projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'cxl-fixes-6.10-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl
[linux-2.6-block.git] / net / core / sock.c
CommitLineData
2874c5fd 1// SPDX-License-Identifier: GPL-2.0-or-later
1da177e4
LT		 2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
9 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT		 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
4ec93edb 35 * code. The ACK stuff can wait and needs major
1da177e4
LT		 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:
1da177e4
LT		 84 */
85
e005d193
JP		 86#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87
80b14dee 88#include <asm/unaligned.h>
4fc268d2 89#include <linux/capability.h>
1da177e4 90#include <linux/errno.h>
cb820f8e 91#include <linux/errqueue.h>
1da177e4
LT		 92#include <linux/types.h>
93#include <linux/socket.h>
94#include <linux/in.h>
95#include <linux/kernel.h>
1da177e4
LT		 96#include <linux/module.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/sched.h>
f1083048 100#include <linux/sched/mm.h>
1da177e4
LT		 101#include <linux/timer.h>
102#include <linux/string.h>
103#include <linux/sockios.h>
104#include <linux/net.h>
105#include <linux/mm.h>
106#include <linux/slab.h>
107#include <linux/interrupt.h>
108#include <linux/poll.h>
109#include <linux/tcp.h>
a54d51fb 110#include <linux/udp.h>
1da177e4 111#include <linux/init.h>
a1f8e7f7 112#include <linux/highmem.h>
3f551f94 113#include <linux/user_namespace.h>
c5905afb 114#include <linux/static_key.h>
3969eb38 115#include <linux/memcontrol.h>
8c1ae10d 116#include <linux/prefetch.h>
a6c0d093 117#include <linux/compat.h>
e1d001fa
BL		 118#include <linux/mroute.h>
119#include <linux/mroute6.h>
120#include <linux/icmpv6.h>
1da177e4 121
7c0f6ba6 122#include <linux/uaccess.h>
1da177e4
LT		 123
124#include <linux/netdevice.h>
125#include <net/protocol.h>
126#include <linux/skbuff.h>
457c4cbc 127#include <net/net_namespace.h>
2e6599cb 128#include <net/request_sock.h>
1da177e4 129#include <net/sock.h>
f3d93817 130#include <net/proto_memory.h>
20d49473 131#include <linux/net_tstamp.h>
1da177e4
LT		 132#include <net/xfrm.h>
133#include <linux/ipsec.h>
f8451725 134#include <net/cls_cgroup.h>
5bc1421e 135#include <net/netprio_cgroup.h>
eb4cb008 136#include <linux/sock_diag.h>
1da177e4
LT		 137
138#include <linux/filter.h>
538950a1 139#include <net/sock_reuseport.h>
6ac99e8f 140#include <net/bpf_sk_storage.h>
1da177e4 141
3847ce32
SM		 142#include <trace/events/sock.h>
143
1da177e4 144#include <net/tcp.h>
076bb0c8 145#include <net/busy_poll.h>
e1d001fa 146#include <net/phonet/phonet.h>
06021292 147
d463126e
YL		 148#include <linux/ethtool.h>
149
6264f58c
JK		 150#include "dev.h"
151
36b77a52 152static DEFINE_MUTEX(proto_list_mutex);
d1a4c0b3
GC		 153static LIST_HEAD(proto_list);
154
0a8afd9f 155static void sock_def_write_space_wfree(struct sock *sk);
052ada09
PB		 156static void sock_def_write_space(struct sock *sk);
157
a3b299da
EB		 158/**
159 * sk_ns_capable - General socket capability test
160 * @sk: Socket to use a capability on or through
161 * @user_ns: The user namespace of the capability to use
162 * @cap: The capability to use
163 *
164 * Test to see if the opener of the socket had when the socket was
165 * created and the current process has the capability @cap in the user
166 * namespace @user_ns.
167 */
168bool sk_ns_capable(const struct sock *sk,
169 struct user_namespace *user_ns, int cap)
170{
171 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
172 ns_capable(user_ns, cap);
173}
174EXPORT_SYMBOL(sk_ns_capable);
175
176/**
177 * sk_capable - Socket global capability test
178 * @sk: Socket to use a capability on or through
e793c0f7 179 * @cap: The global capability to use
a3b299da
EB		 180 *
181 * Test to see if the opener of the socket had when the socket was
182 * created and the current process has the capability @cap in all user
183 * namespaces.
184 */
185bool sk_capable(const struct sock *sk, int cap)
186{
187 return sk_ns_capable(sk, &init_user_ns, cap);
188}
189EXPORT_SYMBOL(sk_capable);
190
191/**
192 * sk_net_capable - Network namespace socket capability test
193 * @sk: Socket to use a capability on or through
194 * @cap: The capability to use
195 *
e793c0f7 196 * Test to see if the opener of the socket had when the socket was created
a3b299da
EB		 197 * and the current process has the capability @cap over the network namespace
198 * the socket is a member of.
199 */
200bool sk_net_capable(const struct sock *sk, int cap)
201{
202 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
203}
204EXPORT_SYMBOL(sk_net_capable);
205
da21f24d
IM		 206/*
207 * Each address family might have different locking rules, so we have
cdfbabfb
DH		 208 * one slock key per address family and separate keys for internal and
209 * userspace sockets.
da21f24d 210 */
a5b5bb9a 211static struct lock_class_key af_family_keys[AF_MAX];
cdfbabfb 212static struct lock_class_key af_family_kern_keys[AF_MAX];
a5b5bb9a 213static struct lock_class_key af_family_slock_keys[AF_MAX];
cdfbabfb 214static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
a5b5bb9a 215
a5b5bb9a
IM		 216/*
217 * Make lock validator output more readable. (we pre-construct these
218 * strings build-time, so that runtime initialization of socket
219 * locks is fast):
220 */
cdfbabfb
DH		 221
222#define _sock_locks(x) \
223 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
224 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
225 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
226 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
227 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
228 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
229 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
230 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
231 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
232 x "27" , x "28" , x "AF_CAN" , \
233 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
234 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
235 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
236 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
68e8b849 237 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
bc49d816 238 x "AF_MCTP" , \
68e8b849 239 x "AF_MAX"
cdfbabfb 240
36cbd3dc 241static const char *const af_family_key_strings[AF_MAX+1] = {
cdfbabfb 242 _sock_locks("sk_lock-")
a5b5bb9a 243};
36cbd3dc 244static const char *const af_family_slock_key_strings[AF_MAX+1] = {
cdfbabfb 245 _sock_locks("slock-")
a5b5bb9a 246};
36cbd3dc 247static const char *const af_family_clock_key_strings[AF_MAX+1] = {
cdfbabfb
DH		 248 _sock_locks("clock-")
249};
250
251static const char *const af_family_kern_key_strings[AF_MAX+1] = {
252 _sock_locks("k-sk_lock-")
253};
254static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
255 _sock_locks("k-slock-")
256};
257static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
258 _sock_locks("k-clock-")
443aef0e 259};
581319c5 260static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
6b431d50 261 _sock_locks("rlock-")
581319c5
PA		 262};
263static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
6b431d50 264 _sock_locks("wlock-")
581319c5
PA		 265};
266static const char *const af_family_elock_key_strings[AF_MAX+1] = {
6b431d50 267 _sock_locks("elock-")
581319c5 268};
da21f24d
IM		 269
270/*
581319c5 271 * sk_callback_lock and sk queues locking rules are per-address-family,
da21f24d
IM		 272 * so split the lock classes by using a per-AF key:
273 */
274static struct lock_class_key af_callback_keys[AF_MAX];
581319c5
PA		 275static struct lock_class_key af_rlock_keys[AF_MAX];
276static struct lock_class_key af_wlock_keys[AF_MAX];
277static struct lock_class_key af_elock_keys[AF_MAX];
cdfbabfb 278static struct lock_class_key af_kern_callback_keys[AF_MAX];
da21f24d 279
1da177e4 280/* Run time adjustable parameters. */
ab32ea5d 281__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
6d8ebc8a 282EXPORT_SYMBOL(sysctl_wmem_max);
ab32ea5d 283__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
6d8ebc8a 284EXPORT_SYMBOL(sysctl_rmem_max);
ab32ea5d
BH		 285__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
286__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4 287
b245be1f
WB		 288int sysctl_tstamp_allow_data __read_mostly = 1;
289
a7950ae8
DB		 290DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
291EXPORT_SYMBOL_GPL(memalloc_socks_key);
c93bdd0e 292
7cb02404
MG		 293/**
294 * sk_set_memalloc - sets %SOCK_MEMALLOC
295 * @sk: socket to set it on
296 *
297 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
298 * It's the responsibility of the admin to adjust min_free_kbytes
299 * to meet the requirements
300 */
301void sk_set_memalloc(struct sock *sk)
302{
303 sock_set_flag(sk, SOCK_MEMALLOC);
304 sk->sk_allocation |= __GFP_MEMALLOC;
a7950ae8 305 static_branch_inc(&memalloc_socks_key);
7cb02404
MG		 306}
307EXPORT_SYMBOL_GPL(sk_set_memalloc);
308
309void sk_clear_memalloc(struct sock *sk)
310{
311 sock_reset_flag(sk, SOCK_MEMALLOC);
312 sk->sk_allocation &= ~__GFP_MEMALLOC;
a7950ae8 313 static_branch_dec(&memalloc_socks_key);
c76562b6
MG		 314
315 /*
316 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
5d753610
MG		 317 * progress of swapping. SOCK_MEMALLOC may be cleared while
318 * it has rmem allocations due to the last swapfile being deactivated
319 * but there is a risk that the socket is unusable due to exceeding
320 * the rmem limits. Reclaim the reserves and obey rmem limits again.
c76562b6 321 */
5d753610 322 sk_mem_reclaim(sk);
7cb02404
MG		 323}
324EXPORT_SYMBOL_GPL(sk_clear_memalloc);
325
b4b9e355
MG		 326int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
327{
328 int ret;
f1083048 329 unsigned int noreclaim_flag;
b4b9e355
MG		 330
331 /* these should have been dropped before queueing */
332 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
333
f1083048 334 noreclaim_flag = memalloc_noreclaim_save();
d2489c7b
ED		 335 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
336 tcp_v6_do_rcv,
337 tcp_v4_do_rcv,
338 sk, skb);
f1083048 339 memalloc_noreclaim_restore(noreclaim_flag);
b4b9e355
MG		 340
341 return ret;
342}
343EXPORT_SYMBOL(__sk_backlog_rcv);
344
e3ae2365
AA		 345void sk_error_report(struct sock *sk)
346{
347 sk->sk_error_report(sk);
e6a3e443
AA		 348
349 switch (sk->sk_family) {
350 case AF_INET:
351 fallthrough;
352 case AF_INET6:
353 trace_inet_sk_error_report(sk);
354 break;
355 default:
356 break;
357 }
e3ae2365
AA		 358}
359EXPORT_SYMBOL(sk_error_report);
360
4c1e34c0 361int sock_get_timeout(long timeo, void *optval, bool old_timeval)
fe0c72f3 362{
a9beb86a 363 struct __kernel_sock_timeval tv;
fe0c72f3
AB		 364
365 if (timeo == MAX_SCHEDULE_TIMEOUT) {
366 tv.tv_sec = 0;
367 tv.tv_usec = 0;
368 } else {
369 tv.tv_sec = timeo / HZ;
370 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
371 }
372
e6986423 373 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
fe0c72f3
AB		 374 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
375 *(struct old_timeval32 *)optval = tv32;
376 return sizeof(tv32);
377 }
378
a9beb86a
DD		 379 if (old_timeval) {
380 struct __kernel_old_timeval old_tv;
381 old_tv.tv_sec = tv.tv_sec;
382 old_tv.tv_usec = tv.tv_usec;
383 *(struct __kernel_old_timeval *)optval = old_tv;
28e72b26 384 return sizeof(old_tv);
a9beb86a
DD		 385 }
386
28e72b26
VC		 387 *(struct __kernel_sock_timeval *)optval = tv;
388 return sizeof(tv);
fe0c72f3 389}
4c1e34c0 390EXPORT_SYMBOL(sock_get_timeout);
fe0c72f3 391
4c1e34c0
RP		 392int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
393 sockptr_t optval, int optlen, bool old_timeval)
1da177e4 394{
e6986423 395 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
fe0c72f3
AB		 396 struct old_timeval32 tv32;
397
398 if (optlen < sizeof(tv32))
399 return -EINVAL;
400
c34645ac 401 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
fe0c72f3 402 return -EFAULT;
4c1e34c0
RP		 403 tv->tv_sec = tv32.tv_sec;
404 tv->tv_usec = tv32.tv_usec;
a9beb86a
DD		 405 } else if (old_timeval) {
406 struct __kernel_old_timeval old_tv;
407
408 if (optlen < sizeof(old_tv))
409 return -EINVAL;
c34645ac 410 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
a9beb86a 411 return -EFAULT;
4c1e34c0
RP		 412 tv->tv_sec = old_tv.tv_sec;
413 tv->tv_usec = old_tv.tv_usec;
fe0c72f3 414 } else {
4c1e34c0 415 if (optlen < sizeof(*tv))
fe0c72f3 416 return -EINVAL;
4c1e34c0 417 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
fe0c72f3
AB		 418 return -EFAULT;
419 }
4c1e34c0
RP		 420
421 return 0;
422}
423EXPORT_SYMBOL(sock_copy_user_timeval);
424
425static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
426 bool old_timeval)
427{
428 struct __kernel_sock_timeval tv;
429 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
285975dd 430 long val;
4c1e34c0
RP		 431
432 if (err)
433 return err;
434
ba78073e
VA		 435 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
436 return -EDOM;
1da177e4 437
ba78073e 438 if (tv.tv_sec < 0) {
6f11df83
AM		 439 static int warned __read_mostly;
440
285975dd 441 WRITE_ONCE(*timeo_p, 0);
50aab54f 442 if (warned < 10 && net_ratelimit()) {
ba78073e 443 warned++;
e005d193
JP		 444 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
445 __func__, current->comm, task_pid_nr(current));
50aab54f 446 }
ba78073e
VA		 447 return 0;
448 }
285975dd
ED		 449 val = MAX_SCHEDULE_TIMEOUT;
450 if ((tv.tv_sec || tv.tv_usec) &&
451 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
452 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
453 USEC_PER_SEC / HZ);
454 WRITE_ONCE(*timeo_p, val);
1da177e4
LT		 455 return 0;
456}
457
080a270f
HFS		 458static bool sock_needs_netstamp(const struct sock *sk)
459{
460 switch (sk->sk_family) {
461 case AF_UNSPEC:
462 case AF_UNIX:
463 return false;
464 default:
465 return true;
466 }
467}
468
08e29af3 469static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
4ec93edb 470{
08e29af3
ED		 471 if (sk->sk_flags & flags) {
472 sk->sk_flags &= ~flags;
080a270f
HFS		 473 if (sock_needs_netstamp(sk) &&
474 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
20d49473 475 net_disable_timestamp();
1da177e4
LT		 476 }
477}
478
479
e6afc8ac 480int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
f0088a50 481{
3b885787
NH		 482 unsigned long flags;
483 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 484
c2deb2e9 485 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) {
766e9037 486 atomic_inc(&sk->sk_drops);
3847ce32 487 trace_sock_rcvqueue_full(sk, skb);
766e9037 488 return -ENOMEM;
f0088a50
DV		 489 }
490
c76562b6 491 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
766e9037
ED		 492 atomic_inc(&sk->sk_drops);
493 return -ENOBUFS;
3ab224be
HA		 494 }
495
f0088a50
DV		 496 skb->dev = NULL;
497 skb_set_owner_r(skb, sk);
49ad9599 498
7fee226a
ED		 499 /* we escape from rcu protected region, make sure we dont leak
500 * a norefcounted dst
501 */
502 skb_dst_force(skb);
503
3b885787 504 spin_lock_irqsave(&list->lock, flags);
3bc3b96f 505 sock_skb_set_dropcount(sk, skb);
3b885787
NH		 506 __skb_queue_tail(list, skb);
507 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 508
509 if (!sock_flag(sk, SOCK_DEAD))
676d2369 510 sk->sk_data_ready(sk);
766e9037 511 return 0;
f0088a50 512}
e6afc8ac 513EXPORT_SYMBOL(__sock_queue_rcv_skb);
514
c1b8a567
MD		 515int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
516 enum skb_drop_reason *reason)
e6afc8ac 517{
c1b8a567 518 enum skb_drop_reason drop_reason;
e6afc8ac 519 int err;
520
521 err = sk_filter(sk, skb);
c1b8a567
MD		 522 if (err) {
523 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
524 goto out;
525 }
526 err = __sock_queue_rcv_skb(sk, skb);
527 switch (err) {
528 case -ENOMEM:
529 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
530 break;
531 case -ENOBUFS:
532 drop_reason = SKB_DROP_REASON_PROTO_MEM;
533 break;
534 default:
535 drop_reason = SKB_NOT_DROPPED_YET;
536 break;
537 }
538out:
539 if (reason)
540 *reason = drop_reason;
541 return err;
e6afc8ac 542}
c1b8a567 543EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
f0088a50 544
4f0c40d9 545int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
c3f24cfb 546 const int nested, unsigned int trim_cap, bool refcounted)
f0088a50
DV		 547{
548 int rc = NET_RX_SUCCESS;
549
4f0c40d9 550 if (sk_filter_trim_cap(sk, skb, trim_cap))
f0088a50
DV		 551 goto discard_and_relse;
552
553 skb->dev = NULL;
554
c2deb2e9 555 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) {
c377411f
ED		 556 atomic_inc(&sk->sk_drops);
557 goto discard_and_relse;
558 }
58a5a7b9
ACM		 559 if (nested)
560 bh_lock_sock_nested(sk);
561 else
562 bh_lock_sock(sk);
a5b5bb9a
IM		 563 if (!sock_owned_by_user(sk)) {
564 /*
565 * trylock + unlock semantics:
566 */
567 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
568
c57943a1 569 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a 570
5facae4f 571 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
8265792b 572 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
8eae939f
ZY		 573 bh_unlock_sock(sk);
574 atomic_inc(&sk->sk_drops);
575 goto discard_and_relse;
576 }
577
f0088a50
DV		 578 bh_unlock_sock(sk);
579out:
c3f24cfb
ED		 580 if (refcounted)
581 sock_put(sk);
f0088a50
DV		 582 return rc;
583discard_and_relse:
584 kfree_skb(skb);
585 goto out;
586}
4f0c40d9 587EXPORT_SYMBOL(__sk_receive_skb);
f0088a50 588
bbd807df
BV		 589INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
590 u32));
591INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
592 u32));
f0088a50
DV		 593struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
594{
b6c6712a 595 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50 596
bbd807df
BV		 597 if (dst && dst->obsolete &&
598 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
599 dst, cookie) == NULL) {
e022f0b4 600 sk_tx_queue_clear(sk);
eb44ad4e 601 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
a9b3cd7f 602 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
f0088a50
DV		 603 dst_release(dst);
604 return NULL;
605 }
606
607 return dst;
608}
609EXPORT_SYMBOL(__sk_dst_check);
610
611struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
612{
613 struct dst_entry *dst = sk_dst_get(sk);
614
bbd807df
BV		 615 if (dst && dst->obsolete &&
616 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
617 dst, cookie) == NULL) {
f0088a50
DV		 618 sk_dst_reset(sk);
619 dst_release(dst);
620 return NULL;
621 }
622
623 return dst;
624}
625EXPORT_SYMBOL(sk_dst_check);
626
7594888c 627static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
4878809f
DM		 628{
629 int ret = -ENOPROTOOPT;
630#ifdef CONFIG_NETDEVICES
3b1e0a65 631 struct net *net = sock_net(sk);
4878809f
DM		 632
633 /* Sorry... */
634 ret = -EPERM;
c427bfec 635 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
4878809f
DM		 636 goto out;
637
f5dd3d0c
DH		 638 ret = -EINVAL;
639 if (ifindex < 0)
640 goto out;
641
e5fccaa1
ED		 642 /* Paired with all READ_ONCE() done locklessly. */
643 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
644
f5dd3d0c
DH		 645 if (sk->sk_prot->rehash)
646 sk->sk_prot->rehash(sk);
647 sk_dst_reset(sk);
648
649 ret = 0;
650
651out:
652#endif
653
654 return ret;
655}
656
8ea204c2 657int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
7594888c
CH		 658{
659 int ret;
660
8ea204c2
FF		 661 if (lock_sk)
662 lock_sock(sk);
7594888c 663 ret = sock_bindtoindex_locked(sk, ifindex);
8ea204c2
FF		 664 if (lock_sk)
665 release_sock(sk);
7594888c
CH		 666
667 return ret;
668}
669EXPORT_SYMBOL(sock_bindtoindex);
670
5790642b 671static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
f5dd3d0c
DH		 672{
673 int ret = -ENOPROTOOPT;
674#ifdef CONFIG_NETDEVICES
675 struct net *net = sock_net(sk);
676 char devname[IFNAMSIZ];
677 int index;
678
4878809f
DM		 679 ret = -EINVAL;
680 if (optlen < 0)
681 goto out;
682
683 /* Bind this socket to a particular device like "eth0",
684 * as specified in the passed interface name. If the
685 * name is "" or the option length is zero the socket
686 * is not bound.
687 */
688 if (optlen > IFNAMSIZ - 1)
689 optlen = IFNAMSIZ - 1;
690 memset(devname, 0, sizeof(devname));
691
692 ret = -EFAULT;
5790642b 693 if (copy_from_sockptr(devname, optval, optlen))
4878809f
DM		 694 goto out;
695
000ba2e4
DM		 696 index = 0;
697 if (devname[0] != '\0') {
bf8e56bf 698 struct net_device *dev;
4878809f 699
bf8e56bf
ED		 700 rcu_read_lock();
701 dev = dev_get_by_name_rcu(net, devname);
702 if (dev)
703 index = dev->ifindex;
704 rcu_read_unlock();
4878809f
DM		 705 ret = -ENODEV;
706 if (!dev)
707 goto out;
4878809f
DM		 708 }
709
24426654
MKL		 710 sockopt_lock_sock(sk);
711 ret = sock_bindtoindex_locked(sk, index);
712 sockopt_release_sock(sk);
4878809f
DM		 713out:
714#endif
715
716 return ret;
717}
718
4ff09db1
MKL		 719static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
720 sockptr_t optlen, int len)
c91f6df2
BH		 721{
722 int ret = -ENOPROTOOPT;
723#ifdef CONFIG_NETDEVICES
e5fccaa1 724 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
c91f6df2 725 struct net *net = sock_net(sk);
c91f6df2 726 char devname[IFNAMSIZ];
c91f6df2 727
e5fccaa1 728 if (bound_dev_if == 0) {
c91f6df2
BH		 729 len = 0;
730 goto zero;
731 }
732
733 ret = -EINVAL;
734 if (len < IFNAMSIZ)
735 goto out;
736
e5fccaa1 737 ret = netdev_get_name(net, devname, bound_dev_if);
5dbe7c17 738 if (ret)
c91f6df2 739 goto out;
c91f6df2
BH		 740
741 len = strlen(devname) + 1;
742
743 ret = -EFAULT;
4ff09db1 744 if (copy_to_sockptr(optval, devname, len))
c91f6df2
BH		 745 goto out;
746
747zero:
748 ret = -EFAULT;
4ff09db1 749 if (copy_to_sockptr(optlen, &len, sizeof(int)))
c91f6df2
BH		 750 goto out;
751
752 ret = 0;
753
754out:
755#endif
756
757 return ret;
758}
759
d986f521 760bool sk_mc_loop(const struct sock *sk)
f60e5990 761{
762 if (dev_recursion_level())
763 return false;
764 if (!sk)
765 return true;
a3e0fdf7
ED		 766 /* IPV6_ADDRFORM can change sk->sk_family under us. */
767 switch (READ_ONCE(sk->sk_family)) {
f60e5990 768 case AF_INET:
b09bde5c 769 return inet_test_bit(MC_LOOP, sk);
f60e5990 770#if IS_ENABLED(CONFIG_IPV6)
771 case AF_INET6:
d986f521 772 return inet6_test_bit(MC6_LOOP, sk);
f60e5990 773#endif
774 }
0ad6f6e7 775 WARN_ON_ONCE(1);
f60e5990 776 return true;
777}
778EXPORT_SYMBOL(sk_mc_loop);
779
b58f0e8f
CH		 780void sock_set_reuseaddr(struct sock *sk)
781{
782 lock_sock(sk);
783 sk->sk_reuse = SK_CAN_REUSE;
784 release_sock(sk);
785}
786EXPORT_SYMBOL(sock_set_reuseaddr);
787
fe31a326
CH		 788void sock_set_reuseport(struct sock *sk)
789{
790 lock_sock(sk);
791 sk->sk_reuseport = true;
792 release_sock(sk);
793}
794EXPORT_SYMBOL(sock_set_reuseport);
795
c433594c
CH		 796void sock_no_linger(struct sock *sk)
797{
798 lock_sock(sk);
bc1fb82a 799 WRITE_ONCE(sk->sk_lingertime, 0);
c433594c
CH		 800 sock_set_flag(sk, SOCK_LINGER);
801 release_sock(sk);
802}
803EXPORT_SYMBOL(sock_no_linger);
804
6e434967
CH		 805void sock_set_priority(struct sock *sk, u32 priority)
806{
8bf43be7 807 WRITE_ONCE(sk->sk_priority, priority);
6e434967
CH		 808}
809EXPORT_SYMBOL(sock_set_priority);
810
76ee0785
CH		 811void sock_set_sndtimeo(struct sock *sk, s64 secs)
812{
813 lock_sock(sk);
814 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
285975dd 815 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
76ee0785 816 else
285975dd 817 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
76ee0785
CH		 818 release_sock(sk);
819}
820EXPORT_SYMBOL(sock_set_sndtimeo);
821
783da70e
CH		 822static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
823{
824 if (val) {
825 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
826 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
827 sock_set_flag(sk, SOCK_RCVTSTAMP);
828 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
829 } else {
830 sock_reset_flag(sk, SOCK_RCVTSTAMP);
831 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
783da70e
CH		 832 }
833}
834
835void sock_enable_timestamps(struct sock *sk)
836{
837 lock_sock(sk);
838 __sock_set_timestamps(sk, true, false, true);
839 release_sock(sk);
840}
841EXPORT_SYMBOL(sock_enable_timestamps);
842
371087aa
FW		 843void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
844{
845 switch (optname) {
846 case SO_TIMESTAMP_OLD:
847 __sock_set_timestamps(sk, valbool, false, false);
848 break;
849 case SO_TIMESTAMP_NEW:
850 __sock_set_timestamps(sk, valbool, true, false);
851 break;
852 case SO_TIMESTAMPNS_OLD:
853 __sock_set_timestamps(sk, valbool, false, true);
854 break;
855 case SO_TIMESTAMPNS_NEW:
856 __sock_set_timestamps(sk, valbool, true, true);
857 break;
858 }
859}
860
d463126e
YL		 861static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
862{
863 struct net *net = sock_net(sk);
864 struct net_device *dev = NULL;
865 bool match = false;
866 int *vclock_index;
867 int i, num;
868
869 if (sk->sk_bound_dev_if)
870 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
871
872 if (!dev) {
873 pr_err("%s: sock not bind to device\n", __func__);
874 return -EOPNOTSUPP;
875 }
876
877 num = ethtool_get_phc_vclocks(dev, &vclock_index);
2a4d75bf
ML		 878 dev_put(dev);
879
d463126e
YL		 880 for (i = 0; i < num; i++) {
881 if (*(vclock_index + i) == phc_index) {
882 match = true;
883 break;
884 }
885 }
886
887 if (num > 0)
888 kfree(vclock_index);
889
890 if (!match)
891 return -EINVAL;
892
251cd405 893 WRITE_ONCE(sk->sk_bind_phc, phc_index);
d463126e
YL		 894
895 return 0;
896}
897
898int sock_set_timestamping(struct sock *sk, int optname,
899 struct so_timestamping timestamping)
ced122d9 900{
d463126e
YL		 901 int val = timestamping.flags;
902 int ret;
903
ced122d9
FW		 904 if (val & ~SOF_TIMESTAMPING_MASK)
905 return -EINVAL;
906
b534dc46
WB		 907 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
908 !(val & SOF_TIMESTAMPING_OPT_ID))
909 return -EINVAL;
910
ced122d9
FW		 911 if (val & SOF_TIMESTAMPING_OPT_ID &&
912 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
42f67eea 913 if (sk_is_tcp(sk)) {
ced122d9
FW		 914 if ((1 << sk->sk_state) &
915 (TCPF_CLOSE | TCPF_LISTEN))
916 return -EINVAL;
b534dc46
WB		 917 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
918 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
919 else
920 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
ced122d9 921 } else {
a1cdec57 922 atomic_set(&sk->sk_tskey, 0);
ced122d9
FW		 923 }
924 }
925
926 if (val & SOF_TIMESTAMPING_OPT_STATS &&
927 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
928 return -EINVAL;
929
d463126e
YL		 930 if (val & SOF_TIMESTAMPING_BIND_PHC) {
931 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
932 if (ret)
933 return ret;
934 }
935
e3390b30 936 WRITE_ONCE(sk->sk_tsflags, val);
ced122d9
FW		 937 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
938
939 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
940 sock_enable_timestamp(sk,
941 SOCK_TIMESTAMPING_RX_SOFTWARE);
942 else
943 sock_disable_timestamp(sk,
944 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
945 return 0;
946}
947
ce3d9544
CH		 948void sock_set_keepalive(struct sock *sk)
949{
950 lock_sock(sk);
951 if (sk->sk_prot->keepalive)
952 sk->sk_prot->keepalive(sk, true);
953 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
954 release_sock(sk);
955}
956EXPORT_SYMBOL(sock_set_keepalive);
957
26cfabf9
CH		 958static void __sock_set_rcvbuf(struct sock *sk, int val)
959{
960 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
961 * as a negative value.
962 */
963 val = min_t(int, val, INT_MAX / 2);
964 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
965
966 /* We double it on the way in to account for "struct sk_buff" etc.
967 * overhead. Applications assume that the SO_RCVBUF setting they make
968 * will allow that much actual data to be received on that socket.
969 *
970 * Applications are unaware that "struct sk_buff" and other overheads
971 * allocate from the receive buffer during socket buffer allocation.
972 *
973 * And after considering the possible alternatives, returning the value
974 * we actually used in getsockopt is the most desirable behavior.
975 */
976 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
977}
978
979void sock_set_rcvbuf(struct sock *sk, int val)
980{
981 lock_sock(sk);
982 __sock_set_rcvbuf(sk, val);
983 release_sock(sk);
984}
985EXPORT_SYMBOL(sock_set_rcvbuf);
986
dd9082f4
AA		 987static void __sock_set_mark(struct sock *sk, u32 val)
988{
989 if (val != sk->sk_mark) {
3c5b4d69 990 WRITE_ONCE(sk->sk_mark, val);
dd9082f4
AA		 991 sk_dst_reset(sk);
992 }
993}
994
84d1c617
AA		 995void sock_set_mark(struct sock *sk, u32 val)
996{
997 lock_sock(sk);
dd9082f4 998 __sock_set_mark(sk, val);
84d1c617
AA		 999 release_sock(sk);
1000}
1001EXPORT_SYMBOL(sock_set_mark);
1002
2bb2f5fb
WW		 1003static void sock_release_reserved_memory(struct sock *sk, int bytes)
1004{
1005 /* Round down bytes to multiple of pages */
100fdd1f 1006 bytes = round_down(bytes, PAGE_SIZE);
2bb2f5fb
WW		 1007
1008 WARN_ON(bytes > sk->sk_reserved_mem);
fe11fdcb 1009 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
2bb2f5fb
WW		 1010 sk_mem_reclaim(sk);
1011}
1012
1013static int sock_reserve_memory(struct sock *sk, int bytes)
1014{
1015 long allocated;
1016 bool charged;
1017 int pages;
1018
d00c8ee3 1019 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
2bb2f5fb
WW		 1020 return -EOPNOTSUPP;
1021
1022 if (!bytes)
1023 return 0;
1024
1025 pages = sk_mem_pages(bytes);
1026
1027 /* pre-charge to memcg */
1028 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1029 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1030 if (!charged)
1031 return -ENOMEM;
1032
1033 /* pre-charge to forward_alloc */
219160be
ED		 1034 sk_memory_allocated_add(sk, pages);
1035 allocated = sk_memory_allocated(sk);
2bb2f5fb
WW		 1036 /* If the system goes into memory pressure with this
1037 * precharge, give up and return error.
1038 */
1039 if (allocated > sk_prot_mem_limits(sk, 1)) {
1040 sk_memory_allocated_sub(sk, pages);
1041 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1042 return -ENOMEM;
1043 }
5e6300e7 1044 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
2bb2f5fb 1045
fe11fdcb
ED		 1046 WRITE_ONCE(sk->sk_reserved_mem,
1047 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
2bb2f5fb
WW		 1048
1049 return 0;
1050}
1051
24426654
MKL		 1052void sockopt_lock_sock(struct sock *sk)
1053{
1054 /* When current->bpf_ctx is set, the setsockopt is called from
1055 * a bpf prog. bpf has ensured the sk lock has been
1056 * acquired before calling setsockopt().
1057 */
1058 if (has_current_bpf_ctx())
1059 return;
1060
1061 lock_sock(sk);
1062}
1063EXPORT_SYMBOL(sockopt_lock_sock);
1064
1065void sockopt_release_sock(struct sock *sk)
1066{
1067 if (has_current_bpf_ctx())
1068 return;
1069
1070 release_sock(sk);
1071}
1072EXPORT_SYMBOL(sockopt_release_sock);
1073
e42c7bee
MKL		 1074bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1075{
1076 return has_current_bpf_ctx() || ns_capable(ns, cap);
1077}
1078EXPORT_SYMBOL(sockopt_ns_capable);
1079
1080bool sockopt_capable(int cap)
1081{
1082 return has_current_bpf_ctx() || capable(cap);
1083}
1084EXPORT_SYMBOL(sockopt_capable);
1085
1da177e4
LT		 1086/*
1087 * This is meant for all protocols to use and covers goings on
1088 * at the socket level. Everything here is generic.
1089 */
1090
29003875
MKL		 1091int sk_setsockopt(struct sock *sk, int level, int optname,
1092 sockptr_t optval, unsigned int optlen)
1da177e4 1093{
d463126e 1094 struct so_timestamping timestamping;
4d748f99 1095 struct socket *sock = sk->sk_socket;
80b14dee 1096 struct sock_txtime sk_txtime;
1da177e4
LT		 1097 int val;
1098 int valbool;
1099 struct linger ling;
1100 int ret = 0;
4ec93edb 1101
1da177e4
LT		 1102 /*
1103 * Options without arguments
1104 */
1105
4878809f 1106 if (optname == SO_BINDTODEVICE)
c91f6df2 1107 return sock_setbindtodevice(sk, optval, optlen);
4878809f 1108
e71a4783
SH		 1109 if (optlen < sizeof(int))
1110 return -EINVAL;
4ec93edb 1111
c8c1bbb6 1112 if (copy_from_sockptr(&val, optval, sizeof(val)))
1da177e4 1113 return -EFAULT;
4ec93edb 1114
2a91525c 1115 valbool = val ? 1 : 0;
1da177e4 1116
10bbf165
ED		 1117 /* handle options which do not require locking the socket. */
1118 switch (optname) {
1119 case SO_PRIORITY:
1120 if ((val >= 0 && val <= 6) ||
1121 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1122 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1123 sock_set_priority(sk, val);
1124 return 0;
1125 }
1126 return -EPERM;
8ebfb6db
ED		 1127 case SO_PASSSEC:
1128 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1129 return 0;
1130 case SO_PASSCRED:
1131 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1132 return 0;
1133 case SO_PASSPIDFD:
1134 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1135 return 0;
b1202515
ED		 1136 case SO_TYPE:
1137 case SO_PROTOCOL:
1138 case SO_DOMAIN:
1139 case SO_ERROR:
1140 return -ENOPROTOOPT;
2a4319cf
ED		 1141#ifdef CONFIG_NET_RX_BUSY_POLL
1142 case SO_BUSY_POLL:
1143 if (val < 0)
1144 return -EINVAL;
1145 WRITE_ONCE(sk->sk_ll_usec, val);
1146 return 0;
1147 case SO_PREFER_BUSY_POLL:
1148 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1149 return -EPERM;
1150 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1151 return 0;
1152 case SO_BUSY_POLL_BUDGET:
1153 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1154 !sockopt_capable(CAP_NET_ADMIN))
1155 return -EPERM;
1156 if (val < 0 || val > U16_MAX)
1157 return -EINVAL;
1158 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1159 return 0;
1160#endif
28b24f90
ED		 1161 case SO_MAX_PACING_RATE:
1162 {
1163 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1164 unsigned long pacing_rate;
1165
1166 if (sizeof(ulval) != sizeof(val) &&
1167 optlen >= sizeof(ulval) &&
1168 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1169 return -EFAULT;
1170 }
1171 if (ulval != ~0UL)
1172 cmpxchg(&sk->sk_pacing_status,
1173 SK_PACING_NONE,
1174 SK_PACING_NEEDED);
1175 /* Pairs with READ_ONCE() from sk_getsockopt() */
1176 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1177 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1178 if (ulval < pacing_rate)
1179 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1180 return 0;
1181 }
5eef0b8d
ED		 1182 case SO_TXREHASH:
1183 if (val < -1 || val > 1)
1184 return -EINVAL;
1185 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1186 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1187 /* Paired with READ_ONCE() in tcp_rtx_synack()
1188 * and sk_getsockopt().
1189 */
1190 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1191 return 0;
56667da7
ED		 1192 case SO_PEEK_OFF:
1193 {
1194 int (*set_peek_off)(struct sock *sk, int val);
1195
1196 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1197 if (set_peek_off)
1198 ret = set_peek_off(sk, val);
1199 else
1200 ret = -EOPNOTSUPP;
1201 return ret;
1202 }
10bbf165
ED		 1203 }
1204
24426654 1205 sockopt_lock_sock(sk);
1da177e4 1206
2a91525c 1207 switch (optname) {
e71a4783 1208 case SO_DEBUG:
e42c7bee 1209 if (val && !sockopt_capable(CAP_NET_ADMIN))
e71a4783 1210 ret = -EACCES;
2a91525c 1211 else
c0ef877b 1212 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 1213 break;
1214 case SO_REUSEADDR:
cdb8744d 1215 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
e71a4783 1216 break;
055dc21a
TH		 1217 case SO_REUSEPORT:
1218 sk->sk_reuseport = valbool;
1219 break;
e71a4783 1220 case SO_DONTROUTE:
c0ef877b 1221 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
0fbe82e6 1222 sk_dst_reset(sk);
e71a4783
SH		 1223 break;
1224 case SO_BROADCAST:
1225 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1226 break;
1227 case SO_SNDBUF:
1228 /* Don't error on this BSD doesn't and if you think
82981930
ED		 1229 * about it this is right. Otherwise apps have to
1230 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1231 * are treated in BSD as hints
1232 */
1227c177 1233 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
b0573dea 1234set_sndbuf:
4057765f
GN		 1235 /* Ensure val * 2 fits into an int, to prevent max_t()
1236 * from treating it as a negative value.
1237 */
1238 val = min_t(int, val, INT_MAX / 2);
e71a4783 1239 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
e292f05e
ED		 1240 WRITE_ONCE(sk->sk_sndbuf,
1241 max_t(int, val * 2, SOCK_MIN_SNDBUF));
82981930 1242 /* Wake up sending tasks if we upped the value. */
e71a4783
SH		 1243 sk->sk_write_space(sk);
1244 break;
1da177e4 1245
e71a4783 1246 case SO_SNDBUFFORCE:
e42c7bee 1247 if (!sockopt_capable(CAP_NET_ADMIN)) {
e71a4783
SH		 1248 ret = -EPERM;
1249 break;
1250 }
4057765f
GN		 1251
1252 /* No negative values (to prevent underflow, as val will be
1253 * multiplied by 2).
1254 */
1255 if (val < 0)
1256 val = 0;
e71a4783 1257 goto set_sndbuf;
b0573dea 1258
e71a4783
SH		 1259 case SO_RCVBUF:
1260 /* Don't error on this BSD doesn't and if you think
82981930
ED		 1261 * about it this is right. Otherwise apps have to
1262 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1263 * are treated in BSD as hints
1264 */
1227c177 1265 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
e71a4783
SH		 1266 break;
1267
1268 case SO_RCVBUFFORCE:
e42c7bee 1269 if (!sockopt_capable(CAP_NET_ADMIN)) {
e71a4783 1270 ret = -EPERM;
1da177e4 1271 break;
e71a4783 1272 }
4057765f
GN		 1273
1274 /* No negative values (to prevent underflow, as val will be
1275 * multiplied by 2).
1276 */
26cfabf9
CH		 1277 __sock_set_rcvbuf(sk, max(val, 0));
1278 break;
1da177e4 1279
e71a4783 1280 case SO_KEEPALIVE:
4b9d07a4
UB		 1281 if (sk->sk_prot->keepalive)
1282 sk->sk_prot->keepalive(sk, valbool);
e71a4783
SH		 1283 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1284 break;
1285
1286 case SO_OOBINLINE:
1287 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1288 break;
1289
1290 case SO_NO_CHECK:
28448b80 1291 sk->sk_no_check_tx = valbool;
e71a4783
SH		 1292 break;
1293
e71a4783
SH		 1294 case SO_LINGER:
1295 if (optlen < sizeof(ling)) {
1296 ret = -EINVAL; /* 1003.1g */
1da177e4 1297 break;
e71a4783 1298 }
c8c1bbb6 1299 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
e71a4783 1300 ret = -EFAULT;
1da177e4 1301 break;
e71a4783 1302 }
bc1fb82a 1303 if (!ling.l_onoff) {
e71a4783 1304 sock_reset_flag(sk, SOCK_LINGER);
bc1fb82a
ED		 1305 } else {
1306 unsigned long t_sec = ling.l_linger;
1307
1308 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1309 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1da177e4 1310 else
bc1fb82a 1311 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
e71a4783
SH		 1312 sock_set_flag(sk, SOCK_LINGER);
1313 }
1314 break;
1315
1316 case SO_BSDCOMPAT:
e71a4783
SH		 1317 break;
1318
7f1bc6e9 1319 case SO_TIMESTAMP_OLD:
887feae3 1320 case SO_TIMESTAMP_NEW:
7f1bc6e9 1321 case SO_TIMESTAMPNS_OLD:
887feae3 1322 case SO_TIMESTAMPNS_NEW:
81b4a0cc 1323 sock_set_timestamp(sk, optname, valbool);
e71a4783 1324 break;
ced122d9 1325
9718475e 1326 case SO_TIMESTAMPING_NEW:
7f1bc6e9 1327 case SO_TIMESTAMPING_OLD:
d463126e
YL		 1328 if (optlen == sizeof(timestamping)) {
1329 if (copy_from_sockptr(&timestamping, optval,
271dbc31
DC		 1330 sizeof(timestamping))) {
1331 ret = -EFAULT;
1332 break;
1333 }
d463126e
YL		 1334 } else {
1335 memset(&timestamping, 0, sizeof(timestamping));
1336 timestamping.flags = val;
1337 }
1338 ret = sock_set_timestamping(sk, optname, timestamping);
20d49473
PO		 1339 break;
1340
e71a4783 1341 case SO_RCVLOWAT:
1ded5e5a
ED		 1342 {
1343 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1344
e71a4783
SH		 1345 if (val < 0)
1346 val = INT_MAX;
1ded5e5a
ED		 1347 if (sock)
1348 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1349 if (set_rcvlowat)
1350 ret = set_rcvlowat(sk, val);
d1361840 1351 else
eac66402 1352 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
e71a4783 1353 break;
1ded5e5a 1354 }
45bdc661 1355 case SO_RCVTIMEO_OLD:
a9beb86a 1356 case SO_RCVTIMEO_NEW:
c8c1bbb6 1357 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
c34645ac 1358 optlen, optname == SO_RCVTIMEO_OLD);
e71a4783
SH		 1359 break;
1360
45bdc661 1361 case SO_SNDTIMEO_OLD:
a9beb86a 1362 case SO_SNDTIMEO_NEW:
c8c1bbb6 1363 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
c34645ac 1364 optlen, optname == SO_SNDTIMEO_OLD);
e71a4783 1365 break;
1da177e4 1366
4d295e54
CH		 1367 case SO_ATTACH_FILTER: {
1368 struct sock_fprog fprog;
e71a4783 1369
c8c1bbb6 1370 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
4d295e54 1371 if (!ret)
e71a4783 1372 ret = sk_attach_filter(&fprog, sk);
e71a4783 1373 break;
4d295e54 1374 }
89aa0758
AS		 1375 case SO_ATTACH_BPF:
1376 ret = -EINVAL;
1377 if (optlen == sizeof(u32)) {
1378 u32 ufd;
1379
1380 ret = -EFAULT;
c8c1bbb6 1381 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
89aa0758
AS		 1382 break;
1383
1384 ret = sk_attach_bpf(ufd, sk);
1385 }
1386 break;
1387
4d295e54
CH		 1388 case SO_ATTACH_REUSEPORT_CBPF: {
1389 struct sock_fprog fprog;
538950a1 1390
c8c1bbb6 1391 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
4d295e54 1392 if (!ret)
538950a1 1393 ret = sk_reuseport_attach_filter(&fprog, sk);
538950a1 1394 break;
4d295e54 1395 }
538950a1
CG		 1396 case SO_ATTACH_REUSEPORT_EBPF:
1397 ret = -EINVAL;
1398 if (optlen == sizeof(u32)) {
1399 u32 ufd;
1400
1401 ret = -EFAULT;
c8c1bbb6 1402 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
538950a1
CG		 1403 break;
1404
1405 ret = sk_reuseport_attach_bpf(ufd, sk);
1406 }
1407 break;
1408
99f3a064
MKL		 1409 case SO_DETACH_REUSEPORT_BPF:
1410 ret = reuseport_detach_prog(sk);
1411 break;
1412
e71a4783 1413 case SO_DETACH_FILTER:
55b33325 1414 ret = sk_detach_filter(sk);
e71a4783 1415 break;
1da177e4 1416
d59577b6
VB		 1417 case SO_LOCK_FILTER:
1418 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1419 ret = -EPERM;
1420 else
1421 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1422 break;
1423
4a19ec58 1424 case SO_MARK:
e42c7bee
MKL		 1425 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1426 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
4a19ec58 1427 ret = -EPERM;
dd9082f4 1428 break;
50254256 1429 }
dd9082f4
AA		 1430
1431 __sock_set_mark(sk, val);
4a19ec58 1432 break;
6fd1d51c
EM		 1433 case SO_RCVMARK:
1434 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1435 break;
877ce7c1 1436
3b885787 1437 case SO_RXQ_OVFL:
8083f0fc 1438 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
3b885787 1439 break;
6e3e939f
JB		 1440
1441 case SO_WIFI_STATUS:
1442 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1443 break;
1444
3bdc0eba
BG		 1445 case SO_NOFCS:
1446 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1447 break;
1448
7d4c04fc
KJ		 1449 case SO_SELECT_ERR_QUEUE:
1450 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1451 break;
1452
62748f32 1453
70da268b 1454 case SO_INCOMING_CPU:
b261eda8 1455 reuseport_update_incoming_cpu(sk, val);
70da268b
ED		 1456 break;
1457
a87cb3e4
TH		 1458 case SO_CNX_ADVICE:
1459 if (val == 1)
1460 dst_negative_advice(sk);
1461 break;
76851d12
WB		 1462
1463 case SO_ZEROCOPY:
28190752 1464 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
42f67eea 1465 if (!(sk_is_tcp(sk) ||
b5947e5d
WB		 1466 (sk->sk_type == SOCK_DGRAM &&
1467 sk->sk_protocol == IPPROTO_UDP)))
869420a8 1468 ret = -EOPNOTSUPP;
28190752 1469 } else if (sk->sk_family != PF_RDS) {
869420a8 1470 ret = -EOPNOTSUPP;
28190752
SV		 1471 }
1472 if (!ret) {
1473 if (val < 0 || val > 1)
1474 ret = -EINVAL;
1475 else
1476 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
28190752 1477 }
334e6413
JSP		 1478 break;
1479
80b14dee 1480 case SO_TXTIME:
790709f2 1481 if (optlen != sizeof(struct sock_txtime)) {
80b14dee 1482 ret = -EINVAL;
790709f2 1483 break;
c8c1bbb6 1484 } else if (copy_from_sockptr(&sk_txtime, optval,
80b14dee
RC		 1485 sizeof(struct sock_txtime))) {
1486 ret = -EFAULT;
790709f2 1487 break;
80b14dee
RC		 1488 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1489 ret = -EINVAL;
790709f2
ED		 1490 break;
1491 }
1492 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1493 * scheduler has enough safe guards.
1494 */
1495 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
e42c7bee 1496 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
790709f2
ED		 1497 ret = -EPERM;
1498 break;
80b14dee 1499 }
790709f2
ED		 1500 sock_valbool_flag(sk, SOCK_TXTIME, true);
1501 sk->sk_clockid = sk_txtime.clockid;
1502 sk->sk_txtime_deadline_mode =
1503 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1504 sk->sk_txtime_report_errors =
1505 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
80b14dee
RC		 1506 break;
1507
f5dd3d0c 1508 case SO_BINDTOIFINDEX:
7594888c 1509 ret = sock_bindtoindex_locked(sk, val);
f5dd3d0c
DH		 1510 break;
1511
04190bf8
PT		 1512 case SO_BUF_LOCK:
1513 if (val & ~SOCK_BUF_LOCK_MASK) {
1514 ret = -EINVAL;
1515 break;
1516 }
1517 sk->sk_userlocks = val | (sk->sk_userlocks &
1518 ~SOCK_BUF_LOCK_MASK);
1519 break;
1520
2bb2f5fb
WW		 1521 case SO_RESERVE_MEM:
1522 {
1523 int delta;
1524
1525 if (val < 0) {
1526 ret = -EINVAL;
1527 break;
1528 }
1529
1530 delta = val - sk->sk_reserved_mem;
1531 if (delta < 0)
1532 sock_release_reserved_memory(sk, -delta);
1533 else
1534 ret = sock_reserve_memory(sk, delta);
1535 break;
1536 }
1537
e71a4783
SH		 1538 default:
1539 ret = -ENOPROTOOPT;
1540 break;
4ec93edb 1541 }
24426654 1542 sockopt_release_sock(sk);
1da177e4
LT		 1543 return ret;
1544}
4d748f99
MKL		 1545
1546int sock_setsockopt(struct socket *sock, int level, int optname,
1547 sockptr_t optval, unsigned int optlen)
1548{
1549 return sk_setsockopt(sock->sk, level, optname,
1550 optval, optlen);
1551}
2a91525c 1552EXPORT_SYMBOL(sock_setsockopt);
1da177e4 1553
35306eb2
ED		 1554static const struct cred *sk_get_peer_cred(struct sock *sk)
1555{
1556 const struct cred *cred;
1557
1558 spin_lock(&sk->sk_peer_lock);
1559 cred = get_cred(sk->sk_peer_cred);
1560 spin_unlock(&sk->sk_peer_lock);
1561
1562 return cred;
1563}
1da177e4 1564
8f09898b 1565static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1566 struct ucred *ucred)
3f551f94
EB		 1567{
1568 ucred->pid = pid_vnr(pid);
1569 ucred->uid = ucred->gid = -1;
1570 if (cred) {
1571 struct user_namespace *current_ns = current_user_ns();
1572
b2e4f544
EB		 1573 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1574 ucred->gid = from_kgid_munged(current_ns, cred->egid);
3f551f94
EB		 1575 }
1576}
1577
4ff09db1 1578static int groups_to_user(sockptr_t dst, const struct group_info *src)
28b5ba2a
DH		 1579{
1580 struct user_namespace *user_ns = current_user_ns();
1581 int i;
1582
4ff09db1
MKL		 1583 for (i = 0; i < src->ngroups; i++) {
1584 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1585
1586 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
28b5ba2a 1587 return -EFAULT;
4ff09db1 1588 }
28b5ba2a
DH		 1589
1590 return 0;
1591}
1592
65ddc82d
MKL		 1593int sk_getsockopt(struct sock *sk, int level, int optname,
1594 sockptr_t optval, sockptr_t optlen)
1da177e4 1595{
ba74a760 1596 struct socket *sock = sk->sk_socket;
4ec93edb 1597
e71a4783 1598 union {
4ec93edb 1599 int val;
5daab9db 1600 u64 val64;
677f136c 1601 unsigned long ulval;
4ec93edb 1602 struct linger ling;
fe0c72f3
AB		 1603 struct old_timeval32 tm32;
1604 struct __kernel_old_timeval tm;
a9beb86a 1605 struct __kernel_sock_timeval stm;
80b14dee 1606 struct sock_txtime txtime;
d463126e 1607 struct so_timestamping timestamping;
1da177e4 1608 } v;
4ec93edb 1609
4d0392be 1610 int lv = sizeof(int);
1da177e4 1611 int len;
4ec93edb 1612
4ff09db1 1613 if (copy_from_sockptr(&len, optlen, sizeof(int)))
4ec93edb 1614 return -EFAULT;
e71a4783 1615 if (len < 0)
1da177e4 1616 return -EINVAL;
4ec93edb 1617
50fee1de 1618 memset(&v, 0, sizeof(v));
df0bca04 1619
2a91525c 1620 switch (optname) {
e71a4783
SH		 1621 case SO_DEBUG:
1622 v.val = sock_flag(sk, SOCK_DBG);
1623 break;
1624
1625 case SO_DONTROUTE:
1626 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1627 break;
1628
1629 case SO_BROADCAST:
1b23a5df 1630 v.val = sock_flag(sk, SOCK_BROADCAST);
e71a4783
SH		 1631 break;
1632
1633 case SO_SNDBUF:
74bc0843 1634 v.val = READ_ONCE(sk->sk_sndbuf);
e71a4783
SH		 1635 break;
1636
1637 case SO_RCVBUF:
b4b55325 1638 v.val = READ_ONCE(sk->sk_rcvbuf);
e71a4783
SH		 1639 break;
1640
1641 case SO_REUSEADDR:
1642 v.val = sk->sk_reuse;
1643 break;
1644
055dc21a
TH		 1645 case SO_REUSEPORT:
1646 v.val = sk->sk_reuseport;
1647 break;
1648
e71a4783 1649 case SO_KEEPALIVE:
1b23a5df 1650 v.val = sock_flag(sk, SOCK_KEEPOPEN);
e71a4783
SH		 1651 break;
1652
1653 case SO_TYPE:
1654 v.val = sk->sk_type;
1655 break;
1656
49c794e9
JE		 1657 case SO_PROTOCOL:
1658 v.val = sk->sk_protocol;
1659 break;
1660
0d6038ee
JE		 1661 case SO_DOMAIN:
1662 v.val = sk->sk_family;
1663 break;
1664
e71a4783
SH		 1665 case SO_ERROR:
1666 v.val = -sock_error(sk);
2a91525c 1667 if (v.val == 0)
e71a4783
SH		 1668 v.val = xchg(&sk->sk_err_soft, 0);
1669 break;
1670
1671 case SO_OOBINLINE:
1b23a5df 1672 v.val = sock_flag(sk, SOCK_URGINLINE);
e71a4783
SH		 1673 break;
1674
1675 case SO_NO_CHECK:
28448b80 1676 v.val = sk->sk_no_check_tx;
e71a4783
SH		 1677 break;
1678
1679 case SO_PRIORITY:
8bf43be7 1680 v.val = READ_ONCE(sk->sk_priority);
e71a4783
SH		 1681 break;
1682
1683 case SO_LINGER:
1684 lv = sizeof(v.ling);
1b23a5df 1685 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
bc1fb82a 1686 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
e71a4783
SH		 1687 break;
1688
1689 case SO_BSDCOMPAT:
e71a4783
SH		 1690 break;
1691
7f1bc6e9 1692 case SO_TIMESTAMP_OLD:
92f37fd2 1693 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
887feae3 1694 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
92f37fd2
ED		 1695 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1696 break;
1697
7f1bc6e9 1698 case SO_TIMESTAMPNS_OLD:
887feae3
DD		 1699 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1700 break;
1701
1702 case SO_TIMESTAMP_NEW:
1703 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1704 break;
1705
1706 case SO_TIMESTAMPNS_NEW:
1707 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
e71a4783
SH		 1708 break;
1709
7f1bc6e9 1710 case SO_TIMESTAMPING_OLD:
7f6ca95d 1711 case SO_TIMESTAMPING_NEW:
d463126e 1712 lv = sizeof(v.timestamping);
7f6ca95d
JTH		 1713 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1714 * returning the flags when they were set through the same option.
1715 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1716 */
1717 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1718 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1719 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1720 }
20d49473
PO		 1721 break;
1722
a9beb86a
DD		 1723 case SO_RCVTIMEO_OLD:
1724 case SO_RCVTIMEO_NEW:
285975dd
ED		 1725 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1726 SO_RCVTIMEO_OLD == optname);
e71a4783
SH		 1727 break;
1728
a9beb86a
DD		 1729 case SO_SNDTIMEO_OLD:
1730 case SO_SNDTIMEO_NEW:
285975dd
ED		 1731 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1732 SO_SNDTIMEO_OLD == optname);
e71a4783 1733 break;
1da177e4 1734
e71a4783 1735 case SO_RCVLOWAT:
e6d12bdb 1736 v.val = READ_ONCE(sk->sk_rcvlowat);
e71a4783 1737 break;
1da177e4 1738
e71a4783 1739 case SO_SNDLOWAT:
2a91525c 1740 v.val = 1;
e71a4783 1741 break;
1da177e4 1742
e71a4783 1743 case SO_PASSCRED:
82981930 1744 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
e71a4783 1745 break;
1da177e4 1746
5e2ff670
AM		 1747 case SO_PASSPIDFD:
1748 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1749 break;
1750
e71a4783 1751 case SO_PEERCRED:
109f6e39
EB		 1752 {
1753 struct ucred peercred;
1754 if (len > sizeof(peercred))
1755 len = sizeof(peercred);
35306eb2
ED		 1756
1757 spin_lock(&sk->sk_peer_lock);
109f6e39 1758 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
35306eb2
ED		 1759 spin_unlock(&sk->sk_peer_lock);
1760
4ff09db1 1761 if (copy_to_sockptr(optval, &peercred, len))
e71a4783
SH		 1762 return -EFAULT;
1763 goto lenout;
109f6e39 1764 }
1da177e4 1765
7b26952a
AM		 1766 case SO_PEERPIDFD:
1767 {
1768 struct pid *peer_pid;
1769 struct file *pidfd_file = NULL;
1770 int pidfd;
1771
1772 if (len > sizeof(pidfd))
1773 len = sizeof(pidfd);
1774
1775 spin_lock(&sk->sk_peer_lock);
1776 peer_pid = get_pid(sk->sk_peer_pid);
1777 spin_unlock(&sk->sk_peer_lock);
1778
1779 if (!peer_pid)
b6f79e82 1780 return -ENODATA;
7b26952a
AM		 1781
1782 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1783 put_pid(peer_pid);
1784 if (pidfd < 0)
1785 return pidfd;
1786
1787 if (copy_to_sockptr(optval, &pidfd, len) ||
1788 copy_to_sockptr(optlen, &len, sizeof(int))) {
1789 put_unused_fd(pidfd);
1790 fput(pidfd_file);
1791
1792 return -EFAULT;
1793 }
1794
1795 fd_install(pidfd, pidfd_file);
1796 return 0;
1797 }
1798
28b5ba2a
DH		 1799 case SO_PEERGROUPS:
1800 {
35306eb2 1801 const struct cred *cred;
28b5ba2a
DH		 1802 int ret, n;
1803
35306eb2
ED		 1804 cred = sk_get_peer_cred(sk);
1805 if (!cred)
28b5ba2a
DH		 1806 return -ENODATA;
1807
35306eb2 1808 n = cred->group_info->ngroups;
28b5ba2a
DH		 1809 if (len < n * sizeof(gid_t)) {
1810 len = n * sizeof(gid_t);
35306eb2 1811 put_cred(cred);
4ff09db1 1812 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
28b5ba2a
DH		 1813 }
1814 len = n * sizeof(gid_t);
1815
4ff09db1 1816 ret = groups_to_user(optval, cred->group_info);
35306eb2 1817 put_cred(cred);
28b5ba2a
DH		 1818 if (ret)
1819 return ret;
1820 goto lenout;
1821 }
1822
e71a4783
SH		 1823 case SO_PEERNAME:
1824 {
8936bf53 1825 struct sockaddr_storage address;
e71a4783 1826
1ded5e5a 1827 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
9b2c45d4 1828 if (lv < 0)
e71a4783
SH		 1829 return -ENOTCONN;
1830 if (lv < len)
1831 return -EINVAL;
8936bf53 1832 if (copy_to_sockptr(optval, &address, len))
e71a4783
SH		 1833 return -EFAULT;
1834 goto lenout;
1835 }
1da177e4 1836
e71a4783
SH		 1837 /* Dubious BSD thing... Probably nobody even uses it, but
1838 * the UNIX standard wants it for whatever reason... -DaveM
1839 */
1840 case SO_ACCEPTCONN:
1841 v.val = sk->sk_state == TCP_LISTEN;
1842 break;
1da177e4 1843
e71a4783 1844 case SO_PASSSEC:
82981930 1845 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
e71a4783 1846 break;
877ce7c1 1847
e71a4783 1848 case SO_PEERSEC:
b10b9c34
PM		 1849 return security_socket_getpeersec_stream(sock,
1850 optval, optlen, len);
1da177e4 1851
4a19ec58 1852 case SO_MARK:
3c5b4d69 1853 v.val = READ_ONCE(sk->sk_mark);
4a19ec58
LAT		 1854 break;
1855
6fd1d51c
EM		 1856 case SO_RCVMARK:
1857 v.val = sock_flag(sk, SOCK_RCVMARK);
1858 break;
1859
3b885787 1860 case SO_RXQ_OVFL:
1b23a5df 1861 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
3b885787
NH		 1862 break;
1863
6e3e939f 1864 case SO_WIFI_STATUS:
1b23a5df 1865 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
6e3e939f
JB		 1866 break;
1867
ef64a54f 1868 case SO_PEEK_OFF:
1ded5e5a 1869 if (!READ_ONCE(sock->ops)->set_peek_off)
ef64a54f
PE		 1870 return -EOPNOTSUPP;
1871
11695c6e 1872 v.val = READ_ONCE(sk->sk_peek_off);
ef64a54f 1873 break;
bc2f7996 1874 case SO_NOFCS:
1b23a5df 1875 v.val = sock_flag(sk, SOCK_NOFCS);
bc2f7996 1876 break;
c91f6df2 1877
f7b86bfe 1878 case SO_BINDTODEVICE:
c91f6df2
BH		 1879 return sock_getbindtodevice(sk, optval, optlen, len);
1880
a8fc9277 1881 case SO_GET_FILTER:
4ff09db1 1882 len = sk_get_filter(sk, optval, len);
a8fc9277
PE		 1883 if (len < 0)
1884 return len;
1885
1886 goto lenout;
c91f6df2 1887
d59577b6
VB		 1888 case SO_LOCK_FILTER:
1889 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1890 break;
1891
ea02f941
MS		 1892 case SO_BPF_EXTENSIONS:
1893 v.val = bpf_tell_extensions();
1894 break;
1895
7d4c04fc
KJ		 1896 case SO_SELECT_ERR_QUEUE:
1897 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1898 break;
1899
e0d1095a 1900#ifdef CONFIG_NET_RX_BUSY_POLL
64b0dc51 1901 case SO_BUSY_POLL:
e5f0d2dd 1902 v.val = READ_ONCE(sk->sk_ll_usec);
dafcc438 1903 break;
7fd3253a
BT		 1904 case SO_PREFER_BUSY_POLL:
1905 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1906 break;
dafcc438
ET		 1907#endif
1908
62748f32 1909 case SO_MAX_PACING_RATE:
ea7f45ef 1910 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
677f136c
ED		 1911 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1912 lv = sizeof(v.ulval);
ea7f45ef 1913 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
677f136c
ED		 1914 } else {
1915 /* 32bit version */
ea7f45ef
ED		 1916 v.val = min_t(unsigned long, ~0U,
1917 READ_ONCE(sk->sk_max_pacing_rate));
677f136c 1918 }
62748f32
ED		 1919 break;
1920
2c8c56e1 1921 case SO_INCOMING_CPU:
7170a977 1922 v.val = READ_ONCE(sk->sk_incoming_cpu);
2c8c56e1
ED		 1923 break;
1924
a2d133b1
JH		 1925 case SO_MEMINFO:
1926 {
1927 u32 meminfo[SK_MEMINFO_VARS];
1928
a2d133b1
JH		 1929 sk_get_meminfo(sk, meminfo);
1930
1931 len = min_t(unsigned int, len, sizeof(meminfo));
4ff09db1 1932 if (copy_to_sockptr(optval, &meminfo, len))
a2d133b1
JH		 1933 return -EFAULT;
1934
1935 goto lenout;
1936 }
6d433902
SS		 1937
1938#ifdef CONFIG_NET_RX_BUSY_POLL
1939 case SO_INCOMING_NAPI_ID:
1940 v.val = READ_ONCE(sk->sk_napi_id);
1941
1942 /* aggregate non-NAPI IDs down to 0 */
1943 if (v.val < MIN_NAPI_ID)
1944 v.val = 0;
1945
1946 break;
1947#endif
1948
5daab9db
CF		 1949 case SO_COOKIE:
1950 lv = sizeof(u64);
1951 if (len < lv)
1952 return -EINVAL;
1953 v.val64 = sock_gen_cookie(sk);
1954 break;
1955
76851d12
WB		 1956 case SO_ZEROCOPY:
1957 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1958 break;
1959
80b14dee
RC		 1960 case SO_TXTIME:
1961 lv = sizeof(v.txtime);
1962 v.txtime.clockid = sk->sk_clockid;
1963 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1964 SOF_TXTIME_DEADLINE_MODE : 0;
4b15c707
JSP		 1965 v.txtime.flags |= sk->sk_txtime_report_errors ?
1966 SOF_TXTIME_REPORT_ERRORS : 0;
80b14dee
RC		 1967 break;
1968
f5dd3d0c 1969 case SO_BINDTOIFINDEX:
e5fccaa1 1970 v.val = READ_ONCE(sk->sk_bound_dev_if);
f5dd3d0c
DH		 1971 break;
1972
e8b9eab9
MP		 1973 case SO_NETNS_COOKIE:
1974 lv = sizeof(u64);
1975 if (len != lv)
1976 return -EINVAL;
1977 v.val64 = sock_net(sk)->net_cookie;
1978 break;
1979
04190bf8
PT		 1980 case SO_BUF_LOCK:
1981 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1982 break;
1983
2bb2f5fb 1984 case SO_RESERVE_MEM:
fe11fdcb 1985 v.val = READ_ONCE(sk->sk_reserved_mem);
2bb2f5fb
WW		 1986 break;
1987
26859240 1988 case SO_TXREHASH:
c76a0328
ED		 1989 /* Paired with WRITE_ONCE() in sk_setsockopt() */
1990 v.val = READ_ONCE(sk->sk_txrehash);
26859240
AK		 1991 break;
1992
e71a4783 1993 default:
443b5991
YH		 1994 /* We implement the SO_SNDLOWAT etc to not be settable
1995 * (1003.1g 7).
1996 */
e71a4783 1997 return -ENOPROTOOPT;
1da177e4 1998 }
e71a4783 1999
1da177e4
LT		 2000 if (len > lv)
2001 len = lv;
4ff09db1 2002 if (copy_to_sockptr(optval, &v, len))
1da177e4
LT		 2003 return -EFAULT;
2004lenout:
4ff09db1 2005 if (copy_to_sockptr(optlen, &len, sizeof(int)))
4ec93edb
YH		 2006 return -EFAULT;
2007 return 0;
1da177e4
LT		 2008}
2009
a5b5bb9a
IM		 2010/*
2011 * Initialize an sk_lock.
2012 *
2013 * (We also register the sk_lock with the lock validator.)
2014 */
b6f99a21 2015static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 2016{
cdfbabfb
DH		 2017 if (sk->sk_kern_sock)
2018 sock_lock_init_class_and_name(
2019 sk,
2020 af_family_kern_slock_key_strings[sk->sk_family],
2021 af_family_kern_slock_keys + sk->sk_family,
2022 af_family_kern_key_strings[sk->sk_family],
2023 af_family_kern_keys + sk->sk_family);
2024 else
2025 sock_lock_init_class_and_name(
2026 sk,
ed07536e
PZ		 2027 af_family_slock_key_strings[sk->sk_family],
2028 af_family_slock_keys + sk->sk_family,
2029 af_family_key_strings[sk->sk_family],
2030 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 2031}
2032
4dc6dc71
ED		 2033/*
2034 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2035 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 2036 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 2037 */
f1a6c4da
PE		 2038static void sock_copy(struct sock *nsk, const struct sock *osk)
2039{
b8e202d1 2040 const struct proto *prot = READ_ONCE(osk->sk_prot);
f1a6c4da
PE		 2041#ifdef CONFIG_SECURITY_NETWORK
2042 void *sptr = nsk->sk_security;
2043#endif
df610cd9
KI		 2044
2045 /* If we move sk_tx_queue_mapping out of the private section,
2046 * we must check if sk_tx_queue_clear() is called after
2047 * sock_copy() in sk_clone_lock().
2048 */
2049 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2050 offsetof(struct sock, sk_dontcopy_begin) ||
2051 offsetof(struct sock, sk_tx_queue_mapping) >=
2052 offsetof(struct sock, sk_dontcopy_end));
2053
68835aba
ED		 2054 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2055
ff73f834
KC		 2056 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2057 prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2058 /* alloc is larger than struct, see sk_prot_alloc() */);
68835aba 2059
f1a6c4da
PE		 2060#ifdef CONFIG_SECURITY_NETWORK
2061 nsk->sk_security = sptr;
2062 security_sk_clone(osk, nsk);
2063#endif
2064}
2065
2e4afe7b
PE		 2066static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2067 int family)
c308c1b2
PE		 2068{
2069 struct sock *sk;
2070 struct kmem_cache *slab;
2071
2072 slab = prot->slab;
e912b114
ED		 2073 if (slab != NULL) {
2074 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2075 if (!sk)
2076 return sk;
6471384a 2077 if (want_init_on_alloc(priority))
ba2489b0 2078 sk_prot_clear_nulls(sk, prot->obj_size);
fcbdf09d 2079 } else
c308c1b2
PE		 2080 sk = kmalloc(prot->obj_size, priority);
2081
2e4afe7b
PE		 2082 if (sk != NULL) {
2083 if (security_sk_alloc(sk, family, priority))
2084 goto out_free;
2085
2086 if (!try_module_get(prot->owner))
2087 goto out_free_sec;
2088 }
2089
c308c1b2 2090 return sk;
2e4afe7b
PE		 2091
2092out_free_sec:
2093 security_sk_free(sk);
2094out_free:
2095 if (slab != NULL)
2096 kmem_cache_free(slab, sk);
2097 else
2098 kfree(sk);
2099 return NULL;
c308c1b2
PE		 2100}
2101
2102static void sk_prot_free(struct proto *prot, struct sock *sk)
2103{
2104 struct kmem_cache *slab;
2e4afe7b 2105 struct module *owner;
c308c1b2 2106
2e4afe7b 2107 owner = prot->owner;
c308c1b2 2108 slab = prot->slab;
2e4afe7b 2109
bd1060a1 2110 cgroup_sk_free(&sk->sk_cgrp_data);
2d758073 2111 mem_cgroup_sk_free(sk);
2e4afe7b 2112 security_sk_free(sk);
c308c1b2
PE		 2113 if (slab != NULL)
2114 kmem_cache_free(slab, sk);
2115 else
2116 kfree(sk);
2e4afe7b 2117 module_put(owner);
c308c1b2
PE		 2118}
2119
1da177e4
LT		 2120/**
2121 * sk_alloc - All socket objects are allocated here
c4ea43c5 2122 * @net: the applicable net namespace
4dc3b16b
PP		 2123 * @family: protocol family
2124 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2125 * @prot: struct proto associated with this new sock instance
11aa9c28 2126 * @kern: is this to be a kernel socket?
1da177e4 2127 */
1b8d7ae4 2128struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
11aa9c28 2129 struct proto *prot, int kern)
1da177e4 2130{
c308c1b2 2131 struct sock *sk;
1da177e4 2132
154adbc8 2133 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 2134 if (sk) {
154adbc8
PE		 2135 sk->sk_family = family;
2136 /*
2137 * See comment in struct sock definition to understand
2138 * why we need sk_prot_creator -acme
2139 */
2140 sk->sk_prot = sk->sk_prot_creator = prot;
cdfbabfb 2141 sk->sk_kern_sock = kern;
154adbc8 2142 sock_lock_init(sk);
26abe143 2143 sk->sk_net_refcnt = kern ? 0 : 1;
648845ab 2144 if (likely(sk->sk_net_refcnt)) {
ffa84b5f 2145 get_net_track(net, &sk->ns_tracker, priority);
648845ab 2146 sock_inuse_add(net, 1);
0cafd77d
ED		 2147 } else {
2148 __netns_tracker_alloc(net, &sk->ns_tracker,
2149 false, priority);
648845ab
TZ		 2150 }
2151
26abe143 2152 sock_net_set(sk, net);
14afee4b 2153 refcount_set(&sk->sk_wmem_alloc, 1);
f8451725 2154
2d758073 2155 mem_cgroup_sk_alloc(sk);
d979a39d 2156 cgroup_sk_alloc(&sk->sk_cgrp_data);
2a56a1fe
TH		 2157 sock_update_classid(&sk->sk_cgrp_data);
2158 sock_update_netprioidx(&sk->sk_cgrp_data);
41b14fb8 2159 sk_tx_queue_clear(sk);
1da177e4 2160 }
a79af59e 2161
2e4afe7b 2162 return sk;
1da177e4 2163}
2a91525c 2164EXPORT_SYMBOL(sk_alloc);
1da177e4 2165
a4298e45
ED		 2166/* Sockets having SOCK_RCU_FREE will call this function after one RCU
2167 * grace period. This is the case for UDP sockets and TCP listeners.
2168 */
2169static void __sk_destruct(struct rcu_head *head)
1da177e4 2170{
a4298e45 2171 struct sock *sk = container_of(head, struct sock, sk_rcu);
1da177e4 2172 struct sk_filter *filter;
1da177e4
LT		 2173
2174 if (sk->sk_destruct)
2175 sk->sk_destruct(sk);
2176
a898def2 2177 filter = rcu_dereference_check(sk->sk_filter,
14afee4b 2178 refcount_read(&sk->sk_wmem_alloc) == 0);
1da177e4 2179 if (filter) {
309dd5fc 2180 sk_filter_uncharge(sk, filter);
a9b3cd7f 2181 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4
LT		 2182 }
2183
08e29af3 2184 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1da177e4 2185
6ac99e8f
MKL		 2186#ifdef CONFIG_BPF_SYSCALL
2187 bpf_sk_storage_free(sk);
2188#endif
2189
1da177e4 2190 if (atomic_read(&sk->sk_omem_alloc))
e005d193
JP		 2191 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2192 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 2193
22a0e18e
ED		 2194 if (sk->sk_frag.page) {
2195 put_page(sk->sk_frag.page);
2196 sk->sk_frag.page = NULL;
2197 }
2198
35306eb2
ED		 2199 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2200 put_cred(sk->sk_peer_cred);
109f6e39 2201 put_pid(sk->sk_peer_pid);
35306eb2 2202
26abe143 2203 if (likely(sk->sk_net_refcnt))
ffa84b5f 2204 put_net_track(sock_net(sk), &sk->ns_tracker);
0cafd77d
ED		 2205 else
2206 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2207
c308c1b2 2208 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 2209}
2b85a34e 2210
a4298e45
ED		 2211void sk_destruct(struct sock *sk)
2212{
8c7138b3
MKL		 2213 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2214
2215 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2216 reuseport_detach_sock(sk);
2217 use_call_rcu = true;
2218 }
2219
2220 if (use_call_rcu)
a4298e45
ED		 2221 call_rcu(&sk->sk_rcu, __sk_destruct);
2222 else
2223 __sk_destruct(&sk->sk_rcu);
2224}
2225
eb4cb008
CG		 2226static void __sk_free(struct sock *sk)
2227{
648845ab
TZ		 2228 if (likely(sk->sk_net_refcnt))
2229 sock_inuse_add(sock_net(sk), -1);
2230
9709020c 2231 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
eb4cb008
CG		 2232 sock_diag_broadcast_destroy(sk);
2233 else
2234 sk_destruct(sk);
2235}
2236
2b85a34e
ED		 2237void sk_free(struct sock *sk)
2238{
2239 /*
25985edc 2240 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 2241 * some packets are still in some tx queue.
2242 * If not null, sock_wfree() will call __sk_free(sk) later
2243 */
14afee4b 2244 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2b85a34e
ED		 2245 __sk_free(sk);
2246}
2a91525c 2247EXPORT_SYMBOL(sk_free);
1da177e4 2248
581319c5
PA		 2249static void sk_init_common(struct sock *sk)
2250{
2251 skb_queue_head_init(&sk->sk_receive_queue);
2252 skb_queue_head_init(&sk->sk_write_queue);
2253 skb_queue_head_init(&sk->sk_error_queue);
2254
2255 rwlock_init(&sk->sk_callback_lock);
2256 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2257 af_rlock_keys + sk->sk_family,
2258 af_family_rlock_key_strings[sk->sk_family]);
2259 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2260 af_wlock_keys + sk->sk_family,
2261 af_family_wlock_key_strings[sk->sk_family]);
2262 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2263 af_elock_keys + sk->sk_family,
2264 af_family_elock_key_strings[sk->sk_family]);
2265 lockdep_set_class_and_name(&sk->sk_callback_lock,
2266 af_callback_keys + sk->sk_family,
2267 af_family_clock_key_strings[sk->sk_family]);
2268}
2269
e56c57d0
ED		 2270/**
2271 * sk_clone_lock - clone a socket, and lock its clone
2272 * @sk: the socket to clone
2273 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2274 *
2275 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2276 */
2277struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
87d11ceb 2278{
b8e202d1 2279 struct proto *prot = READ_ONCE(sk->sk_prot);
bbc20b70 2280 struct sk_filter *filter;
278571ba 2281 bool is_charged = true;
bbc20b70 2282 struct sock *newsk;
87d11ceb 2283
b8e202d1 2284 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
bbc20b70
ED		 2285 if (!newsk)
2286 goto out;
87d11ceb 2287
bbc20b70 2288 sock_copy(newsk, sk);
9d538fa6 2289
bbc20b70 2290 newsk->sk_prot_creator = prot;
87d11ceb 2291
bbc20b70 2292 /* SANITY */
938cca9e 2293 if (likely(newsk->sk_net_refcnt)) {
ffa84b5f 2294 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
938cca9e 2295 sock_inuse_add(sock_net(newsk), 1);
0cafd77d
ED		 2296 } else {
2297 /* Kernel sockets are not elevating the struct net refcount.
2298 * Instead, use a tracker to more easily detect if a layer
2299 * is not properly dismantling its kernel sockets at netns
2300 * destroy time.
2301 */
2302 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2303 false, priority);
938cca9e 2304 }
bbc20b70
ED		 2305 sk_node_init(&newsk->sk_node);
2306 sock_lock_init(newsk);
2307 bh_lock_sock(newsk);
2308 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2309 newsk->sk_backlog.len = 0;
87d11ceb 2310
bbc20b70 2311 atomic_set(&newsk->sk_rmem_alloc, 0);
87d11ceb 2312
bbc20b70
ED		 2313 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2314 refcount_set(&newsk->sk_wmem_alloc, 1);
d752a498 2315
bbc20b70
ED		 2316 atomic_set(&newsk->sk_omem_alloc, 0);
2317 sk_init_common(newsk);
d752a498 2318
bbc20b70
ED		 2319 newsk->sk_dst_cache = NULL;
2320 newsk->sk_dst_pending_confirm = 0;
2321 newsk->sk_wmem_queued = 0;
2322 newsk->sk_forward_alloc = 0;
2bb2f5fb 2323 newsk->sk_reserved_mem = 0;
bbc20b70
ED		 2324 atomic_set(&newsk->sk_drops, 0);
2325 newsk->sk_send_head = NULL;
2326 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2327 atomic_set(&newsk->sk_zckey, 0);
87d11ceb 2328
bbc20b70 2329 sock_reset_flag(newsk, SOCK_DONE);
87d11ceb 2330
bbc20b70
ED		 2331 /* sk->sk_memcg will be populated at accept() time */
2332 newsk->sk_memcg = NULL;
8f51dfc7 2333
bbc20b70 2334 cgroup_sk_clone(&newsk->sk_cgrp_data);
87d11ceb 2335
bbc20b70
ED		 2336 rcu_read_lock();
2337 filter = rcu_dereference(sk->sk_filter);
2338 if (filter != NULL)
2339 /* though it's an empty new sock, the charging may fail
2340 * if sysctl_optmem_max was changed between creation of
2341 * original socket and cloning
2342 */
2343 is_charged = sk_filter_charge(newsk, filter);
2344 RCU_INIT_POINTER(newsk->sk_filter, filter);
2345 rcu_read_unlock();
2346
2347 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2348 /* We need to make sure that we don't uncharge the new
2349 * socket if we couldn't charge it in the first place
2350 * as otherwise we uncharge the parent's filter.
f1ff5ce2 2351 */
bbc20b70
ED		 2352 if (!is_charged)
2353 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2354 sk_free_unlock_clone(newsk);
2355 newsk = NULL;
2356 goto out;
2357 }
2358 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
f1ff5ce2 2359
bbc20b70
ED		 2360 if (bpf_sk_storage_clone(sk, newsk)) {
2361 sk_free_unlock_clone(newsk);
2362 newsk = NULL;
2363 goto out;
2364 }
d979a39d 2365
bbc20b70
ED		 2366 /* Clear sk_user_data if parent had the pointer tagged
2367 * as not suitable for copying when cloning.
2368 */
2369 if (sk_user_data_is_nocopy(newsk))
2370 newsk->sk_user_data = NULL;
2371
2372 newsk->sk_err = 0;
2373 newsk->sk_err_soft = 0;
2374 newsk->sk_priority = 0;
2375 newsk->sk_incoming_cpu = raw_smp_processor_id();
bbc20b70
ED		 2376
2377 /* Before updating sk_refcnt, we must commit prior changes to memory
2378 * (Documentation/RCU/rculist_nulls.rst for details)
2379 */
2380 smp_wmb();
2381 refcount_set(&newsk->sk_refcnt, 2);
87d11ceb 2382
bbc20b70
ED		 2383 sk_set_socket(newsk, NULL);
2384 sk_tx_queue_clear(newsk);
2385 RCU_INIT_POINTER(newsk->sk_wq, NULL);
87d11ceb 2386
bbc20b70
ED		 2387 if (newsk->sk_prot->sockets_allocated)
2388 sk_sockets_allocated_inc(newsk);
704da560 2389
bbc20b70
ED		 2390 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2391 net_enable_timestamp();
87d11ceb
ACM		 2392out:
2393 return newsk;
2394}
e56c57d0 2395EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 2396
94352d45
ACM		 2397void sk_free_unlock_clone(struct sock *sk)
2398{
2399 /* It is still raw copy of parent, so invalidate
2400 * destructor and make plain sk_free() */
2401 sk->sk_destruct = NULL;
2402 bh_unlock_sock(sk);
2403 sk_free(sk);
2404}
2405EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2406
b1a78b9b 2407static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
7c4e983c 2408{
b1a78b9b
XL		 2409 bool is_ipv6 = false;
2410 u32 max_size;
2411
7c4e983c 2412#if IS_ENABLED(CONFIG_IPV6)
b1a78b9b
XL		 2413 is_ipv6 = (sk->sk_family == AF_INET6 &&
2414 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
7c4e983c 2415#endif
b1a78b9b
XL		 2416 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2417 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2418 READ_ONCE(dst->dev->gso_ipv4_max_size);
2419 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2420 max_size = GSO_LEGACY_MAX_SIZE;
2421
2422 return max_size - (MAX_TCP_HEADER + 1);
7c4e983c
AD		 2423}
2424
9958089a
AK		 2425void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2426{
d6a4e26a
ED		 2427 u32 max_segs = 1;
2428
d0d598ca
ED		 2429 sk->sk_route_caps = dst->dev->features;
2430 if (sk_is_tcp(sk))
2431 sk->sk_route_caps |= NETIF_F_GSO;
9958089a 2432 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 2433 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
aba54656
ED		 2434 if (unlikely(sk->sk_gso_disabled))
2435 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
9958089a 2436 if (sk_can_gso(sk)) {
f70f250a 2437 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
9958089a 2438 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 2439 } else {
9958089a 2440 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
b1a78b9b 2441 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
6d872df3
ED		 2442 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2443 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
82cc1a7a 2444 }
9958089a 2445 }
d6a4e26a 2446 sk->sk_gso_max_segs = max_segs;
448a5ce1 2447 sk_dst_set(sk, dst);
9958089a
AK		 2448}
2449EXPORT_SYMBOL_GPL(sk_setup_caps);
2450
1da177e4
LT		 2451/*
2452 * Simple resource managers for sockets.
2453 */
2454
2455
4ec93edb
YH		 2456/*
2457 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 2458 */
2459void sock_wfree(struct sk_buff *skb)
2460{
2461 struct sock *sk = skb->sk;
d99927f4 2462 unsigned int len = skb->truesize;
052ada09 2463 bool free;
1da177e4 2464
d99927f4 2465 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
052ada09
PB		 2466 if (sock_flag(sk, SOCK_RCU_FREE) &&
2467 sk->sk_write_space == sock_def_write_space) {
2468 rcu_read_lock();
2469 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
0a8afd9f 2470 sock_def_write_space_wfree(sk);
052ada09
PB		 2471 rcu_read_unlock();
2472 if (unlikely(free))
2473 __sk_free(sk);
2474 return;
2475 }
2476
d99927f4
ED		 2477 /*
2478 * Keep a reference on sk_wmem_alloc, this will be released
2479 * after sk_write_space() call
2480 */
14afee4b 2481 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1da177e4 2482 sk->sk_write_space(sk);
d99927f4
ED		 2483 len = 1;
2484 }
2b85a34e 2485 /*
d99927f4
ED		 2486 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2487 * could not do because of in-flight packets
2b85a34e 2488 */
14afee4b 2489 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 2490 __sk_free(sk);
1da177e4 2491}
2a91525c 2492EXPORT_SYMBOL(sock_wfree);
1da177e4 2493
1d2077ac
ED		 2494/* This variant of sock_wfree() is used by TCP,
2495 * since it sets SOCK_USE_WRITE_QUEUE.
2496 */
2497void __sock_wfree(struct sk_buff *skb)
2498{
2499 struct sock *sk = skb->sk;
2500
14afee4b 2501 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1d2077ac
ED		 2502 __sk_free(sk);
2503}
2504
9e17f8a4
ED		 2505void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2506{
2507 skb_orphan(skb);
2508 skb->sk = sk;
2509#ifdef CONFIG_INET
2510 if (unlikely(!sk_fullsock(sk))) {
2511 skb->destructor = sock_edemux;
2512 sock_hold(sk);
2513 return;
2514 }
2515#endif
2516 skb->destructor = sock_wfree;
2517 skb_set_hash_from_sk(skb, sk);
2518 /*
2519 * We used to take a refcount on sk, but following operation
2520 * is enough to guarantee sk_free() wont free this sock until
2521 * all in-flight packets are completed
2522 */
14afee4b 2523 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
9e17f8a4
ED		 2524}
2525EXPORT_SYMBOL(skb_set_owner_w);
2526
41477662
JK		 2527static bool can_skb_orphan_partial(const struct sk_buff *skb)
2528{
41477662
JK		 2529 /* Drivers depend on in-order delivery for crypto offload,
2530 * partial orphan breaks out-of-order-OK logic.
2531 */
9f06f87f 2532 if (skb_is_decrypted(skb))
41477662 2533 return false;
9f06f87f 2534
41477662
JK		 2535 return (skb->destructor == sock_wfree ||
2536 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2537}
2538
1d2077ac
ED		 2539/* This helper is used by netem, as it can hold packets in its
2540 * delay queue. We want to allow the owner socket to send more
2541 * packets, as if they were already TX completed by a typical driver.
2542 * But we also want to keep skb->sk set because some packet schedulers
f6ba8d33 2543 * rely on it (sch_fq for example).
1d2077ac 2544 */
f2f872f9
ED		 2545void skb_orphan_partial(struct sk_buff *skb)
2546{
f6ba8d33 2547 if (skb_is_tcp_pure_ack(skb))
1d2077ac
ED		 2548 return;
2549
098116e7
PA		 2550 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2551 return;
2552
2553 skb_orphan(skb);
f2f872f9
ED		 2554}
2555EXPORT_SYMBOL(skb_orphan_partial);
2556
4ec93edb
YH		 2557/*
2558 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 2559 */
2560void sock_rfree(struct sk_buff *skb)
2561{
2562 struct sock *sk = skb->sk;
d361fd59 2563 unsigned int len = skb->truesize;
1da177e4 2564
d361fd59
ED		 2565 atomic_sub(len, &sk->sk_rmem_alloc);
2566 sk_mem_uncharge(sk, len);
1da177e4 2567}
2a91525c 2568EXPORT_SYMBOL(sock_rfree);
1da177e4 2569
7768eed8
OH		 2570/*
2571 * Buffer destructor for skbs that are not used directly in read or write
2572 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2573 */
62bccb8c
AD		 2574void sock_efree(struct sk_buff *skb)
2575{
2576 sock_put(skb->sk);
2577}
2578EXPORT_SYMBOL(sock_efree);
2579
cf7fbe66
JS		 2580/* Buffer destructor for prefetch/receive path where reference count may
2581 * not be held, e.g. for listen sockets.
2582 */
2583#ifdef CONFIG_INET
2584void sock_pfree(struct sk_buff *skb)
2585{
e472f888
KI		 2586 struct sock *sk = skb->sk;
2587
2588 if (!sk_is_refcounted(sk))
2589 return;
2590
2591 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2592 inet_reqsk(sk)->rsk_listener = NULL;
2593 reqsk_free(inet_reqsk(sk));
2594 return;
2595 }
2596
2597 sock_gen_put(sk);
cf7fbe66
JS		 2598}
2599EXPORT_SYMBOL(sock_pfree);
2600#endif /* CONFIG_INET */
2601
976d0201 2602kuid_t sock_i_uid(struct sock *sk)
1da177e4 2603{
976d0201 2604 kuid_t uid;
1da177e4 2605
f064af1e 2606 read_lock_bh(&sk->sk_callback_lock);
976d0201 2607 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
f064af1e 2608 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 2609 return uid;
2610}
2a91525c 2611EXPORT_SYMBOL(sock_i_uid);
1da177e4 2612
25a9c8a4 2613unsigned long __sock_i_ino(struct sock *sk)
1da177e4
LT		 2614{
2615 unsigned long ino;
2616
25a9c8a4 2617 read_lock(&sk->sk_callback_lock);
1da177e4 2618 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
25a9c8a4
KI		 2619 read_unlock(&sk->sk_callback_lock);
2620 return ino;
2621}
2622EXPORT_SYMBOL(__sock_i_ino);
2623
2624unsigned long sock_i_ino(struct sock *sk)
2625{
2626 unsigned long ino;
2627
2628 local_bh_disable();
2629 ino = __sock_i_ino(sk);
2630 local_bh_enable();
1da177e4
LT		 2631 return ino;
2632}
2a91525c 2633EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 2634
2635/*
2636 * Allocate a skb from the socket's send buffer.
2637 */
86a76caf 2638struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 2639 gfp_t priority)
1da177e4 2640{
e292f05e
ED		 2641 if (force ||
2642 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2a91525c 2643 struct sk_buff *skb = alloc_skb(size, priority);
e292f05e 2644
1da177e4
LT		 2645 if (skb) {
2646 skb_set_owner_w(skb, sk);
2647 return skb;
2648 }
2649 }
2650 return NULL;
2651}
2a91525c 2652EXPORT_SYMBOL(sock_wmalloc);
1da177e4 2653
98ba0bd5
WB		 2654static void sock_ofree(struct sk_buff *skb)
2655{
2656 struct sock *sk = skb->sk;
2657
2658 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2659}
2660
2661struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2662 gfp_t priority)
2663{
2664 struct sk_buff *skb;
2665
2666 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2667 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
f5769fae 2668 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
98ba0bd5
WB		 2669 return NULL;
2670
2671 skb = alloc_skb(size, priority);
2672 if (!skb)
2673 return NULL;
2674
2675 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2676 skb->sk = sk;
2677 skb->destructor = sock_ofree;
2678 return skb;
2679}
2680
4ec93edb 2681/*
1da177e4 2682 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 2683 */
dd0fc66f 2684void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4 2685{
f5769fae 2686 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
7de6d09f
KI		 2687
2688 if ((unsigned int)size <= optmem_max &&
2689 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
1da177e4
LT		 2690 void *mem;
2691 /* First do the add, to avoid the race if kmalloc
4ec93edb 2692 * might sleep.
1da177e4
LT		 2693 */
2694 atomic_add(size, &sk->sk_omem_alloc);
2695 mem = kmalloc(size, priority);
2696 if (mem)
2697 return mem;
2698 atomic_sub(size, &sk->sk_omem_alloc);
2699 }
2700 return NULL;
2701}
2a91525c 2702EXPORT_SYMBOL(sock_kmalloc);
1da177e4 2703
79e88659
DB		 2704/* Free an option memory block. Note, we actually want the inline
2705 * here as this allows gcc to detect the nullify and fold away the
2706 * condition entirely.
1da177e4 2707 */
79e88659
DB		 2708static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2709 const bool nullify)
1da177e4 2710{
e53da5fb
DM		 2711 if (WARN_ON_ONCE(!mem))
2712 return;
79e88659 2713 if (nullify)
453431a5 2714 kfree_sensitive(mem);
79e88659
DB		 2715 else
2716 kfree(mem);
1da177e4
LT		 2717 atomic_sub(size, &sk->sk_omem_alloc);
2718}
79e88659
DB		 2719
2720void sock_kfree_s(struct sock *sk, void *mem, int size)
2721{
2722 __sock_kfree_s(sk, mem, size, false);
2723}
2a91525c 2724EXPORT_SYMBOL(sock_kfree_s);
1da177e4 2725
79e88659
DB		 2726void sock_kzfree_s(struct sock *sk, void *mem, int size)
2727{
2728 __sock_kfree_s(sk, mem, size, true);
2729}
2730EXPORT_SYMBOL(sock_kzfree_s);
2731
1da177e4
LT		 2732/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2733 I think, these locks should be removed for datagram sockets.
2734 */
2a91525c 2735static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 2736{
2737 DEFINE_WAIT(wait);
2738
9cd3e072 2739 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1da177e4
LT		 2740 for (;;) {
2741 if (!timeo)
2742 break;
2743 if (signal_pending(current))
2744 break;
2745 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 2746 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
e292f05e 2747 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
1da177e4 2748 break;
afe8764f 2749 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
1da177e4 2750 break;
b1928129 2751 if (READ_ONCE(sk->sk_err))
1da177e4
LT		 2752 break;
2753 timeo = schedule_timeout(timeo);
2754 }
aa395145 2755 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 2756 return timeo;
2757}
2758
2759
2760/*
2761 * Generic send/receive buffer handlers
2762 */
2763
4cc7f68d
HX		 2764struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2765 unsigned long data_len, int noblock,
28d64271 2766 int *errcode, int max_page_order)
1da177e4 2767{
2e4e4410 2768 struct sk_buff *skb;
1da177e4
LT		 2769 long timeo;
2770 int err;
2771
1da177e4 2772 timeo = sock_sndtimeo(sk, noblock);
2e4e4410 2773 for (;;) {
1da177e4
LT		 2774 err = sock_error(sk);
2775 if (err != 0)
2776 goto failure;
2777
2778 err = -EPIPE;
afe8764f 2779 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
1da177e4
LT		 2780 goto failure;
2781
e292f05e 2782 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2e4e4410 2783 break;
28d64271 2784
9cd3e072 2785 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2e4e4410
ED		 2786 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2787 err = -EAGAIN;
2788 if (!timeo)
1da177e4 2789 goto failure;
2e4e4410
ED		 2790 if (signal_pending(current))
2791 goto interrupted;
2792 timeo = sock_wait_for_wmem(sk, timeo);
1da177e4 2793 }
2e4e4410
ED		 2794 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2795 errcode, sk->sk_allocation);
2796 if (skb)
2797 skb_set_owner_w(skb, sk);
1da177e4
LT		 2798 return skb;
2799
2800interrupted:
2801 err = sock_intr_errno(timeo);
2802failure:
2803 *errcode = err;
2804 return NULL;
2805}
4cc7f68d 2806EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 2807
233baf9a 2808int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
39771b12
WB		 2809 struct sockcm_cookie *sockc)
2810{
3dd17e63
SHY		 2811 u32 tsflags;
2812
39771b12
WB		 2813 switch (cmsg->cmsg_type) {
2814 case SO_MARK:
91f0d8a4
JK		 2815 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2816 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
39771b12
WB		 2817 return -EPERM;
2818 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2819 return -EINVAL;
2820 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2821 break;
7f1bc6e9 2822 case SO_TIMESTAMPING_OLD:
382a3201 2823 case SO_TIMESTAMPING_NEW:
3dd17e63
SHY		 2824 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2825 return -EINVAL;
2826
2827 tsflags = *(u32 *)CMSG_DATA(cmsg);
2828 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2829 return -EINVAL;
2830
2831 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2832 sockc->tsflags |= tsflags;
2833 break;
80b14dee
RC		 2834 case SCM_TXTIME:
2835 if (!sock_flag(sk, SOCK_TXTIME))
2836 return -EINVAL;
2837 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2838 return -EINVAL;
2839 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2840 break;
779f1ede
SHY		 2841 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2842 case SCM_RIGHTS:
2843 case SCM_CREDENTIALS:
2844 break;
39771b12
WB		 2845 default:
2846 return -EINVAL;
2847 }
2848 return 0;
2849}
2850EXPORT_SYMBOL(__sock_cmsg_send);
2851
f28ea365
EJ		 2852int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2853 struct sockcm_cookie *sockc)
2854{
2855 struct cmsghdr *cmsg;
39771b12 2856 int ret;
f28ea365
EJ		 2857
2858 for_each_cmsghdr(cmsg, msg) {
2859 if (!CMSG_OK(msg, cmsg))
2860 return -EINVAL;
2861 if (cmsg->cmsg_level != SOL_SOCKET)
2862 continue;
233baf9a 2863 ret = __sock_cmsg_send(sk, cmsg, sockc);
39771b12
WB		 2864 if (ret)
2865 return ret;
f28ea365
EJ		 2866 }
2867 return 0;
2868}
2869EXPORT_SYMBOL(sock_cmsg_send);
2870
06044751
ED		 2871static void sk_enter_memory_pressure(struct sock *sk)
2872{
2873 if (!sk->sk_prot->enter_memory_pressure)
2874 return;
2875
2876 sk->sk_prot->enter_memory_pressure(sk);
2877}
2878
2879static void sk_leave_memory_pressure(struct sock *sk)
2880{
2881 if (sk->sk_prot->leave_memory_pressure) {
5c1ebbfa
BV		 2882 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2883 tcp_leave_memory_pressure, sk);
06044751
ED		 2884 } else {
2885 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2886
503978ac
ED		 2887 if (memory_pressure && READ_ONCE(*memory_pressure))
2888 WRITE_ONCE(*memory_pressure, 0);
06044751
ED		 2889 }
2890}
2891
ce27ec60 2892DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
5640f768 2893
400dfd3a
ED		 2894/**
2895 * skb_page_frag_refill - check that a page_frag contains enough room
2896 * @sz: minimum size of the fragment we want to get
2897 * @pfrag: pointer to page_frag
82d5e2b8 2898 * @gfp: priority for memory allocation
400dfd3a
ED		 2899 *
2900 * Note: While this allocator tries to use high order pages, there is
2901 * no guarantee that allocations succeed. Therefore, @sz MUST be
2902 * less or equal than PAGE_SIZE.
2903 */
d9b2938a 2904bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
5640f768 2905{
5640f768 2906 if (pfrag->page) {
fe896d18 2907 if (page_ref_count(pfrag->page) == 1) {
5640f768
ED		 2908 pfrag->offset = 0;
2909 return true;
2910 }
400dfd3a 2911 if (pfrag->offset + sz <= pfrag->size)
5640f768
ED		 2912 return true;
2913 put_page(pfrag->page);
2914 }
2915
d9b2938a 2916 pfrag->offset = 0;
ce27ec60
ED		 2917 if (SKB_FRAG_PAGE_ORDER &&
2918 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
d0164adc
MG		 2919 /* Avoid direct reclaim but allow kswapd to wake */
2920 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2921 __GFP_COMP | __GFP_NOWARN |
2922 __GFP_NORETRY,
d9b2938a 2923 SKB_FRAG_PAGE_ORDER);
5640f768 2924 if (likely(pfrag->page)) {
d9b2938a 2925 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
5640f768
ED		 2926 return true;
2927 }
d9b2938a
ED		 2928 }
2929 pfrag->page = alloc_page(gfp);
2930 if (likely(pfrag->page)) {
2931 pfrag->size = PAGE_SIZE;
2932 return true;
2933 }
400dfd3a
ED		 2934 return false;
2935}
2936EXPORT_SYMBOL(skb_page_frag_refill);
2937
2938bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2939{
2940 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2941 return true;
2942
5640f768
ED		 2943 sk_enter_memory_pressure(sk);
2944 sk_stream_moderate_sndbuf(sk);
2945 return false;
2946}
2947EXPORT_SYMBOL(sk_page_frag_refill);
2948
ad80b0fc 2949void __lock_sock(struct sock *sk)
f39234d6
NK		 2950 __releases(&sk->sk_lock.slock)
2951 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 2952{
2953 DEFINE_WAIT(wait);
2954
e71a4783 2955 for (;;) {
1da177e4
LT		 2956 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2957 TASK_UNINTERRUPTIBLE);
2958 spin_unlock_bh(&sk->sk_lock.slock);
2959 schedule();
2960 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 2961 if (!sock_owned_by_user(sk))
1da177e4
LT		 2962 break;
2963 }
2964 finish_wait(&sk->sk_lock.wq, &wait);
2965}
2966
8873c064 2967void __release_sock(struct sock *sk)
f39234d6
NK		 2968 __releases(&sk->sk_lock.slock)
2969 __acquires(&sk->sk_lock.slock)
1da177e4 2970{
5413d1ba 2971 struct sk_buff *skb, *next;
1da177e4 2972
5413d1ba 2973 while ((skb = sk->sk_backlog.head) != NULL) {
1da177e4 2974 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1da177e4 2975
5413d1ba 2976 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4 2977
5413d1ba
ED		 2978 do {
2979 next = skb->next;
e4cbb02a 2980 prefetch(next);
63fbdd3c 2981 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
a8305bff 2982 skb_mark_not_on_list(skb);
c57943a1 2983 sk_backlog_rcv(sk, skb);
1da177e4 2984
5413d1ba 2985 cond_resched();
1da177e4
LT		 2986
2987 skb = next;
2988 } while (skb != NULL);
2989
5413d1ba
ED		 2990 spin_lock_bh(&sk->sk_lock.slock);
2991 }
8eae939f
ZY		 2992
2993 /*
2994 * Doing the zeroing here guarantee we can not loop forever
2995 * while a wild producer attempts to flood us.
2996 */
2997 sk->sk_backlog.len = 0;
1da177e4
LT		 2998}
2999
d41a69f1
ED		 3000void __sk_flush_backlog(struct sock *sk)
3001{
3002 spin_lock_bh(&sk->sk_lock.slock);
3003 __release_sock(sk);
4505dc2a
ED		 3004
3005 if (sk->sk_prot->release_cb)
41862d12
ED		 3006 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3007 tcp_release_cb, sk);
3008
d41a69f1
ED		 3009 spin_unlock_bh(&sk->sk_lock.slock);
3010}
c46b0183 3011EXPORT_SYMBOL_GPL(__sk_flush_backlog);
d41a69f1 3012
1da177e4
LT		 3013/**
3014 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 3015 * @sk: sock to wait on
3016 * @timeo: for how long
dfbafc99 3017 * @skb: last skb seen on sk_receive_queue
1da177e4
LT		 3018 *
3019 * Now socket state including sk->sk_err is changed only under lock,
3020 * hence we may omit checks after joining wait queue.
3021 * We check receive queue before schedule() only as optimization;
3022 * it is very likely that release_sock() added new data.
3023 */
dfbafc99 3024int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
1da177e4 3025{
d9dc8b0f 3026 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1da177e4 3027 int rc;
1da177e4 3028
d9dc8b0f 3029 add_wait_queue(sk_sleep(sk), &wait);
9cd3e072 3030 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
d9dc8b0f 3031 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
9cd3e072 3032 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
d9dc8b0f 3033 remove_wait_queue(sk_sleep(sk), &wait);
1da177e4
LT		 3034 return rc;
3035}
1da177e4
LT		 3036EXPORT_SYMBOL(sk_wait_data);
3037
3ab224be 3038/**
f8c3bf00 3039 * __sk_mem_raise_allocated - increase memory_allocated
3ab224be
HA		 3040 * @sk: socket
3041 * @size: memory size to allocate
f8c3bf00 3042 * @amt: pages to allocate
3ab224be
HA		 3043 * @kind: allocation type
3044 *
66e6369e
AW		 3045 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3046 *
3047 * Unlike the globally shared limits among the sockets under same protocol,
3048 * consuming the budget of a memcg won't have direct effect on other ones.
3049 * So be optimistic about memcg's tolerance, and leave the callers to decide
3050 * whether or not to raise allocated through sk_under_memory_pressure() or
3051 * its variants.
3ab224be 3052 */
f8c3bf00 3053int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3ab224be 3054{
2def8ff3 3055 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
219160be 3056 struct proto *prot = sk->sk_prot;
2def8ff3 3057 bool charged = false;
219160be 3058 long allocated;
e805605c 3059
219160be
ED		 3060 sk_memory_allocated_add(sk, amt);
3061 allocated = sk_memory_allocated(sk);
2def8ff3
AW		 3062
3063 if (memcg) {
3064 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3065 goto suppress_allocation;
3066 charged = true;
3067 }
3ab224be
HA		 3068
3069 /* Under limit. */
e805605c 3070 if (allocated <= sk_prot_mem_limits(sk, 0)) {
180d8cd9 3071 sk_leave_memory_pressure(sk);
3ab224be
HA		 3072 return 1;
3073 }
3074
e805605c
JW		 3075 /* Under pressure. */
3076 if (allocated > sk_prot_mem_limits(sk, 1))
180d8cd9 3077 sk_enter_memory_pressure(sk);
3ab224be 3078
e805605c
JW		 3079 /* Over hard limit. */
3080 if (allocated > sk_prot_mem_limits(sk, 2))
3ab224be
HA		 3081 goto suppress_allocation;
3082
2e12072c
AW		 3083 /* Guarantee minimum buffer size under pressure (either global
3084 * or memcg) to make sure features described in RFC 7323 (TCP
3085 * Extensions for High Performance) work properly.
3086 *
3087 * This rule does NOT stand when exceeds global or memcg's hard
3088 * limit, or else a DoS attack can be taken place by spawning
3089 * lots of sockets whose usage are under minimum buffer size.
3090 */
3ab224be 3091 if (kind == SK_MEM_RECV) {
a3dcaf17 3092 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3ab224be 3093 return 1;
180d8cd9 3094
3ab224be 3095 } else { /* SK_MEM_SEND */
a3dcaf17
ED		 3096 int wmem0 = sk_get_wmem0(sk, prot);
3097
3ab224be 3098 if (sk->sk_type == SOCK_STREAM) {
a3dcaf17 3099 if (sk->sk_wmem_queued < wmem0)
3ab224be 3100 return 1;
a3dcaf17 3101 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3ab224be 3102 return 1;
a3dcaf17 3103 }
3ab224be
HA		 3104 }
3105
180d8cd9 3106 if (sk_has_memory_pressure(sk)) {
5bf325a5 3107 u64 alloc;
1748376b 3108
66e6369e
AW		 3109 /* The following 'average' heuristic is within the
3110 * scope of global accounting, so it only makes
3111 * sense for global memory pressure.
3112 */
3113 if (!sk_under_global_memory_pressure(sk))
1748376b 3114 return 1;
2e12072c
AW		 3115
3116 /* Try to be fair among all the sockets under global
3117 * pressure by allowing the ones that below average
3118 * usage to raise.
3119 */
180d8cd9
GC		 3120 alloc = sk_sockets_allocated_read_positive(sk);
3121 if (sk_prot_mem_limits(sk, 2) > alloc *
3ab224be
HA		 3122 sk_mem_pages(sk->sk_wmem_queued +
3123 atomic_read(&sk->sk_rmem_alloc) +
3124 sk->sk_forward_alloc))
3125 return 1;
3126 }
3127
3128suppress_allocation:
3129
3130 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3131 sk_stream_moderate_sndbuf(sk);
3132
3133 /* Fail only if socket is _under_ its sndbuf.
3134 * In this case we cannot block, so that we have to fail.
3135 */
4b1327be
WW		 3136 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3137 /* Force charge with __GFP_NOFAIL */
2def8ff3
AW		 3138 if (memcg && !charged) {
3139 mem_cgroup_charge_skmem(memcg, amt,
4b1327be
WW		 3140 gfp_memcg_charge() | __GFP_NOFAIL);
3141 }
3ab224be 3142 return 1;
4b1327be 3143 }
3ab224be
HA		 3144 }
3145
d6f19938
YS		 3146 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3147 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3847ce32 3148
0e90b31f 3149 sk_memory_allocated_sub(sk, amt);
180d8cd9 3150
2def8ff3
AW		 3151 if (charged)
3152 mem_cgroup_uncharge_skmem(memcg, amt);
e805605c 3153
3ab224be
HA		 3154 return 0;
3155}
f8c3bf00
PA		 3156
3157/**
3158 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3159 * @sk: socket
3160 * @size: memory size to allocate
3161 * @kind: allocation type
3162 *
3163 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3164 * rmem allocation. This function assumes that protocols which have
3165 * memory_pressure use sk_wmem_queued as write buffer accounting.
3166 */
3167int __sk_mem_schedule(struct sock *sk, int size, int kind)
3168{
3169 int ret, amt = sk_mem_pages(size);
3170
5e6300e7 3171 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
f8c3bf00
PA		 3172 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3173 if (!ret)
5e6300e7 3174 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
f8c3bf00
PA		 3175 return ret;
3176}
3ab224be
HA		 3177EXPORT_SYMBOL(__sk_mem_schedule);
3178
3179/**
f8c3bf00 3180 * __sk_mem_reduce_allocated - reclaim memory_allocated
3ab224be 3181 * @sk: socket
f8c3bf00
PA		 3182 * @amount: number of quanta
3183 *
3184 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3ab224be 3185 */
f8c3bf00 3186void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3ab224be 3187{
1a24e04e 3188 sk_memory_allocated_sub(sk, amount);
3ab224be 3189
baac50bb
JW		 3190 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3191 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
e805605c 3192
2d0c88e8 3193 if (sk_under_global_memory_pressure(sk) &&
180d8cd9
GC		 3194 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3195 sk_leave_memory_pressure(sk);
3ab224be 3196}
f8c3bf00
PA		 3197
3198/**
3199 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3200 * @sk: socket
100fdd1f 3201 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
f8c3bf00
PA		 3202 */
3203void __sk_mem_reclaim(struct sock *sk, int amount)
3204{
100fdd1f 3205 amount >>= PAGE_SHIFT;
5e6300e7 3206 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
f8c3bf00
PA		 3207 __sk_mem_reduce_allocated(sk, amount);
3208}
3ab224be
HA		 3209EXPORT_SYMBOL(__sk_mem_reclaim);
3210
627d2d6b 3211int sk_set_peek_off(struct sock *sk, int val)
3212{
11695c6e 3213 WRITE_ONCE(sk->sk_peek_off, val);
627d2d6b 3214 return 0;
3215}
3216EXPORT_SYMBOL_GPL(sk_set_peek_off);
3ab224be 3217
1da177e4
LT		 3218/*
3219 * Set of default routines for initialising struct proto_ops when
3220 * the protocol does not support a particular function. In certain
3221 * cases where it makes no sense for a protocol to have a "do nothing"
3222 * function, some default processing is provided.
3223 */
3224
3225int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3226{
3227 return -EOPNOTSUPP;
3228}
2a91525c 3229EXPORT_SYMBOL(sock_no_bind);
1da177e4 3230
4ec93edb 3231int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 3232 int len, int flags)
3233{
3234 return -EOPNOTSUPP;
3235}
2a91525c 3236EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 3237
3238int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3239{
3240 return -EOPNOTSUPP;
3241}
2a91525c 3242EXPORT_SYMBOL(sock_no_socketpair);
1da177e4 3243
92ef0fd5
JA		 3244int sock_no_accept(struct socket *sock, struct socket *newsock,
3245 struct proto_accept_arg *arg)
1da177e4
LT		 3246{
3247 return -EOPNOTSUPP;
3248}
2a91525c 3249EXPORT_SYMBOL(sock_no_accept);
1da177e4 3250
4ec93edb 3251int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
9b2c45d4 3252 int peer)
1da177e4
LT		 3253{
3254 return -EOPNOTSUPP;
3255}
2a91525c 3256EXPORT_SYMBOL(sock_no_getname);
1da177e4 3257
1da177e4
LT		 3258int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3259{
3260 return -EOPNOTSUPP;
3261}
2a91525c 3262EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 3263
3264int sock_no_listen(struct socket *sock, int backlog)
3265{
3266 return -EOPNOTSUPP;
3267}
2a91525c 3268EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 3269
3270int sock_no_shutdown(struct socket *sock, int how)
3271{
3272 return -EOPNOTSUPP;
3273}
2a91525c 3274EXPORT_SYMBOL(sock_no_shutdown);
1da177e4 3275
1b784140 3276int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
1da177e4
LT		 3277{
3278 return -EOPNOTSUPP;
3279}
2a91525c 3280EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4 3281
306b13eb
TH		 3282int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3283{
3284 return -EOPNOTSUPP;
3285}
3286EXPORT_SYMBOL(sock_no_sendmsg_locked);
3287
1b784140
YX		 3288int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3289 int flags)
1da177e4
LT		 3290{
3291 return -EOPNOTSUPP;
3292}
2a91525c 3293EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 3294
3295int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3296{
3297 /* Mirror missing mmap method error code */
3298 return -ENODEV;
3299}
2a91525c 3300EXPORT_SYMBOL(sock_no_mmap);
1da177e4 3301
d9539752
KC		 3302/*
3303 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3304 * various sock-based usage counts.
3305 */
3306void __receive_sock(struct file *file)
3307{
3308 struct socket *sock;
d9539752 3309
dba4a925 3310 sock = sock_from_file(file);
d9539752
KC		 3311 if (sock) {
3312 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3313 sock_update_classid(&sock->sk->sk_cgrp_data);
3314 }
3315}
3316
1da177e4
LT		 3317/*
3318 * Default Socket Callbacks
3319 */
3320
3321static void sock_def_wakeup(struct sock *sk)
3322{
43815482
ED		 3323 struct socket_wq *wq;
3324
3325 rcu_read_lock();
3326 wq = rcu_dereference(sk->sk_wq);
1ce0bf50 3327 if (skwq_has_sleeper(wq))
43815482
ED		 3328 wake_up_interruptible_all(&wq->wait);
3329 rcu_read_unlock();
1da177e4
LT		 3330}
3331
3332static void sock_def_error_report(struct sock *sk)
3333{
43815482
ED		 3334 struct socket_wq *wq;
3335
3336 rcu_read_lock();
3337 wq = rcu_dereference(sk->sk_wq);
1ce0bf50 3338 if (skwq_has_sleeper(wq))
a9a08845 3339 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
1abe267f 3340 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 3341 rcu_read_unlock();
1da177e4
LT		 3342}
3343
43a825af 3344void sock_def_readable(struct sock *sk)
1da177e4 3345{
43815482
ED		 3346 struct socket_wq *wq;
3347
40e0b090
PY		 3348 trace_sk_data_ready(sk);
3349
43815482
ED		 3350 rcu_read_lock();
3351 wq = rcu_dereference(sk->sk_wq);
1ce0bf50 3352 if (skwq_has_sleeper(wq))
a9a08845
LT		 3353 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3354 EPOLLRDNORM | EPOLLRDBAND);
1abe267f 3355 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 3356 rcu_read_unlock();
1da177e4
LT		 3357}
3358
3359static void sock_def_write_space(struct sock *sk)
3360{
43815482
ED		 3361 struct socket_wq *wq;
3362
3363 rcu_read_lock();
1da177e4
LT		 3364
3365 /* Do not wake up a writer until he can make "significant"
3366 * progress. --DaveM
3367 */
14bfee9b 3368 if (sock_writeable(sk)) {
43815482 3369 wq = rcu_dereference(sk->sk_wq);
1ce0bf50 3370 if (skwq_has_sleeper(wq))
a9a08845
LT		 3371 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3372 EPOLLWRNORM | EPOLLWRBAND);
1da177e4
LT		 3373
3374 /* Should agree with poll, otherwise some programs break */
1abe267f 3375 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 3376 }
3377
43815482 3378 rcu_read_unlock();
1da177e4
LT		 3379}
3380
0a8afd9f
PB		 3381/* An optimised version of sock_def_write_space(), should only be called
3382 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3383 * ->sk_wmem_alloc.
3384 */
3385static void sock_def_write_space_wfree(struct sock *sk)
3386{
3387 /* Do not wake up a writer until he can make "significant"
3388 * progress. --DaveM
3389 */
3390 if (sock_writeable(sk)) {
3391 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3392
3393 /* rely on refcount_sub from sock_wfree() */
3394 smp_mb__after_atomic();
3395 if (wq && waitqueue_active(&wq->wait))
3396 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3397 EPOLLWRNORM | EPOLLWRBAND);
3398
3399 /* Should agree with poll, otherwise some programs break */
1abe267f 3400 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
0a8afd9f
PB		 3401 }
3402}
3403
1da177e4
LT		 3404static void sock_def_destruct(struct sock *sk)
3405{
1da177e4
LT		 3406}
3407
3408void sk_send_sigurg(struct sock *sk)
3409{
3410 if (sk->sk_socket && sk->sk_socket->file)
3411 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 3412 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 3413}
2a91525c 3414EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 3415
3416void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3417 unsigned long expires)
3418{
3419 if (!mod_timer(timer, expires))
3420 sock_hold(sk);
3421}
1da177e4
LT		 3422EXPORT_SYMBOL(sk_reset_timer);
3423
3424void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3425{
25cc4ae9 3426 if (del_timer(timer))
1da177e4
LT		 3427 __sock_put(sk);
3428}
1da177e4
LT		 3429EXPORT_SYMBOL(sk_stop_timer);
3430
08b81d87
GT		 3431void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3432{
3433 if (del_timer_sync(timer))
3434 __sock_put(sk);
3435}
3436EXPORT_SYMBOL(sk_stop_timer_sync);
3437
584f3742 3438void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
1da177e4 3439{
581319c5 3440 sk_init_common(sk);
1da177e4
LT		 3441 sk->sk_send_head = NULL;
3442
99767f27 3443 timer_setup(&sk->sk_timer, NULL, 0);
4ec93edb 3444
1da177e4 3445 sk->sk_allocation = GFP_KERNEL;
1227c177
KI		 3446 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3447 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
1da177e4 3448 sk->sk_state = TCP_CLOSE;
fb87bd47 3449 sk->sk_use_task_frag = true;
972692e0 3450 sk_set_socket(sk, sock);
1da177e4
LT		 3451
3452 sock_set_flag(sk, SOCK_ZAPPED);
3453
e71a4783 3454 if (sock) {
1da177e4 3455 sk->sk_type = sock->type;
333f7909 3456 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
1da177e4 3457 sock->sk = sk;
86741ec2 3458 } else {
c2f26e8f 3459 RCU_INIT_POINTER(sk->sk_wq, NULL);
86741ec2 3460 }
584f3742 3461 sk->sk_uid = uid;
1da177e4 3462
1da177e4 3463 rwlock_init(&sk->sk_callback_lock);
cdfbabfb
DH		 3464 if (sk->sk_kern_sock)
3465 lockdep_set_class_and_name(
3466 &sk->sk_callback_lock,
3467 af_kern_callback_keys + sk->sk_family,
3468 af_family_kern_clock_key_strings[sk->sk_family]);
3469 else
3470 lockdep_set_class_and_name(
3471 &sk->sk_callback_lock,
443aef0e
PZ		 3472 af_callback_keys + sk->sk_family,
3473 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 3474
3475 sk->sk_state_change = sock_def_wakeup;
3476 sk->sk_data_ready = sock_def_readable;
3477 sk->sk_write_space = sock_def_write_space;
3478 sk->sk_error_report = sock_def_error_report;
3479 sk->sk_destruct = sock_def_destruct;
3480
5640f768
ED		 3481 sk->sk_frag.page = NULL;
3482 sk->sk_frag.offset = 0;
ef64a54f 3483 sk->sk_peek_off = -1;
1da177e4 3484
109f6e39
EB		 3485 sk->sk_peer_pid = NULL;
3486 sk->sk_peer_cred = NULL;
35306eb2
ED		 3487 spin_lock_init(&sk->sk_peer_lock);
3488
1da177e4
LT		 3489 sk->sk_write_pending = 0;
3490 sk->sk_rcvlowat = 1;
3491 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3492 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3493
6c7c98ba 3494 sk->sk_stamp = SK_DEFAULT_STAMP;
3a0ed3e9
DD		 3495#if BITS_PER_LONG==32
3496 seqlock_init(&sk->sk_stamp_seq);
3497#endif
52267790 3498 atomic_set(&sk->sk_zckey, 0);
1da177e4 3499
e0d1095a 3500#ifdef CONFIG_NET_RX_BUSY_POLL
06021292 3501 sk->sk_napi_id = 0;
e59ef36f 3502 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
06021292
ET		 3503#endif
3504
76a9ebe8
ED		 3505 sk->sk_max_pacing_rate = ~0UL;
3506 sk->sk_pacing_rate = ~0UL;
7c68fa2b 3507 WRITE_ONCE(sk->sk_pacing_shift, 10);
70da268b 3508 sk->sk_incoming_cpu = -1;
c6345ce7
AN		 3509
3510 sk_rx_queue_clear(sk);
4dc6dc71
ED		 3511 /*
3512 * Before updating sk_refcnt, we must commit prior changes to memory
2cdb54c9 3513 * (Documentation/RCU/rculist_nulls.rst for details)
4dc6dc71
ED		 3514 */
3515 smp_wmb();
41c6d650 3516 refcount_set(&sk->sk_refcnt, 1);
33c732c3 3517 atomic_set(&sk->sk_drops, 0);
1da177e4 3518}
584f3742
PB		 3519EXPORT_SYMBOL(sock_init_data_uid);
3520
3521void sock_init_data(struct socket *sock, struct sock *sk)
3522{
3523 kuid_t uid = sock ?
3524 SOCK_INODE(sock)->i_uid :
3525 make_kuid(sock_net(sk)->user_ns, 0);
3526
3527 sock_init_data_uid(sock, sk, uid);
3528}
2a91525c 3529EXPORT_SYMBOL(sock_init_data);
1da177e4 3530
b5606c2d 3531void lock_sock_nested(struct sock *sk, int subclass)
1da177e4 3532{
2dcb96ba
TG		 3533 /* The sk_lock has mutex_lock() semantics here. */
3534 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3535
1da177e4 3536 might_sleep();
a5b5bb9a 3537 spin_lock_bh(&sk->sk_lock.slock);
33d60fbd 3538 if (sock_owned_by_user_nocheck(sk))
1da177e4 3539 __lock_sock(sk);
d2e9117c 3540 sk->sk_lock.owned = 1;
2dcb96ba 3541 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4 3542}
fcc70d5f 3543EXPORT_SYMBOL(lock_sock_nested);
1da177e4 3544
b5606c2d 3545void release_sock(struct sock *sk)
1da177e4 3546{
a5b5bb9a 3547 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 3548 if (sk->sk_backlog.tail)
3549 __release_sock(sk);
46d3ceab
ED		 3550
3551 if (sk->sk_prot->release_cb)
41862d12
ED		 3552 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3553 tcp_release_cb, sk);
46d3ceab 3554
c3f9b018 3555 sock_release_ownership(sk);
a5b5bb9a
IM		 3556 if (waitqueue_active(&sk->sk_lock.wq))
3557 wake_up(&sk->sk_lock.wq);
3558 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 3559}
3560EXPORT_SYMBOL(release_sock);
3561
49054556 3562bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
8a74ad60
ED		 3563{
3564 might_sleep();
3565 spin_lock_bh(&sk->sk_lock.slock);
3566
33d60fbd 3567 if (!sock_owned_by_user_nocheck(sk)) {
8a74ad60 3568 /*
2dcb96ba
TG		 3569 * Fast path return with bottom halves disabled and
3570 * sock::sk_lock.slock held.
3571 *
3572 * The 'mutex' is not contended and holding
3573 * sock::sk_lock.slock prevents all other lockers to
3574 * proceed so the corresponding unlock_sock_fast() can
3575 * avoid the slow path of release_sock() completely and
3576 * just release slock.
3577 *
3578 * From a semantical POV this is equivalent to 'acquiring'
3579 * the 'mutex', hence the corresponding lockdep
3580 * mutex_release() has to happen in the fast path of
3581 * unlock_sock_fast().
8a74ad60
ED		 3582 */
3583 return false;
2dcb96ba 3584 }
8a74ad60
ED		 3585
3586 __lock_sock(sk);
3587 sk->sk_lock.owned = 1;
12f4bd86 3588 __acquire(&sk->sk_lock.slock);
2dcb96ba 3589 spin_unlock_bh(&sk->sk_lock.slock);
8a74ad60
ED		 3590 return true;
3591}
49054556 3592EXPORT_SYMBOL(__lock_sock_fast);
8a74ad60 3593
c7cbdbf2
AB		 3594int sock_gettstamp(struct socket *sock, void __user *userstamp,
3595 bool timeval, bool time32)
4ec93edb 3596{
c7cbdbf2
AB		 3597 struct sock *sk = sock->sk;
3598 struct timespec64 ts;
9dae3497
YS		 3599
3600 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
c7cbdbf2
AB		 3601 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3602 if (ts.tv_sec == -1)
1da177e4 3603 return -ENOENT;
c7cbdbf2 3604 if (ts.tv_sec == 0) {
3a0ed3e9 3605 ktime_t kt = ktime_get_real();
f95f96a4 3606 sock_write_timestamp(sk, kt);
c7cbdbf2 3607 ts = ktime_to_timespec64(kt);
b7aa0bf7 3608 }
1da177e4 3609
c7cbdbf2
AB		 3610 if (timeval)
3611 ts.tv_nsec /= 1000;
9dae3497 3612
c7cbdbf2
AB		 3613#ifdef CONFIG_COMPAT_32BIT_TIME
3614 if (time32)
3615 return put_old_timespec32(&ts, userstamp);
3616#endif
3617#ifdef CONFIG_SPARC64
3618 /* beware of padding in sparc64 timeval */
3619 if (timeval && !in_compat_syscall()) {
3620 struct __kernel_old_timeval __user tv = {
c98f4822
SR		 3621 .tv_sec = ts.tv_sec,
3622 .tv_usec = ts.tv_nsec,
c7cbdbf2 3623 };
c98f4822 3624 if (copy_to_user(userstamp, &tv, sizeof(tv)))
c7cbdbf2
AB		 3625 return -EFAULT;
3626 return 0;
ae40eb1e 3627 }
c7cbdbf2
AB		 3628#endif
3629 return put_timespec64(&ts, userstamp);
ae40eb1e 3630}
c7cbdbf2 3631EXPORT_SYMBOL(sock_gettstamp);
ae40eb1e 3632
193d357d 3633void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
4ec93edb 3634{
20d49473 3635 if (!sock_flag(sk, flag)) {
08e29af3
ED		 3636 unsigned long previous_flags = sk->sk_flags;
3637
20d49473
PO		 3638 sock_set_flag(sk, flag);
3639 /*
3640 * we just set one of the two flags which require net
3641 * time stamping, but time stamping might have been on
3642 * already because of the other one
3643 */
080a270f
HFS		 3644 if (sock_needs_netstamp(sk) &&
3645 !(previous_flags & SK_FLAGS_TIMESTAMP))
20d49473 3646 net_enable_timestamp();
1da177e4
LT		 3647 }
3648}
1da177e4 3649
cb820f8e
RC		 3650int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3651 int level, int type)
3652{
3653 struct sock_exterr_skb *serr;
364a9e93 3654 struct sk_buff *skb;
cb820f8e
RC		 3655 int copied, err;
3656
3657 err = -EAGAIN;
364a9e93 3658 skb = sock_dequeue_err_skb(sk);
cb820f8e
RC		 3659 if (skb == NULL)
3660 goto out;
3661
3662 copied = skb->len;
3663 if (copied > len) {
3664 msg->msg_flags |= MSG_TRUNC;
3665 copied = len;
3666 }
51f3d02b 3667 err = skb_copy_datagram_msg(skb, 0, msg, copied);
cb820f8e
RC		 3668 if (err)
3669 goto out_free_skb;
3670
3671 sock_recv_timestamp(msg, sk, skb);
3672
3673 serr = SKB_EXT_ERR(skb);
3674 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3675
3676 msg->msg_flags |= MSG_ERRQUEUE;
3677 err = copied;
3678
cb820f8e
RC		 3679out_free_skb:
3680 kfree_skb(skb);
3681out:
3682 return err;
3683}
3684EXPORT_SYMBOL(sock_recv_errqueue);
3685
1da177e4
LT		 3686/*
3687 * Get a socket option on an socket.
3688 *
3689 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3690 * asynchronous errors should be reported by getsockopt. We assume
3691 * this means if you specify SO_ERROR (otherwise whats the point of it).
3692 */
3693int sock_common_getsockopt(struct socket *sock, int level, int optname,
3694 char __user *optval, int __user *optlen)
3695{
3696 struct sock *sk = sock->sk;
3697
364f997b
KI		 3698 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3699 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
1da177e4 3700}
1da177e4
LT		 3701EXPORT_SYMBOL(sock_common_getsockopt);
3702
1b784140
YX		 3703int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3704 int flags)
1da177e4
LT		 3705{
3706 struct sock *sk = sock->sk;
3707 int addr_len = 0;
3708 int err;
3709
ec095263 3710 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
1da177e4
LT		 3711 if (err >= 0)
3712 msg->msg_namelen = addr_len;
3713 return err;
3714}
1da177e4
LT		 3715EXPORT_SYMBOL(sock_common_recvmsg);
3716
3717/*
3718 * Set socket options on an inet socket.
3719 */
3720int sock_common_setsockopt(struct socket *sock, int level, int optname,
a7b75c5a 3721 sockptr_t optval, unsigned int optlen)
1da177e4
LT		 3722{
3723 struct sock *sk = sock->sk;
3724
364f997b
KI		 3725 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3726 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
1da177e4 3727}
1da177e4
LT		 3728EXPORT_SYMBOL(sock_common_setsockopt);
3729
3730void sk_common_release(struct sock *sk)
3731{
3732 if (sk->sk_prot->destroy)
3733 sk->sk_prot->destroy(sk);
3734
3735 /*
645f0897 3736 * Observation: when sk_common_release is called, processes have
1da177e4
LT		 3737 * no access to socket. But net still has.
3738 * Step one, detach it from networking:
3739 *
3740 * A. Remove from hash tables.
3741 */
3742
3743 sk->sk_prot->unhash(sk);
3744
3745 /*
3746 * In this point socket cannot receive new packets, but it is possible
3747 * that some packets are in flight because some CPU runs receiver and
3748 * did hash table lookup before we unhashed socket. They will achieve
3749 * receive queue and will be purged by socket destructor.
3750 *
3751 * Also we still have packets pending on receive queue and probably,
3752 * our own packets waiting in device queues. sock_destroy will drain
3753 * receive queue, but transmitted packets will delay socket destruction
3754 * until the last reference will be released.
3755 */
3756
3757 sock_orphan(sk);
3758
3759 xfrm_sk_free_policy(sk);
3760
1da177e4
LT		 3761 sock_put(sk);
3762}
1da177e4
LT		 3763EXPORT_SYMBOL(sk_common_release);
3764
a2d133b1
JH		 3765void sk_get_meminfo(const struct sock *sk, u32 *mem)
3766{
3767 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3768
3769 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
ebb3b78d 3770 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
a2d133b1 3771 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
e292f05e 3772 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
66d58f04 3773 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
ab4e846a 3774 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
a2d133b1 3775 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
70c26558 3776 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
a2d133b1
JH		 3777 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3778}
3779
13ff3d6f 3780#ifdef CONFIG_PROC_FS
13ff3d6f 3781static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159 3782
70ee1159
PE		 3783int sock_prot_inuse_get(struct net *net, struct proto *prot)
3784{
3785 int cpu, idx = prot->inuse_idx;
3786 int res = 0;
3787
3788 for_each_possible_cpu(cpu)
08fc7f81 3789 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
70ee1159
PE		 3790
3791 return res >= 0 ? res : 0;
3792}
3793EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3794
648845ab
TZ		 3795int sock_inuse_get(struct net *net)
3796{
3797 int cpu, res = 0;
3798
3799 for_each_possible_cpu(cpu)
4199bae1 3800 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
648845ab
TZ		 3801
3802 return res;
3803}
3804
3805EXPORT_SYMBOL_GPL(sock_inuse_get);
3806
2c8c1e72 3807static int __net_init sock_inuse_init_net(struct net *net)
70ee1159 3808{
08fc7f81 3809 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
648845ab
TZ		 3810 if (net->core.prot_inuse == NULL)
3811 return -ENOMEM;
648845ab 3812 return 0;
70ee1159
PE		 3813}
3814
2c8c1e72 3815static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159 3816{
08fc7f81 3817 free_percpu(net->core.prot_inuse);
70ee1159
PE		 3818}
3819
3820static struct pernet_operations net_inuse_ops = {
3821 .init = sock_inuse_init_net,
3822 .exit = sock_inuse_exit_net,
3823};
3824
3825static __init int net_inuse_init(void)
3826{
3827 if (register_pernet_subsys(&net_inuse_ops))
3828 panic("Cannot initialize net inuse counters");
3829
3830 return 0;
3831}
3832
3833core_initcall(net_inuse_init);
13ff3d6f 3834
b45ce321 3835static int assign_proto_idx(struct proto *prot)
13ff3d6f
PE		 3836{
3837 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3838
3839 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
e005d193 3840 pr_err("PROTO_INUSE_NR exhausted\n");
b45ce321 3841 return -ENOSPC;
13ff3d6f
PE		 3842 }
3843
3844 set_bit(prot->inuse_idx, proto_inuse_idx);
b45ce321 3845 return 0;
13ff3d6f
PE		 3846}
3847
3848static void release_proto_idx(struct proto *prot)
3849{
3850 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3851 clear_bit(prot->inuse_idx, proto_inuse_idx);
3852}
3853#else
b45ce321 3854static inline int assign_proto_idx(struct proto *prot)
13ff3d6f 3855{
b45ce321 3856 return 0;
13ff3d6f
PE		 3857}
3858
3859static inline void release_proto_idx(struct proto *prot)
3860{
3861}
648845ab 3862
13ff3d6f
PE		 3863#endif
3864
0f5907af
ML		 3865static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3866{
3867 if (!twsk_prot)
3868 return;
3869 kfree(twsk_prot->twsk_slab_name);
3870 twsk_prot->twsk_slab_name = NULL;
3871 kmem_cache_destroy(twsk_prot->twsk_slab);
3872 twsk_prot->twsk_slab = NULL;
3873}
3874
b80350f3
TZ		 3875static int tw_prot_init(const struct proto *prot)
3876{
3877 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3878
3879 if (!twsk_prot)
3880 return 0;
3881
3882 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3883 prot->name);
3884 if (!twsk_prot->twsk_slab_name)
3885 return -ENOMEM;
3886
3887 twsk_prot->twsk_slab =
3888 kmem_cache_create(twsk_prot->twsk_slab_name,
3889 twsk_prot->twsk_obj_size, 0,
3890 SLAB_ACCOUNT | prot->slab_flags,
3891 NULL);
3892 if (!twsk_prot->twsk_slab) {
3893 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3894 prot->name);
3895 return -ENOMEM;
3896 }
3897
3898 return 0;
3899}
3900
0159dfd3
ED		 3901static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3902{
3903 if (!rsk_prot)
3904 return;
3905 kfree(rsk_prot->slab_name);
3906 rsk_prot->slab_name = NULL;
adf78eda
JL		 3907 kmem_cache_destroy(rsk_prot->slab);
3908 rsk_prot->slab = NULL;
0159dfd3
ED		 3909}
3910
3911static int req_prot_init(const struct proto *prot)
3912{
3913 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3914
3915 if (!rsk_prot)
3916 return 0;
3917
3918 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3919 prot->name);
3920 if (!rsk_prot->slab_name)
3921 return -ENOMEM;
3922
3923 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3924 rsk_prot->obj_size, 0,
e699e2c6
SB		 3925 SLAB_ACCOUNT | prot->slab_flags,
3926 NULL);
0159dfd3
ED		 3927
3928 if (!rsk_prot->slab) {
3929 pr_crit("%s: Can't create request sock SLAB cache!\n",
3930 prot->name);
3931 return -ENOMEM;
3932 }
3933 return 0;
3934}
3935
b733c007
PE		 3936int proto_register(struct proto *prot, int alloc_slab)
3937{
b45ce321 3938 int ret = -ENOBUFS;
3939
f20cfd66
ED		 3940 if (prot->memory_allocated && !prot->sysctl_mem) {
3941 pr_err("%s: missing sysctl_mem\n", prot->name);
3942 return -EINVAL;
3943 }
0defbb0a
ED		 3944 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3945 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3946 return -EINVAL;
3947 }
1da177e4 3948 if (alloc_slab) {
30c2c9f1
DW		 3949 prot->slab = kmem_cache_create_usercopy(prot->name,
3950 prot->obj_size, 0,
e699e2c6
SB		 3951 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3952 prot->slab_flags,
289a4860 3953 prot->useroffset, prot->usersize,
271b72c7 3954 NULL);
1da177e4
LT		 3955
3956 if (prot->slab == NULL) {
e005d193
JP		 3957 pr_crit("%s: Can't create sock SLAB cache!\n",
3958 prot->name);
60e7663d 3959 goto out;
1da177e4 3960 }
2e6599cb 3961
0159dfd3
ED		 3962 if (req_prot_init(prot))
3963 goto out_free_request_sock_slab;
8feaf0c0 3964
b80350f3
TZ		 3965 if (tw_prot_init(prot))
3966 goto out_free_timewait_sock_slab;
1da177e4
LT		 3967 }
3968
36b77a52 3969 mutex_lock(&proto_list_mutex);
b45ce321 3970 ret = assign_proto_idx(prot);
3971 if (ret) {
3972 mutex_unlock(&proto_list_mutex);
0f5907af 3973 goto out_free_timewait_sock_slab;
b45ce321 3974 }
1da177e4 3975 list_add(&prot->node, &proto_list);
36b77a52 3976 mutex_unlock(&proto_list_mutex);
b45ce321 3977 return ret;
b733c007 3978
0f5907af 3979out_free_timewait_sock_slab:
ed744d81 3980 if (alloc_slab)
0f5907af 3981 tw_prot_cleanup(prot->twsk_prot);
8feaf0c0 3982out_free_request_sock_slab:
b45ce321 3983 if (alloc_slab) {
3984 req_prot_cleanup(prot->rsk_prot);
0159dfd3 3985
b45ce321 3986 kmem_cache_destroy(prot->slab);
3987 prot->slab = NULL;
3988 }
b733c007 3989out:
b45ce321 3990 return ret;
1da177e4 3991}
1da177e4
LT		 3992EXPORT_SYMBOL(proto_register);
3993
3994void proto_unregister(struct proto *prot)
3995{
36b77a52 3996 mutex_lock(&proto_list_mutex);
13ff3d6f 3997 release_proto_idx(prot);
0a3f4358 3998 list_del(&prot->node);
36b77a52 3999 mutex_unlock(&proto_list_mutex);
1da177e4 4000
adf78eda
JL		 4001 kmem_cache_destroy(prot->slab);
4002 prot->slab = NULL;
1da177e4 4003
0159dfd3 4004 req_prot_cleanup(prot->rsk_prot);
0f5907af 4005 tw_prot_cleanup(prot->twsk_prot);
1da177e4 4006}
1da177e4
LT		 4007EXPORT_SYMBOL(proto_unregister);
4008
bf2ae2e4
XL		 4009int sock_load_diag_module(int family, int protocol)
4010{
4011 if (!protocol) {
4012 if (!sock_is_registered(family))
4013 return -ENOENT;
4014
4015 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4016 NETLINK_SOCK_DIAG, family);
4017 }
4018
4019#ifdef CONFIG_INET
4020 if (family == AF_INET &&
c34c1287 4021 protocol != IPPROTO_RAW &&
3f935c75 4022 protocol < MAX_INET_PROTOS &&
bf2ae2e4
XL		 4023 !rcu_access_pointer(inet_protos[protocol]))
4024 return -ENOENT;
4025#endif
4026
4027 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4028 NETLINK_SOCK_DIAG, family, protocol);
4029}
4030EXPORT_SYMBOL(sock_load_diag_module);
4031
1da177e4 4032#ifdef CONFIG_PROC_FS
1da177e4 4033static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
36b77a52 4034 __acquires(proto_list_mutex)
1da177e4 4035{
36b77a52 4036 mutex_lock(&proto_list_mutex);
60f0438a 4037 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 4038}
4039
4040static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4041{
60f0438a 4042 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 4043}
4044
4045static void proto_seq_stop(struct seq_file *seq, void *v)
36b77a52 4046 __releases(proto_list_mutex)
1da177e4 4047{
36b77a52 4048 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 4049}
4050
4051static char proto_method_implemented(const void *method)
4052{
4053 return method == NULL ? 'n' : 'y';
4054}
180d8cd9
GC		 4055static long sock_prot_memory_allocated(struct proto *proto)
4056{
cb75a36c 4057 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
180d8cd9
GC		 4058}
4059
7a512eb8 4060static const char *sock_prot_memory_pressure(struct proto *proto)
180d8cd9
GC		 4061{
4062 return proto->memory_pressure != NULL ?
4063 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4064}
1da177e4
LT		 4065
4066static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4067{
180d8cd9 4068
8d987e5c 4069 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
dc97391e 4070 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1da177e4
LT		 4071 proto->name,
4072 proto->obj_size,
14e943db 4073 sock_prot_inuse_get(seq_file_net(seq), proto),
180d8cd9
GC		 4074 sock_prot_memory_allocated(proto),
4075 sock_prot_memory_pressure(proto),
1da177e4
LT		 4076 proto->max_header,
4077 proto->slab == NULL ? "no" : "yes",
4078 module_name(proto->owner),
4079 proto_method_implemented(proto->close),
4080 proto_method_implemented(proto->connect),
4081 proto_method_implemented(proto->disconnect),
4082 proto_method_implemented(proto->accept),
4083 proto_method_implemented(proto->ioctl),
4084 proto_method_implemented(proto->init),
4085 proto_method_implemented(proto->destroy),
4086 proto_method_implemented(proto->shutdown),
4087 proto_method_implemented(proto->setsockopt),
4088 proto_method_implemented(proto->getsockopt),
4089 proto_method_implemented(proto->sendmsg),
4090 proto_method_implemented(proto->recvmsg),
1da177e4
LT		 4091 proto_method_implemented(proto->bind),
4092 proto_method_implemented(proto->backlog_rcv),
4093 proto_method_implemented(proto->hash),
4094 proto_method_implemented(proto->unhash),
4095 proto_method_implemented(proto->get_port),
4096 proto_method_implemented(proto->enter_memory_pressure));
4097}
4098
4099static int proto_seq_show(struct seq_file *seq, void *v)
4100{
60f0438a 4101 if (v == &proto_list)
1da177e4
LT		 4102 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4103 "protocol",
4104 "size",
4105 "sockets",
4106 "memory",
4107 "press",
4108 "maxhdr",
4109 "slab",
4110 "module",
dc97391e 4111 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
1da177e4 4112 else
60f0438a 4113 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 4114 return 0;
4115}
4116
f690808e 4117static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 4118 .start = proto_seq_start,
4119 .next = proto_seq_next,
4120 .stop = proto_seq_stop,
4121 .show = proto_seq_show,
4122};
4123
14e943db
ED		 4124static __net_init int proto_init_net(struct net *net)
4125{
c3506372
CH		 4126 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4127 sizeof(struct seq_net_private)))
14e943db
ED		 4128 return -ENOMEM;
4129
4130 return 0;
4131}
4132
4133static __net_exit void proto_exit_net(struct net *net)
4134{
ece31ffd 4135 remove_proc_entry("protocols", net->proc_net);
14e943db
ED		 4136}
4137
4138
4139static __net_initdata struct pernet_operations proto_net_ops = {
4140 .init = proto_init_net,
4141 .exit = proto_exit_net,
1da177e4
LT		 4142};
4143
4144static int __init proto_init(void)
4145{
14e943db 4146 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 4147}
4148
4149subsys_initcall(proto_init);
4150
4151#endif /* PROC_FS */
7db6b048
SS		 4152
4153#ifdef CONFIG_NET_RX_BUSY_POLL
4154bool sk_busy_loop_end(void *p, unsigned long start_time)
4155{
4156 struct sock *sk = p;
4157
a54d51fb
ED		 4158 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4159 return true;
4160
4161 if (sk_is_udp(sk) &&
4162 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4163 return true;
4164
4165 return sk_busy_loop_timeout(sk, start_time);
7db6b048
SS		 4166}
4167EXPORT_SYMBOL(sk_busy_loop_end);
4168#endif /* CONFIG_NET_RX_BUSY_POLL */
c0425a42
CH		 4169
4170int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4171{
4172 if (!sk->sk_prot->bind_add)
4173 return -EOPNOTSUPP;
4174 return sk->sk_prot->bind_add(sk, addr, addr_len);
4175}
4176EXPORT_SYMBOL(sock_bind_add);
e1d001fa
BL		 4177
4178/* Copy 'size' bytes from userspace and return `size` back to userspace */
4179int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4180 void __user *arg, void *karg, size_t size)
4181{
4182 int ret;
4183
4184 if (copy_from_user(karg, arg, size))
4185 return -EFAULT;
4186
4187 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4188 if (ret)
4189 return ret;
4190
4191 if (copy_to_user(arg, karg, size))
4192 return -EFAULT;
4193
4194 return 0;
4195}
4196EXPORT_SYMBOL(sock_ioctl_inout);
4197
4198/* This is the most common ioctl prep function, where the result (4 bytes) is
4199 * copied back to userspace if the ioctl() returns successfully. No input is
4200 * copied from userspace as input argument.
4201 */
4202static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4203{
4204 int ret, karg = 0;
4205
4206 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4207 if (ret)
4208 return ret;
4209
4210 return put_user(karg, (int __user *)arg);
4211}
4212
4213/* A wrapper around sock ioctls, which copies the data from userspace
4214 * (depending on the protocol/ioctl), and copies back the result to userspace.
4215 * The main motivation for this function is to pass kernel memory to the
4216 * protocol ioctl callbacks, instead of userspace memory.
4217 */
4218int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4219{
4220 int rc = 1;
4221
634236b3 4222 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
e1d001fa 4223 rc = ipmr_sk_ioctl(sk, cmd, arg);
634236b3 4224 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
e1d001fa
BL		 4225 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4226 else if (sk_is_phonet(sk))
4227 rc = phonet_sk_ioctl(sk, cmd, arg);
4228
4229 /* If ioctl was processed, returns its value */
4230 if (rc <= 0)
4231 return rc;
4232
4233 /* Otherwise call the default handler */
4234 return sock_ioctl_out(sk, cmd, arg);
4235}
4236EXPORT_SYMBOL(sk_ioctl);
5d4cc874
ED		 4237
4238static int __init sock_struct_check(void)
4239{
4240 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4241 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4242 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4243 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4244 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4245
4246 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4247 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4248 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4249 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4250 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4251 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4252 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4253 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4254 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4255
4256 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4257 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4258 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4259
4260 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4261 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4262 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4263 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4264
4265 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4266 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4267 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf);
4268 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4269 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4270 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4271 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4272 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4273 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4274 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm);
4275 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status);
4276 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4277 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4278 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4279 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4280 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4281
4282 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4283 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4284 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4285 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4286 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4287 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4288 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4289 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4290 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4291 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4292 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4293 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4294 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4295 return 0;
4296}
4297
4298core_initcall(sock_struct_check);
Linux 6.x block layer and io_uring tree(s)