]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/core/sock.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md
[mirror_ubuntu-bionic-kernel.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 <linux/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 and separate keys for internal and
201 * userspace sockets.
202 */
203 static struct lock_class_key af_family_keys[AF_MAX];
204 static struct lock_class_key af_family_kern_keys[AF_MAX];
205 static struct lock_class_key af_family_slock_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
207
208 /*
209 * Make lock validator output more readable. (we pre-construct these
210 * strings build-time, so that runtime initialization of socket
211 * locks is fast):
212 */
213
214 #define _sock_locks(x) \
215 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
216 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
217 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
218 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
219 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
220 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
221 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
222 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
223 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
224 x "27" , x "28" , x "AF_CAN" , \
225 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
226 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
227 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
228 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
229 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX"
230
231 static const char *const af_family_key_strings[AF_MAX+1] = {
232 _sock_locks("sk_lock-")
233 };
234 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
235 _sock_locks("slock-")
236 };
237 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
238 _sock_locks("clock-")
239 };
240
241 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
242 _sock_locks("k-sk_lock-")
243 };
244 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
245 _sock_locks("k-slock-")
246 };
247 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
248 _sock_locks("k-clock-")
249 };
250
251 /*
252 * sk_callback_lock locking rules are per-address-family,
253 * so split the lock classes by using a per-AF key:
254 */
255 static struct lock_class_key af_callback_keys[AF_MAX];
256 static struct lock_class_key af_kern_callback_keys[AF_MAX];
257
258 /* Take into consideration the size of the struct sk_buff overhead in the
259 * determination of these values, since that is non-constant across
260 * platforms. This makes socket queueing behavior and performance
261 * not depend upon such differences.
262 */
263 #define _SK_MEM_PACKETS 256
264 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
265 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
266 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
267
268 /* Run time adjustable parameters. */
269 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
270 EXPORT_SYMBOL(sysctl_wmem_max);
271 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
272 EXPORT_SYMBOL(sysctl_rmem_max);
273 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
274 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
275
276 /* Maximal space eaten by iovec or ancillary data plus some space */
277 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
278 EXPORT_SYMBOL(sysctl_optmem_max);
279
280 int sysctl_tstamp_allow_data __read_mostly = 1;
281
282 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
283 EXPORT_SYMBOL_GPL(memalloc_socks);
284
285 /**
286 * sk_set_memalloc - sets %SOCK_MEMALLOC
287 * @sk: socket to set it on
288 *
289 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
290 * It's the responsibility of the admin to adjust min_free_kbytes
291 * to meet the requirements
292 */
293 void sk_set_memalloc(struct sock *sk)
294 {
295 sock_set_flag(sk, SOCK_MEMALLOC);
296 sk->sk_allocation |= __GFP_MEMALLOC;
297 static_key_slow_inc(&memalloc_socks);
298 }
299 EXPORT_SYMBOL_GPL(sk_set_memalloc);
300
301 void sk_clear_memalloc(struct sock *sk)
302 {
303 sock_reset_flag(sk, SOCK_MEMALLOC);
304 sk->sk_allocation &= ~__GFP_MEMALLOC;
305 static_key_slow_dec(&memalloc_socks);
306
307 /*
308 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
309 * progress of swapping. SOCK_MEMALLOC may be cleared while
310 * it has rmem allocations due to the last swapfile being deactivated
311 * but there is a risk that the socket is unusable due to exceeding
312 * the rmem limits. Reclaim the reserves and obey rmem limits again.
313 */
314 sk_mem_reclaim(sk);
315 }
316 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
317
318 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
319 {
320 int ret;
321 unsigned long pflags = current->flags;
322
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
325
326 current->flags |= PF_MEMALLOC;
327 ret = sk->sk_backlog_rcv(sk, skb);
328 tsk_restore_flags(current, pflags, PF_MEMALLOC);
329
330 return ret;
331 }
332 EXPORT_SYMBOL(__sk_backlog_rcv);
333
334 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
335 {
336 struct timeval tv;
337
338 if (optlen < sizeof(tv))
339 return -EINVAL;
340 if (copy_from_user(&tv, optval, sizeof(tv)))
341 return -EFAULT;
342 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
343 return -EDOM;
344
345 if (tv.tv_sec < 0) {
346 static int warned __read_mostly;
347
348 *timeo_p = 0;
349 if (warned < 10 && net_ratelimit()) {
350 warned++;
351 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
352 __func__, current->comm, task_pid_nr(current));
353 }
354 return 0;
355 }
356 *timeo_p = MAX_SCHEDULE_TIMEOUT;
357 if (tv.tv_sec == 0 && tv.tv_usec == 0)
358 return 0;
359 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
360 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
361 return 0;
362 }
363
364 static void sock_warn_obsolete_bsdism(const char *name)
365 {
366 static int warned;
367 static char warncomm[TASK_COMM_LEN];
368 if (strcmp(warncomm, current->comm) && warned < 5) {
369 strcpy(warncomm, current->comm);
370 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
371 warncomm, name);
372 warned++;
373 }
374 }
375
376 static bool sock_needs_netstamp(const struct sock *sk)
377 {
378 switch (sk->sk_family) {
379 case AF_UNSPEC:
380 case AF_UNIX:
381 return false;
382 default:
383 return true;
384 }
385 }
386
387 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
388 {
389 if (sk->sk_flags & flags) {
390 sk->sk_flags &= ~flags;
391 if (sock_needs_netstamp(sk) &&
392 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
393 net_disable_timestamp();
394 }
395 }
396
397
398 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
399 {
400 unsigned long flags;
401 struct sk_buff_head *list = &sk->sk_receive_queue;
402
403 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
404 atomic_inc(&sk->sk_drops);
405 trace_sock_rcvqueue_full(sk, skb);
406 return -ENOMEM;
407 }
408
409 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
410 atomic_inc(&sk->sk_drops);
411 return -ENOBUFS;
412 }
413
414 skb->dev = NULL;
415 skb_set_owner_r(skb, sk);
416
417 /* we escape from rcu protected region, make sure we dont leak
418 * a norefcounted dst
419 */
420 skb_dst_force(skb);
421
422 spin_lock_irqsave(&list->lock, flags);
423 sock_skb_set_dropcount(sk, skb);
424 __skb_queue_tail(list, skb);
425 spin_unlock_irqrestore(&list->lock, flags);
426
427 if (!sock_flag(sk, SOCK_DEAD))
428 sk->sk_data_ready(sk);
429 return 0;
430 }
431 EXPORT_SYMBOL(__sock_queue_rcv_skb);
432
433 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
434 {
435 int err;
436
437 err = sk_filter(sk, skb);
438 if (err)
439 return err;
440
441 return __sock_queue_rcv_skb(sk, skb);
442 }
443 EXPORT_SYMBOL(sock_queue_rcv_skb);
444
445 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
446 const int nested, unsigned int trim_cap, bool refcounted)
447 {
448 int rc = NET_RX_SUCCESS;
449
450 if (sk_filter_trim_cap(sk, skb, trim_cap))
451 goto discard_and_relse;
452
453 skb->dev = NULL;
454
455 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
456 atomic_inc(&sk->sk_drops);
457 goto discard_and_relse;
458 }
459 if (nested)
460 bh_lock_sock_nested(sk);
461 else
462 bh_lock_sock(sk);
463 if (!sock_owned_by_user(sk)) {
464 /*
465 * trylock + unlock semantics:
466 */
467 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
468
469 rc = sk_backlog_rcv(sk, skb);
470
471 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
472 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
473 bh_unlock_sock(sk);
474 atomic_inc(&sk->sk_drops);
475 goto discard_and_relse;
476 }
477
478 bh_unlock_sock(sk);
479 out:
480 if (refcounted)
481 sock_put(sk);
482 return rc;
483 discard_and_relse:
484 kfree_skb(skb);
485 goto out;
486 }
487 EXPORT_SYMBOL(__sk_receive_skb);
488
489 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
490 {
491 struct dst_entry *dst = __sk_dst_get(sk);
492
493 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
494 sk_tx_queue_clear(sk);
495 sk->sk_dst_pending_confirm = 0;
496 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
497 dst_release(dst);
498 return NULL;
499 }
500
501 return dst;
502 }
503 EXPORT_SYMBOL(__sk_dst_check);
504
505 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
506 {
507 struct dst_entry *dst = sk_dst_get(sk);
508
509 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
510 sk_dst_reset(sk);
511 dst_release(dst);
512 return NULL;
513 }
514
515 return dst;
516 }
517 EXPORT_SYMBOL(sk_dst_check);
518
519 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
520 int optlen)
521 {
522 int ret = -ENOPROTOOPT;
523 #ifdef CONFIG_NETDEVICES
524 struct net *net = sock_net(sk);
525 char devname[IFNAMSIZ];
526 int index;
527
528 /* Sorry... */
529 ret = -EPERM;
530 if (!ns_capable(net->user_ns, CAP_NET_RAW))
531 goto out;
532
533 ret = -EINVAL;
534 if (optlen < 0)
535 goto out;
536
537 /* Bind this socket to a particular device like "eth0",
538 * as specified in the passed interface name. If the
539 * name is "" or the option length is zero the socket
540 * is not bound.
541 */
542 if (optlen > IFNAMSIZ - 1)
543 optlen = IFNAMSIZ - 1;
544 memset(devname, 0, sizeof(devname));
545
546 ret = -EFAULT;
547 if (copy_from_user(devname, optval, optlen))
548 goto out;
549
550 index = 0;
551 if (devname[0] != '\0') {
552 struct net_device *dev;
553
554 rcu_read_lock();
555 dev = dev_get_by_name_rcu(net, devname);
556 if (dev)
557 index = dev->ifindex;
558 rcu_read_unlock();
559 ret = -ENODEV;
560 if (!dev)
561 goto out;
562 }
563
564 lock_sock(sk);
565 sk->sk_bound_dev_if = index;
566 sk_dst_reset(sk);
567 release_sock(sk);
568
569 ret = 0;
570
571 out:
572 #endif
573
574 return ret;
575 }
576
577 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
578 int __user *optlen, int len)
579 {
580 int ret = -ENOPROTOOPT;
581 #ifdef CONFIG_NETDEVICES
582 struct net *net = sock_net(sk);
583 char devname[IFNAMSIZ];
584
585 if (sk->sk_bound_dev_if == 0) {
586 len = 0;
587 goto zero;
588 }
589
590 ret = -EINVAL;
591 if (len < IFNAMSIZ)
592 goto out;
593
594 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
595 if (ret)
596 goto out;
597
598 len = strlen(devname) + 1;
599
600 ret = -EFAULT;
601 if (copy_to_user(optval, devname, len))
602 goto out;
603
604 zero:
605 ret = -EFAULT;
606 if (put_user(len, optlen))
607 goto out;
608
609 ret = 0;
610
611 out:
612 #endif
613
614 return ret;
615 }
616
617 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
618 {
619 if (valbool)
620 sock_set_flag(sk, bit);
621 else
622 sock_reset_flag(sk, bit);
623 }
624
625 bool sk_mc_loop(struct sock *sk)
626 {
627 if (dev_recursion_level())
628 return false;
629 if (!sk)
630 return true;
631 switch (sk->sk_family) {
632 case AF_INET:
633 return inet_sk(sk)->mc_loop;
634 #if IS_ENABLED(CONFIG_IPV6)
635 case AF_INET6:
636 return inet6_sk(sk)->mc_loop;
637 #endif
638 }
639 WARN_ON(1);
640 return true;
641 }
642 EXPORT_SYMBOL(sk_mc_loop);
643
644 /*
645 * This is meant for all protocols to use and covers goings on
646 * at the socket level. Everything here is generic.
647 */
648
649 int sock_setsockopt(struct socket *sock, int level, int optname,
650 char __user *optval, unsigned int optlen)
651 {
652 struct sock *sk = sock->sk;
653 int val;
654 int valbool;
655 struct linger ling;
656 int ret = 0;
657
658 /*
659 * Options without arguments
660 */
661
662 if (optname == SO_BINDTODEVICE)
663 return sock_setbindtodevice(sk, optval, optlen);
664
665 if (optlen < sizeof(int))
666 return -EINVAL;
667
668 if (get_user(val, (int __user *)optval))
669 return -EFAULT;
670
671 valbool = val ? 1 : 0;
672
673 lock_sock(sk);
674
675 switch (optname) {
676 case SO_DEBUG:
677 if (val && !capable(CAP_NET_ADMIN))
678 ret = -EACCES;
679 else
680 sock_valbool_flag(sk, SOCK_DBG, valbool);
681 break;
682 case SO_REUSEADDR:
683 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
684 break;
685 case SO_REUSEPORT:
686 sk->sk_reuseport = valbool;
687 break;
688 case SO_TYPE:
689 case SO_PROTOCOL:
690 case SO_DOMAIN:
691 case SO_ERROR:
692 ret = -ENOPROTOOPT;
693 break;
694 case SO_DONTROUTE:
695 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
696 break;
697 case SO_BROADCAST:
698 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
699 break;
700 case SO_SNDBUF:
701 /* Don't error on this BSD doesn't and if you think
702 * about it this is right. Otherwise apps have to
703 * play 'guess the biggest size' games. RCVBUF/SNDBUF
704 * are treated in BSD as hints
705 */
706 val = min_t(u32, val, sysctl_wmem_max);
707 set_sndbuf:
708 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
709 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
710 /* Wake up sending tasks if we upped the value. */
711 sk->sk_write_space(sk);
712 break;
713
714 case SO_SNDBUFFORCE:
715 if (!capable(CAP_NET_ADMIN)) {
716 ret = -EPERM;
717 break;
718 }
719 goto set_sndbuf;
720
721 case SO_RCVBUF:
722 /* Don't error on this BSD doesn't and if you think
723 * about it this is right. Otherwise apps have to
724 * play 'guess the biggest size' games. RCVBUF/SNDBUF
725 * are treated in BSD as hints
726 */
727 val = min_t(u32, val, sysctl_rmem_max);
728 set_rcvbuf:
729 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
730 /*
731 * We double it on the way in to account for
732 * "struct sk_buff" etc. overhead. Applications
733 * assume that the SO_RCVBUF setting they make will
734 * allow that much actual data to be received on that
735 * socket.
736 *
737 * Applications are unaware that "struct sk_buff" and
738 * other overheads allocate from the receive buffer
739 * during socket buffer allocation.
740 *
741 * And after considering the possible alternatives,
742 * returning the value we actually used in getsockopt
743 * is the most desirable behavior.
744 */
745 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
746 break;
747
748 case SO_RCVBUFFORCE:
749 if (!capable(CAP_NET_ADMIN)) {
750 ret = -EPERM;
751 break;
752 }
753 goto set_rcvbuf;
754
755 case SO_KEEPALIVE:
756 if (sk->sk_prot->keepalive)
757 sk->sk_prot->keepalive(sk, valbool);
758 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
759 break;
760
761 case SO_OOBINLINE:
762 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
763 break;
764
765 case SO_NO_CHECK:
766 sk->sk_no_check_tx = valbool;
767 break;
768
769 case SO_PRIORITY:
770 if ((val >= 0 && val <= 6) ||
771 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
772 sk->sk_priority = val;
773 else
774 ret = -EPERM;
775 break;
776
777 case SO_LINGER:
778 if (optlen < sizeof(ling)) {
779 ret = -EINVAL; /* 1003.1g */
780 break;
781 }
782 if (copy_from_user(&ling, optval, sizeof(ling))) {
783 ret = -EFAULT;
784 break;
785 }
786 if (!ling.l_onoff)
787 sock_reset_flag(sk, SOCK_LINGER);
788 else {
789 #if (BITS_PER_LONG == 32)
790 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
791 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
792 else
793 #endif
794 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
795 sock_set_flag(sk, SOCK_LINGER);
796 }
797 break;
798
799 case SO_BSDCOMPAT:
800 sock_warn_obsolete_bsdism("setsockopt");
801 break;
802
803 case SO_PASSCRED:
804 if (valbool)
805 set_bit(SOCK_PASSCRED, &sock->flags);
806 else
807 clear_bit(SOCK_PASSCRED, &sock->flags);
808 break;
809
810 case SO_TIMESTAMP:
811 case SO_TIMESTAMPNS:
812 if (valbool) {
813 if (optname == SO_TIMESTAMP)
814 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
815 else
816 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
817 sock_set_flag(sk, SOCK_RCVTSTAMP);
818 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
819 } else {
820 sock_reset_flag(sk, SOCK_RCVTSTAMP);
821 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
822 }
823 break;
824
825 case SO_TIMESTAMPING:
826 if (val & ~SOF_TIMESTAMPING_MASK) {
827 ret = -EINVAL;
828 break;
829 }
830
831 if (val & SOF_TIMESTAMPING_OPT_ID &&
832 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
833 if (sk->sk_protocol == IPPROTO_TCP &&
834 sk->sk_type == SOCK_STREAM) {
835 if ((1 << sk->sk_state) &
836 (TCPF_CLOSE | TCPF_LISTEN)) {
837 ret = -EINVAL;
838 break;
839 }
840 sk->sk_tskey = tcp_sk(sk)->snd_una;
841 } else {
842 sk->sk_tskey = 0;
843 }
844 }
845
846 if (val & SOF_TIMESTAMPING_OPT_STATS &&
847 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
848 ret = -EINVAL;
849 break;
850 }
851
852 sk->sk_tsflags = val;
853 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
854 sock_enable_timestamp(sk,
855 SOCK_TIMESTAMPING_RX_SOFTWARE);
856 else
857 sock_disable_timestamp(sk,
858 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
859 break;
860
861 case SO_RCVLOWAT:
862 if (val < 0)
863 val = INT_MAX;
864 sk->sk_rcvlowat = val ? : 1;
865 break;
866
867 case SO_RCVTIMEO:
868 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
869 break;
870
871 case SO_SNDTIMEO:
872 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
873 break;
874
875 case SO_ATTACH_FILTER:
876 ret = -EINVAL;
877 if (optlen == sizeof(struct sock_fprog)) {
878 struct sock_fprog fprog;
879
880 ret = -EFAULT;
881 if (copy_from_user(&fprog, optval, sizeof(fprog)))
882 break;
883
884 ret = sk_attach_filter(&fprog, sk);
885 }
886 break;
887
888 case SO_ATTACH_BPF:
889 ret = -EINVAL;
890 if (optlen == sizeof(u32)) {
891 u32 ufd;
892
893 ret = -EFAULT;
894 if (copy_from_user(&ufd, optval, sizeof(ufd)))
895 break;
896
897 ret = sk_attach_bpf(ufd, sk);
898 }
899 break;
900
901 case SO_ATTACH_REUSEPORT_CBPF:
902 ret = -EINVAL;
903 if (optlen == sizeof(struct sock_fprog)) {
904 struct sock_fprog fprog;
905
906 ret = -EFAULT;
907 if (copy_from_user(&fprog, optval, sizeof(fprog)))
908 break;
909
910 ret = sk_reuseport_attach_filter(&fprog, sk);
911 }
912 break;
913
914 case SO_ATTACH_REUSEPORT_EBPF:
915 ret = -EINVAL;
916 if (optlen == sizeof(u32)) {
917 u32 ufd;
918
919 ret = -EFAULT;
920 if (copy_from_user(&ufd, optval, sizeof(ufd)))
921 break;
922
923 ret = sk_reuseport_attach_bpf(ufd, sk);
924 }
925 break;
926
927 case SO_DETACH_FILTER:
928 ret = sk_detach_filter(sk);
929 break;
930
931 case SO_LOCK_FILTER:
932 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
933 ret = -EPERM;
934 else
935 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
936 break;
937
938 case SO_PASSSEC:
939 if (valbool)
940 set_bit(SOCK_PASSSEC, &sock->flags);
941 else
942 clear_bit(SOCK_PASSSEC, &sock->flags);
943 break;
944 case SO_MARK:
945 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
946 ret = -EPERM;
947 else
948 sk->sk_mark = val;
949 break;
950
951 case SO_RXQ_OVFL:
952 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
953 break;
954
955 case SO_WIFI_STATUS:
956 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
957 break;
958
959 case SO_PEEK_OFF:
960 if (sock->ops->set_peek_off)
961 ret = sock->ops->set_peek_off(sk, val);
962 else
963 ret = -EOPNOTSUPP;
964 break;
965
966 case SO_NOFCS:
967 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
968 break;
969
970 case SO_SELECT_ERR_QUEUE:
971 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
972 break;
973
974 #ifdef CONFIG_NET_RX_BUSY_POLL
975 case SO_BUSY_POLL:
976 /* allow unprivileged users to decrease the value */
977 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
978 ret = -EPERM;
979 else {
980 if (val < 0)
981 ret = -EINVAL;
982 else
983 sk->sk_ll_usec = val;
984 }
985 break;
986 #endif
987
988 case SO_MAX_PACING_RATE:
989 sk->sk_max_pacing_rate = val;
990 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
991 sk->sk_max_pacing_rate);
992 break;
993
994 case SO_INCOMING_CPU:
995 sk->sk_incoming_cpu = val;
996 break;
997
998 case SO_CNX_ADVICE:
999 if (val == 1)
1000 dst_negative_advice(sk);
1001 break;
1002 default:
1003 ret = -ENOPROTOOPT;
1004 break;
1005 }
1006 release_sock(sk);
1007 return ret;
1008 }
1009 EXPORT_SYMBOL(sock_setsockopt);
1010
1011
1012 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1013 struct ucred *ucred)
1014 {
1015 ucred->pid = pid_vnr(pid);
1016 ucred->uid = ucred->gid = -1;
1017 if (cred) {
1018 struct user_namespace *current_ns = current_user_ns();
1019
1020 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1021 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1022 }
1023 }
1024
1025 int sock_getsockopt(struct socket *sock, int level, int optname,
1026 char __user *optval, int __user *optlen)
1027 {
1028 struct sock *sk = sock->sk;
1029
1030 union {
1031 int val;
1032 struct linger ling;
1033 struct timeval tm;
1034 } v;
1035
1036 int lv = sizeof(int);
1037 int len;
1038
1039 if (get_user(len, optlen))
1040 return -EFAULT;
1041 if (len < 0)
1042 return -EINVAL;
1043
1044 memset(&v, 0, sizeof(v));
1045
1046 switch (optname) {
1047 case SO_DEBUG:
1048 v.val = sock_flag(sk, SOCK_DBG);
1049 break;
1050
1051 case SO_DONTROUTE:
1052 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1053 break;
1054
1055 case SO_BROADCAST:
1056 v.val = sock_flag(sk, SOCK_BROADCAST);
1057 break;
1058
1059 case SO_SNDBUF:
1060 v.val = sk->sk_sndbuf;
1061 break;
1062
1063 case SO_RCVBUF:
1064 v.val = sk->sk_rcvbuf;
1065 break;
1066
1067 case SO_REUSEADDR:
1068 v.val = sk->sk_reuse;
1069 break;
1070
1071 case SO_REUSEPORT:
1072 v.val = sk->sk_reuseport;
1073 break;
1074
1075 case SO_KEEPALIVE:
1076 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1077 break;
1078
1079 case SO_TYPE:
1080 v.val = sk->sk_type;
1081 break;
1082
1083 case SO_PROTOCOL:
1084 v.val = sk->sk_protocol;
1085 break;
1086
1087 case SO_DOMAIN:
1088 v.val = sk->sk_family;
1089 break;
1090
1091 case SO_ERROR:
1092 v.val = -sock_error(sk);
1093 if (v.val == 0)
1094 v.val = xchg(&sk->sk_err_soft, 0);
1095 break;
1096
1097 case SO_OOBINLINE:
1098 v.val = sock_flag(sk, SOCK_URGINLINE);
1099 break;
1100
1101 case SO_NO_CHECK:
1102 v.val = sk->sk_no_check_tx;
1103 break;
1104
1105 case SO_PRIORITY:
1106 v.val = sk->sk_priority;
1107 break;
1108
1109 case SO_LINGER:
1110 lv = sizeof(v.ling);
1111 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1112 v.ling.l_linger = sk->sk_lingertime / HZ;
1113 break;
1114
1115 case SO_BSDCOMPAT:
1116 sock_warn_obsolete_bsdism("getsockopt");
1117 break;
1118
1119 case SO_TIMESTAMP:
1120 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1121 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1122 break;
1123
1124 case SO_TIMESTAMPNS:
1125 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1126 break;
1127
1128 case SO_TIMESTAMPING:
1129 v.val = sk->sk_tsflags;
1130 break;
1131
1132 case SO_RCVTIMEO:
1133 lv = sizeof(struct timeval);
1134 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1135 v.tm.tv_sec = 0;
1136 v.tm.tv_usec = 0;
1137 } else {
1138 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1139 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1140 }
1141 break;
1142
1143 case SO_SNDTIMEO:
1144 lv = sizeof(struct timeval);
1145 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1146 v.tm.tv_sec = 0;
1147 v.tm.tv_usec = 0;
1148 } else {
1149 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1150 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1151 }
1152 break;
1153
1154 case SO_RCVLOWAT:
1155 v.val = sk->sk_rcvlowat;
1156 break;
1157
1158 case SO_SNDLOWAT:
1159 v.val = 1;
1160 break;
1161
1162 case SO_PASSCRED:
1163 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1164 break;
1165
1166 case SO_PEERCRED:
1167 {
1168 struct ucred peercred;
1169 if (len > sizeof(peercred))
1170 len = sizeof(peercred);
1171 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1172 if (copy_to_user(optval, &peercred, len))
1173 return -EFAULT;
1174 goto lenout;
1175 }
1176
1177 case SO_PEERNAME:
1178 {
1179 char address[128];
1180
1181 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1182 return -ENOTCONN;
1183 if (lv < len)
1184 return -EINVAL;
1185 if (copy_to_user(optval, address, len))
1186 return -EFAULT;
1187 goto lenout;
1188 }
1189
1190 /* Dubious BSD thing... Probably nobody even uses it, but
1191 * the UNIX standard wants it for whatever reason... -DaveM
1192 */
1193 case SO_ACCEPTCONN:
1194 v.val = sk->sk_state == TCP_LISTEN;
1195 break;
1196
1197 case SO_PASSSEC:
1198 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1199 break;
1200
1201 case SO_PEERSEC:
1202 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1203
1204 case SO_MARK:
1205 v.val = sk->sk_mark;
1206 break;
1207
1208 case SO_RXQ_OVFL:
1209 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1210 break;
1211
1212 case SO_WIFI_STATUS:
1213 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1214 break;
1215
1216 case SO_PEEK_OFF:
1217 if (!sock->ops->set_peek_off)
1218 return -EOPNOTSUPP;
1219
1220 v.val = sk->sk_peek_off;
1221 break;
1222 case SO_NOFCS:
1223 v.val = sock_flag(sk, SOCK_NOFCS);
1224 break;
1225
1226 case SO_BINDTODEVICE:
1227 return sock_getbindtodevice(sk, optval, optlen, len);
1228
1229 case SO_GET_FILTER:
1230 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1231 if (len < 0)
1232 return len;
1233
1234 goto lenout;
1235
1236 case SO_LOCK_FILTER:
1237 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1238 break;
1239
1240 case SO_BPF_EXTENSIONS:
1241 v.val = bpf_tell_extensions();
1242 break;
1243
1244 case SO_SELECT_ERR_QUEUE:
1245 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1246 break;
1247
1248 #ifdef CONFIG_NET_RX_BUSY_POLL
1249 case SO_BUSY_POLL:
1250 v.val = sk->sk_ll_usec;
1251 break;
1252 #endif
1253
1254 case SO_MAX_PACING_RATE:
1255 v.val = sk->sk_max_pacing_rate;
1256 break;
1257
1258 case SO_INCOMING_CPU:
1259 v.val = sk->sk_incoming_cpu;
1260 break;
1261
1262 default:
1263 /* We implement the SO_SNDLOWAT etc to not be settable
1264 * (1003.1g 7).
1265 */
1266 return -ENOPROTOOPT;
1267 }
1268
1269 if (len > lv)
1270 len = lv;
1271 if (copy_to_user(optval, &v, len))
1272 return -EFAULT;
1273 lenout:
1274 if (put_user(len, optlen))
1275 return -EFAULT;
1276 return 0;
1277 }
1278
1279 /*
1280 * Initialize an sk_lock.
1281 *
1282 * (We also register the sk_lock with the lock validator.)
1283 */
1284 static inline void sock_lock_init(struct sock *sk)
1285 {
1286 if (sk->sk_kern_sock)
1287 sock_lock_init_class_and_name(
1288 sk,
1289 af_family_kern_slock_key_strings[sk->sk_family],
1290 af_family_kern_slock_keys + sk->sk_family,
1291 af_family_kern_key_strings[sk->sk_family],
1292 af_family_kern_keys + sk->sk_family);
1293 else
1294 sock_lock_init_class_and_name(
1295 sk,
1296 af_family_slock_key_strings[sk->sk_family],
1297 af_family_slock_keys + sk->sk_family,
1298 af_family_key_strings[sk->sk_family],
1299 af_family_keys + sk->sk_family);
1300 }
1301
1302 /*
1303 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1304 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1305 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1306 */
1307 static void sock_copy(struct sock *nsk, const struct sock *osk)
1308 {
1309 #ifdef CONFIG_SECURITY_NETWORK
1310 void *sptr = nsk->sk_security;
1311 #endif
1312 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1313
1314 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1315 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1316
1317 #ifdef CONFIG_SECURITY_NETWORK
1318 nsk->sk_security = sptr;
1319 security_sk_clone(osk, nsk);
1320 #endif
1321 }
1322
1323 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1324 int family)
1325 {
1326 struct sock *sk;
1327 struct kmem_cache *slab;
1328
1329 slab = prot->slab;
1330 if (slab != NULL) {
1331 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1332 if (!sk)
1333 return sk;
1334 if (priority & __GFP_ZERO)
1335 sk_prot_clear_nulls(sk, prot->obj_size);
1336 } else
1337 sk = kmalloc(prot->obj_size, priority);
1338
1339 if (sk != NULL) {
1340 kmemcheck_annotate_bitfield(sk, flags);
1341
1342 if (security_sk_alloc(sk, family, priority))
1343 goto out_free;
1344
1345 if (!try_module_get(prot->owner))
1346 goto out_free_sec;
1347 sk_tx_queue_clear(sk);
1348 }
1349
1350 return sk;
1351
1352 out_free_sec:
1353 security_sk_free(sk);
1354 out_free:
1355 if (slab != NULL)
1356 kmem_cache_free(slab, sk);
1357 else
1358 kfree(sk);
1359 return NULL;
1360 }
1361
1362 static void sk_prot_free(struct proto *prot, struct sock *sk)
1363 {
1364 struct kmem_cache *slab;
1365 struct module *owner;
1366
1367 owner = prot->owner;
1368 slab = prot->slab;
1369
1370 cgroup_sk_free(&sk->sk_cgrp_data);
1371 mem_cgroup_sk_free(sk);
1372 security_sk_free(sk);
1373 if (slab != NULL)
1374 kmem_cache_free(slab, sk);
1375 else
1376 kfree(sk);
1377 module_put(owner);
1378 }
1379
1380 /**
1381 * sk_alloc - All socket objects are allocated here
1382 * @net: the applicable net namespace
1383 * @family: protocol family
1384 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1385 * @prot: struct proto associated with this new sock instance
1386 * @kern: is this to be a kernel socket?
1387 */
1388 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1389 struct proto *prot, int kern)
1390 {
1391 struct sock *sk;
1392
1393 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1394 if (sk) {
1395 sk->sk_family = family;
1396 /*
1397 * See comment in struct sock definition to understand
1398 * why we need sk_prot_creator -acme
1399 */
1400 sk->sk_prot = sk->sk_prot_creator = prot;
1401 sk->sk_kern_sock = kern;
1402 sock_lock_init(sk);
1403 sk->sk_net_refcnt = kern ? 0 : 1;
1404 if (likely(sk->sk_net_refcnt))
1405 get_net(net);
1406 sock_net_set(sk, net);
1407 atomic_set(&sk->sk_wmem_alloc, 1);
1408
1409 mem_cgroup_sk_alloc(sk);
1410 cgroup_sk_alloc(&sk->sk_cgrp_data);
1411 sock_update_classid(&sk->sk_cgrp_data);
1412 sock_update_netprioidx(&sk->sk_cgrp_data);
1413 }
1414
1415 return sk;
1416 }
1417 EXPORT_SYMBOL(sk_alloc);
1418
1419 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1420 * grace period. This is the case for UDP sockets and TCP listeners.
1421 */
1422 static void __sk_destruct(struct rcu_head *head)
1423 {
1424 struct sock *sk = container_of(head, struct sock, sk_rcu);
1425 struct sk_filter *filter;
1426
1427 if (sk->sk_destruct)
1428 sk->sk_destruct(sk);
1429
1430 filter = rcu_dereference_check(sk->sk_filter,
1431 atomic_read(&sk->sk_wmem_alloc) == 0);
1432 if (filter) {
1433 sk_filter_uncharge(sk, filter);
1434 RCU_INIT_POINTER(sk->sk_filter, NULL);
1435 }
1436 if (rcu_access_pointer(sk->sk_reuseport_cb))
1437 reuseport_detach_sock(sk);
1438
1439 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1440
1441 if (atomic_read(&sk->sk_omem_alloc))
1442 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1443 __func__, atomic_read(&sk->sk_omem_alloc));
1444
1445 if (sk->sk_peer_cred)
1446 put_cred(sk->sk_peer_cred);
1447 put_pid(sk->sk_peer_pid);
1448 if (likely(sk->sk_net_refcnt))
1449 put_net(sock_net(sk));
1450 sk_prot_free(sk->sk_prot_creator, sk);
1451 }
1452
1453 void sk_destruct(struct sock *sk)
1454 {
1455 if (sock_flag(sk, SOCK_RCU_FREE))
1456 call_rcu(&sk->sk_rcu, __sk_destruct);
1457 else
1458 __sk_destruct(&sk->sk_rcu);
1459 }
1460
1461 static void __sk_free(struct sock *sk)
1462 {
1463 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1464 sock_diag_broadcast_destroy(sk);
1465 else
1466 sk_destruct(sk);
1467 }
1468
1469 void sk_free(struct sock *sk)
1470 {
1471 /*
1472 * We subtract one from sk_wmem_alloc and can know if
1473 * some packets are still in some tx queue.
1474 * If not null, sock_wfree() will call __sk_free(sk) later
1475 */
1476 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1477 __sk_free(sk);
1478 }
1479 EXPORT_SYMBOL(sk_free);
1480
1481 /**
1482 * sk_clone_lock - clone a socket, and lock its clone
1483 * @sk: the socket to clone
1484 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1485 *
1486 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1487 */
1488 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1489 {
1490 struct sock *newsk;
1491 bool is_charged = true;
1492
1493 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1494 if (newsk != NULL) {
1495 struct sk_filter *filter;
1496
1497 sock_copy(newsk, sk);
1498
1499 /* SANITY */
1500 if (likely(newsk->sk_net_refcnt))
1501 get_net(sock_net(newsk));
1502 sk_node_init(&newsk->sk_node);
1503 sock_lock_init(newsk);
1504 bh_lock_sock(newsk);
1505 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1506 newsk->sk_backlog.len = 0;
1507
1508 atomic_set(&newsk->sk_rmem_alloc, 0);
1509 /*
1510 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1511 */
1512 atomic_set(&newsk->sk_wmem_alloc, 1);
1513 atomic_set(&newsk->sk_omem_alloc, 0);
1514 skb_queue_head_init(&newsk->sk_receive_queue);
1515 skb_queue_head_init(&newsk->sk_write_queue);
1516
1517 rwlock_init(&newsk->sk_callback_lock);
1518 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1519 af_callback_keys + newsk->sk_family,
1520 af_family_clock_key_strings[newsk->sk_family]);
1521
1522 newsk->sk_dst_cache = NULL;
1523 newsk->sk_dst_pending_confirm = 0;
1524 newsk->sk_wmem_queued = 0;
1525 newsk->sk_forward_alloc = 0;
1526 atomic_set(&newsk->sk_drops, 0);
1527 newsk->sk_send_head = NULL;
1528 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1529
1530 sock_reset_flag(newsk, SOCK_DONE);
1531 skb_queue_head_init(&newsk->sk_error_queue);
1532
1533 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1534 if (filter != NULL)
1535 /* though it's an empty new sock, the charging may fail
1536 * if sysctl_optmem_max was changed between creation of
1537 * original socket and cloning
1538 */
1539 is_charged = sk_filter_charge(newsk, filter);
1540
1541 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1542 sk_free_unlock_clone(newsk);
1543 newsk = NULL;
1544 goto out;
1545 }
1546 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1547
1548 newsk->sk_err = 0;
1549 newsk->sk_err_soft = 0;
1550 newsk->sk_priority = 0;
1551 newsk->sk_incoming_cpu = raw_smp_processor_id();
1552 atomic64_set(&newsk->sk_cookie, 0);
1553
1554 mem_cgroup_sk_alloc(newsk);
1555 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1556
1557 /*
1558 * Before updating sk_refcnt, we must commit prior changes to memory
1559 * (Documentation/RCU/rculist_nulls.txt for details)
1560 */
1561 smp_wmb();
1562 atomic_set(&newsk->sk_refcnt, 2);
1563
1564 /*
1565 * Increment the counter in the same struct proto as the master
1566 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1567 * is the same as sk->sk_prot->socks, as this field was copied
1568 * with memcpy).
1569 *
1570 * This _changes_ the previous behaviour, where
1571 * tcp_create_openreq_child always was incrementing the
1572 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1573 * to be taken into account in all callers. -acme
1574 */
1575 sk_refcnt_debug_inc(newsk);
1576 sk_set_socket(newsk, NULL);
1577 newsk->sk_wq = NULL;
1578
1579 if (newsk->sk_prot->sockets_allocated)
1580 sk_sockets_allocated_inc(newsk);
1581
1582 if (sock_needs_netstamp(sk) &&
1583 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1584 net_enable_timestamp();
1585 }
1586 out:
1587 return newsk;
1588 }
1589 EXPORT_SYMBOL_GPL(sk_clone_lock);
1590
1591 void sk_free_unlock_clone(struct sock *sk)
1592 {
1593 /* It is still raw copy of parent, so invalidate
1594 * destructor and make plain sk_free() */
1595 sk->sk_destruct = NULL;
1596 bh_unlock_sock(sk);
1597 sk_free(sk);
1598 }
1599 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1600
1601 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1602 {
1603 u32 max_segs = 1;
1604
1605 sk_dst_set(sk, dst);
1606 sk->sk_route_caps = dst->dev->features;
1607 if (sk->sk_route_caps & NETIF_F_GSO)
1608 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1609 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1610 if (sk_can_gso(sk)) {
1611 if (dst->header_len) {
1612 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1613 } else {
1614 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1615 sk->sk_gso_max_size = dst->dev->gso_max_size;
1616 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1617 }
1618 }
1619 sk->sk_gso_max_segs = max_segs;
1620 }
1621 EXPORT_SYMBOL_GPL(sk_setup_caps);
1622
1623 /*
1624 * Simple resource managers for sockets.
1625 */
1626
1627
1628 /*
1629 * Write buffer destructor automatically called from kfree_skb.
1630 */
1631 void sock_wfree(struct sk_buff *skb)
1632 {
1633 struct sock *sk = skb->sk;
1634 unsigned int len = skb->truesize;
1635
1636 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1637 /*
1638 * Keep a reference on sk_wmem_alloc, this will be released
1639 * after sk_write_space() call
1640 */
1641 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1642 sk->sk_write_space(sk);
1643 len = 1;
1644 }
1645 /*
1646 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1647 * could not do because of in-flight packets
1648 */
1649 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1650 __sk_free(sk);
1651 }
1652 EXPORT_SYMBOL(sock_wfree);
1653
1654 /* This variant of sock_wfree() is used by TCP,
1655 * since it sets SOCK_USE_WRITE_QUEUE.
1656 */
1657 void __sock_wfree(struct sk_buff *skb)
1658 {
1659 struct sock *sk = skb->sk;
1660
1661 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1662 __sk_free(sk);
1663 }
1664
1665 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1666 {
1667 skb_orphan(skb);
1668 skb->sk = sk;
1669 #ifdef CONFIG_INET
1670 if (unlikely(!sk_fullsock(sk))) {
1671 skb->destructor = sock_edemux;
1672 sock_hold(sk);
1673 return;
1674 }
1675 #endif
1676 skb->destructor = sock_wfree;
1677 skb_set_hash_from_sk(skb, sk);
1678 /*
1679 * We used to take a refcount on sk, but following operation
1680 * is enough to guarantee sk_free() wont free this sock until
1681 * all in-flight packets are completed
1682 */
1683 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1684 }
1685 EXPORT_SYMBOL(skb_set_owner_w);
1686
1687 /* This helper is used by netem, as it can hold packets in its
1688 * delay queue. We want to allow the owner socket to send more
1689 * packets, as if they were already TX completed by a typical driver.
1690 * But we also want to keep skb->sk set because some packet schedulers
1691 * rely on it (sch_fq for example). So we set skb->truesize to a small
1692 * amount (1) and decrease sk_wmem_alloc accordingly.
1693 */
1694 void skb_orphan_partial(struct sk_buff *skb)
1695 {
1696 /* If this skb is a TCP pure ACK or already went here,
1697 * we have nothing to do. 2 is already a very small truesize.
1698 */
1699 if (skb->truesize <= 2)
1700 return;
1701
1702 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1703 * so we do not completely orphan skb, but transfert all
1704 * accounted bytes but one, to avoid unexpected reorders.
1705 */
1706 if (skb->destructor == sock_wfree
1707 #ifdef CONFIG_INET
1708 || skb->destructor == tcp_wfree
1709 #endif
1710 ) {
1711 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1712 skb->truesize = 1;
1713 } else {
1714 skb_orphan(skb);
1715 }
1716 }
1717 EXPORT_SYMBOL(skb_orphan_partial);
1718
1719 /*
1720 * Read buffer destructor automatically called from kfree_skb.
1721 */
1722 void sock_rfree(struct sk_buff *skb)
1723 {
1724 struct sock *sk = skb->sk;
1725 unsigned int len = skb->truesize;
1726
1727 atomic_sub(len, &sk->sk_rmem_alloc);
1728 sk_mem_uncharge(sk, len);
1729 }
1730 EXPORT_SYMBOL(sock_rfree);
1731
1732 /*
1733 * Buffer destructor for skbs that are not used directly in read or write
1734 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1735 */
1736 void sock_efree(struct sk_buff *skb)
1737 {
1738 sock_put(skb->sk);
1739 }
1740 EXPORT_SYMBOL(sock_efree);
1741
1742 kuid_t sock_i_uid(struct sock *sk)
1743 {
1744 kuid_t uid;
1745
1746 read_lock_bh(&sk->sk_callback_lock);
1747 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1748 read_unlock_bh(&sk->sk_callback_lock);
1749 return uid;
1750 }
1751 EXPORT_SYMBOL(sock_i_uid);
1752
1753 unsigned long sock_i_ino(struct sock *sk)
1754 {
1755 unsigned long ino;
1756
1757 read_lock_bh(&sk->sk_callback_lock);
1758 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1759 read_unlock_bh(&sk->sk_callback_lock);
1760 return ino;
1761 }
1762 EXPORT_SYMBOL(sock_i_ino);
1763
1764 /*
1765 * Allocate a skb from the socket's send buffer.
1766 */
1767 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1768 gfp_t priority)
1769 {
1770 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1771 struct sk_buff *skb = alloc_skb(size, priority);
1772 if (skb) {
1773 skb_set_owner_w(skb, sk);
1774 return skb;
1775 }
1776 }
1777 return NULL;
1778 }
1779 EXPORT_SYMBOL(sock_wmalloc);
1780
1781 /*
1782 * Allocate a memory block from the socket's option memory buffer.
1783 */
1784 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1785 {
1786 if ((unsigned int)size <= sysctl_optmem_max &&
1787 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1788 void *mem;
1789 /* First do the add, to avoid the race if kmalloc
1790 * might sleep.
1791 */
1792 atomic_add(size, &sk->sk_omem_alloc);
1793 mem = kmalloc(size, priority);
1794 if (mem)
1795 return mem;
1796 atomic_sub(size, &sk->sk_omem_alloc);
1797 }
1798 return NULL;
1799 }
1800 EXPORT_SYMBOL(sock_kmalloc);
1801
1802 /* Free an option memory block. Note, we actually want the inline
1803 * here as this allows gcc to detect the nullify and fold away the
1804 * condition entirely.
1805 */
1806 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1807 const bool nullify)
1808 {
1809 if (WARN_ON_ONCE(!mem))
1810 return;
1811 if (nullify)
1812 kzfree(mem);
1813 else
1814 kfree(mem);
1815 atomic_sub(size, &sk->sk_omem_alloc);
1816 }
1817
1818 void sock_kfree_s(struct sock *sk, void *mem, int size)
1819 {
1820 __sock_kfree_s(sk, mem, size, false);
1821 }
1822 EXPORT_SYMBOL(sock_kfree_s);
1823
1824 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1825 {
1826 __sock_kfree_s(sk, mem, size, true);
1827 }
1828 EXPORT_SYMBOL(sock_kzfree_s);
1829
1830 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1831 I think, these locks should be removed for datagram sockets.
1832 */
1833 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1834 {
1835 DEFINE_WAIT(wait);
1836
1837 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1838 for (;;) {
1839 if (!timeo)
1840 break;
1841 if (signal_pending(current))
1842 break;
1843 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1844 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1845 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1846 break;
1847 if (sk->sk_shutdown & SEND_SHUTDOWN)
1848 break;
1849 if (sk->sk_err)
1850 break;
1851 timeo = schedule_timeout(timeo);
1852 }
1853 finish_wait(sk_sleep(sk), &wait);
1854 return timeo;
1855 }
1856
1857
1858 /*
1859 * Generic send/receive buffer handlers
1860 */
1861
1862 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1863 unsigned long data_len, int noblock,
1864 int *errcode, int max_page_order)
1865 {
1866 struct sk_buff *skb;
1867 long timeo;
1868 int err;
1869
1870 timeo = sock_sndtimeo(sk, noblock);
1871 for (;;) {
1872 err = sock_error(sk);
1873 if (err != 0)
1874 goto failure;
1875
1876 err = -EPIPE;
1877 if (sk->sk_shutdown & SEND_SHUTDOWN)
1878 goto failure;
1879
1880 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1881 break;
1882
1883 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1884 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1885 err = -EAGAIN;
1886 if (!timeo)
1887 goto failure;
1888 if (signal_pending(current))
1889 goto interrupted;
1890 timeo = sock_wait_for_wmem(sk, timeo);
1891 }
1892 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1893 errcode, sk->sk_allocation);
1894 if (skb)
1895 skb_set_owner_w(skb, sk);
1896 return skb;
1897
1898 interrupted:
1899 err = sock_intr_errno(timeo);
1900 failure:
1901 *errcode = err;
1902 return NULL;
1903 }
1904 EXPORT_SYMBOL(sock_alloc_send_pskb);
1905
1906 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1907 int noblock, int *errcode)
1908 {
1909 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1910 }
1911 EXPORT_SYMBOL(sock_alloc_send_skb);
1912
1913 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1914 struct sockcm_cookie *sockc)
1915 {
1916 u32 tsflags;
1917
1918 switch (cmsg->cmsg_type) {
1919 case SO_MARK:
1920 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1921 return -EPERM;
1922 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1923 return -EINVAL;
1924 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1925 break;
1926 case SO_TIMESTAMPING:
1927 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1928 return -EINVAL;
1929
1930 tsflags = *(u32 *)CMSG_DATA(cmsg);
1931 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1932 return -EINVAL;
1933
1934 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1935 sockc->tsflags |= tsflags;
1936 break;
1937 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1938 case SCM_RIGHTS:
1939 case SCM_CREDENTIALS:
1940 break;
1941 default:
1942 return -EINVAL;
1943 }
1944 return 0;
1945 }
1946 EXPORT_SYMBOL(__sock_cmsg_send);
1947
1948 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1949 struct sockcm_cookie *sockc)
1950 {
1951 struct cmsghdr *cmsg;
1952 int ret;
1953
1954 for_each_cmsghdr(cmsg, msg) {
1955 if (!CMSG_OK(msg, cmsg))
1956 return -EINVAL;
1957 if (cmsg->cmsg_level != SOL_SOCKET)
1958 continue;
1959 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1960 if (ret)
1961 return ret;
1962 }
1963 return 0;
1964 }
1965 EXPORT_SYMBOL(sock_cmsg_send);
1966
1967 /* On 32bit arches, an skb frag is limited to 2^15 */
1968 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1969
1970 /**
1971 * skb_page_frag_refill - check that a page_frag contains enough room
1972 * @sz: minimum size of the fragment we want to get
1973 * @pfrag: pointer to page_frag
1974 * @gfp: priority for memory allocation
1975 *
1976 * Note: While this allocator tries to use high order pages, there is
1977 * no guarantee that allocations succeed. Therefore, @sz MUST be
1978 * less or equal than PAGE_SIZE.
1979 */
1980 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1981 {
1982 if (pfrag->page) {
1983 if (page_ref_count(pfrag->page) == 1) {
1984 pfrag->offset = 0;
1985 return true;
1986 }
1987 if (pfrag->offset + sz <= pfrag->size)
1988 return true;
1989 put_page(pfrag->page);
1990 }
1991
1992 pfrag->offset = 0;
1993 if (SKB_FRAG_PAGE_ORDER) {
1994 /* Avoid direct reclaim but allow kswapd to wake */
1995 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1996 __GFP_COMP | __GFP_NOWARN |
1997 __GFP_NORETRY,
1998 SKB_FRAG_PAGE_ORDER);
1999 if (likely(pfrag->page)) {
2000 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2001 return true;
2002 }
2003 }
2004 pfrag->page = alloc_page(gfp);
2005 if (likely(pfrag->page)) {
2006 pfrag->size = PAGE_SIZE;
2007 return true;
2008 }
2009 return false;
2010 }
2011 EXPORT_SYMBOL(skb_page_frag_refill);
2012
2013 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2014 {
2015 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2016 return true;
2017
2018 sk_enter_memory_pressure(sk);
2019 sk_stream_moderate_sndbuf(sk);
2020 return false;
2021 }
2022 EXPORT_SYMBOL(sk_page_frag_refill);
2023
2024 static void __lock_sock(struct sock *sk)
2025 __releases(&sk->sk_lock.slock)
2026 __acquires(&sk->sk_lock.slock)
2027 {
2028 DEFINE_WAIT(wait);
2029
2030 for (;;) {
2031 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2032 TASK_UNINTERRUPTIBLE);
2033 spin_unlock_bh(&sk->sk_lock.slock);
2034 schedule();
2035 spin_lock_bh(&sk->sk_lock.slock);
2036 if (!sock_owned_by_user(sk))
2037 break;
2038 }
2039 finish_wait(&sk->sk_lock.wq, &wait);
2040 }
2041
2042 static void __release_sock(struct sock *sk)
2043 __releases(&sk->sk_lock.slock)
2044 __acquires(&sk->sk_lock.slock)
2045 {
2046 struct sk_buff *skb, *next;
2047
2048 while ((skb = sk->sk_backlog.head) != NULL) {
2049 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2050
2051 spin_unlock_bh(&sk->sk_lock.slock);
2052
2053 do {
2054 next = skb->next;
2055 prefetch(next);
2056 WARN_ON_ONCE(skb_dst_is_noref(skb));
2057 skb->next = NULL;
2058 sk_backlog_rcv(sk, skb);
2059
2060 cond_resched();
2061
2062 skb = next;
2063 } while (skb != NULL);
2064
2065 spin_lock_bh(&sk->sk_lock.slock);
2066 }
2067
2068 /*
2069 * Doing the zeroing here guarantee we can not loop forever
2070 * while a wild producer attempts to flood us.
2071 */
2072 sk->sk_backlog.len = 0;
2073 }
2074
2075 void __sk_flush_backlog(struct sock *sk)
2076 {
2077 spin_lock_bh(&sk->sk_lock.slock);
2078 __release_sock(sk);
2079 spin_unlock_bh(&sk->sk_lock.slock);
2080 }
2081
2082 /**
2083 * sk_wait_data - wait for data to arrive at sk_receive_queue
2084 * @sk: sock to wait on
2085 * @timeo: for how long
2086 * @skb: last skb seen on sk_receive_queue
2087 *
2088 * Now socket state including sk->sk_err is changed only under lock,
2089 * hence we may omit checks after joining wait queue.
2090 * We check receive queue before schedule() only as optimization;
2091 * it is very likely that release_sock() added new data.
2092 */
2093 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2094 {
2095 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2096 int rc;
2097
2098 add_wait_queue(sk_sleep(sk), &wait);
2099 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2100 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2101 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2102 remove_wait_queue(sk_sleep(sk), &wait);
2103 return rc;
2104 }
2105 EXPORT_SYMBOL(sk_wait_data);
2106
2107 /**
2108 * __sk_mem_raise_allocated - increase memory_allocated
2109 * @sk: socket
2110 * @size: memory size to allocate
2111 * @amt: pages to allocate
2112 * @kind: allocation type
2113 *
2114 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2115 */
2116 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2117 {
2118 struct proto *prot = sk->sk_prot;
2119 long allocated = sk_memory_allocated_add(sk, amt);
2120
2121 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2122 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2123 goto suppress_allocation;
2124
2125 /* Under limit. */
2126 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2127 sk_leave_memory_pressure(sk);
2128 return 1;
2129 }
2130
2131 /* Under pressure. */
2132 if (allocated > sk_prot_mem_limits(sk, 1))
2133 sk_enter_memory_pressure(sk);
2134
2135 /* Over hard limit. */
2136 if (allocated > sk_prot_mem_limits(sk, 2))
2137 goto suppress_allocation;
2138
2139 /* guarantee minimum buffer size under pressure */
2140 if (kind == SK_MEM_RECV) {
2141 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2142 return 1;
2143
2144 } else { /* SK_MEM_SEND */
2145 if (sk->sk_type == SOCK_STREAM) {
2146 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2147 return 1;
2148 } else if (atomic_read(&sk->sk_wmem_alloc) <
2149 prot->sysctl_wmem[0])
2150 return 1;
2151 }
2152
2153 if (sk_has_memory_pressure(sk)) {
2154 int alloc;
2155
2156 if (!sk_under_memory_pressure(sk))
2157 return 1;
2158 alloc = sk_sockets_allocated_read_positive(sk);
2159 if (sk_prot_mem_limits(sk, 2) > alloc *
2160 sk_mem_pages(sk->sk_wmem_queued +
2161 atomic_read(&sk->sk_rmem_alloc) +
2162 sk->sk_forward_alloc))
2163 return 1;
2164 }
2165
2166 suppress_allocation:
2167
2168 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2169 sk_stream_moderate_sndbuf(sk);
2170
2171 /* Fail only if socket is _under_ its sndbuf.
2172 * In this case we cannot block, so that we have to fail.
2173 */
2174 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2175 return 1;
2176 }
2177
2178 trace_sock_exceed_buf_limit(sk, prot, allocated);
2179
2180 sk_memory_allocated_sub(sk, amt);
2181
2182 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2183 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2184
2185 return 0;
2186 }
2187 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2188
2189 /**
2190 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2191 * @sk: socket
2192 * @size: memory size to allocate
2193 * @kind: allocation type
2194 *
2195 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2196 * rmem allocation. This function assumes that protocols which have
2197 * memory_pressure use sk_wmem_queued as write buffer accounting.
2198 */
2199 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2200 {
2201 int ret, amt = sk_mem_pages(size);
2202
2203 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2204 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2205 if (!ret)
2206 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2207 return ret;
2208 }
2209 EXPORT_SYMBOL(__sk_mem_schedule);
2210
2211 /**
2212 * __sk_mem_reduce_allocated - reclaim memory_allocated
2213 * @sk: socket
2214 * @amount: number of quanta
2215 *
2216 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2217 */
2218 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2219 {
2220 sk_memory_allocated_sub(sk, amount);
2221
2222 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2223 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2224
2225 if (sk_under_memory_pressure(sk) &&
2226 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2227 sk_leave_memory_pressure(sk);
2228 }
2229 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2230
2231 /**
2232 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2233 * @sk: socket
2234 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2235 */
2236 void __sk_mem_reclaim(struct sock *sk, int amount)
2237 {
2238 amount >>= SK_MEM_QUANTUM_SHIFT;
2239 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2240 __sk_mem_reduce_allocated(sk, amount);
2241 }
2242 EXPORT_SYMBOL(__sk_mem_reclaim);
2243
2244 int sk_set_peek_off(struct sock *sk, int val)
2245 {
2246 if (val < 0)
2247 return -EINVAL;
2248
2249 sk->sk_peek_off = val;
2250 return 0;
2251 }
2252 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2253
2254 /*
2255 * Set of default routines for initialising struct proto_ops when
2256 * the protocol does not support a particular function. In certain
2257 * cases where it makes no sense for a protocol to have a "do nothing"
2258 * function, some default processing is provided.
2259 */
2260
2261 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2262 {
2263 return -EOPNOTSUPP;
2264 }
2265 EXPORT_SYMBOL(sock_no_bind);
2266
2267 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2268 int len, int flags)
2269 {
2270 return -EOPNOTSUPP;
2271 }
2272 EXPORT_SYMBOL(sock_no_connect);
2273
2274 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2275 {
2276 return -EOPNOTSUPP;
2277 }
2278 EXPORT_SYMBOL(sock_no_socketpair);
2279
2280 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2281 bool kern)
2282 {
2283 return -EOPNOTSUPP;
2284 }
2285 EXPORT_SYMBOL(sock_no_accept);
2286
2287 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2288 int *len, int peer)
2289 {
2290 return -EOPNOTSUPP;
2291 }
2292 EXPORT_SYMBOL(sock_no_getname);
2293
2294 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2295 {
2296 return 0;
2297 }
2298 EXPORT_SYMBOL(sock_no_poll);
2299
2300 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2301 {
2302 return -EOPNOTSUPP;
2303 }
2304 EXPORT_SYMBOL(sock_no_ioctl);
2305
2306 int sock_no_listen(struct socket *sock, int backlog)
2307 {
2308 return -EOPNOTSUPP;
2309 }
2310 EXPORT_SYMBOL(sock_no_listen);
2311
2312 int sock_no_shutdown(struct socket *sock, int how)
2313 {
2314 return -EOPNOTSUPP;
2315 }
2316 EXPORT_SYMBOL(sock_no_shutdown);
2317
2318 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2319 char __user *optval, unsigned int optlen)
2320 {
2321 return -EOPNOTSUPP;
2322 }
2323 EXPORT_SYMBOL(sock_no_setsockopt);
2324
2325 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2326 char __user *optval, int __user *optlen)
2327 {
2328 return -EOPNOTSUPP;
2329 }
2330 EXPORT_SYMBOL(sock_no_getsockopt);
2331
2332 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2333 {
2334 return -EOPNOTSUPP;
2335 }
2336 EXPORT_SYMBOL(sock_no_sendmsg);
2337
2338 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2339 int flags)
2340 {
2341 return -EOPNOTSUPP;
2342 }
2343 EXPORT_SYMBOL(sock_no_recvmsg);
2344
2345 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2346 {
2347 /* Mirror missing mmap method error code */
2348 return -ENODEV;
2349 }
2350 EXPORT_SYMBOL(sock_no_mmap);
2351
2352 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2353 {
2354 ssize_t res;
2355 struct msghdr msg = {.msg_flags = flags};
2356 struct kvec iov;
2357 char *kaddr = kmap(page);
2358 iov.iov_base = kaddr + offset;
2359 iov.iov_len = size;
2360 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2361 kunmap(page);
2362 return res;
2363 }
2364 EXPORT_SYMBOL(sock_no_sendpage);
2365
2366 /*
2367 * Default Socket Callbacks
2368 */
2369
2370 static void sock_def_wakeup(struct sock *sk)
2371 {
2372 struct socket_wq *wq;
2373
2374 rcu_read_lock();
2375 wq = rcu_dereference(sk->sk_wq);
2376 if (skwq_has_sleeper(wq))
2377 wake_up_interruptible_all(&wq->wait);
2378 rcu_read_unlock();
2379 }
2380
2381 static void sock_def_error_report(struct sock *sk)
2382 {
2383 struct socket_wq *wq;
2384
2385 rcu_read_lock();
2386 wq = rcu_dereference(sk->sk_wq);
2387 if (skwq_has_sleeper(wq))
2388 wake_up_interruptible_poll(&wq->wait, POLLERR);
2389 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2390 rcu_read_unlock();
2391 }
2392
2393 static void sock_def_readable(struct sock *sk)
2394 {
2395 struct socket_wq *wq;
2396
2397 rcu_read_lock();
2398 wq = rcu_dereference(sk->sk_wq);
2399 if (skwq_has_sleeper(wq))
2400 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2401 POLLRDNORM | POLLRDBAND);
2402 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2403 rcu_read_unlock();
2404 }
2405
2406 static void sock_def_write_space(struct sock *sk)
2407 {
2408 struct socket_wq *wq;
2409
2410 rcu_read_lock();
2411
2412 /* Do not wake up a writer until he can make "significant"
2413 * progress. --DaveM
2414 */
2415 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2416 wq = rcu_dereference(sk->sk_wq);
2417 if (skwq_has_sleeper(wq))
2418 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2419 POLLWRNORM | POLLWRBAND);
2420
2421 /* Should agree with poll, otherwise some programs break */
2422 if (sock_writeable(sk))
2423 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2424 }
2425
2426 rcu_read_unlock();
2427 }
2428
2429 static void sock_def_destruct(struct sock *sk)
2430 {
2431 }
2432
2433 void sk_send_sigurg(struct sock *sk)
2434 {
2435 if (sk->sk_socket && sk->sk_socket->file)
2436 if (send_sigurg(&sk->sk_socket->file->f_owner))
2437 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2438 }
2439 EXPORT_SYMBOL(sk_send_sigurg);
2440
2441 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2442 unsigned long expires)
2443 {
2444 if (!mod_timer(timer, expires))
2445 sock_hold(sk);
2446 }
2447 EXPORT_SYMBOL(sk_reset_timer);
2448
2449 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2450 {
2451 if (del_timer(timer))
2452 __sock_put(sk);
2453 }
2454 EXPORT_SYMBOL(sk_stop_timer);
2455
2456 void sock_init_data(struct socket *sock, struct sock *sk)
2457 {
2458 skb_queue_head_init(&sk->sk_receive_queue);
2459 skb_queue_head_init(&sk->sk_write_queue);
2460 skb_queue_head_init(&sk->sk_error_queue);
2461
2462 sk->sk_send_head = NULL;
2463
2464 init_timer(&sk->sk_timer);
2465
2466 sk->sk_allocation = GFP_KERNEL;
2467 sk->sk_rcvbuf = sysctl_rmem_default;
2468 sk->sk_sndbuf = sysctl_wmem_default;
2469 sk->sk_state = TCP_CLOSE;
2470 sk_set_socket(sk, sock);
2471
2472 sock_set_flag(sk, SOCK_ZAPPED);
2473
2474 if (sock) {
2475 sk->sk_type = sock->type;
2476 sk->sk_wq = sock->wq;
2477 sock->sk = sk;
2478 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2479 } else {
2480 sk->sk_wq = NULL;
2481 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2482 }
2483
2484 rwlock_init(&sk->sk_callback_lock);
2485 if (sk->sk_kern_sock)
2486 lockdep_set_class_and_name(
2487 &sk->sk_callback_lock,
2488 af_kern_callback_keys + sk->sk_family,
2489 af_family_kern_clock_key_strings[sk->sk_family]);
2490 else
2491 lockdep_set_class_and_name(
2492 &sk->sk_callback_lock,
2493 af_callback_keys + sk->sk_family,
2494 af_family_clock_key_strings[sk->sk_family]);
2495
2496 sk->sk_state_change = sock_def_wakeup;
2497 sk->sk_data_ready = sock_def_readable;
2498 sk->sk_write_space = sock_def_write_space;
2499 sk->sk_error_report = sock_def_error_report;
2500 sk->sk_destruct = sock_def_destruct;
2501
2502 sk->sk_frag.page = NULL;
2503 sk->sk_frag.offset = 0;
2504 sk->sk_peek_off = -1;
2505
2506 sk->sk_peer_pid = NULL;
2507 sk->sk_peer_cred = NULL;
2508 sk->sk_write_pending = 0;
2509 sk->sk_rcvlowat = 1;
2510 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2511 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2512
2513 sk->sk_stamp = ktime_set(-1L, 0);
2514
2515 #ifdef CONFIG_NET_RX_BUSY_POLL
2516 sk->sk_napi_id = 0;
2517 sk->sk_ll_usec = sysctl_net_busy_read;
2518 #endif
2519
2520 sk->sk_max_pacing_rate = ~0U;
2521 sk->sk_pacing_rate = ~0U;
2522 sk->sk_incoming_cpu = -1;
2523 /*
2524 * Before updating sk_refcnt, we must commit prior changes to memory
2525 * (Documentation/RCU/rculist_nulls.txt for details)
2526 */
2527 smp_wmb();
2528 atomic_set(&sk->sk_refcnt, 1);
2529 atomic_set(&sk->sk_drops, 0);
2530 }
2531 EXPORT_SYMBOL(sock_init_data);
2532
2533 void lock_sock_nested(struct sock *sk, int subclass)
2534 {
2535 might_sleep();
2536 spin_lock_bh(&sk->sk_lock.slock);
2537 if (sk->sk_lock.owned)
2538 __lock_sock(sk);
2539 sk->sk_lock.owned = 1;
2540 spin_unlock(&sk->sk_lock.slock);
2541 /*
2542 * The sk_lock has mutex_lock() semantics here:
2543 */
2544 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2545 local_bh_enable();
2546 }
2547 EXPORT_SYMBOL(lock_sock_nested);
2548
2549 void release_sock(struct sock *sk)
2550 {
2551 spin_lock_bh(&sk->sk_lock.slock);
2552 if (sk->sk_backlog.tail)
2553 __release_sock(sk);
2554
2555 /* Warning : release_cb() might need to release sk ownership,
2556 * ie call sock_release_ownership(sk) before us.
2557 */
2558 if (sk->sk_prot->release_cb)
2559 sk->sk_prot->release_cb(sk);
2560
2561 sock_release_ownership(sk);
2562 if (waitqueue_active(&sk->sk_lock.wq))
2563 wake_up(&sk->sk_lock.wq);
2564 spin_unlock_bh(&sk->sk_lock.slock);
2565 }
2566 EXPORT_SYMBOL(release_sock);
2567
2568 /**
2569 * lock_sock_fast - fast version of lock_sock
2570 * @sk: socket
2571 *
2572 * This version should be used for very small section, where process wont block
2573 * return false if fast path is taken
2574 * sk_lock.slock locked, owned = 0, BH disabled
2575 * return true if slow path is taken
2576 * sk_lock.slock unlocked, owned = 1, BH enabled
2577 */
2578 bool lock_sock_fast(struct sock *sk)
2579 {
2580 might_sleep();
2581 spin_lock_bh(&sk->sk_lock.slock);
2582
2583 if (!sk->sk_lock.owned)
2584 /*
2585 * Note : We must disable BH
2586 */
2587 return false;
2588
2589 __lock_sock(sk);
2590 sk->sk_lock.owned = 1;
2591 spin_unlock(&sk->sk_lock.slock);
2592 /*
2593 * The sk_lock has mutex_lock() semantics here:
2594 */
2595 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2596 local_bh_enable();
2597 return true;
2598 }
2599 EXPORT_SYMBOL(lock_sock_fast);
2600
2601 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2602 {
2603 struct timeval tv;
2604 if (!sock_flag(sk, SOCK_TIMESTAMP))
2605 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2606 tv = ktime_to_timeval(sk->sk_stamp);
2607 if (tv.tv_sec == -1)
2608 return -ENOENT;
2609 if (tv.tv_sec == 0) {
2610 sk->sk_stamp = ktime_get_real();
2611 tv = ktime_to_timeval(sk->sk_stamp);
2612 }
2613 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2614 }
2615 EXPORT_SYMBOL(sock_get_timestamp);
2616
2617 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2618 {
2619 struct timespec ts;
2620 if (!sock_flag(sk, SOCK_TIMESTAMP))
2621 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2622 ts = ktime_to_timespec(sk->sk_stamp);
2623 if (ts.tv_sec == -1)
2624 return -ENOENT;
2625 if (ts.tv_sec == 0) {
2626 sk->sk_stamp = ktime_get_real();
2627 ts = ktime_to_timespec(sk->sk_stamp);
2628 }
2629 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2630 }
2631 EXPORT_SYMBOL(sock_get_timestampns);
2632
2633 void sock_enable_timestamp(struct sock *sk, int flag)
2634 {
2635 if (!sock_flag(sk, flag)) {
2636 unsigned long previous_flags = sk->sk_flags;
2637
2638 sock_set_flag(sk, flag);
2639 /*
2640 * we just set one of the two flags which require net
2641 * time stamping, but time stamping might have been on
2642 * already because of the other one
2643 */
2644 if (sock_needs_netstamp(sk) &&
2645 !(previous_flags & SK_FLAGS_TIMESTAMP))
2646 net_enable_timestamp();
2647 }
2648 }
2649
2650 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2651 int level, int type)
2652 {
2653 struct sock_exterr_skb *serr;
2654 struct sk_buff *skb;
2655 int copied, err;
2656
2657 err = -EAGAIN;
2658 skb = sock_dequeue_err_skb(sk);
2659 if (skb == NULL)
2660 goto out;
2661
2662 copied = skb->len;
2663 if (copied > len) {
2664 msg->msg_flags |= MSG_TRUNC;
2665 copied = len;
2666 }
2667 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2668 if (err)
2669 goto out_free_skb;
2670
2671 sock_recv_timestamp(msg, sk, skb);
2672
2673 serr = SKB_EXT_ERR(skb);
2674 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2675
2676 msg->msg_flags |= MSG_ERRQUEUE;
2677 err = copied;
2678
2679 out_free_skb:
2680 kfree_skb(skb);
2681 out:
2682 return err;
2683 }
2684 EXPORT_SYMBOL(sock_recv_errqueue);
2685
2686 /*
2687 * Get a socket option on an socket.
2688 *
2689 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2690 * asynchronous errors should be reported by getsockopt. We assume
2691 * this means if you specify SO_ERROR (otherwise whats the point of it).
2692 */
2693 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2694 char __user *optval, int __user *optlen)
2695 {
2696 struct sock *sk = sock->sk;
2697
2698 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2699 }
2700 EXPORT_SYMBOL(sock_common_getsockopt);
2701
2702 #ifdef CONFIG_COMPAT
2703 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2704 char __user *optval, int __user *optlen)
2705 {
2706 struct sock *sk = sock->sk;
2707
2708 if (sk->sk_prot->compat_getsockopt != NULL)
2709 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2710 optval, optlen);
2711 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2712 }
2713 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2714 #endif
2715
2716 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2717 int flags)
2718 {
2719 struct sock *sk = sock->sk;
2720 int addr_len = 0;
2721 int err;
2722
2723 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2724 flags & ~MSG_DONTWAIT, &addr_len);
2725 if (err >= 0)
2726 msg->msg_namelen = addr_len;
2727 return err;
2728 }
2729 EXPORT_SYMBOL(sock_common_recvmsg);
2730
2731 /*
2732 * Set socket options on an inet socket.
2733 */
2734 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2735 char __user *optval, unsigned int optlen)
2736 {
2737 struct sock *sk = sock->sk;
2738
2739 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2740 }
2741 EXPORT_SYMBOL(sock_common_setsockopt);
2742
2743 #ifdef CONFIG_COMPAT
2744 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2745 char __user *optval, unsigned int optlen)
2746 {
2747 struct sock *sk = sock->sk;
2748
2749 if (sk->sk_prot->compat_setsockopt != NULL)
2750 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2751 optval, optlen);
2752 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2753 }
2754 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2755 #endif
2756
2757 void sk_common_release(struct sock *sk)
2758 {
2759 if (sk->sk_prot->destroy)
2760 sk->sk_prot->destroy(sk);
2761
2762 /*
2763 * Observation: when sock_common_release is called, processes have
2764 * no access to socket. But net still has.
2765 * Step one, detach it from networking:
2766 *
2767 * A. Remove from hash tables.
2768 */
2769
2770 sk->sk_prot->unhash(sk);
2771
2772 /*
2773 * In this point socket cannot receive new packets, but it is possible
2774 * that some packets are in flight because some CPU runs receiver and
2775 * did hash table lookup before we unhashed socket. They will achieve
2776 * receive queue and will be purged by socket destructor.
2777 *
2778 * Also we still have packets pending on receive queue and probably,
2779 * our own packets waiting in device queues. sock_destroy will drain
2780 * receive queue, but transmitted packets will delay socket destruction
2781 * until the last reference will be released.
2782 */
2783
2784 sock_orphan(sk);
2785
2786 xfrm_sk_free_policy(sk);
2787
2788 sk_refcnt_debug_release(sk);
2789
2790 if (sk->sk_frag.page) {
2791 put_page(sk->sk_frag.page);
2792 sk->sk_frag.page = NULL;
2793 }
2794
2795 sock_put(sk);
2796 }
2797 EXPORT_SYMBOL(sk_common_release);
2798
2799 #ifdef CONFIG_PROC_FS
2800 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2801 struct prot_inuse {
2802 int val[PROTO_INUSE_NR];
2803 };
2804
2805 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2806
2807 #ifdef CONFIG_NET_NS
2808 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2809 {
2810 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2811 }
2812 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2813
2814 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2815 {
2816 int cpu, idx = prot->inuse_idx;
2817 int res = 0;
2818
2819 for_each_possible_cpu(cpu)
2820 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2821
2822 return res >= 0 ? res : 0;
2823 }
2824 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2825
2826 static int __net_init sock_inuse_init_net(struct net *net)
2827 {
2828 net->core.inuse = alloc_percpu(struct prot_inuse);
2829 return net->core.inuse ? 0 : -ENOMEM;
2830 }
2831
2832 static void __net_exit sock_inuse_exit_net(struct net *net)
2833 {
2834 free_percpu(net->core.inuse);
2835 }
2836
2837 static struct pernet_operations net_inuse_ops = {
2838 .init = sock_inuse_init_net,
2839 .exit = sock_inuse_exit_net,
2840 };
2841
2842 static __init int net_inuse_init(void)
2843 {
2844 if (register_pernet_subsys(&net_inuse_ops))
2845 panic("Cannot initialize net inuse counters");
2846
2847 return 0;
2848 }
2849
2850 core_initcall(net_inuse_init);
2851 #else
2852 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2853
2854 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2855 {
2856 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2857 }
2858 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2859
2860 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2861 {
2862 int cpu, idx = prot->inuse_idx;
2863 int res = 0;
2864
2865 for_each_possible_cpu(cpu)
2866 res += per_cpu(prot_inuse, cpu).val[idx];
2867
2868 return res >= 0 ? res : 0;
2869 }
2870 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2871 #endif
2872
2873 static void assign_proto_idx(struct proto *prot)
2874 {
2875 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2876
2877 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2878 pr_err("PROTO_INUSE_NR exhausted\n");
2879 return;
2880 }
2881
2882 set_bit(prot->inuse_idx, proto_inuse_idx);
2883 }
2884
2885 static void release_proto_idx(struct proto *prot)
2886 {
2887 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2888 clear_bit(prot->inuse_idx, proto_inuse_idx);
2889 }
2890 #else
2891 static inline void assign_proto_idx(struct proto *prot)
2892 {
2893 }
2894
2895 static inline void release_proto_idx(struct proto *prot)
2896 {
2897 }
2898 #endif
2899
2900 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2901 {
2902 if (!rsk_prot)
2903 return;
2904 kfree(rsk_prot->slab_name);
2905 rsk_prot->slab_name = NULL;
2906 kmem_cache_destroy(rsk_prot->slab);
2907 rsk_prot->slab = NULL;
2908 }
2909
2910 static int req_prot_init(const struct proto *prot)
2911 {
2912 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2913
2914 if (!rsk_prot)
2915 return 0;
2916
2917 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2918 prot->name);
2919 if (!rsk_prot->slab_name)
2920 return -ENOMEM;
2921
2922 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2923 rsk_prot->obj_size, 0,
2924 prot->slab_flags, NULL);
2925
2926 if (!rsk_prot->slab) {
2927 pr_crit("%s: Can't create request sock SLAB cache!\n",
2928 prot->name);
2929 return -ENOMEM;
2930 }
2931 return 0;
2932 }
2933
2934 int proto_register(struct proto *prot, int alloc_slab)
2935 {
2936 if (alloc_slab) {
2937 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2938 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2939 NULL);
2940
2941 if (prot->slab == NULL) {
2942 pr_crit("%s: Can't create sock SLAB cache!\n",
2943 prot->name);
2944 goto out;
2945 }
2946
2947 if (req_prot_init(prot))
2948 goto out_free_request_sock_slab;
2949
2950 if (prot->twsk_prot != NULL) {
2951 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2952
2953 if (prot->twsk_prot->twsk_slab_name == NULL)
2954 goto out_free_request_sock_slab;
2955
2956 prot->twsk_prot->twsk_slab =
2957 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2958 prot->twsk_prot->twsk_obj_size,
2959 0,
2960 prot->slab_flags,
2961 NULL);
2962 if (prot->twsk_prot->twsk_slab == NULL)
2963 goto out_free_timewait_sock_slab_name;
2964 }
2965 }
2966
2967 mutex_lock(&proto_list_mutex);
2968 list_add(&prot->node, &proto_list);
2969 assign_proto_idx(prot);
2970 mutex_unlock(&proto_list_mutex);
2971 return 0;
2972
2973 out_free_timewait_sock_slab_name:
2974 kfree(prot->twsk_prot->twsk_slab_name);
2975 out_free_request_sock_slab:
2976 req_prot_cleanup(prot->rsk_prot);
2977
2978 kmem_cache_destroy(prot->slab);
2979 prot->slab = NULL;
2980 out:
2981 return -ENOBUFS;
2982 }
2983 EXPORT_SYMBOL(proto_register);
2984
2985 void proto_unregister(struct proto *prot)
2986 {
2987 mutex_lock(&proto_list_mutex);
2988 release_proto_idx(prot);
2989 list_del(&prot->node);
2990 mutex_unlock(&proto_list_mutex);
2991
2992 kmem_cache_destroy(prot->slab);
2993 prot->slab = NULL;
2994
2995 req_prot_cleanup(prot->rsk_prot);
2996
2997 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2998 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2999 kfree(prot->twsk_prot->twsk_slab_name);
3000 prot->twsk_prot->twsk_slab = NULL;
3001 }
3002 }
3003 EXPORT_SYMBOL(proto_unregister);
3004
3005 #ifdef CONFIG_PROC_FS
3006 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3007 __acquires(proto_list_mutex)
3008 {
3009 mutex_lock(&proto_list_mutex);
3010 return seq_list_start_head(&proto_list, *pos);
3011 }
3012
3013 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3014 {
3015 return seq_list_next(v, &proto_list, pos);
3016 }
3017
3018 static void proto_seq_stop(struct seq_file *seq, void *v)
3019 __releases(proto_list_mutex)
3020 {
3021 mutex_unlock(&proto_list_mutex);
3022 }
3023
3024 static char proto_method_implemented(const void *method)
3025 {
3026 return method == NULL ? 'n' : 'y';
3027 }
3028 static long sock_prot_memory_allocated(struct proto *proto)
3029 {
3030 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3031 }
3032
3033 static char *sock_prot_memory_pressure(struct proto *proto)
3034 {
3035 return proto->memory_pressure != NULL ?
3036 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3037 }
3038
3039 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3040 {
3041
3042 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3043 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3044 proto->name,
3045 proto->obj_size,
3046 sock_prot_inuse_get(seq_file_net(seq), proto),
3047 sock_prot_memory_allocated(proto),
3048 sock_prot_memory_pressure(proto),
3049 proto->max_header,
3050 proto->slab == NULL ? "no" : "yes",
3051 module_name(proto->owner),
3052 proto_method_implemented(proto->close),
3053 proto_method_implemented(proto->connect),
3054 proto_method_implemented(proto->disconnect),
3055 proto_method_implemented(proto->accept),
3056 proto_method_implemented(proto->ioctl),
3057 proto_method_implemented(proto->init),
3058 proto_method_implemented(proto->destroy),
3059 proto_method_implemented(proto->shutdown),
3060 proto_method_implemented(proto->setsockopt),
3061 proto_method_implemented(proto->getsockopt),
3062 proto_method_implemented(proto->sendmsg),
3063 proto_method_implemented(proto->recvmsg),
3064 proto_method_implemented(proto->sendpage),
3065 proto_method_implemented(proto->bind),
3066 proto_method_implemented(proto->backlog_rcv),
3067 proto_method_implemented(proto->hash),
3068 proto_method_implemented(proto->unhash),
3069 proto_method_implemented(proto->get_port),
3070 proto_method_implemented(proto->enter_memory_pressure));
3071 }
3072
3073 static int proto_seq_show(struct seq_file *seq, void *v)
3074 {
3075 if (v == &proto_list)
3076 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3077 "protocol",
3078 "size",
3079 "sockets",
3080 "memory",
3081 "press",
3082 "maxhdr",
3083 "slab",
3084 "module",
3085 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3086 else
3087 proto_seq_printf(seq, list_entry(v, struct proto, node));
3088 return 0;
3089 }
3090
3091 static const struct seq_operations proto_seq_ops = {
3092 .start = proto_seq_start,
3093 .next = proto_seq_next,
3094 .stop = proto_seq_stop,
3095 .show = proto_seq_show,
3096 };
3097
3098 static int proto_seq_open(struct inode *inode, struct file *file)
3099 {
3100 return seq_open_net(inode, file, &proto_seq_ops,
3101 sizeof(struct seq_net_private));
3102 }
3103
3104 static const struct file_operations proto_seq_fops = {
3105 .owner = THIS_MODULE,
3106 .open = proto_seq_open,
3107 .read = seq_read,
3108 .llseek = seq_lseek,
3109 .release = seq_release_net,
3110 };
3111
3112 static __net_init int proto_init_net(struct net *net)
3113 {
3114 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3115 return -ENOMEM;
3116
3117 return 0;
3118 }
3119
3120 static __net_exit void proto_exit_net(struct net *net)
3121 {
3122 remove_proc_entry("protocols", net->proc_net);
3123 }
3124
3125
3126 static __net_initdata struct pernet_operations proto_net_ops = {
3127 .init = proto_init_net,
3128 .exit = proto_exit_net,
3129 };
3130
3131 static int __init proto_init(void)
3132 {
3133 return register_pernet_subsys(&proto_net_ops);
3134 }
3135
3136 subsys_initcall(proto_init);
3137
3138 #endif /* PROC_FS */
ubuntu-bionic-kernel mirror