
2729
2730 static void assign_proto_idx(struct proto *prot)
2731 {
2732 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2733
2734 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2735 pr_err("PROTO_INUSE_NR exhausted\n");
2736 return;
2737 }
2738
2739 set_bit(prot->inuse_idx, proto_inuse_idx);
2740 }
2741
2742 static void release_proto_idx(struct proto *prot)
2743 {
2744 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2745 clear_bit(prot->inuse_idx, proto_inuse_idx);
2746 }
2747 #else
2748 static inline void assign_proto_idx(struct proto *prot)
2749 {
2750 }
2751
2752 static inline void release_proto_idx(struct proto *prot)
2753 {
2754 }
2755 #endif
2756
2757 int proto_register(struct proto *prot, int alloc_slab)
2758 {
2759 if (alloc_slab) {
2760 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2761 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2762 NULL);
2763
2764 if (prot->slab == NULL) {
2765 pr_crit("%s: Can't create sock SLAB cache!\n",
2766 prot->name);
2767 goto out;
2768 }
2769
2770 if (prot->rsk_prot != NULL) {
2771 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2772 if (prot->rsk_prot->slab_name == NULL)
2773 goto out_free_sock_slab;
2774
2775 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2776 prot->rsk_prot->obj_size, 0,
2777 SLAB_HWCACHE_ALIGN, NULL);
2778
2779 if (prot->rsk_prot->slab == NULL) {
2780 pr_crit("%s: Can't create request sock SLAB cache!\n",
2781 prot->name);
2782 goto out_free_request_sock_slab_name;
2783 }
2784 }
2785
2786 if (prot->twsk_prot != NULL) {
2787 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2788
2789 if (prot->twsk_prot->twsk_slab_name == NULL)
2790 goto out_free_request_sock_slab;
2791
2792 prot->twsk_prot->twsk_slab =
2793 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2794 prot->twsk_prot->twsk_obj_size,
2795 0,
2796 SLAB_HWCACHE_ALIGN |
2797 prot->slab_flags,
2798 NULL);
2799 if (prot->twsk_prot->twsk_slab == NULL)
2800 goto out_free_timewait_sock_slab_name;
2801 }
2802 }
2803
2804 mutex_lock(&proto_list_mutex);
2805 list_add(&prot->node, &proto_list);
2806 assign_proto_idx(prot);
2807 mutex_unlock(&proto_list_mutex);
2808 return 0;
2809
2810 out_free_timewait_sock_slab_name:
2811 kfree(prot->twsk_prot->twsk_slab_name);
2812 out_free_request_sock_slab:
2813 if (prot->rsk_prot && prot->rsk_prot->slab) {
2814 kmem_cache_destroy(prot->rsk_prot->slab);
2815 prot->rsk_prot->slab = NULL;
2816 }
2817 out_free_request_sock_slab_name:
2818 if (prot->rsk_prot)
2819 kfree(prot->rsk_prot->slab_name);
2820 out_free_sock_slab:
2821 kmem_cache_destroy(prot->slab);
2822 prot->slab = NULL;
2823 out:
2824 return -ENOBUFS;
2825 }
2826 EXPORT_SYMBOL(proto_register);
2827
2828 void proto_unregister(struct proto *prot)
2829 {
2830 mutex_lock(&proto_list_mutex);
2831 release_proto_idx(prot);
2832 list_del(&prot->node);
2833 mutex_unlock(&proto_list_mutex);
2834
2835 if (prot->slab != NULL) {
2836 kmem_cache_destroy(prot->slab);
2837 prot->slab = NULL;
2838 }
2839
2840 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2841 kmem_cache_destroy(prot->rsk_prot->slab);
2842 kfree(prot->rsk_prot->slab_name);
2843 prot->rsk_prot->slab = NULL;
2844 }
2845
2846 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2847 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2848 kfree(prot->twsk_prot->twsk_slab_name);
2849 prot->twsk_prot->twsk_slab = NULL;
2850 }
2851 }
2852 EXPORT_SYMBOL(proto_unregister);
2853
2854 #ifdef CONFIG_PROC_FS
2855 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2856 __acquires(proto_list_mutex)
2857 {
2858 mutex_lock(&proto_list_mutex);
2859 return seq_list_start_head(&proto_list, *pos);
2860 }
2861
2862 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2863 {
2864 return seq_list_next(v, &proto_list, pos);
2865 }
2866
2867 static void proto_seq_stop(struct seq_file *seq, void *v)
2868 __releases(proto_list_mutex)
2869 {
2870 mutex_unlock(&proto_list_mutex);
2871 }
2872
2873 static char proto_method_implemented(const void *method)
2874 {
2875 return method == NULL ? 'n' : 'y';
2876 }
2877 static long sock_prot_memory_allocated(struct proto *proto)
2878 {
2879 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2880 }
2881
2882 static char *sock_prot_memory_pressure(struct proto *proto)
2883 {
2884 return proto->memory_pressure != NULL ?
2885 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2886 }
2887
2888 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2889 {
2890
2891 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2892 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2893 proto->name,
2894 proto->obj_size,
2895 sock_prot_inuse_get(seq_file_net(seq), proto),
2896 sock_prot_memory_allocated(proto),
2897 sock_prot_memory_pressure(proto),
2898 proto->max_header,
2899 proto->slab == NULL ? "no" : "yes",
2900 module_name(proto->owner),
2901 proto_method_implemented(proto->close),
2902 proto_method_implemented(proto->connect),
2903 proto_method_implemented(proto->disconnect),
2904 proto_method_implemented(proto->accept),
2905 proto_method_implemented(proto->ioctl),
2906 proto_method_implemented(proto->init),
2907 proto_method_implemented(proto->destroy),
2908 proto_method_implemented(proto->shutdown),
2909 proto_method_implemented(proto->setsockopt),
2910 proto_method_implemented(proto->getsockopt),
2911 proto_method_implemented(proto->sendmsg),
2912 proto_method_implemented(proto->recvmsg),
2913 proto_method_implemented(proto->sendpage),
2914 proto_method_implemented(proto->bind),
2915 proto_method_implemented(proto->backlog_rcv),
2916 proto_method_implemented(proto->hash),
2917 proto_method_implemented(proto->unhash),
2918 proto_method_implemented(proto->get_port),
2919 proto_method_implemented(proto->enter_memory_pressure));
2920 }
2921
2922 static int proto_seq_show(struct seq_file *seq, void *v)
2923 {
2924 if (v == &proto_list)
2925 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2926 "protocol",
2927 "size",
2928 "sockets",
2929 "memory",
2930 "press",
2931 "maxhdr",
2932 "slab",
2933 "module",
2934 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2935 else
2936 proto_seq_printf(seq, list_entry(v, struct proto, node));
2937 return 0;
2938 }
2939
2940 static const struct seq_operations proto_seq_ops = {
2941 .start = proto_seq_start,
2942 .next = proto_seq_next,
2943 .stop = proto_seq_stop,
2944 .show = proto_seq_show,
2945 };
2946
2947 static int proto_seq_open(struct inode *inode, struct file *file)
2948 {
2949 return seq_open_net(inode, file, &proto_seq_ops,
2950 sizeof(struct seq_net_private));
2951 }
2952
2953 static const struct file_operations proto_seq_fops = {
2954 .owner = THIS_MODULE,
2955 .open = proto_seq_open,
2956 .read = seq_read,
2957 .llseek = seq_lseek,
2958 .release = seq_release_net,
2959 };
2960
2961 static __net_init int proto_init_net(struct net *net)
2962 {
2963 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2964 return -ENOMEM;
2965
2966 return 0;
2967 }
2968
2969 static __net_exit void proto_exit_net(struct net *net)
2970 {
2971 remove_proc_entry("protocols", net->proc_net);
2972 }
2973
2974
2975 static __net_initdata struct pernet_operations proto_net_ops = {
2976 .init = proto_init_net,
2977 .exit = proto_exit_net,
2978 };
2979
2980 static int __init proto_init(void)
2981 {
2982 return register_pernet_subsys(&proto_net_ops);
2983 }
2984
2985 subsys_initcall(proto_init);
2986
2987 #endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.14238 seconds and 5 git commands to generate.