projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge remote-tracking branch 'vfio/next'
[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 #include <linux/sock_diag.h>
135
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
138
139 #include <trace/events/sock.h>
140
141 #ifdef CONFIG_INET
142 #include <net/tcp.h>
143 #endif
144
145 #include <net/busy_poll.h>
146
147 static DEFINE_MUTEX(proto_list_mutex);
148 static LIST_HEAD(proto_list);
149
150 /**
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
155 *
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
159 */
160 bool sk_ns_capable(const struct sock *sk,
161 struct user_namespace *user_ns, int cap)
162 {
163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
164 ns_capable(user_ns, cap);
165 }
166 EXPORT_SYMBOL(sk_ns_capable);
167
168 /**
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
172 *
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
175 * namespaces.
176 */
177 bool sk_capable(const struct sock *sk, int cap)
178 {
179 return sk_ns_capable(sk, &init_user_ns, cap);
180 }
181 EXPORT_SYMBOL(sk_capable);
182
183 /**
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
187 *
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
191 */
192 bool sk_net_capable(const struct sock *sk, int cap)
193 {
194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195 }
196 EXPORT_SYMBOL(sk_net_capable);
197
198 /*
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family:
201 */
202 static struct lock_class_key af_family_keys[AF_MAX];
203 static struct lock_class_key af_family_slock_keys[AF_MAX];
204
205 /*
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
208 * locks is fast):
209 */
210 static const char *const af_family_key_strings[AF_MAX+1] = {
211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_KCM" ,
225 "sk_lock-AF_MAX"
226 };
227 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
228 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
229 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
230 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
231 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
232 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
233 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
234 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
235 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
236 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
237 "slock-27" , "slock-28" , "slock-AF_CAN" ,
238 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
239 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
240 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
241 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" ,
242 "slock-AF_MAX"
243 };
244 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
245 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
246 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
247 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
248 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
249 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
250 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
251 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
252 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
253 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
254 "clock-27" , "clock-28" , "clock-AF_CAN" ,
255 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
256 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
257 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
258 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" ,
259 "clock-AF_MAX"
260 };
261
262 /*
263 * sk_callback_lock locking rules are per-address-family,
264 * so split the lock classes by using a per-AF key:
265 */
266 static struct lock_class_key af_callback_keys[AF_MAX];
267
268 /* Take into consideration the size of the struct sk_buff overhead in the
269 * determination of these values, since that is non-constant across
270 * platforms. This makes socket queueing behavior and performance
271 * not depend upon such differences.
272 */
273 #define _SK_MEM_PACKETS 256
274 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
275 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
276 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
277
278 /* Run time adjustable parameters. */
279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
280 EXPORT_SYMBOL(sysctl_wmem_max);
281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
282 EXPORT_SYMBOL(sysctl_rmem_max);
283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
285
286 /* Maximal space eaten by iovec or ancillary data plus some space */
287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
288 EXPORT_SYMBOL(sysctl_optmem_max);
289
290 int sysctl_tstamp_allow_data __read_mostly = 1;
291
292 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
293 EXPORT_SYMBOL_GPL(memalloc_socks);
294
295 /**
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
298 *
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
302 */
303 void sk_set_memalloc(struct sock *sk)
304 {
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
307 static_key_slow_inc(&memalloc_socks);
308 }
309 EXPORT_SYMBOL_GPL(sk_set_memalloc);
310
311 void sk_clear_memalloc(struct sock *sk)
312 {
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
315 static_key_slow_dec(&memalloc_socks);
316
317 /*
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
323 */
324 sk_mem_reclaim(sk);
325 }
326 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
327
328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
329 {
330 int ret;
331 unsigned long pflags = current->flags;
332
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
335
336 current->flags |= PF_MEMALLOC;
337 ret = sk->sk_backlog_rcv(sk, skb);
338 tsk_restore_flags(current, pflags, PF_MEMALLOC);
339
340 return ret;
341 }
342 EXPORT_SYMBOL(__sk_backlog_rcv);
343
344 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
345 {
346 struct timeval tv;
347
348 if (optlen < sizeof(tv))
349 return -EINVAL;
350 if (copy_from_user(&tv, optval, sizeof(tv)))
351 return -EFAULT;
352 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
353 return -EDOM;
354
355 if (tv.tv_sec < 0) {
356 static int warned __read_mostly;
357
358 *timeo_p = 0;
359 if (warned < 10 && net_ratelimit()) {
360 warned++;
361 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
362 __func__, current->comm, task_pid_nr(current));
363 }
364 return 0;
365 }
366 *timeo_p = MAX_SCHEDULE_TIMEOUT;
367 if (tv.tv_sec == 0 && tv.tv_usec == 0)
368 return 0;
369 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
370 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
371 return 0;
372 }
373
374 static void sock_warn_obsolete_bsdism(const char *name)
375 {
376 static int warned;
377 static char warncomm[TASK_COMM_LEN];
378 if (strcmp(warncomm, current->comm) && warned < 5) {
379 strcpy(warncomm, current->comm);
380 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
381 warncomm, name);
382 warned++;
383 }
384 }
385
386 static bool sock_needs_netstamp(const struct sock *sk)
387 {
388 switch (sk->sk_family) {
389 case AF_UNSPEC:
390 case AF_UNIX:
391 return false;
392 default:
393 return true;
394 }
395 }
396
397 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
398 {
399 if (sk->sk_flags & flags) {
400 sk->sk_flags &= ~flags;
401 if (sock_needs_netstamp(sk) &&
402 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
403 net_disable_timestamp();
404 }
405 }
406
407
408 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
409 {
410 unsigned long flags;
411 struct sk_buff_head *list = &sk->sk_receive_queue;
412
413 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
414 atomic_inc(&sk->sk_drops);
415 trace_sock_rcvqueue_full(sk, skb);
416 return -ENOMEM;
417 }
418
419 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
420 atomic_inc(&sk->sk_drops);
421 return -ENOBUFS;
422 }
423
424 skb->dev = NULL;
425 skb_set_owner_r(skb, sk);
426
427 /* we escape from rcu protected region, make sure we dont leak
428 * a norefcounted dst
429 */
430 skb_dst_force(skb);
431
432 spin_lock_irqsave(&list->lock, flags);
433 sock_skb_set_dropcount(sk, skb);
434 __skb_queue_tail(list, skb);
435 spin_unlock_irqrestore(&list->lock, flags);
436
437 if (!sock_flag(sk, SOCK_DEAD))
438 sk->sk_data_ready(sk);
439 return 0;
440 }
441 EXPORT_SYMBOL(__sock_queue_rcv_skb);
442
443 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
444 {
445 int err;
446
447 err = sk_filter(sk, skb);
448 if (err)
449 return err;
450
451 return __sock_queue_rcv_skb(sk, skb);
452 }
453 EXPORT_SYMBOL(sock_queue_rcv_skb);
454
455 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
456 const int nested, unsigned int trim_cap)
457 {
458 int rc = NET_RX_SUCCESS;
459
460 if (sk_filter_trim_cap(sk, skb, trim_cap))
461 goto discard_and_relse;
462
463 skb->dev = NULL;
464
465 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
466 atomic_inc(&sk->sk_drops);
467 goto discard_and_relse;
468 }
469 if (nested)
470 bh_lock_sock_nested(sk);
471 else
472 bh_lock_sock(sk);
473 if (!sock_owned_by_user(sk)) {
474 /*
475 * trylock + unlock semantics:
476 */
477 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
478
479 rc = sk_backlog_rcv(sk, skb);
480
481 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
482 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
483 bh_unlock_sock(sk);
484 atomic_inc(&sk->sk_drops);
485 goto discard_and_relse;
486 }
487
488 bh_unlock_sock(sk);
489 out:
490 sock_put(sk);
491 return rc;
492 discard_and_relse:
493 kfree_skb(skb);
494 goto out;
495 }
496 EXPORT_SYMBOL(__sk_receive_skb);
497
498 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
499 {
500 struct dst_entry *dst = __sk_dst_get(sk);
501
502 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
503 sk_tx_queue_clear(sk);
504 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
505 dst_release(dst);
506 return NULL;
507 }
508
509 return dst;
510 }
511 EXPORT_SYMBOL(__sk_dst_check);
512
513 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
514 {
515 struct dst_entry *dst = sk_dst_get(sk);
516
517 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
518 sk_dst_reset(sk);
519 dst_release(dst);
520 return NULL;
521 }
522
523 return dst;
524 }
525 EXPORT_SYMBOL(sk_dst_check);
526
527 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
528 int optlen)
529 {
530 int ret = -ENOPROTOOPT;
531 #ifdef CONFIG_NETDEVICES
532 struct net *net = sock_net(sk);
533 char devname[IFNAMSIZ];
534 int index;
535
536 /* Sorry... */
537 ret = -EPERM;
538 if (!ns_capable(net->user_ns, CAP_NET_RAW))
539 goto out;
540
541 ret = -EINVAL;
542 if (optlen < 0)
543 goto out;
544
545 /* Bind this socket to a particular device like "eth0",
546 * as specified in the passed interface name. If the
547 * name is "" or the option length is zero the socket
548 * is not bound.
549 */
550 if (optlen > IFNAMSIZ - 1)
551 optlen = IFNAMSIZ - 1;
552 memset(devname, 0, sizeof(devname));
553
554 ret = -EFAULT;
555 if (copy_from_user(devname, optval, optlen))
556 goto out;
557
558 index = 0;
559 if (devname[0] != '\0') {
560 struct net_device *dev;
561
562 rcu_read_lock();
563 dev = dev_get_by_name_rcu(net, devname);
564 if (dev)
565 index = dev->ifindex;
566 rcu_read_unlock();
567 ret = -ENODEV;
568 if (!dev)
569 goto out;
570 }
571
572 lock_sock(sk);
573 sk->sk_bound_dev_if = index;
574 sk_dst_reset(sk);
575 release_sock(sk);
576
577 ret = 0;
578
579 out:
580 #endif
581
582 return ret;
583 }
584
585 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
586 int __user *optlen, int len)
587 {
588 int ret = -ENOPROTOOPT;
589 #ifdef CONFIG_NETDEVICES
590 struct net *net = sock_net(sk);
591 char devname[IFNAMSIZ];
592
593 if (sk->sk_bound_dev_if == 0) {
594 len = 0;
595 goto zero;
596 }
597
598 ret = -EINVAL;
599 if (len < IFNAMSIZ)
600 goto out;
601
602 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
603 if (ret)
604 goto out;
605
606 len = strlen(devname) + 1;
607
608 ret = -EFAULT;
609 if (copy_to_user(optval, devname, len))
610 goto out;
611
612 zero:
613 ret = -EFAULT;
614 if (put_user(len, optlen))
615 goto out;
616
617 ret = 0;
618
619 out:
620 #endif
621
622 return ret;
623 }
624
625 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
626 {
627 if (valbool)
628 sock_set_flag(sk, bit);
629 else
630 sock_reset_flag(sk, bit);
631 }
632
633 bool sk_mc_loop(struct sock *sk)
634 {
635 if (dev_recursion_level())
636 return false;
637 if (!sk)
638 return true;
639 switch (sk->sk_family) {
640 case AF_INET:
641 return inet_sk(sk)->mc_loop;
642 #if IS_ENABLED(CONFIG_IPV6)
643 case AF_INET6:
644 return inet6_sk(sk)->mc_loop;
645 #endif
646 }
647 WARN_ON(1);
648 return true;
649 }
650 EXPORT_SYMBOL(sk_mc_loop);
651
652 /*
653 * This is meant for all protocols to use and covers goings on
654 * at the socket level. Everything here is generic.
655 */
656
657 int sock_setsockopt(struct socket *sock, int level, int optname,
658 char __user *optval, unsigned int optlen)
659 {
660 struct sock *sk = sock->sk;
661 int val;
662 int valbool;
663 struct linger ling;
664 int ret = 0;
665
666 /*
667 * Options without arguments
668 */
669
670 if (optname == SO_BINDTODEVICE)
671 return sock_setbindtodevice(sk, optval, optlen);
672
673 if (optlen < sizeof(int))
674 return -EINVAL;
675
676 if (get_user(val, (int __user *)optval))
677 return -EFAULT;
678
679 valbool = val ? 1 : 0;
680
681 lock_sock(sk);
682
683 switch (optname) {
684 case SO_DEBUG:
685 if (val && !capable(CAP_NET_ADMIN))
686 ret = -EACCES;
687 else
688 sock_valbool_flag(sk, SOCK_DBG, valbool);
689 break;
690 case SO_REUSEADDR:
691 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
692 break;
693 case SO_REUSEPORT:
694 sk->sk_reuseport = valbool;
695 break;
696 case SO_TYPE:
697 case SO_PROTOCOL:
698 case SO_DOMAIN:
699 case SO_ERROR:
700 ret = -ENOPROTOOPT;
701 break;
702 case SO_DONTROUTE:
703 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
704 break;
705 case SO_BROADCAST:
706 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
707 break;
708 case SO_SNDBUF:
709 /* Don't error on this BSD doesn't and if you think
710 * about it this is right. Otherwise apps have to
711 * play 'guess the biggest size' games. RCVBUF/SNDBUF
712 * are treated in BSD as hints
713 */
714 val = min_t(u32, val, sysctl_wmem_max);
715 set_sndbuf:
716 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
717 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
718 /* Wake up sending tasks if we upped the value. */
719 sk->sk_write_space(sk);
720 break;
721
722 case SO_SNDBUFFORCE:
723 if (!capable(CAP_NET_ADMIN)) {
724 ret = -EPERM;
725 break;
726 }
727 goto set_sndbuf;
728
729 case SO_RCVBUF:
730 /* Don't error on this BSD doesn't and if you think
731 * about it this is right. Otherwise apps have to
732 * play 'guess the biggest size' games. RCVBUF/SNDBUF
733 * are treated in BSD as hints
734 */
735 val = min_t(u32, val, sysctl_rmem_max);
736 set_rcvbuf:
737 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
738 /*
739 * We double it on the way in to account for
740 * "struct sk_buff" etc. overhead. Applications
741 * assume that the SO_RCVBUF setting they make will
742 * allow that much actual data to be received on that
743 * socket.
744 *
745 * Applications are unaware that "struct sk_buff" and
746 * other overheads allocate from the receive buffer
747 * during socket buffer allocation.
748 *
749 * And after considering the possible alternatives,
750 * returning the value we actually used in getsockopt
751 * is the most desirable behavior.
752 */
753 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
754 break;
755
756 case SO_RCVBUFFORCE:
757 if (!capable(CAP_NET_ADMIN)) {
758 ret = -EPERM;
759 break;
760 }
761 goto set_rcvbuf;
762
763 case SO_KEEPALIVE:
764 #ifdef CONFIG_INET
765 if (sk->sk_protocol == IPPROTO_TCP &&
766 sk->sk_type == SOCK_STREAM)
767 tcp_set_keepalive(sk, valbool);
768 #endif
769 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
770 break;
771
772 case SO_OOBINLINE:
773 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
774 break;
775
776 case SO_NO_CHECK:
777 sk->sk_no_check_tx = valbool;
778 break;
779
780 case SO_PRIORITY:
781 if ((val >= 0 && val <= 6) ||
782 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
783 sk->sk_priority = val;
784 else
785 ret = -EPERM;
786 break;
787
788 case SO_LINGER:
789 if (optlen < sizeof(ling)) {
790 ret = -EINVAL; /* 1003.1g */
791 break;
792 }
793 if (copy_from_user(&ling, optval, sizeof(ling))) {
794 ret = -EFAULT;
795 break;
796 }
797 if (!ling.l_onoff)
798 sock_reset_flag(sk, SOCK_LINGER);
799 else {
800 #if (BITS_PER_LONG == 32)
801 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
802 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
803 else
804 #endif
805 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
806 sock_set_flag(sk, SOCK_LINGER);
807 }
808 break;
809
810 case SO_BSDCOMPAT:
811 sock_warn_obsolete_bsdism("setsockopt");
812 break;
813
814 case SO_PASSCRED:
815 if (valbool)
816 set_bit(SOCK_PASSCRED, &sock->flags);
817 else
818 clear_bit(SOCK_PASSCRED, &sock->flags);
819 break;
820
821 case SO_TIMESTAMP:
822 case SO_TIMESTAMPNS:
823 if (valbool) {
824 if (optname == SO_TIMESTAMP)
825 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
826 else
827 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
828 sock_set_flag(sk, SOCK_RCVTSTAMP);
829 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
830 } else {
831 sock_reset_flag(sk, SOCK_RCVTSTAMP);
832 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
833 }
834 break;
835
836 case SO_TIMESTAMPING:
837 if (val & ~SOF_TIMESTAMPING_MASK) {
838 ret = -EINVAL;
839 break;
840 }
841
842 if (val & SOF_TIMESTAMPING_OPT_ID &&
843 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
844 if (sk->sk_protocol == IPPROTO_TCP &&
845 sk->sk_type == SOCK_STREAM) {
846 if ((1 << sk->sk_state) &
847 (TCPF_CLOSE | TCPF_LISTEN)) {
848 ret = -EINVAL;
849 break;
850 }
851 sk->sk_tskey = tcp_sk(sk)->snd_una;
852 } else {
853 sk->sk_tskey = 0;
854 }
855 }
856 sk->sk_tsflags = val;
857 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
858 sock_enable_timestamp(sk,
859 SOCK_TIMESTAMPING_RX_SOFTWARE);
860 else
861 sock_disable_timestamp(sk,
862 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
863 break;
864
865 case SO_RCVLOWAT:
866 if (val < 0)
867 val = INT_MAX;
868 sk->sk_rcvlowat = val ? : 1;
869 break;
870
871 case SO_RCVTIMEO:
872 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
873 break;
874
875 case SO_SNDTIMEO:
876 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
877 break;
878
879 case SO_ATTACH_FILTER:
880 ret = -EINVAL;
881 if (optlen == sizeof(struct sock_fprog)) {
882 struct sock_fprog fprog;
883
884 ret = -EFAULT;
885 if (copy_from_user(&fprog, optval, sizeof(fprog)))
886 break;
887
888 ret = sk_attach_filter(&fprog, sk);
889 }
890 break;
891
892 case SO_ATTACH_BPF:
893 ret = -EINVAL;
894 if (optlen == sizeof(u32)) {
895 u32 ufd;
896
897 ret = -EFAULT;
898 if (copy_from_user(&ufd, optval, sizeof(ufd)))
899 break;
900
901 ret = sk_attach_bpf(ufd, sk);
902 }
903 break;
904
905 case SO_ATTACH_REUSEPORT_CBPF:
906 ret = -EINVAL;
907 if (optlen == sizeof(struct sock_fprog)) {
908 struct sock_fprog fprog;
909
910 ret = -EFAULT;
911 if (copy_from_user(&fprog, optval, sizeof(fprog)))
912 break;
913
914 ret = sk_reuseport_attach_filter(&fprog, sk);
915 }
916 break;
917
918 case SO_ATTACH_REUSEPORT_EBPF:
919 ret = -EINVAL;
920 if (optlen == sizeof(u32)) {
921 u32 ufd;
922
923 ret = -EFAULT;
924 if (copy_from_user(&ufd, optval, sizeof(ufd)))
925 break;
926
927 ret = sk_reuseport_attach_bpf(ufd, sk);
928 }
929 break;
930
931 case SO_DETACH_FILTER:
932 ret = sk_detach_filter(sk);
933 break;
934
935 case SO_LOCK_FILTER:
936 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
937 ret = -EPERM;
938 else
939 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
940 break;
941
942 case SO_PASSSEC:
943 if (valbool)
944 set_bit(SOCK_PASSSEC, &sock->flags);
945 else
946 clear_bit(SOCK_PASSSEC, &sock->flags);
947 break;
948 case SO_MARK:
949 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
950 ret = -EPERM;
951 else
952 sk->sk_mark = val;
953 break;
954
955 case SO_RXQ_OVFL:
956 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
957 break;
958
959 case SO_WIFI_STATUS:
960 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
961 break;
962
963 case SO_PEEK_OFF:
964 if (sock->ops->set_peek_off)
965 ret = sock->ops->set_peek_off(sk, val);
966 else
967 ret = -EOPNOTSUPP;
968 break;
969
970 case SO_NOFCS:
971 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
972 break;
973
974 case SO_SELECT_ERR_QUEUE:
975 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
976 break;
977
978 #ifdef CONFIG_NET_RX_BUSY_POLL
979 case SO_BUSY_POLL:
980 /* allow unprivileged users to decrease the value */
981 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
982 ret = -EPERM;
983 else {
984 if (val < 0)
985 ret = -EINVAL;
986 else
987 sk->sk_ll_usec = val;
988 }
989 break;
990 #endif
991
992 case SO_MAX_PACING_RATE:
993 sk->sk_max_pacing_rate = val;
994 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
995 sk->sk_max_pacing_rate);
996 break;
997
998 case SO_INCOMING_CPU:
999 sk->sk_incoming_cpu = val;
1000 break;
1001
1002 case SO_CNX_ADVICE:
1003 if (val == 1)
1004 dst_negative_advice(sk);
1005 break;
1006 default:
1007 ret = -ENOPROTOOPT;
1008 break;
1009 }
1010 release_sock(sk);
1011 return ret;
1012 }
1013 EXPORT_SYMBOL(sock_setsockopt);
1014
1015
1016 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1017 struct ucred *ucred)
1018 {
1019 ucred->pid = pid_vnr(pid);
1020 ucred->uid = ucred->gid = -1;
1021 if (cred) {
1022 struct user_namespace *current_ns = current_user_ns();
1023
1024 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1025 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1026 }
1027 }
1028
1029 int sock_getsockopt(struct socket *sock, int level, int optname,
1030 char __user *optval, int __user *optlen)
1031 {
1032 struct sock *sk = sock->sk;
1033
1034 union {
1035 int val;
1036 struct linger ling;
1037 struct timeval tm;
1038 } v;
1039
1040 int lv = sizeof(int);
1041 int len;
1042
1043 if (get_user(len, optlen))
1044 return -EFAULT;
1045 if (len < 0)
1046 return -EINVAL;
1047
1048 memset(&v, 0, sizeof(v));
1049
1050 switch (optname) {
1051 case SO_DEBUG:
1052 v.val = sock_flag(sk, SOCK_DBG);
1053 break;
1054
1055 case SO_DONTROUTE:
1056 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1057 break;
1058
1059 case SO_BROADCAST:
1060 v.val = sock_flag(sk, SOCK_BROADCAST);
1061 break;
1062
1063 case SO_SNDBUF:
1064 v.val = sk->sk_sndbuf;
1065 break;
1066
1067 case SO_RCVBUF:
1068 v.val = sk->sk_rcvbuf;
1069 break;
1070
1071 case SO_REUSEADDR:
1072 v.val = sk->sk_reuse;
1073 break;
1074
1075 case SO_REUSEPORT:
1076 v.val = sk->sk_reuseport;
1077 break;
1078
1079 case SO_KEEPALIVE:
1080 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1081 break;
1082
1083 case SO_TYPE:
1084 v.val = sk->sk_type;
1085 break;
1086
1087 case SO_PROTOCOL:
1088 v.val = sk->sk_protocol;
1089 break;
1090
1091 case SO_DOMAIN:
1092 v.val = sk->sk_family;
1093 break;
1094
1095 case SO_ERROR:
1096 v.val = -sock_error(sk);
1097 if (v.val == 0)
1098 v.val = xchg(&sk->sk_err_soft, 0);
1099 break;
1100
1101 case SO_OOBINLINE:
1102 v.val = sock_flag(sk, SOCK_URGINLINE);
1103 break;
1104
1105 case SO_NO_CHECK:
1106 v.val = sk->sk_no_check_tx;
1107 break;
1108
1109 case SO_PRIORITY:
1110 v.val = sk->sk_priority;
1111 break;
1112
1113 case SO_LINGER:
1114 lv = sizeof(v.ling);
1115 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1116 v.ling.l_linger = sk->sk_lingertime / HZ;
1117 break;
1118
1119 case SO_BSDCOMPAT:
1120 sock_warn_obsolete_bsdism("getsockopt");
1121 break;
1122
1123 case SO_TIMESTAMP:
1124 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1125 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1126 break;
1127
1128 case SO_TIMESTAMPNS:
1129 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1130 break;
1131
1132 case SO_TIMESTAMPING:
1133 v.val = sk->sk_tsflags;
1134 break;
1135
1136 case SO_RCVTIMEO:
1137 lv = sizeof(struct timeval);
1138 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1139 v.tm.tv_sec = 0;
1140 v.tm.tv_usec = 0;
1141 } else {
1142 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1143 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1144 }
1145 break;
1146
1147 case SO_SNDTIMEO:
1148 lv = sizeof(struct timeval);
1149 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1150 v.tm.tv_sec = 0;
1151 v.tm.tv_usec = 0;
1152 } else {
1153 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1154 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1155 }
1156 break;
1157
1158 case SO_RCVLOWAT:
1159 v.val = sk->sk_rcvlowat;
1160 break;
1161
1162 case SO_SNDLOWAT:
1163 v.val = 1;
1164 break;
1165
1166 case SO_PASSCRED:
1167 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1168 break;
1169
1170 case SO_PEERCRED:
1171 {
1172 struct ucred peercred;
1173 if (len > sizeof(peercred))
1174 len = sizeof(peercred);
1175 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1176 if (copy_to_user(optval, &peercred, len))
1177 return -EFAULT;
1178 goto lenout;
1179 }
1180
1181 case SO_PEERNAME:
1182 {
1183 char address[128];
1184
1185 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1186 return -ENOTCONN;
1187 if (lv < len)
1188 return -EINVAL;
1189 if (copy_to_user(optval, address, len))
1190 return -EFAULT;
1191 goto lenout;
1192 }
1193
1194 /* Dubious BSD thing... Probably nobody even uses it, but
1195 * the UNIX standard wants it for whatever reason... -DaveM
1196 */
1197 case SO_ACCEPTCONN:
1198 v.val = sk->sk_state == TCP_LISTEN;
1199 break;
1200
1201 case SO_PASSSEC:
1202 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1203 break;
1204
1205 case SO_PEERSEC:
1206 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1207
1208 case SO_MARK:
1209 v.val = sk->sk_mark;
1210 break;
1211
1212 case SO_RXQ_OVFL:
1213 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1214 break;
1215
1216 case SO_WIFI_STATUS:
1217 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1218 break;
1219
1220 case SO_PEEK_OFF:
1221 if (!sock->ops->set_peek_off)
1222 return -EOPNOTSUPP;
1223
1224 v.val = sk->sk_peek_off;
1225 break;
1226 case SO_NOFCS:
1227 v.val = sock_flag(sk, SOCK_NOFCS);
1228 break;
1229
1230 case SO_BINDTODEVICE:
1231 return sock_getbindtodevice(sk, optval, optlen, len);
1232
1233 case SO_GET_FILTER:
1234 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1235 if (len < 0)
1236 return len;
1237
1238 goto lenout;
1239
1240 case SO_LOCK_FILTER:
1241 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1242 break;
1243
1244 case SO_BPF_EXTENSIONS:
1245 v.val = bpf_tell_extensions();
1246 break;
1247
1248 case SO_SELECT_ERR_QUEUE:
1249 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1250 break;
1251
1252 #ifdef CONFIG_NET_RX_BUSY_POLL
1253 case SO_BUSY_POLL:
1254 v.val = sk->sk_ll_usec;
1255 break;
1256 #endif
1257
1258 case SO_MAX_PACING_RATE:
1259 v.val = sk->sk_max_pacing_rate;
1260 break;
1261
1262 case SO_INCOMING_CPU:
1263 v.val = sk->sk_incoming_cpu;
1264 break;
1265
1266 default:
1267 /* We implement the SO_SNDLOWAT etc to not be settable
1268 * (1003.1g 7).
1269 */
1270 return -ENOPROTOOPT;
1271 }
1272
1273 if (len > lv)
1274 len = lv;
1275 if (copy_to_user(optval, &v, len))
1276 return -EFAULT;
1277 lenout:
1278 if (put_user(len, optlen))
1279 return -EFAULT;
1280 return 0;
1281 }
1282
1283 /*
1284 * Initialize an sk_lock.
1285 *
1286 * (We also register the sk_lock with the lock validator.)
1287 */
1288 static inline void sock_lock_init(struct sock *sk)
1289 {
1290 sock_lock_init_class_and_name(sk,
1291 af_family_slock_key_strings[sk->sk_family],
1292 af_family_slock_keys + sk->sk_family,
1293 af_family_key_strings[sk->sk_family],
1294 af_family_keys + sk->sk_family);
1295 }
1296
1297 /*
1298 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1299 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1300 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1301 */
1302 static void sock_copy(struct sock *nsk, const struct sock *osk)
1303 {
1304 #ifdef CONFIG_SECURITY_NETWORK
1305 void *sptr = nsk->sk_security;
1306 #endif
1307 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1308
1309 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1310 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1311
1312 #ifdef CONFIG_SECURITY_NETWORK
1313 nsk->sk_security = sptr;
1314 security_sk_clone(osk, nsk);
1315 #endif
1316 }
1317
1318 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1319 int family)
1320 {
1321 struct sock *sk;
1322 struct kmem_cache *slab;
1323
1324 slab = prot->slab;
1325 if (slab != NULL) {
1326 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1327 if (!sk)
1328 return sk;
1329 if (priority & __GFP_ZERO)
1330 sk_prot_clear_nulls(sk, prot->obj_size);
1331 } else
1332 sk = kmalloc(prot->obj_size, priority);
1333
1334 if (sk != NULL) {
1335 kmemcheck_annotate_bitfield(sk, flags);
1336
1337 if (security_sk_alloc(sk, family, priority))
1338 goto out_free;
1339
1340 if (!try_module_get(prot->owner))
1341 goto out_free_sec;
1342 sk_tx_queue_clear(sk);
1343 cgroup_sk_alloc(&sk->sk_cgrp_data);
1344 }
1345
1346 return sk;
1347
1348 out_free_sec:
1349 security_sk_free(sk);
1350 out_free:
1351 if (slab != NULL)
1352 kmem_cache_free(slab, sk);
1353 else
1354 kfree(sk);
1355 return NULL;
1356 }
1357
1358 static void sk_prot_free(struct proto *prot, struct sock *sk)
1359 {
1360 struct kmem_cache *slab;
1361 struct module *owner;
1362
1363 owner = prot->owner;
1364 slab = prot->slab;
1365
1366 cgroup_sk_free(&sk->sk_cgrp_data);
1367 security_sk_free(sk);
1368 if (slab != NULL)
1369 kmem_cache_free(slab, sk);
1370 else
1371 kfree(sk);
1372 module_put(owner);
1373 }
1374
1375 /**
1376 * sk_alloc - All socket objects are allocated here
1377 * @net: the applicable net namespace
1378 * @family: protocol family
1379 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1380 * @prot: struct proto associated with this new sock instance
1381 * @kern: is this to be a kernel socket?
1382 */
1383 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1384 struct proto *prot, int kern)
1385 {
1386 struct sock *sk;
1387
1388 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1389 if (sk) {
1390 sk->sk_family = family;
1391 /*
1392 * See comment in struct sock definition to understand
1393 * why we need sk_prot_creator -acme
1394 */
1395 sk->sk_prot = sk->sk_prot_creator = prot;
1396 sock_lock_init(sk);
1397 sk->sk_net_refcnt = kern ? 0 : 1;
1398 if (likely(sk->sk_net_refcnt))
1399 get_net(net);
1400 sock_net_set(sk, net);
1401 atomic_set(&sk->sk_wmem_alloc, 1);
1402
1403 sock_update_classid(&sk->sk_cgrp_data);
1404 sock_update_netprioidx(&sk->sk_cgrp_data);
1405 }
1406
1407 return sk;
1408 }
1409 EXPORT_SYMBOL(sk_alloc);
1410
1411 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1412 * grace period. This is the case for UDP sockets and TCP listeners.
1413 */
1414 static void __sk_destruct(struct rcu_head *head)
1415 {
1416 struct sock *sk = container_of(head, struct sock, sk_rcu);
1417 struct sk_filter *filter;
1418
1419 if (sk->sk_destruct)
1420 sk->sk_destruct(sk);
1421
1422 filter = rcu_dereference_check(sk->sk_filter,
1423 atomic_read(&sk->sk_wmem_alloc) == 0);
1424 if (filter) {
1425 sk_filter_uncharge(sk, filter);
1426 RCU_INIT_POINTER(sk->sk_filter, NULL);
1427 }
1428 if (rcu_access_pointer(sk->sk_reuseport_cb))
1429 reuseport_detach_sock(sk);
1430
1431 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1432
1433 if (atomic_read(&sk->sk_omem_alloc))
1434 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1435 __func__, atomic_read(&sk->sk_omem_alloc));
1436
1437 if (sk->sk_peer_cred)
1438 put_cred(sk->sk_peer_cred);
1439 put_pid(sk->sk_peer_pid);
1440 if (likely(sk->sk_net_refcnt))
1441 put_net(sock_net(sk));
1442 sk_prot_free(sk->sk_prot_creator, sk);
1443 }
1444
1445 void sk_destruct(struct sock *sk)
1446 {
1447 if (sock_flag(sk, SOCK_RCU_FREE))
1448 call_rcu(&sk->sk_rcu, __sk_destruct);
1449 else
1450 __sk_destruct(&sk->sk_rcu);
1451 }
1452
1453 static void __sk_free(struct sock *sk)
1454 {
1455 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1456 sock_diag_broadcast_destroy(sk);
1457 else
1458 sk_destruct(sk);
1459 }
1460
1461 void sk_free(struct sock *sk)
1462 {
1463 /*
1464 * We subtract one from sk_wmem_alloc and can know if
1465 * some packets are still in some tx queue.
1466 * If not null, sock_wfree() will call __sk_free(sk) later
1467 */
1468 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1469 __sk_free(sk);
1470 }
1471 EXPORT_SYMBOL(sk_free);
1472
1473 /**
1474 * sk_clone_lock - clone a socket, and lock its clone
1475 * @sk: the socket to clone
1476 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1477 *
1478 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1479 */
1480 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1481 {
1482 struct sock *newsk;
1483 bool is_charged = true;
1484
1485 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1486 if (newsk != NULL) {
1487 struct sk_filter *filter;
1488
1489 sock_copy(newsk, sk);
1490
1491 /* SANITY */
1492 if (likely(newsk->sk_net_refcnt))
1493 get_net(sock_net(newsk));
1494 sk_node_init(&newsk->sk_node);
1495 sock_lock_init(newsk);
1496 bh_lock_sock(newsk);
1497 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1498 newsk->sk_backlog.len = 0;
1499
1500 atomic_set(&newsk->sk_rmem_alloc, 0);
1501 /*
1502 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1503 */
1504 atomic_set(&newsk->sk_wmem_alloc, 1);
1505 atomic_set(&newsk->sk_omem_alloc, 0);
1506 skb_queue_head_init(&newsk->sk_receive_queue);
1507 skb_queue_head_init(&newsk->sk_write_queue);
1508
1509 rwlock_init(&newsk->sk_callback_lock);
1510 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1511 af_callback_keys + newsk->sk_family,
1512 af_family_clock_key_strings[newsk->sk_family]);
1513
1514 newsk->sk_dst_cache = NULL;
1515 newsk->sk_wmem_queued = 0;
1516 newsk->sk_forward_alloc = 0;
1517 atomic_set(&newsk->sk_drops, 0);
1518 newsk->sk_send_head = NULL;
1519 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1520
1521 sock_reset_flag(newsk, SOCK_DONE);
1522 skb_queue_head_init(&newsk->sk_error_queue);
1523
1524 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1525 if (filter != NULL)
1526 /* though it's an empty new sock, the charging may fail
1527 * if sysctl_optmem_max was changed between creation of
1528 * original socket and cloning
1529 */
1530 is_charged = sk_filter_charge(newsk, filter);
1531
1532 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1533 /* It is still raw copy of parent, so invalidate
1534 * destructor and make plain sk_free() */
1535 newsk->sk_destruct = NULL;
1536 bh_unlock_sock(newsk);
1537 sk_free(newsk);
1538 newsk = NULL;
1539 goto out;
1540 }
1541 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1542
1543 newsk->sk_err = 0;
1544 newsk->sk_priority = 0;
1545 newsk->sk_incoming_cpu = raw_smp_processor_id();
1546 atomic64_set(&newsk->sk_cookie, 0);
1547 /*
1548 * Before updating sk_refcnt, we must commit prior changes to memory
1549 * (Documentation/RCU/rculist_nulls.txt for details)
1550 */
1551 smp_wmb();
1552 atomic_set(&newsk->sk_refcnt, 2);
1553
1554 /*
1555 * Increment the counter in the same struct proto as the master
1556 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1557 * is the same as sk->sk_prot->socks, as this field was copied
1558 * with memcpy).
1559 *
1560 * This _changes_ the previous behaviour, where
1561 * tcp_create_openreq_child always was incrementing the
1562 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1563 * to be taken into account in all callers. -acme
1564 */
1565 sk_refcnt_debug_inc(newsk);
1566 sk_set_socket(newsk, NULL);
1567 newsk->sk_wq = NULL;
1568
1569 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1570 sock_update_memcg(newsk);
1571
1572 if (newsk->sk_prot->sockets_allocated)
1573 sk_sockets_allocated_inc(newsk);
1574
1575 if (sock_needs_netstamp(sk) &&
1576 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1577 net_enable_timestamp();
1578 }
1579 out:
1580 return newsk;
1581 }
1582 EXPORT_SYMBOL_GPL(sk_clone_lock);
1583
1584 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1585 {
1586 u32 max_segs = 1;
1587
1588 sk_dst_set(sk, dst);
1589 sk->sk_route_caps = dst->dev->features;
1590 if (sk->sk_route_caps & NETIF_F_GSO)
1591 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1592 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1593 if (sk_can_gso(sk)) {
1594 if (dst->header_len) {
1595 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1596 } else {
1597 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1598 sk->sk_gso_max_size = dst->dev->gso_max_size;
1599 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1600 }
1601 }
1602 sk->sk_gso_max_segs = max_segs;
1603 }
1604 EXPORT_SYMBOL_GPL(sk_setup_caps);
1605
1606 /*
1607 * Simple resource managers for sockets.
1608 */
1609
1610
1611 /*
1612 * Write buffer destructor automatically called from kfree_skb.
1613 */
1614 void sock_wfree(struct sk_buff *skb)
1615 {
1616 struct sock *sk = skb->sk;
1617 unsigned int len = skb->truesize;
1618
1619 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1620 /*
1621 * Keep a reference on sk_wmem_alloc, this will be released
1622 * after sk_write_space() call
1623 */
1624 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1625 sk->sk_write_space(sk);
1626 len = 1;
1627 }
1628 /*
1629 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1630 * could not do because of in-flight packets
1631 */
1632 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1633 __sk_free(sk);
1634 }
1635 EXPORT_SYMBOL(sock_wfree);
1636
1637 /* This variant of sock_wfree() is used by TCP,
1638 * since it sets SOCK_USE_WRITE_QUEUE.
1639 */
1640 void __sock_wfree(struct sk_buff *skb)
1641 {
1642 struct sock *sk = skb->sk;
1643
1644 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1645 __sk_free(sk);
1646 }
1647
1648 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1649 {
1650 skb_orphan(skb);
1651 skb->sk = sk;
1652 #ifdef CONFIG_INET
1653 if (unlikely(!sk_fullsock(sk))) {
1654 skb->destructor = sock_edemux;
1655 sock_hold(sk);
1656 return;
1657 }
1658 #endif
1659 skb->destructor = sock_wfree;
1660 skb_set_hash_from_sk(skb, sk);
1661 /*
1662 * We used to take a refcount on sk, but following operation
1663 * is enough to guarantee sk_free() wont free this sock until
1664 * all in-flight packets are completed
1665 */
1666 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1667 }
1668 EXPORT_SYMBOL(skb_set_owner_w);
1669
1670 /* This helper is used by netem, as it can hold packets in its
1671 * delay queue. We want to allow the owner socket to send more
1672 * packets, as if they were already TX completed by a typical driver.
1673 * But we also want to keep skb->sk set because some packet schedulers
1674 * rely on it (sch_fq for example). So we set skb->truesize to a small
1675 * amount (1) and decrease sk_wmem_alloc accordingly.
1676 */
1677 void skb_orphan_partial(struct sk_buff *skb)
1678 {
1679 /* If this skb is a TCP pure ACK or already went here,
1680 * we have nothing to do. 2 is already a very small truesize.
1681 */
1682 if (skb->truesize <= 2)
1683 return;
1684
1685 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1686 * so we do not completely orphan skb, but transfert all
1687 * accounted bytes but one, to avoid unexpected reorders.
1688 */
1689 if (skb->destructor == sock_wfree
1690 #ifdef CONFIG_INET
1691 || skb->destructor == tcp_wfree
1692 #endif
1693 ) {
1694 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1695 skb->truesize = 1;
1696 } else {
1697 skb_orphan(skb);
1698 }
1699 }
1700 EXPORT_SYMBOL(skb_orphan_partial);
1701
1702 /*
1703 * Read buffer destructor automatically called from kfree_skb.
1704 */
1705 void sock_rfree(struct sk_buff *skb)
1706 {
1707 struct sock *sk = skb->sk;
1708 unsigned int len = skb->truesize;
1709
1710 atomic_sub(len, &sk->sk_rmem_alloc);
1711 sk_mem_uncharge(sk, len);
1712 }
1713 EXPORT_SYMBOL(sock_rfree);
1714
1715 /*
1716 * Buffer destructor for skbs that are not used directly in read or write
1717 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1718 */
1719 void sock_efree(struct sk_buff *skb)
1720 {
1721 sock_put(skb->sk);
1722 }
1723 EXPORT_SYMBOL(sock_efree);
1724
1725 kuid_t sock_i_uid(struct sock *sk)
1726 {
1727 kuid_t uid;
1728
1729 read_lock_bh(&sk->sk_callback_lock);
1730 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1731 read_unlock_bh(&sk->sk_callback_lock);
1732 return uid;
1733 }
1734 EXPORT_SYMBOL(sock_i_uid);
1735
1736 unsigned long sock_i_ino(struct sock *sk)
1737 {
1738 unsigned long ino;
1739
1740 read_lock_bh(&sk->sk_callback_lock);
1741 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1742 read_unlock_bh(&sk->sk_callback_lock);
1743 return ino;
1744 }
1745 EXPORT_SYMBOL(sock_i_ino);
1746
1747 /*
1748 * Allocate a skb from the socket's send buffer.
1749 */
1750 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1751 gfp_t priority)
1752 {
1753 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1754 struct sk_buff *skb = alloc_skb(size, priority);
1755 if (skb) {
1756 skb_set_owner_w(skb, sk);
1757 return skb;
1758 }
1759 }
1760 return NULL;
1761 }
1762 EXPORT_SYMBOL(sock_wmalloc);
1763
1764 /*
1765 * Allocate a memory block from the socket's option memory buffer.
1766 */
1767 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1768 {
1769 if ((unsigned int)size <= sysctl_optmem_max &&
1770 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1771 void *mem;
1772 /* First do the add, to avoid the race if kmalloc
1773 * might sleep.
1774 */
1775 atomic_add(size, &sk->sk_omem_alloc);
1776 mem = kmalloc(size, priority);
1777 if (mem)
1778 return mem;
1779 atomic_sub(size, &sk->sk_omem_alloc);
1780 }
1781 return NULL;
1782 }
1783 EXPORT_SYMBOL(sock_kmalloc);
1784
1785 /* Free an option memory block. Note, we actually want the inline
1786 * here as this allows gcc to detect the nullify and fold away the
1787 * condition entirely.
1788 */
1789 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1790 const bool nullify)
1791 {
1792 if (WARN_ON_ONCE(!mem))
1793 return;
1794 if (nullify)
1795 kzfree(mem);
1796 else
1797 kfree(mem);
1798 atomic_sub(size, &sk->sk_omem_alloc);
1799 }
1800
1801 void sock_kfree_s(struct sock *sk, void *mem, int size)
1802 {
1803 __sock_kfree_s(sk, mem, size, false);
1804 }
1805 EXPORT_SYMBOL(sock_kfree_s);
1806
1807 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1808 {
1809 __sock_kfree_s(sk, mem, size, true);
1810 }
1811 EXPORT_SYMBOL(sock_kzfree_s);
1812
1813 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1814 I think, these locks should be removed for datagram sockets.
1815 */
1816 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1817 {
1818 DEFINE_WAIT(wait);
1819
1820 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1821 for (;;) {
1822 if (!timeo)
1823 break;
1824 if (signal_pending(current))
1825 break;
1826 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1827 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1828 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1829 break;
1830 if (sk->sk_shutdown & SEND_SHUTDOWN)
1831 break;
1832 if (sk->sk_err)
1833 break;
1834 timeo = schedule_timeout(timeo);
1835 }
1836 finish_wait(sk_sleep(sk), &wait);
1837 return timeo;
1838 }
1839
1840
1841 /*
1842 * Generic send/receive buffer handlers
1843 */
1844
1845 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1846 unsigned long data_len, int noblock,
1847 int *errcode, int max_page_order)
1848 {
1849 struct sk_buff *skb;
1850 long timeo;
1851 int err;
1852
1853 timeo = sock_sndtimeo(sk, noblock);
1854 for (;;) {
1855 err = sock_error(sk);
1856 if (err != 0)
1857 goto failure;
1858
1859 err = -EPIPE;
1860 if (sk->sk_shutdown & SEND_SHUTDOWN)
1861 goto failure;
1862
1863 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1864 break;
1865
1866 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1867 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1868 err = -EAGAIN;
1869 if (!timeo)
1870 goto failure;
1871 if (signal_pending(current))
1872 goto interrupted;
1873 timeo = sock_wait_for_wmem(sk, timeo);
1874 }
1875 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1876 errcode, sk->sk_allocation);
1877 if (skb)
1878 skb_set_owner_w(skb, sk);
1879 return skb;
1880
1881 interrupted:
1882 err = sock_intr_errno(timeo);
1883 failure:
1884 *errcode = err;
1885 return NULL;
1886 }
1887 EXPORT_SYMBOL(sock_alloc_send_pskb);
1888
1889 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1890 int noblock, int *errcode)
1891 {
1892 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1893 }
1894 EXPORT_SYMBOL(sock_alloc_send_skb);
1895
1896 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1897 struct sockcm_cookie *sockc)
1898 {
1899 u32 tsflags;
1900
1901 switch (cmsg->cmsg_type) {
1902 case SO_MARK:
1903 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1904 return -EPERM;
1905 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1906 return -EINVAL;
1907 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1908 break;
1909 case SO_TIMESTAMPING:
1910 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1911 return -EINVAL;
1912
1913 tsflags = *(u32 *)CMSG_DATA(cmsg);
1914 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1915 return -EINVAL;
1916
1917 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1918 sockc->tsflags |= tsflags;
1919 break;
1920 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1921 case SCM_RIGHTS:
1922 case SCM_CREDENTIALS:
1923 break;
1924 default:
1925 return -EINVAL;
1926 }
1927 return 0;
1928 }
1929 EXPORT_SYMBOL(__sock_cmsg_send);
1930
1931 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1932 struct sockcm_cookie *sockc)
1933 {
1934 struct cmsghdr *cmsg;
1935 int ret;
1936
1937 for_each_cmsghdr(cmsg, msg) {
1938 if (!CMSG_OK(msg, cmsg))
1939 return -EINVAL;
1940 if (cmsg->cmsg_level != SOL_SOCKET)
1941 continue;
1942 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1943 if (ret)
1944 return ret;
1945 }
1946 return 0;
1947 }
1948 EXPORT_SYMBOL(sock_cmsg_send);
1949
1950 /* On 32bit arches, an skb frag is limited to 2^15 */
1951 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1952
1953 /**
1954 * skb_page_frag_refill - check that a page_frag contains enough room
1955 * @sz: minimum size of the fragment we want to get
1956 * @pfrag: pointer to page_frag
1957 * @gfp: priority for memory allocation
1958 *
1959 * Note: While this allocator tries to use high order pages, there is
1960 * no guarantee that allocations succeed. Therefore, @sz MUST be
1961 * less or equal than PAGE_SIZE.
1962 */
1963 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1964 {
1965 if (pfrag->page) {
1966 if (page_ref_count(pfrag->page) == 1) {
1967 pfrag->offset = 0;
1968 return true;
1969 }
1970 if (pfrag->offset + sz <= pfrag->size)
1971 return true;
1972 put_page(pfrag->page);
1973 }
1974
1975 pfrag->offset = 0;
1976 if (SKB_FRAG_PAGE_ORDER) {
1977 /* Avoid direct reclaim but allow kswapd to wake */
1978 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1979 __GFP_COMP | __GFP_NOWARN |
1980 __GFP_NORETRY,
1981 SKB_FRAG_PAGE_ORDER);
1982 if (likely(pfrag->page)) {
1983 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1984 return true;
1985 }
1986 }
1987 pfrag->page = alloc_page(gfp);
1988 if (likely(pfrag->page)) {
1989 pfrag->size = PAGE_SIZE;
1990 return true;
1991 }
1992 return false;
1993 }
1994 EXPORT_SYMBOL(skb_page_frag_refill);
1995
1996 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1997 {
1998 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1999 return true;
2000
2001 sk_enter_memory_pressure(sk);
2002 sk_stream_moderate_sndbuf(sk);
2003 return false;
2004 }
2005 EXPORT_SYMBOL(sk_page_frag_refill);
2006
2007 static void __lock_sock(struct sock *sk)
2008 __releases(&sk->sk_lock.slock)
2009 __acquires(&sk->sk_lock.slock)
2010 {
2011 DEFINE_WAIT(wait);
2012
2013 for (;;) {
2014 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2015 TASK_UNINTERRUPTIBLE);
2016 spin_unlock_bh(&sk->sk_lock.slock);
2017 schedule();
2018 spin_lock_bh(&sk->sk_lock.slock);
2019 if (!sock_owned_by_user(sk))
2020 break;
2021 }
2022 finish_wait(&sk->sk_lock.wq, &wait);
2023 }
2024
2025 static void __release_sock(struct sock *sk)
2026 __releases(&sk->sk_lock.slock)
2027 __acquires(&sk->sk_lock.slock)
2028 {
2029 struct sk_buff *skb, *next;
2030
2031 while ((skb = sk->sk_backlog.head) != NULL) {
2032 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2033
2034 spin_unlock_bh(&sk->sk_lock.slock);
2035
2036 do {
2037 next = skb->next;
2038 prefetch(next);
2039 WARN_ON_ONCE(skb_dst_is_noref(skb));
2040 skb->next = NULL;
2041 sk_backlog_rcv(sk, skb);
2042
2043 cond_resched();
2044
2045 skb = next;
2046 } while (skb != NULL);
2047
2048 spin_lock_bh(&sk->sk_lock.slock);
2049 }
2050
2051 /*
2052 * Doing the zeroing here guarantee we can not loop forever
2053 * while a wild producer attempts to flood us.
2054 */
2055 sk->sk_backlog.len = 0;
2056 }
2057
2058 void __sk_flush_backlog(struct sock *sk)
2059 {
2060 spin_lock_bh(&sk->sk_lock.slock);
2061 __release_sock(sk);
2062 spin_unlock_bh(&sk->sk_lock.slock);
2063 }
2064
2065 /**
2066 * sk_wait_data - wait for data to arrive at sk_receive_queue
2067 * @sk: sock to wait on
2068 * @timeo: for how long
2069 * @skb: last skb seen on sk_receive_queue
2070 *
2071 * Now socket state including sk->sk_err is changed only under lock,
2072 * hence we may omit checks after joining wait queue.
2073 * We check receive queue before schedule() only as optimization;
2074 * it is very likely that release_sock() added new data.
2075 */
2076 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2077 {
2078 int rc;
2079 DEFINE_WAIT(wait);
2080
2081 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2082 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2083 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2084 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2085 finish_wait(sk_sleep(sk), &wait);
2086 return rc;
2087 }
2088 EXPORT_SYMBOL(sk_wait_data);
2089
2090 /**
2091 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2092 * @sk: socket
2093 * @size: memory size to allocate
2094 * @kind: allocation type
2095 *
2096 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2097 * rmem allocation. This function assumes that protocols which have
2098 * memory_pressure use sk_wmem_queued as write buffer accounting.
2099 */
2100 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2101 {
2102 struct proto *prot = sk->sk_prot;
2103 int amt = sk_mem_pages(size);
2104 long allocated;
2105
2106 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2107
2108 allocated = sk_memory_allocated_add(sk, amt);
2109
2110 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2111 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2112 goto suppress_allocation;
2113
2114 /* Under limit. */
2115 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2116 sk_leave_memory_pressure(sk);
2117 return 1;
2118 }
2119
2120 /* Under pressure. */
2121 if (allocated > sk_prot_mem_limits(sk, 1))
2122 sk_enter_memory_pressure(sk);
2123
2124 /* Over hard limit. */
2125 if (allocated > sk_prot_mem_limits(sk, 2))
2126 goto suppress_allocation;
2127
2128 /* guarantee minimum buffer size under pressure */
2129 if (kind == SK_MEM_RECV) {
2130 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2131 return 1;
2132
2133 } else { /* SK_MEM_SEND */
2134 if (sk->sk_type == SOCK_STREAM) {
2135 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2136 return 1;
2137 } else if (atomic_read(&sk->sk_wmem_alloc) <
2138 prot->sysctl_wmem[0])
2139 return 1;
2140 }
2141
2142 if (sk_has_memory_pressure(sk)) {
2143 int alloc;
2144
2145 if (!sk_under_memory_pressure(sk))
2146 return 1;
2147 alloc = sk_sockets_allocated_read_positive(sk);
2148 if (sk_prot_mem_limits(sk, 2) > alloc *
2149 sk_mem_pages(sk->sk_wmem_queued +
2150 atomic_read(&sk->sk_rmem_alloc) +
2151 sk->sk_forward_alloc))
2152 return 1;
2153 }
2154
2155 suppress_allocation:
2156
2157 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2158 sk_stream_moderate_sndbuf(sk);
2159
2160 /* Fail only if socket is _under_ its sndbuf.
2161 * In this case we cannot block, so that we have to fail.
2162 */
2163 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2164 return 1;
2165 }
2166
2167 trace_sock_exceed_buf_limit(sk, prot, allocated);
2168
2169 /* Alas. Undo changes. */
2170 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2171
2172 sk_memory_allocated_sub(sk, amt);
2173
2174 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2175 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2176
2177 return 0;
2178 }
2179 EXPORT_SYMBOL(__sk_mem_schedule);
2180
2181 /**
2182 * __sk_mem_reclaim - reclaim memory_allocated
2183 * @sk: socket
2184 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2185 */
2186 void __sk_mem_reclaim(struct sock *sk, int amount)
2187 {
2188 amount >>= SK_MEM_QUANTUM_SHIFT;
2189 sk_memory_allocated_sub(sk, amount);
2190 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2191
2192 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2193 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2194
2195 if (sk_under_memory_pressure(sk) &&
2196 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2197 sk_leave_memory_pressure(sk);
2198 }
2199 EXPORT_SYMBOL(__sk_mem_reclaim);
2200
2201 int sk_set_peek_off(struct sock *sk, int val)
2202 {
2203 if (val < 0)
2204 return -EINVAL;
2205
2206 sk->sk_peek_off = val;
2207 return 0;
2208 }
2209 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2210
2211 /*
2212 * Set of default routines for initialising struct proto_ops when
2213 * the protocol does not support a particular function. In certain
2214 * cases where it makes no sense for a protocol to have a "do nothing"
2215 * function, some default processing is provided.
2216 */
2217
2218 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2219 {
2220 return -EOPNOTSUPP;
2221 }
2222 EXPORT_SYMBOL(sock_no_bind);
2223
2224 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2225 int len, int flags)
2226 {
2227 return -EOPNOTSUPP;
2228 }
2229 EXPORT_SYMBOL(sock_no_connect);
2230
2231 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2232 {
2233 return -EOPNOTSUPP;
2234 }
2235 EXPORT_SYMBOL(sock_no_socketpair);
2236
2237 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2238 {
2239 return -EOPNOTSUPP;
2240 }
2241 EXPORT_SYMBOL(sock_no_accept);
2242
2243 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2244 int *len, int peer)
2245 {
2246 return -EOPNOTSUPP;
2247 }
2248 EXPORT_SYMBOL(sock_no_getname);
2249
2250 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2251 {
2252 return 0;
2253 }
2254 EXPORT_SYMBOL(sock_no_poll);
2255
2256 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2257 {
2258 return -EOPNOTSUPP;
2259 }
2260 EXPORT_SYMBOL(sock_no_ioctl);
2261
2262 int sock_no_listen(struct socket *sock, int backlog)
2263 {
2264 return -EOPNOTSUPP;
2265 }
2266 EXPORT_SYMBOL(sock_no_listen);
2267
2268 int sock_no_shutdown(struct socket *sock, int how)
2269 {
2270 return -EOPNOTSUPP;
2271 }
2272 EXPORT_SYMBOL(sock_no_shutdown);
2273
2274 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2275 char __user *optval, unsigned int optlen)
2276 {
2277 return -EOPNOTSUPP;
2278 }
2279 EXPORT_SYMBOL(sock_no_setsockopt);
2280
2281 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2282 char __user *optval, int __user *optlen)
2283 {
2284 return -EOPNOTSUPP;
2285 }
2286 EXPORT_SYMBOL(sock_no_getsockopt);
2287
2288 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2289 {
2290 return -EOPNOTSUPP;
2291 }
2292 EXPORT_SYMBOL(sock_no_sendmsg);
2293
2294 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2295 int flags)
2296 {
2297 return -EOPNOTSUPP;
2298 }
2299 EXPORT_SYMBOL(sock_no_recvmsg);
2300
2301 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2302 {
2303 /* Mirror missing mmap method error code */
2304 return -ENODEV;
2305 }
2306 EXPORT_SYMBOL(sock_no_mmap);
2307
2308 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2309 {
2310 ssize_t res;
2311 struct msghdr msg = {.msg_flags = flags};
2312 struct kvec iov;
2313 char *kaddr = kmap(page);
2314 iov.iov_base = kaddr + offset;
2315 iov.iov_len = size;
2316 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2317 kunmap(page);
2318 return res;
2319 }
2320 EXPORT_SYMBOL(sock_no_sendpage);
2321
2322 /*
2323 * Default Socket Callbacks
2324 */
2325
2326 static void sock_def_wakeup(struct sock *sk)
2327 {
2328 struct socket_wq *wq;
2329
2330 rcu_read_lock();
2331 wq = rcu_dereference(sk->sk_wq);
2332 if (skwq_has_sleeper(wq))
2333 wake_up_interruptible_all(&wq->wait);
2334 rcu_read_unlock();
2335 }
2336
2337 static void sock_def_error_report(struct sock *sk)
2338 {
2339 struct socket_wq *wq;
2340
2341 rcu_read_lock();
2342 wq = rcu_dereference(sk->sk_wq);
2343 if (skwq_has_sleeper(wq))
2344 wake_up_interruptible_poll(&wq->wait, POLLERR);
2345 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2346 rcu_read_unlock();
2347 }
2348
2349 static void sock_def_readable(struct sock *sk)
2350 {
2351 struct socket_wq *wq;
2352
2353 rcu_read_lock();
2354 wq = rcu_dereference(sk->sk_wq);
2355 if (skwq_has_sleeper(wq))
2356 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2357 POLLRDNORM | POLLRDBAND);
2358 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2359 rcu_read_unlock();
2360 }
2361
2362 static void sock_def_write_space(struct sock *sk)
2363 {
2364 struct socket_wq *wq;
2365
2366 rcu_read_lock();
2367
2368 /* Do not wake up a writer until he can make "significant"
2369 * progress. --DaveM
2370 */
2371 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2372 wq = rcu_dereference(sk->sk_wq);
2373 if (skwq_has_sleeper(wq))
2374 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2375 POLLWRNORM | POLLWRBAND);
2376
2377 /* Should agree with poll, otherwise some programs break */
2378 if (sock_writeable(sk))
2379 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2380 }
2381
2382 rcu_read_unlock();
2383 }
2384
2385 static void sock_def_destruct(struct sock *sk)
2386 {
2387 }
2388
2389 void sk_send_sigurg(struct sock *sk)
2390 {
2391 if (sk->sk_socket && sk->sk_socket->file)
2392 if (send_sigurg(&sk->sk_socket->file->f_owner))
2393 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2394 }
2395 EXPORT_SYMBOL(sk_send_sigurg);
2396
2397 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2398 unsigned long expires)
2399 {
2400 if (!mod_timer(timer, expires))
2401 sock_hold(sk);
2402 }
2403 EXPORT_SYMBOL(sk_reset_timer);
2404
2405 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2406 {
2407 if (del_timer(timer))
2408 __sock_put(sk);
2409 }
2410 EXPORT_SYMBOL(sk_stop_timer);
2411
2412 void sock_init_data(struct socket *sock, struct sock *sk)
2413 {
2414 skb_queue_head_init(&sk->sk_receive_queue);
2415 skb_queue_head_init(&sk->sk_write_queue);
2416 skb_queue_head_init(&sk->sk_error_queue);
2417
2418 sk->sk_send_head = NULL;
2419
2420 init_timer(&sk->sk_timer);
2421
2422 sk->sk_allocation = GFP_KERNEL;
2423 sk->sk_rcvbuf = sysctl_rmem_default;
2424 sk->sk_sndbuf = sysctl_wmem_default;
2425 sk->sk_state = TCP_CLOSE;
2426 sk_set_socket(sk, sock);
2427
2428 sock_set_flag(sk, SOCK_ZAPPED);
2429
2430 if (sock) {
2431 sk->sk_type = sock->type;
2432 sk->sk_wq = sock->wq;
2433 sock->sk = sk;
2434 } else
2435 sk->sk_wq = NULL;
2436
2437 rwlock_init(&sk->sk_callback_lock);
2438 lockdep_set_class_and_name(&sk->sk_callback_lock,
2439 af_callback_keys + sk->sk_family,
2440 af_family_clock_key_strings[sk->sk_family]);
2441
2442 sk->sk_state_change = sock_def_wakeup;
2443 sk->sk_data_ready = sock_def_readable;
2444 sk->sk_write_space = sock_def_write_space;
2445 sk->sk_error_report = sock_def_error_report;
2446 sk->sk_destruct = sock_def_destruct;
2447
2448 sk->sk_frag.page = NULL;
2449 sk->sk_frag.offset = 0;
2450 sk->sk_peek_off = -1;
2451
2452 sk->sk_peer_pid = NULL;
2453 sk->sk_peer_cred = NULL;
2454 sk->sk_write_pending = 0;
2455 sk->sk_rcvlowat = 1;
2456 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2457 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2458
2459 sk->sk_stamp = ktime_set(-1L, 0);
2460
2461 #ifdef CONFIG_NET_RX_BUSY_POLL
2462 sk->sk_napi_id = 0;
2463 sk->sk_ll_usec = sysctl_net_busy_read;
2464 #endif
2465
2466 sk->sk_max_pacing_rate = ~0U;
2467 sk->sk_pacing_rate = ~0U;
2468 sk->sk_incoming_cpu = -1;
2469 /*
2470 * Before updating sk_refcnt, we must commit prior changes to memory
2471 * (Documentation/RCU/rculist_nulls.txt for details)
2472 */
2473 smp_wmb();
2474 atomic_set(&sk->sk_refcnt, 1);
2475 atomic_set(&sk->sk_drops, 0);
2476 }
2477 EXPORT_SYMBOL(sock_init_data);
2478
2479 void lock_sock_nested(struct sock *sk, int subclass)
2480 {
2481 might_sleep();
2482 spin_lock_bh(&sk->sk_lock.slock);
2483 if (sk->sk_lock.owned)
2484 __lock_sock(sk);
2485 sk->sk_lock.owned = 1;
2486 spin_unlock(&sk->sk_lock.slock);
2487 /*
2488 * The sk_lock has mutex_lock() semantics here:
2489 */
2490 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2491 local_bh_enable();
2492 }
2493 EXPORT_SYMBOL(lock_sock_nested);
2494
2495 void release_sock(struct sock *sk)
2496 {
2497 spin_lock_bh(&sk->sk_lock.slock);
2498 if (sk->sk_backlog.tail)
2499 __release_sock(sk);
2500
2501 /* Warning : release_cb() might need to release sk ownership,
2502 * ie call sock_release_ownership(sk) before us.
2503 */
2504 if (sk->sk_prot->release_cb)
2505 sk->sk_prot->release_cb(sk);
2506
2507 sock_release_ownership(sk);
2508 if (waitqueue_active(&sk->sk_lock.wq))
2509 wake_up(&sk->sk_lock.wq);
2510 spin_unlock_bh(&sk->sk_lock.slock);
2511 }
2512 EXPORT_SYMBOL(release_sock);
2513
2514 /**
2515 * lock_sock_fast - fast version of lock_sock
2516 * @sk: socket
2517 *
2518 * This version should be used for very small section, where process wont block
2519 * return false if fast path is taken
2520 * sk_lock.slock locked, owned = 0, BH disabled
2521 * return true if slow path is taken
2522 * sk_lock.slock unlocked, owned = 1, BH enabled
2523 */
2524 bool lock_sock_fast(struct sock *sk)
2525 {
2526 might_sleep();
2527 spin_lock_bh(&sk->sk_lock.slock);
2528
2529 if (!sk->sk_lock.owned)
2530 /*
2531 * Note : We must disable BH
2532 */
2533 return false;
2534
2535 __lock_sock(sk);
2536 sk->sk_lock.owned = 1;
2537 spin_unlock(&sk->sk_lock.slock);
2538 /*
2539 * The sk_lock has mutex_lock() semantics here:
2540 */
2541 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2542 local_bh_enable();
2543 return true;
2544 }
2545 EXPORT_SYMBOL(lock_sock_fast);
2546
2547 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2548 {
2549 struct timeval tv;
2550 if (!sock_flag(sk, SOCK_TIMESTAMP))
2551 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2552 tv = ktime_to_timeval(sk->sk_stamp);
2553 if (tv.tv_sec == -1)
2554 return -ENOENT;
2555 if (tv.tv_sec == 0) {
2556 sk->sk_stamp = ktime_get_real();
2557 tv = ktime_to_timeval(sk->sk_stamp);
2558 }
2559 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2560 }
2561 EXPORT_SYMBOL(sock_get_timestamp);
2562
2563 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2564 {
2565 struct timespec ts;
2566 if (!sock_flag(sk, SOCK_TIMESTAMP))
2567 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2568 ts = ktime_to_timespec(sk->sk_stamp);
2569 if (ts.tv_sec == -1)
2570 return -ENOENT;
2571 if (ts.tv_sec == 0) {
2572 sk->sk_stamp = ktime_get_real();
2573 ts = ktime_to_timespec(sk->sk_stamp);
2574 }
2575 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2576 }
2577 EXPORT_SYMBOL(sock_get_timestampns);
2578
2579 void sock_enable_timestamp(struct sock *sk, int flag)
2580 {
2581 if (!sock_flag(sk, flag)) {
2582 unsigned long previous_flags = sk->sk_flags;
2583
2584 sock_set_flag(sk, flag);
2585 /*
2586 * we just set one of the two flags which require net
2587 * time stamping, but time stamping might have been on
2588 * already because of the other one
2589 */
2590 if (sock_needs_netstamp(sk) &&
2591 !(previous_flags & SK_FLAGS_TIMESTAMP))
2592 net_enable_timestamp();
2593 }
2594 }
2595
2596 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2597 int level, int type)
2598 {
2599 struct sock_exterr_skb *serr;
2600 struct sk_buff *skb;
2601 int copied, err;
2602
2603 err = -EAGAIN;
2604 skb = sock_dequeue_err_skb(sk);
2605 if (skb == NULL)
2606 goto out;
2607
2608 copied = skb->len;
2609 if (copied > len) {
2610 msg->msg_flags |= MSG_TRUNC;
2611 copied = len;
2612 }
2613 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2614 if (err)
2615 goto out_free_skb;
2616
2617 sock_recv_timestamp(msg, sk, skb);
2618
2619 serr = SKB_EXT_ERR(skb);
2620 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2621
2622 msg->msg_flags |= MSG_ERRQUEUE;
2623 err = copied;
2624
2625 out_free_skb:
2626 kfree_skb(skb);
2627 out:
2628 return err;
2629 }
2630 EXPORT_SYMBOL(sock_recv_errqueue);
2631
2632 /*
2633 * Get a socket option on an socket.
2634 *
2635 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2636 * asynchronous errors should be reported by getsockopt. We assume
2637 * this means if you specify SO_ERROR (otherwise whats the point of it).
2638 */
2639 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2640 char __user *optval, int __user *optlen)
2641 {
2642 struct sock *sk = sock->sk;
2643
2644 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2645 }
2646 EXPORT_SYMBOL(sock_common_getsockopt);
2647
2648 #ifdef CONFIG_COMPAT
2649 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2650 char __user *optval, int __user *optlen)
2651 {
2652 struct sock *sk = sock->sk;
2653
2654 if (sk->sk_prot->compat_getsockopt != NULL)
2655 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2656 optval, optlen);
2657 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2658 }
2659 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2660 #endif
2661
2662 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2663 int flags)
2664 {
2665 struct sock *sk = sock->sk;
2666 int addr_len = 0;
2667 int err;
2668
2669 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2670 flags & ~MSG_DONTWAIT, &addr_len);
2671 if (err >= 0)
2672 msg->msg_namelen = addr_len;
2673 return err;
2674 }
2675 EXPORT_SYMBOL(sock_common_recvmsg);
2676
2677 /*
2678 * Set socket options on an inet socket.
2679 */
2680 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2681 char __user *optval, unsigned int optlen)
2682 {
2683 struct sock *sk = sock->sk;
2684
2685 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2686 }
2687 EXPORT_SYMBOL(sock_common_setsockopt);
2688
2689 #ifdef CONFIG_COMPAT
2690 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2691 char __user *optval, unsigned int optlen)
2692 {
2693 struct sock *sk = sock->sk;
2694
2695 if (sk->sk_prot->compat_setsockopt != NULL)
2696 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2697 optval, optlen);
2698 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2699 }
2700 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2701 #endif
2702
2703 void sk_common_release(struct sock *sk)
2704 {
2705 if (sk->sk_prot->destroy)
2706 sk->sk_prot->destroy(sk);
2707
2708 /*
2709 * Observation: when sock_common_release is called, processes have
2710 * no access to socket. But net still has.
2711 * Step one, detach it from networking:
2712 *
2713 * A. Remove from hash tables.
2714 */
2715
2716 sk->sk_prot->unhash(sk);
2717
2718 /*
2719 * In this point socket cannot receive new packets, but it is possible
2720 * that some packets are in flight because some CPU runs receiver and
2721 * did hash table lookup before we unhashed socket. They will achieve
2722 * receive queue and will be purged by socket destructor.
2723 *
2724 * Also we still have packets pending on receive queue and probably,
2725 * our own packets waiting in device queues. sock_destroy will drain
2726 * receive queue, but transmitted packets will delay socket destruction
2727 * until the last reference will be released.
2728 */
2729
2730 sock_orphan(sk);
2731
2732 xfrm_sk_free_policy(sk);
2733
2734 sk_refcnt_debug_release(sk);
2735
2736 if (sk->sk_frag.page) {
2737 put_page(sk->sk_frag.page);
2738 sk->sk_frag.page = NULL;
2739 }
2740
2741 sock_put(sk);
2742 }
2743 EXPORT_SYMBOL(sk_common_release);
2744
2745 #ifdef CONFIG_PROC_FS
2746 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2747 struct prot_inuse {
2748 int val[PROTO_INUSE_NR];
2749 };
2750
2751 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2752
2753 #ifdef CONFIG_NET_NS
2754 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2755 {
2756 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2757 }
2758 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2759
2760 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2761 {
2762 int cpu, idx = prot->inuse_idx;
2763 int res = 0;
2764
2765 for_each_possible_cpu(cpu)
2766 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2767
2768 return res >= 0 ? res : 0;
2769 }
2770 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2771
2772 static int __net_init sock_inuse_init_net(struct net *net)
2773 {
2774 net->core.inuse = alloc_percpu(struct prot_inuse);
2775 return net->core.inuse ? 0 : -ENOMEM;
2776 }
2777
2778 static void __net_exit sock_inuse_exit_net(struct net *net)
2779 {
2780 free_percpu(net->core.inuse);
2781 }
2782
2783 static struct pernet_operations net_inuse_ops = {
2784 .init = sock_inuse_init_net,
2785 .exit = sock_inuse_exit_net,
2786 };
2787
2788 static __init int net_inuse_init(void)
2789 {
2790 if (register_pernet_subsys(&net_inuse_ops))
2791 panic("Cannot initialize net inuse counters");
2792
2793 return 0;
2794 }
2795
2796 core_initcall(net_inuse_init);
2797 #else
2798 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2799
2800 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2801 {
2802 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2803 }
2804 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2805
2806 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2807 {
2808 int cpu, idx = prot->inuse_idx;
2809 int res = 0;
2810
2811 for_each_possible_cpu(cpu)
2812 res += per_cpu(prot_inuse, cpu).val[idx];
2813
2814 return res >= 0 ? res : 0;
2815 }
2816 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2817 #endif
2818
2819 static void assign_proto_idx(struct proto *prot)
2820 {
2821 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2822
2823 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2824 pr_err("PROTO_INUSE_NR exhausted\n");
2825 return;
2826 }
2827
2828 set_bit(prot->inuse_idx, proto_inuse_idx);
2829 }
2830
2831 static void release_proto_idx(struct proto *prot)
2832 {
2833 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2834 clear_bit(prot->inuse_idx, proto_inuse_idx);
2835 }
2836 #else
2837 static inline void assign_proto_idx(struct proto *prot)
2838 {
2839 }
2840
2841 static inline void release_proto_idx(struct proto *prot)
2842 {
2843 }
2844 #endif
2845
2846 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2847 {
2848 if (!rsk_prot)
2849 return;
2850 kfree(rsk_prot->slab_name);
2851 rsk_prot->slab_name = NULL;
2852 kmem_cache_destroy(rsk_prot->slab);
2853 rsk_prot->slab = NULL;
2854 }
2855
2856 static int req_prot_init(const struct proto *prot)
2857 {
2858 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2859
2860 if (!rsk_prot)
2861 return 0;
2862
2863 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2864 prot->name);
2865 if (!rsk_prot->slab_name)
2866 return -ENOMEM;
2867
2868 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2869 rsk_prot->obj_size, 0,
2870 prot->slab_flags, NULL);
2871
2872 if (!rsk_prot->slab) {
2873 pr_crit("%s: Can't create request sock SLAB cache!\n",
2874 prot->name);
2875 return -ENOMEM;
2876 }
2877 return 0;
2878 }
2879
2880 int proto_register(struct proto *prot, int alloc_slab)
2881 {
2882 if (alloc_slab) {
2883 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2884 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2885 NULL);
2886
2887 if (prot->slab == NULL) {
2888 pr_crit("%s: Can't create sock SLAB cache!\n",
2889 prot->name);
2890 goto out;
2891 }
2892
2893 if (req_prot_init(prot))
2894 goto out_free_request_sock_slab;
2895
2896 if (prot->twsk_prot != NULL) {
2897 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2898
2899 if (prot->twsk_prot->twsk_slab_name == NULL)
2900 goto out_free_request_sock_slab;
2901
2902 prot->twsk_prot->twsk_slab =
2903 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2904 prot->twsk_prot->twsk_obj_size,
2905 0,
2906 prot->slab_flags,
2907 NULL);
2908 if (prot->twsk_prot->twsk_slab == NULL)
2909 goto out_free_timewait_sock_slab_name;
2910 }
2911 }
2912
2913 mutex_lock(&proto_list_mutex);
2914 list_add(&prot->node, &proto_list);
2915 assign_proto_idx(prot);
2916 mutex_unlock(&proto_list_mutex);
2917 return 0;
2918
2919 out_free_timewait_sock_slab_name:
2920 kfree(prot->twsk_prot->twsk_slab_name);
2921 out_free_request_sock_slab:
2922 req_prot_cleanup(prot->rsk_prot);
2923
2924 kmem_cache_destroy(prot->slab);
2925 prot->slab = NULL;
2926 out:
2927 return -ENOBUFS;
2928 }
2929 EXPORT_SYMBOL(proto_register);
2930
2931 void proto_unregister(struct proto *prot)
2932 {
2933 mutex_lock(&proto_list_mutex);
2934 release_proto_idx(prot);
2935 list_del(&prot->node);
2936 mutex_unlock(&proto_list_mutex);
2937
2938 kmem_cache_destroy(prot->slab);
2939 prot->slab = NULL;
2940
2941 req_prot_cleanup(prot->rsk_prot);
2942
2943 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2944 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2945 kfree(prot->twsk_prot->twsk_slab_name);
2946 prot->twsk_prot->twsk_slab = NULL;
2947 }
2948 }
2949 EXPORT_SYMBOL(proto_unregister);
2950
2951 #ifdef CONFIG_PROC_FS
2952 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2953 __acquires(proto_list_mutex)
2954 {
2955 mutex_lock(&proto_list_mutex);
2956 return seq_list_start_head(&proto_list, *pos);
2957 }
2958
2959 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2960 {
2961 return seq_list_next(v, &proto_list, pos);
2962 }
2963
2964 static void proto_seq_stop(struct seq_file *seq, void *v)
2965 __releases(proto_list_mutex)
2966 {
2967 mutex_unlock(&proto_list_mutex);
2968 }
2969
2970 static char proto_method_implemented(const void *method)
2971 {
2972 return method == NULL ? 'n' : 'y';
2973 }
2974 static long sock_prot_memory_allocated(struct proto *proto)
2975 {
2976 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2977 }
2978
2979 static char *sock_prot_memory_pressure(struct proto *proto)
2980 {
2981 return proto->memory_pressure != NULL ?
2982 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2983 }
2984
2985 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2986 {
2987
2988 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2989 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2990 proto->name,
2991 proto->obj_size,
2992 sock_prot_inuse_get(seq_file_net(seq), proto),
2993 sock_prot_memory_allocated(proto),
2994 sock_prot_memory_pressure(proto),
2995 proto->max_header,
2996 proto->slab == NULL ? "no" : "yes",
2997 module_name(proto->owner),
2998 proto_method_implemented(proto->close),
2999 proto_method_implemented(proto->connect),
3000 proto_method_implemented(proto->disconnect),
3001 proto_method_implemented(proto->accept),
3002 proto_method_implemented(proto->ioctl),
3003 proto_method_implemented(proto->init),
3004 proto_method_implemented(proto->destroy),
3005 proto_method_implemented(proto->shutdown),
3006 proto_method_implemented(proto->setsockopt),
3007 proto_method_implemented(proto->getsockopt),
3008 proto_method_implemented(proto->sendmsg),
3009 proto_method_implemented(proto->recvmsg),
3010 proto_method_implemented(proto->sendpage),
3011 proto_method_implemented(proto->bind),
3012 proto_method_implemented(proto->backlog_rcv),
3013 proto_method_implemented(proto->hash),
3014 proto_method_implemented(proto->unhash),
3015 proto_method_implemented(proto->get_port),
3016 proto_method_implemented(proto->enter_memory_pressure));
3017 }
3018
3019 static int proto_seq_show(struct seq_file *seq, void *v)
3020 {
3021 if (v == &proto_list)
3022 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3023 "protocol",
3024 "size",
3025 "sockets",
3026 "memory",
3027 "press",
3028 "maxhdr",
3029 "slab",
3030 "module",
3031 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3032 else
3033 proto_seq_printf(seq, list_entry(v, struct proto, node));
3034 return 0;
3035 }
3036
3037 static const struct seq_operations proto_seq_ops = {
3038 .start = proto_seq_start,
3039 .next = proto_seq_next,
3040 .stop = proto_seq_stop,
3041 .show = proto_seq_show,
3042 };
3043
3044 static int proto_seq_open(struct inode *inode, struct file *file)
3045 {
3046 return seq_open_net(inode, file, &proto_seq_ops,
3047 sizeof(struct seq_net_private));
3048 }
3049
3050 static const struct file_operations proto_seq_fops = {
3051 .owner = THIS_MODULE,
3052 .open = proto_seq_open,
3053 .read = seq_read,
3054 .llseek = seq_lseek,
3055 .release = seq_release_net,
3056 };
3057
3058 static __net_init int proto_init_net(struct net *net)
3059 {
3060 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3061 return -ENOMEM;
3062
3063 return 0;
3064 }
3065
3066 static __net_exit void proto_exit_net(struct net *net)
3067 {
3068 remove_proc_entry("protocols", net->proc_net);
3069 }
3070
3071
3072 static __net_initdata struct pernet_operations proto_net_ops = {
3073 .init = proto_init_net,
3074 .exit = proto_exit_net,
3075 };
3076
3077 static int __init proto_init(void)
3078 {
3079 return register_pernet_subsys(&proto_net_ops);
3080 }
3081
3082 subsys_initcall(proto_init);
3083
3084 #endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.101538 seconds and 5 git commands to generate.