1 // SPDX-License-Identifier: GPL-2.0-or-later
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.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
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>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
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
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
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
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>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
117 #include <linux/mroute.h>
118 #include <linux/mroute6.h>
119 #include <linux/icmpv6.h>
121 #include <linux/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
138 #include <net/bpf_sk_storage.h>
140 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
144 #include <net/phonet/phonet.h>
146 #include <linux/ethtool.h>
150 static DEFINE_MUTEX(proto_list_mutex);
151 static LIST_HEAD(proto_list);
153 static void sock_def_write_space_wfree(struct sock *sk);
154 static void sock_def_write_space(struct sock *sk);
157 * sk_ns_capable - General socket capability test
158 * @sk: Socket to use a capability on or through
159 * @user_ns: The user namespace of the capability to use
160 * @cap: The capability to use
162 * Test to see if the opener of the socket had when the socket was
163 * created and the current process has the capability @cap in the user
164 * namespace @user_ns.
166 bool sk_ns_capable(const struct sock *sk,
167 struct user_namespace *user_ns, int cap)
169 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
170 ns_capable(user_ns, cap);
172 EXPORT_SYMBOL(sk_ns_capable);
175 * sk_capable - Socket global capability test
176 * @sk: Socket to use a capability on or through
177 * @cap: The global capability to use
179 * Test to see if the opener of the socket had when the socket was
180 * created and the current process has the capability @cap in all user
183 bool sk_capable(const struct sock *sk, int cap)
185 return sk_ns_capable(sk, &init_user_ns, cap);
187 EXPORT_SYMBOL(sk_capable);
190 * sk_net_capable - Network namespace socket capability test
191 * @sk: Socket to use a capability on or through
192 * @cap: The capability to use
194 * Test to see if the opener of the socket had when the socket was created
195 * and the current process has the capability @cap over the network namespace
196 * the socket is a member of.
198 bool sk_net_capable(const struct sock *sk, int cap)
200 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
202 EXPORT_SYMBOL(sk_net_capable);
205 * Each address family might have different locking rules, so we have
206 * one slock key per address family and separate keys for internal and
209 static struct lock_class_key af_family_keys[AF_MAX];
210 static struct lock_class_key af_family_kern_keys[AF_MAX];
211 static struct lock_class_key af_family_slock_keys[AF_MAX];
212 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
215 * Make lock validator output more readable. (we pre-construct these
216 * strings build-time, so that runtime initialization of socket
220 #define _sock_locks(x) \
221 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
222 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
223 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
224 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
225 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
226 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
227 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
228 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
229 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
230 x "27" , x "28" , x "AF_CAN" , \
231 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
232 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
233 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
234 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
235 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
239 static const char *const af_family_key_strings[AF_MAX+1] = {
240 _sock_locks("sk_lock-")
242 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("slock-")
245 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("clock-")
249 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
250 _sock_locks("k-sk_lock-")
252 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
253 _sock_locks("k-slock-")
255 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
256 _sock_locks("k-clock-")
258 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
259 _sock_locks("rlock-")
261 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
262 _sock_locks("wlock-")
264 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
265 _sock_locks("elock-")
269 * sk_callback_lock and sk queues locking rules are per-address-family,
270 * so split the lock classes by using a per-AF key:
272 static struct lock_class_key af_callback_keys[AF_MAX];
273 static struct lock_class_key af_rlock_keys[AF_MAX];
274 static struct lock_class_key af_wlock_keys[AF_MAX];
275 static struct lock_class_key af_elock_keys[AF_MAX];
276 static struct lock_class_key af_kern_callback_keys[AF_MAX];
278 /* Run time adjustable parameters. */
279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
280 EXPORT_SYMBOL(sysctl_wmem_max);
281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
282 EXPORT_SYMBOL(sysctl_rmem_max);
283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
286 /* Maximal space eaten by iovec or ancillary data plus some space */
287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
288 EXPORT_SYMBOL(sysctl_optmem_max);
290 int sysctl_tstamp_allow_data __read_mostly = 1;
292 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
293 EXPORT_SYMBOL_GPL(memalloc_socks_key);
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
303 void sk_set_memalloc(struct sock *sk)
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
307 static_branch_inc(&memalloc_socks_key);
309 EXPORT_SYMBOL_GPL(sk_set_memalloc);
311 void sk_clear_memalloc(struct sock *sk)
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
315 static_branch_dec(&memalloc_socks_key);
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
326 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
331 unsigned int noreclaim_flag;
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
336 noreclaim_flag = memalloc_noreclaim_save();
337 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
341 memalloc_noreclaim_restore(noreclaim_flag);
345 EXPORT_SYMBOL(__sk_backlog_rcv);
347 void sk_error_report(struct sock *sk)
349 sk->sk_error_report(sk);
351 switch (sk->sk_family) {
355 trace_inet_sk_error_report(sk);
361 EXPORT_SYMBOL(sk_error_report);
363 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
365 struct __kernel_sock_timeval tv;
367 if (timeo == MAX_SCHEDULE_TIMEOUT) {
371 tv.tv_sec = timeo / HZ;
372 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
375 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
376 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
377 *(struct old_timeval32 *)optval = tv32;
382 struct __kernel_old_timeval old_tv;
383 old_tv.tv_sec = tv.tv_sec;
384 old_tv.tv_usec = tv.tv_usec;
385 *(struct __kernel_old_timeval *)optval = old_tv;
386 return sizeof(old_tv);
389 *(struct __kernel_sock_timeval *)optval = tv;
392 EXPORT_SYMBOL(sock_get_timeout);
394 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
395 sockptr_t optval, int optlen, bool old_timeval)
397 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
398 struct old_timeval32 tv32;
400 if (optlen < sizeof(tv32))
403 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
405 tv->tv_sec = tv32.tv_sec;
406 tv->tv_usec = tv32.tv_usec;
407 } else if (old_timeval) {
408 struct __kernel_old_timeval old_tv;
410 if (optlen < sizeof(old_tv))
412 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
414 tv->tv_sec = old_tv.tv_sec;
415 tv->tv_usec = old_tv.tv_usec;
417 if (optlen < sizeof(*tv))
419 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
425 EXPORT_SYMBOL(sock_copy_user_timeval);
427 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
430 struct __kernel_sock_timeval tv;
431 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
437 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
441 static int warned __read_mostly;
443 WRITE_ONCE(*timeo_p, 0);
444 if (warned < 10 && net_ratelimit()) {
446 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
447 __func__, current->comm, task_pid_nr(current));
451 val = MAX_SCHEDULE_TIMEOUT;
452 if ((tv.tv_sec || tv.tv_usec) &&
453 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
454 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
456 WRITE_ONCE(*timeo_p, val);
460 static bool sock_needs_netstamp(const struct sock *sk)
462 switch (sk->sk_family) {
471 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
473 if (sk->sk_flags & flags) {
474 sk->sk_flags &= ~flags;
475 if (sock_needs_netstamp(sk) &&
476 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
477 net_disable_timestamp();
482 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
485 struct sk_buff_head *list = &sk->sk_receive_queue;
487 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
488 atomic_inc(&sk->sk_drops);
489 trace_sock_rcvqueue_full(sk, skb);
493 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
494 atomic_inc(&sk->sk_drops);
499 skb_set_owner_r(skb, sk);
501 /* we escape from rcu protected region, make sure we dont leak
506 spin_lock_irqsave(&list->lock, flags);
507 sock_skb_set_dropcount(sk, skb);
508 __skb_queue_tail(list, skb);
509 spin_unlock_irqrestore(&list->lock, flags);
511 if (!sock_flag(sk, SOCK_DEAD))
512 sk->sk_data_ready(sk);
515 EXPORT_SYMBOL(__sock_queue_rcv_skb);
517 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
518 enum skb_drop_reason *reason)
520 enum skb_drop_reason drop_reason;
523 err = sk_filter(sk, skb);
525 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
528 err = __sock_queue_rcv_skb(sk, skb);
531 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
534 drop_reason = SKB_DROP_REASON_PROTO_MEM;
537 drop_reason = SKB_NOT_DROPPED_YET;
542 *reason = drop_reason;
545 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
547 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
548 const int nested, unsigned int trim_cap, bool refcounted)
550 int rc = NET_RX_SUCCESS;
552 if (sk_filter_trim_cap(sk, skb, trim_cap))
553 goto discard_and_relse;
557 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
558 atomic_inc(&sk->sk_drops);
559 goto discard_and_relse;
562 bh_lock_sock_nested(sk);
565 if (!sock_owned_by_user(sk)) {
567 * trylock + unlock semantics:
569 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
571 rc = sk_backlog_rcv(sk, skb);
573 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
574 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
576 atomic_inc(&sk->sk_drops);
577 goto discard_and_relse;
589 EXPORT_SYMBOL(__sk_receive_skb);
591 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
593 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
595 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
597 struct dst_entry *dst = __sk_dst_get(sk);
599 if (dst && dst->obsolete &&
600 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
601 dst, cookie) == NULL) {
602 sk_tx_queue_clear(sk);
603 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
604 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
611 EXPORT_SYMBOL(__sk_dst_check);
613 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
615 struct dst_entry *dst = sk_dst_get(sk);
617 if (dst && dst->obsolete &&
618 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
619 dst, cookie) == NULL) {
627 EXPORT_SYMBOL(sk_dst_check);
629 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
631 int ret = -ENOPROTOOPT;
632 #ifdef CONFIG_NETDEVICES
633 struct net *net = sock_net(sk);
637 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
644 /* Paired with all READ_ONCE() done locklessly. */
645 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
647 if (sk->sk_prot->rehash)
648 sk->sk_prot->rehash(sk);
659 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
665 ret = sock_bindtoindex_locked(sk, ifindex);
671 EXPORT_SYMBOL(sock_bindtoindex);
673 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
675 int ret = -ENOPROTOOPT;
676 #ifdef CONFIG_NETDEVICES
677 struct net *net = sock_net(sk);
678 char devname[IFNAMSIZ];
685 /* Bind this socket to a particular device like "eth0",
686 * as specified in the passed interface name. If the
687 * name is "" or the option length is zero the socket
690 if (optlen > IFNAMSIZ - 1)
691 optlen = IFNAMSIZ - 1;
692 memset(devname, 0, sizeof(devname));
695 if (copy_from_sockptr(devname, optval, optlen))
699 if (devname[0] != '\0') {
700 struct net_device *dev;
703 dev = dev_get_by_name_rcu(net, devname);
705 index = dev->ifindex;
712 sockopt_lock_sock(sk);
713 ret = sock_bindtoindex_locked(sk, index);
714 sockopt_release_sock(sk);
721 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
722 sockptr_t optlen, int len)
724 int ret = -ENOPROTOOPT;
725 #ifdef CONFIG_NETDEVICES
726 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
727 struct net *net = sock_net(sk);
728 char devname[IFNAMSIZ];
730 if (bound_dev_if == 0) {
739 ret = netdev_get_name(net, devname, bound_dev_if);
743 len = strlen(devname) + 1;
746 if (copy_to_sockptr(optval, devname, len))
751 if (copy_to_sockptr(optlen, &len, sizeof(int)))
762 bool sk_mc_loop(const struct sock *sk)
764 if (dev_recursion_level())
768 /* IPV6_ADDRFORM can change sk->sk_family under us. */
769 switch (READ_ONCE(sk->sk_family)) {
771 return inet_test_bit(MC_LOOP, sk);
772 #if IS_ENABLED(CONFIG_IPV6)
774 return inet6_test_bit(MC6_LOOP, sk);
780 EXPORT_SYMBOL(sk_mc_loop);
782 void sock_set_reuseaddr(struct sock *sk)
785 sk->sk_reuse = SK_CAN_REUSE;
788 EXPORT_SYMBOL(sock_set_reuseaddr);
790 void sock_set_reuseport(struct sock *sk)
793 sk->sk_reuseport = true;
796 EXPORT_SYMBOL(sock_set_reuseport);
798 void sock_no_linger(struct sock *sk)
801 WRITE_ONCE(sk->sk_lingertime, 0);
802 sock_set_flag(sk, SOCK_LINGER);
805 EXPORT_SYMBOL(sock_no_linger);
807 void sock_set_priority(struct sock *sk, u32 priority)
809 WRITE_ONCE(sk->sk_priority, priority);
811 EXPORT_SYMBOL(sock_set_priority);
813 void sock_set_sndtimeo(struct sock *sk, s64 secs)
816 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
817 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
819 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
822 EXPORT_SYMBOL(sock_set_sndtimeo);
824 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
827 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
828 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
829 sock_set_flag(sk, SOCK_RCVTSTAMP);
830 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
832 sock_reset_flag(sk, SOCK_RCVTSTAMP);
833 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
837 void sock_enable_timestamps(struct sock *sk)
840 __sock_set_timestamps(sk, true, false, true);
843 EXPORT_SYMBOL(sock_enable_timestamps);
845 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
848 case SO_TIMESTAMP_OLD:
849 __sock_set_timestamps(sk, valbool, false, false);
851 case SO_TIMESTAMP_NEW:
852 __sock_set_timestamps(sk, valbool, true, false);
854 case SO_TIMESTAMPNS_OLD:
855 __sock_set_timestamps(sk, valbool, false, true);
857 case SO_TIMESTAMPNS_NEW:
858 __sock_set_timestamps(sk, valbool, true, true);
863 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
865 struct net *net = sock_net(sk);
866 struct net_device *dev = NULL;
871 if (sk->sk_bound_dev_if)
872 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
875 pr_err("%s: sock not bind to device\n", __func__);
879 num = ethtool_get_phc_vclocks(dev, &vclock_index);
882 for (i = 0; i < num; i++) {
883 if (*(vclock_index + i) == phc_index) {
895 WRITE_ONCE(sk->sk_bind_phc, phc_index);
900 int sock_set_timestamping(struct sock *sk, int optname,
901 struct so_timestamping timestamping)
903 int val = timestamping.flags;
906 if (val & ~SOF_TIMESTAMPING_MASK)
909 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
910 !(val & SOF_TIMESTAMPING_OPT_ID))
913 if (val & SOF_TIMESTAMPING_OPT_ID &&
914 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
916 if ((1 << sk->sk_state) &
917 (TCPF_CLOSE | TCPF_LISTEN))
919 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
920 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
922 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
924 atomic_set(&sk->sk_tskey, 0);
928 if (val & SOF_TIMESTAMPING_OPT_STATS &&
929 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
932 if (val & SOF_TIMESTAMPING_BIND_PHC) {
933 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
938 WRITE_ONCE(sk->sk_tsflags, val);
939 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
941 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
942 sock_enable_timestamp(sk,
943 SOCK_TIMESTAMPING_RX_SOFTWARE);
945 sock_disable_timestamp(sk,
946 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
950 void sock_set_keepalive(struct sock *sk)
953 if (sk->sk_prot->keepalive)
954 sk->sk_prot->keepalive(sk, true);
955 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
958 EXPORT_SYMBOL(sock_set_keepalive);
960 static void __sock_set_rcvbuf(struct sock *sk, int val)
962 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
963 * as a negative value.
965 val = min_t(int, val, INT_MAX / 2);
966 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
968 /* We double it on the way in to account for "struct sk_buff" etc.
969 * overhead. Applications assume that the SO_RCVBUF setting they make
970 * will allow that much actual data to be received on that socket.
972 * Applications are unaware that "struct sk_buff" and other overheads
973 * allocate from the receive buffer during socket buffer allocation.
975 * And after considering the possible alternatives, returning the value
976 * we actually used in getsockopt is the most desirable behavior.
978 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
981 void sock_set_rcvbuf(struct sock *sk, int val)
984 __sock_set_rcvbuf(sk, val);
987 EXPORT_SYMBOL(sock_set_rcvbuf);
989 static void __sock_set_mark(struct sock *sk, u32 val)
991 if (val != sk->sk_mark) {
992 WRITE_ONCE(sk->sk_mark, val);
997 void sock_set_mark(struct sock *sk, u32 val)
1000 __sock_set_mark(sk, val);
1003 EXPORT_SYMBOL(sock_set_mark);
1005 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1007 /* Round down bytes to multiple of pages */
1008 bytes = round_down(bytes, PAGE_SIZE);
1010 WARN_ON(bytes > sk->sk_reserved_mem);
1011 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1015 static int sock_reserve_memory(struct sock *sk, int bytes)
1021 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1027 pages = sk_mem_pages(bytes);
1029 /* pre-charge to memcg */
1030 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1031 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1035 /* pre-charge to forward_alloc */
1036 sk_memory_allocated_add(sk, pages);
1037 allocated = sk_memory_allocated(sk);
1038 /* If the system goes into memory pressure with this
1039 * precharge, give up and return error.
1041 if (allocated > sk_prot_mem_limits(sk, 1)) {
1042 sk_memory_allocated_sub(sk, pages);
1043 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1046 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1048 WRITE_ONCE(sk->sk_reserved_mem,
1049 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1054 void sockopt_lock_sock(struct sock *sk)
1056 /* When current->bpf_ctx is set, the setsockopt is called from
1057 * a bpf prog. bpf has ensured the sk lock has been
1058 * acquired before calling setsockopt().
1060 if (has_current_bpf_ctx())
1065 EXPORT_SYMBOL(sockopt_lock_sock);
1067 void sockopt_release_sock(struct sock *sk)
1069 if (has_current_bpf_ctx())
1074 EXPORT_SYMBOL(sockopt_release_sock);
1076 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1078 return has_current_bpf_ctx() || ns_capable(ns, cap);
1080 EXPORT_SYMBOL(sockopt_ns_capable);
1082 bool sockopt_capable(int cap)
1084 return has_current_bpf_ctx() || capable(cap);
1086 EXPORT_SYMBOL(sockopt_capable);
1089 * This is meant for all protocols to use and covers goings on
1090 * at the socket level. Everything here is generic.
1093 int sk_setsockopt(struct sock *sk, int level, int optname,
1094 sockptr_t optval, unsigned int optlen)
1096 struct so_timestamping timestamping;
1097 struct socket *sock = sk->sk_socket;
1098 struct sock_txtime sk_txtime;
1105 * Options without arguments
1108 if (optname == SO_BINDTODEVICE)
1109 return sock_setbindtodevice(sk, optval, optlen);
1111 if (optlen < sizeof(int))
1114 if (copy_from_sockptr(&val, optval, sizeof(val)))
1117 valbool = val ? 1 : 0;
1119 /* handle options which do not require locking the socket. */
1122 if ((val >= 0 && val <= 6) ||
1123 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1124 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1125 sock_set_priority(sk, val);
1130 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1133 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1136 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1142 return -ENOPROTOOPT;
1143 #ifdef CONFIG_NET_RX_BUSY_POLL
1147 WRITE_ONCE(sk->sk_ll_usec, val);
1149 case SO_PREFER_BUSY_POLL:
1150 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1152 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1154 case SO_BUSY_POLL_BUDGET:
1155 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1156 !sockopt_capable(CAP_NET_ADMIN))
1158 if (val < 0 || val > U16_MAX)
1160 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1163 case SO_MAX_PACING_RATE:
1165 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1166 unsigned long pacing_rate;
1168 if (sizeof(ulval) != sizeof(val) &&
1169 optlen >= sizeof(ulval) &&
1170 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1174 cmpxchg(&sk->sk_pacing_status,
1177 /* Pairs with READ_ONCE() from sk_getsockopt() */
1178 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1179 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1180 if (ulval < pacing_rate)
1181 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1185 if (val < -1 || val > 1)
1187 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1188 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1189 /* Paired with READ_ONCE() in tcp_rtx_synack()
1190 * and sk_getsockopt().
1192 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1196 sockopt_lock_sock(sk);
1200 if (val && !sockopt_capable(CAP_NET_ADMIN))
1203 sock_valbool_flag(sk, SOCK_DBG, valbool);
1206 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1209 sk->sk_reuseport = valbool;
1212 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1216 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1219 /* Don't error on this BSD doesn't and if you think
1220 * about it this is right. Otherwise apps have to
1221 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1222 * are treated in BSD as hints
1224 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1226 /* Ensure val * 2 fits into an int, to prevent max_t()
1227 * from treating it as a negative value.
1229 val = min_t(int, val, INT_MAX / 2);
1230 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1231 WRITE_ONCE(sk->sk_sndbuf,
1232 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1233 /* Wake up sending tasks if we upped the value. */
1234 sk->sk_write_space(sk);
1237 case SO_SNDBUFFORCE:
1238 if (!sockopt_capable(CAP_NET_ADMIN)) {
1243 /* No negative values (to prevent underflow, as val will be
1251 /* Don't error on this BSD doesn't and if you think
1252 * about it this is right. Otherwise apps have to
1253 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1254 * are treated in BSD as hints
1256 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1259 case SO_RCVBUFFORCE:
1260 if (!sockopt_capable(CAP_NET_ADMIN)) {
1265 /* No negative values (to prevent underflow, as val will be
1268 __sock_set_rcvbuf(sk, max(val, 0));
1272 if (sk->sk_prot->keepalive)
1273 sk->sk_prot->keepalive(sk, valbool);
1274 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1278 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1282 sk->sk_no_check_tx = valbool;
1286 if (optlen < sizeof(ling)) {
1287 ret = -EINVAL; /* 1003.1g */
1290 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1294 if (!ling.l_onoff) {
1295 sock_reset_flag(sk, SOCK_LINGER);
1297 unsigned long t_sec = ling.l_linger;
1299 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1300 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1302 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1303 sock_set_flag(sk, SOCK_LINGER);
1310 case SO_TIMESTAMP_OLD:
1311 case SO_TIMESTAMP_NEW:
1312 case SO_TIMESTAMPNS_OLD:
1313 case SO_TIMESTAMPNS_NEW:
1314 sock_set_timestamp(sk, optname, valbool);
1317 case SO_TIMESTAMPING_NEW:
1318 case SO_TIMESTAMPING_OLD:
1319 if (optlen == sizeof(timestamping)) {
1320 if (copy_from_sockptr(×tamping, optval,
1321 sizeof(timestamping))) {
1326 memset(×tamping, 0, sizeof(timestamping));
1327 timestamping.flags = val;
1329 ret = sock_set_timestamping(sk, optname, timestamping);
1334 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1339 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1341 ret = set_rcvlowat(sk, val);
1343 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1346 case SO_RCVTIMEO_OLD:
1347 case SO_RCVTIMEO_NEW:
1348 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1349 optlen, optname == SO_RCVTIMEO_OLD);
1352 case SO_SNDTIMEO_OLD:
1353 case SO_SNDTIMEO_NEW:
1354 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1355 optlen, optname == SO_SNDTIMEO_OLD);
1358 case SO_ATTACH_FILTER: {
1359 struct sock_fprog fprog;
1361 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1363 ret = sk_attach_filter(&fprog, sk);
1368 if (optlen == sizeof(u32)) {
1372 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1375 ret = sk_attach_bpf(ufd, sk);
1379 case SO_ATTACH_REUSEPORT_CBPF: {
1380 struct sock_fprog fprog;
1382 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1384 ret = sk_reuseport_attach_filter(&fprog, sk);
1387 case SO_ATTACH_REUSEPORT_EBPF:
1389 if (optlen == sizeof(u32)) {
1393 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1396 ret = sk_reuseport_attach_bpf(ufd, sk);
1400 case SO_DETACH_REUSEPORT_BPF:
1401 ret = reuseport_detach_prog(sk);
1404 case SO_DETACH_FILTER:
1405 ret = sk_detach_filter(sk);
1408 case SO_LOCK_FILTER:
1409 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1412 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1416 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1417 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1422 __sock_set_mark(sk, val);
1425 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1429 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1432 case SO_WIFI_STATUS:
1433 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1438 int (*set_peek_off)(struct sock *sk, int val);
1440 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1442 ret = set_peek_off(sk, val);
1449 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1452 case SO_SELECT_ERR_QUEUE:
1453 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1457 case SO_INCOMING_CPU:
1458 reuseport_update_incoming_cpu(sk, val);
1463 dst_negative_advice(sk);
1467 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1468 if (!(sk_is_tcp(sk) ||
1469 (sk->sk_type == SOCK_DGRAM &&
1470 sk->sk_protocol == IPPROTO_UDP)))
1472 } else if (sk->sk_family != PF_RDS) {
1476 if (val < 0 || val > 1)
1479 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1484 if (optlen != sizeof(struct sock_txtime)) {
1487 } else if (copy_from_sockptr(&sk_txtime, optval,
1488 sizeof(struct sock_txtime))) {
1491 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1495 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1496 * scheduler has enough safe guards.
1498 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1499 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1503 sock_valbool_flag(sk, SOCK_TXTIME, true);
1504 sk->sk_clockid = sk_txtime.clockid;
1505 sk->sk_txtime_deadline_mode =
1506 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1507 sk->sk_txtime_report_errors =
1508 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1511 case SO_BINDTOIFINDEX:
1512 ret = sock_bindtoindex_locked(sk, val);
1516 if (val & ~SOCK_BUF_LOCK_MASK) {
1520 sk->sk_userlocks = val | (sk->sk_userlocks &
1521 ~SOCK_BUF_LOCK_MASK);
1524 case SO_RESERVE_MEM:
1533 delta = val - sk->sk_reserved_mem;
1535 sock_release_reserved_memory(sk, -delta);
1537 ret = sock_reserve_memory(sk, delta);
1545 sockopt_release_sock(sk);
1549 int sock_setsockopt(struct socket *sock, int level, int optname,
1550 sockptr_t optval, unsigned int optlen)
1552 return sk_setsockopt(sock->sk, level, optname,
1555 EXPORT_SYMBOL(sock_setsockopt);
1557 static const struct cred *sk_get_peer_cred(struct sock *sk)
1559 const struct cred *cred;
1561 spin_lock(&sk->sk_peer_lock);
1562 cred = get_cred(sk->sk_peer_cred);
1563 spin_unlock(&sk->sk_peer_lock);
1568 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1569 struct ucred *ucred)
1571 ucred->pid = pid_vnr(pid);
1572 ucred->uid = ucred->gid = -1;
1574 struct user_namespace *current_ns = current_user_ns();
1576 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1577 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1581 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1583 struct user_namespace *user_ns = current_user_ns();
1586 for (i = 0; i < src->ngroups; i++) {
1587 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1589 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1596 int sk_getsockopt(struct sock *sk, int level, int optname,
1597 sockptr_t optval, sockptr_t optlen)
1599 struct socket *sock = sk->sk_socket;
1604 unsigned long ulval;
1606 struct old_timeval32 tm32;
1607 struct __kernel_old_timeval tm;
1608 struct __kernel_sock_timeval stm;
1609 struct sock_txtime txtime;
1610 struct so_timestamping timestamping;
1613 int lv = sizeof(int);
1616 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1621 memset(&v, 0, sizeof(v));
1625 v.val = sock_flag(sk, SOCK_DBG);
1629 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1633 v.val = sock_flag(sk, SOCK_BROADCAST);
1637 v.val = READ_ONCE(sk->sk_sndbuf);
1641 v.val = READ_ONCE(sk->sk_rcvbuf);
1645 v.val = sk->sk_reuse;
1649 v.val = sk->sk_reuseport;
1653 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1657 v.val = sk->sk_type;
1661 v.val = sk->sk_protocol;
1665 v.val = sk->sk_family;
1669 v.val = -sock_error(sk);
1671 v.val = xchg(&sk->sk_err_soft, 0);
1675 v.val = sock_flag(sk, SOCK_URGINLINE);
1679 v.val = sk->sk_no_check_tx;
1683 v.val = READ_ONCE(sk->sk_priority);
1687 lv = sizeof(v.ling);
1688 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1689 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1695 case SO_TIMESTAMP_OLD:
1696 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1697 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1698 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1701 case SO_TIMESTAMPNS_OLD:
1702 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1705 case SO_TIMESTAMP_NEW:
1706 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1709 case SO_TIMESTAMPNS_NEW:
1710 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1713 case SO_TIMESTAMPING_OLD:
1714 case SO_TIMESTAMPING_NEW:
1715 lv = sizeof(v.timestamping);
1716 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1717 * returning the flags when they were set through the same option.
1718 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1720 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1721 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1722 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1726 case SO_RCVTIMEO_OLD:
1727 case SO_RCVTIMEO_NEW:
1728 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1729 SO_RCVTIMEO_OLD == optname);
1732 case SO_SNDTIMEO_OLD:
1733 case SO_SNDTIMEO_NEW:
1734 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1735 SO_SNDTIMEO_OLD == optname);
1739 v.val = READ_ONCE(sk->sk_rcvlowat);
1747 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1751 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1756 struct ucred peercred;
1757 if (len > sizeof(peercred))
1758 len = sizeof(peercred);
1760 spin_lock(&sk->sk_peer_lock);
1761 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1762 spin_unlock(&sk->sk_peer_lock);
1764 if (copy_to_sockptr(optval, &peercred, len))
1771 struct pid *peer_pid;
1772 struct file *pidfd_file = NULL;
1775 if (len > sizeof(pidfd))
1776 len = sizeof(pidfd);
1778 spin_lock(&sk->sk_peer_lock);
1779 peer_pid = get_pid(sk->sk_peer_pid);
1780 spin_unlock(&sk->sk_peer_lock);
1785 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1790 if (copy_to_sockptr(optval, &pidfd, len) ||
1791 copy_to_sockptr(optlen, &len, sizeof(int))) {
1792 put_unused_fd(pidfd);
1798 fd_install(pidfd, pidfd_file);
1804 const struct cred *cred;
1807 cred = sk_get_peer_cred(sk);
1811 n = cred->group_info->ngroups;
1812 if (len < n * sizeof(gid_t)) {
1813 len = n * sizeof(gid_t);
1815 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1817 len = n * sizeof(gid_t);
1819 ret = groups_to_user(optval, cred->group_info);
1828 struct sockaddr_storage address;
1830 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1835 if (copy_to_sockptr(optval, &address, len))
1840 /* Dubious BSD thing... Probably nobody even uses it, but
1841 * the UNIX standard wants it for whatever reason... -DaveM
1844 v.val = sk->sk_state == TCP_LISTEN;
1848 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1852 return security_socket_getpeersec_stream(sock,
1853 optval, optlen, len);
1856 v.val = READ_ONCE(sk->sk_mark);
1860 v.val = sock_flag(sk, SOCK_RCVMARK);
1864 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1867 case SO_WIFI_STATUS:
1868 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1872 if (!READ_ONCE(sock->ops)->set_peek_off)
1875 v.val = READ_ONCE(sk->sk_peek_off);
1878 v.val = sock_flag(sk, SOCK_NOFCS);
1881 case SO_BINDTODEVICE:
1882 return sock_getbindtodevice(sk, optval, optlen, len);
1885 len = sk_get_filter(sk, optval, len);
1891 case SO_LOCK_FILTER:
1892 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1895 case SO_BPF_EXTENSIONS:
1896 v.val = bpf_tell_extensions();
1899 case SO_SELECT_ERR_QUEUE:
1900 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1903 #ifdef CONFIG_NET_RX_BUSY_POLL
1905 v.val = READ_ONCE(sk->sk_ll_usec);
1907 case SO_PREFER_BUSY_POLL:
1908 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1912 case SO_MAX_PACING_RATE:
1913 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1914 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1915 lv = sizeof(v.ulval);
1916 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1919 v.val = min_t(unsigned long, ~0U,
1920 READ_ONCE(sk->sk_max_pacing_rate));
1924 case SO_INCOMING_CPU:
1925 v.val = READ_ONCE(sk->sk_incoming_cpu);
1930 u32 meminfo[SK_MEMINFO_VARS];
1932 sk_get_meminfo(sk, meminfo);
1934 len = min_t(unsigned int, len, sizeof(meminfo));
1935 if (copy_to_sockptr(optval, &meminfo, len))
1941 #ifdef CONFIG_NET_RX_BUSY_POLL
1942 case SO_INCOMING_NAPI_ID:
1943 v.val = READ_ONCE(sk->sk_napi_id);
1945 /* aggregate non-NAPI IDs down to 0 */
1946 if (v.val < MIN_NAPI_ID)
1956 v.val64 = sock_gen_cookie(sk);
1960 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1964 lv = sizeof(v.txtime);
1965 v.txtime.clockid = sk->sk_clockid;
1966 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1967 SOF_TXTIME_DEADLINE_MODE : 0;
1968 v.txtime.flags |= sk->sk_txtime_report_errors ?
1969 SOF_TXTIME_REPORT_ERRORS : 0;
1972 case SO_BINDTOIFINDEX:
1973 v.val = READ_ONCE(sk->sk_bound_dev_if);
1976 case SO_NETNS_COOKIE:
1980 v.val64 = sock_net(sk)->net_cookie;
1984 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1987 case SO_RESERVE_MEM:
1988 v.val = READ_ONCE(sk->sk_reserved_mem);
1992 /* Paired with WRITE_ONCE() in sk_setsockopt() */
1993 v.val = READ_ONCE(sk->sk_txrehash);
1997 /* We implement the SO_SNDLOWAT etc to not be settable
2000 return -ENOPROTOOPT;
2005 if (copy_to_sockptr(optval, &v, len))
2008 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2014 * Initialize an sk_lock.
2016 * (We also register the sk_lock with the lock validator.)
2018 static inline void sock_lock_init(struct sock *sk)
2020 if (sk->sk_kern_sock)
2021 sock_lock_init_class_and_name(
2023 af_family_kern_slock_key_strings[sk->sk_family],
2024 af_family_kern_slock_keys + sk->sk_family,
2025 af_family_kern_key_strings[sk->sk_family],
2026 af_family_kern_keys + sk->sk_family);
2028 sock_lock_init_class_and_name(
2030 af_family_slock_key_strings[sk->sk_family],
2031 af_family_slock_keys + sk->sk_family,
2032 af_family_key_strings[sk->sk_family],
2033 af_family_keys + sk->sk_family);
2037 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2038 * even temporarly, because of RCU lookups. sk_node should also be left as is.
2039 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2041 static void sock_copy(struct sock *nsk, const struct sock *osk)
2043 const struct proto *prot = READ_ONCE(osk->sk_prot);
2044 #ifdef CONFIG_SECURITY_NETWORK
2045 void *sptr = nsk->sk_security;
2048 /* If we move sk_tx_queue_mapping out of the private section,
2049 * we must check if sk_tx_queue_clear() is called after
2050 * sock_copy() in sk_clone_lock().
2052 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2053 offsetof(struct sock, sk_dontcopy_begin) ||
2054 offsetof(struct sock, sk_tx_queue_mapping) >=
2055 offsetof(struct sock, sk_dontcopy_end));
2057 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2059 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2060 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2062 #ifdef CONFIG_SECURITY_NETWORK
2063 nsk->sk_security = sptr;
2064 security_sk_clone(osk, nsk);
2068 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2072 struct kmem_cache *slab;
2076 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2079 if (want_init_on_alloc(priority))
2080 sk_prot_clear_nulls(sk, prot->obj_size);
2082 sk = kmalloc(prot->obj_size, priority);
2085 if (security_sk_alloc(sk, family, priority))
2088 if (!try_module_get(prot->owner))
2095 security_sk_free(sk);
2098 kmem_cache_free(slab, sk);
2104 static void sk_prot_free(struct proto *prot, struct sock *sk)
2106 struct kmem_cache *slab;
2107 struct module *owner;
2109 owner = prot->owner;
2112 cgroup_sk_free(&sk->sk_cgrp_data);
2113 mem_cgroup_sk_free(sk);
2114 security_sk_free(sk);
2116 kmem_cache_free(slab, sk);
2123 * sk_alloc - All socket objects are allocated here
2124 * @net: the applicable net namespace
2125 * @family: protocol family
2126 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2127 * @prot: struct proto associated with this new sock instance
2128 * @kern: is this to be a kernel socket?
2130 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2131 struct proto *prot, int kern)
2135 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2137 sk->sk_family = family;
2139 * See comment in struct sock definition to understand
2140 * why we need sk_prot_creator -acme
2142 sk->sk_prot = sk->sk_prot_creator = prot;
2143 sk->sk_kern_sock = kern;
2145 sk->sk_net_refcnt = kern ? 0 : 1;
2146 if (likely(sk->sk_net_refcnt)) {
2147 get_net_track(net, &sk->ns_tracker, priority);
2148 sock_inuse_add(net, 1);
2150 __netns_tracker_alloc(net, &sk->ns_tracker,
2154 sock_net_set(sk, net);
2155 refcount_set(&sk->sk_wmem_alloc, 1);
2157 mem_cgroup_sk_alloc(sk);
2158 cgroup_sk_alloc(&sk->sk_cgrp_data);
2159 sock_update_classid(&sk->sk_cgrp_data);
2160 sock_update_netprioidx(&sk->sk_cgrp_data);
2161 sk_tx_queue_clear(sk);
2166 EXPORT_SYMBOL(sk_alloc);
2168 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2169 * grace period. This is the case for UDP sockets and TCP listeners.
2171 static void __sk_destruct(struct rcu_head *head)
2173 struct sock *sk = container_of(head, struct sock, sk_rcu);
2174 struct sk_filter *filter;
2176 if (sk->sk_destruct)
2177 sk->sk_destruct(sk);
2179 filter = rcu_dereference_check(sk->sk_filter,
2180 refcount_read(&sk->sk_wmem_alloc) == 0);
2182 sk_filter_uncharge(sk, filter);
2183 RCU_INIT_POINTER(sk->sk_filter, NULL);
2186 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2188 #ifdef CONFIG_BPF_SYSCALL
2189 bpf_sk_storage_free(sk);
2192 if (atomic_read(&sk->sk_omem_alloc))
2193 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2194 __func__, atomic_read(&sk->sk_omem_alloc));
2196 if (sk->sk_frag.page) {
2197 put_page(sk->sk_frag.page);
2198 sk->sk_frag.page = NULL;
2201 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2202 put_cred(sk->sk_peer_cred);
2203 put_pid(sk->sk_peer_pid);
2205 if (likely(sk->sk_net_refcnt))
2206 put_net_track(sock_net(sk), &sk->ns_tracker);
2208 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2210 sk_prot_free(sk->sk_prot_creator, sk);
2213 void sk_destruct(struct sock *sk)
2215 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2217 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2218 reuseport_detach_sock(sk);
2219 use_call_rcu = true;
2223 call_rcu(&sk->sk_rcu, __sk_destruct);
2225 __sk_destruct(&sk->sk_rcu);
2228 static void __sk_free(struct sock *sk)
2230 if (likely(sk->sk_net_refcnt))
2231 sock_inuse_add(sock_net(sk), -1);
2233 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2234 sock_diag_broadcast_destroy(sk);
2239 void sk_free(struct sock *sk)
2242 * We subtract one from sk_wmem_alloc and can know if
2243 * some packets are still in some tx queue.
2244 * If not null, sock_wfree() will call __sk_free(sk) later
2246 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2249 EXPORT_SYMBOL(sk_free);
2251 static void sk_init_common(struct sock *sk)
2253 skb_queue_head_init(&sk->sk_receive_queue);
2254 skb_queue_head_init(&sk->sk_write_queue);
2255 skb_queue_head_init(&sk->sk_error_queue);
2257 rwlock_init(&sk->sk_callback_lock);
2258 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2259 af_rlock_keys + sk->sk_family,
2260 af_family_rlock_key_strings[sk->sk_family]);
2261 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2262 af_wlock_keys + sk->sk_family,
2263 af_family_wlock_key_strings[sk->sk_family]);
2264 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2265 af_elock_keys + sk->sk_family,
2266 af_family_elock_key_strings[sk->sk_family]);
2267 lockdep_set_class_and_name(&sk->sk_callback_lock,
2268 af_callback_keys + sk->sk_family,
2269 af_family_clock_key_strings[sk->sk_family]);
2273 * sk_clone_lock - clone a socket, and lock its clone
2274 * @sk: the socket to clone
2275 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2277 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2279 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2281 struct proto *prot = READ_ONCE(sk->sk_prot);
2282 struct sk_filter *filter;
2283 bool is_charged = true;
2286 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2290 sock_copy(newsk, sk);
2292 newsk->sk_prot_creator = prot;
2295 if (likely(newsk->sk_net_refcnt)) {
2296 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2297 sock_inuse_add(sock_net(newsk), 1);
2299 /* Kernel sockets are not elevating the struct net refcount.
2300 * Instead, use a tracker to more easily detect if a layer
2301 * is not properly dismantling its kernel sockets at netns
2304 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2307 sk_node_init(&newsk->sk_node);
2308 sock_lock_init(newsk);
2309 bh_lock_sock(newsk);
2310 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2311 newsk->sk_backlog.len = 0;
2313 atomic_set(&newsk->sk_rmem_alloc, 0);
2315 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2316 refcount_set(&newsk->sk_wmem_alloc, 1);
2318 atomic_set(&newsk->sk_omem_alloc, 0);
2319 sk_init_common(newsk);
2321 newsk->sk_dst_cache = NULL;
2322 newsk->sk_dst_pending_confirm = 0;
2323 newsk->sk_wmem_queued = 0;
2324 newsk->sk_forward_alloc = 0;
2325 newsk->sk_reserved_mem = 0;
2326 atomic_set(&newsk->sk_drops, 0);
2327 newsk->sk_send_head = NULL;
2328 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2329 atomic_set(&newsk->sk_zckey, 0);
2331 sock_reset_flag(newsk, SOCK_DONE);
2333 /* sk->sk_memcg will be populated at accept() time */
2334 newsk->sk_memcg = NULL;
2336 cgroup_sk_clone(&newsk->sk_cgrp_data);
2339 filter = rcu_dereference(sk->sk_filter);
2341 /* though it's an empty new sock, the charging may fail
2342 * if sysctl_optmem_max was changed between creation of
2343 * original socket and cloning
2345 is_charged = sk_filter_charge(newsk, filter);
2346 RCU_INIT_POINTER(newsk->sk_filter, filter);
2349 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2350 /* We need to make sure that we don't uncharge the new
2351 * socket if we couldn't charge it in the first place
2352 * as otherwise we uncharge the parent's filter.
2355 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2356 sk_free_unlock_clone(newsk);
2360 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2362 if (bpf_sk_storage_clone(sk, newsk)) {
2363 sk_free_unlock_clone(newsk);
2368 /* Clear sk_user_data if parent had the pointer tagged
2369 * as not suitable for copying when cloning.
2371 if (sk_user_data_is_nocopy(newsk))
2372 newsk->sk_user_data = NULL;
2375 newsk->sk_err_soft = 0;
2376 newsk->sk_priority = 0;
2377 newsk->sk_incoming_cpu = raw_smp_processor_id();
2379 /* Before updating sk_refcnt, we must commit prior changes to memory
2380 * (Documentation/RCU/rculist_nulls.rst for details)
2383 refcount_set(&newsk->sk_refcnt, 2);
2385 sk_set_socket(newsk, NULL);
2386 sk_tx_queue_clear(newsk);
2387 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2389 if (newsk->sk_prot->sockets_allocated)
2390 sk_sockets_allocated_inc(newsk);
2392 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2393 net_enable_timestamp();
2397 EXPORT_SYMBOL_GPL(sk_clone_lock);
2399 void sk_free_unlock_clone(struct sock *sk)
2401 /* It is still raw copy of parent, so invalidate
2402 * destructor and make plain sk_free() */
2403 sk->sk_destruct = NULL;
2407 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2409 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2411 bool is_ipv6 = false;
2414 #if IS_ENABLED(CONFIG_IPV6)
2415 is_ipv6 = (sk->sk_family == AF_INET6 &&
2416 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2418 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2419 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2420 READ_ONCE(dst->dev->gso_ipv4_max_size);
2421 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2422 max_size = GSO_LEGACY_MAX_SIZE;
2424 return max_size - (MAX_TCP_HEADER + 1);
2427 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2431 sk->sk_route_caps = dst->dev->features;
2433 sk->sk_route_caps |= NETIF_F_GSO;
2434 if (sk->sk_route_caps & NETIF_F_GSO)
2435 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2436 if (unlikely(sk->sk_gso_disabled))
2437 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2438 if (sk_can_gso(sk)) {
2439 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2440 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2442 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2443 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2444 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2445 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2448 sk->sk_gso_max_segs = max_segs;
2449 sk_dst_set(sk, dst);
2451 EXPORT_SYMBOL_GPL(sk_setup_caps);
2454 * Simple resource managers for sockets.
2459 * Write buffer destructor automatically called from kfree_skb.
2461 void sock_wfree(struct sk_buff *skb)
2463 struct sock *sk = skb->sk;
2464 unsigned int len = skb->truesize;
2467 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2468 if (sock_flag(sk, SOCK_RCU_FREE) &&
2469 sk->sk_write_space == sock_def_write_space) {
2471 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2472 sock_def_write_space_wfree(sk);
2480 * Keep a reference on sk_wmem_alloc, this will be released
2481 * after sk_write_space() call
2483 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2484 sk->sk_write_space(sk);
2488 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2489 * could not do because of in-flight packets
2491 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2494 EXPORT_SYMBOL(sock_wfree);
2496 /* This variant of sock_wfree() is used by TCP,
2497 * since it sets SOCK_USE_WRITE_QUEUE.
2499 void __sock_wfree(struct sk_buff *skb)
2501 struct sock *sk = skb->sk;
2503 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2507 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2512 if (unlikely(!sk_fullsock(sk))) {
2513 skb->destructor = sock_edemux;
2518 skb->destructor = sock_wfree;
2519 skb_set_hash_from_sk(skb, sk);
2521 * We used to take a refcount on sk, but following operation
2522 * is enough to guarantee sk_free() wont free this sock until
2523 * all in-flight packets are completed
2525 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2527 EXPORT_SYMBOL(skb_set_owner_w);
2529 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2531 #ifdef CONFIG_TLS_DEVICE
2532 /* Drivers depend on in-order delivery for crypto offload,
2533 * partial orphan breaks out-of-order-OK logic.
2538 return (skb->destructor == sock_wfree ||
2539 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2542 /* This helper is used by netem, as it can hold packets in its
2543 * delay queue. We want to allow the owner socket to send more
2544 * packets, as if they were already TX completed by a typical driver.
2545 * But we also want to keep skb->sk set because some packet schedulers
2546 * rely on it (sch_fq for example).
2548 void skb_orphan_partial(struct sk_buff *skb)
2550 if (skb_is_tcp_pure_ack(skb))
2553 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2558 EXPORT_SYMBOL(skb_orphan_partial);
2561 * Read buffer destructor automatically called from kfree_skb.
2563 void sock_rfree(struct sk_buff *skb)
2565 struct sock *sk = skb->sk;
2566 unsigned int len = skb->truesize;
2568 atomic_sub(len, &sk->sk_rmem_alloc);
2569 sk_mem_uncharge(sk, len);
2571 EXPORT_SYMBOL(sock_rfree);
2574 * Buffer destructor for skbs that are not used directly in read or write
2575 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2577 void sock_efree(struct sk_buff *skb)
2581 EXPORT_SYMBOL(sock_efree);
2583 /* Buffer destructor for prefetch/receive path where reference count may
2584 * not be held, e.g. for listen sockets.
2587 void sock_pfree(struct sk_buff *skb)
2589 if (sk_is_refcounted(skb->sk))
2590 sock_gen_put(skb->sk);
2592 EXPORT_SYMBOL(sock_pfree);
2593 #endif /* CONFIG_INET */
2595 kuid_t sock_i_uid(struct sock *sk)
2599 read_lock_bh(&sk->sk_callback_lock);
2600 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2601 read_unlock_bh(&sk->sk_callback_lock);
2604 EXPORT_SYMBOL(sock_i_uid);
2606 unsigned long __sock_i_ino(struct sock *sk)
2610 read_lock(&sk->sk_callback_lock);
2611 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2612 read_unlock(&sk->sk_callback_lock);
2615 EXPORT_SYMBOL(__sock_i_ino);
2617 unsigned long sock_i_ino(struct sock *sk)
2622 ino = __sock_i_ino(sk);
2626 EXPORT_SYMBOL(sock_i_ino);
2629 * Allocate a skb from the socket's send buffer.
2631 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2635 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2636 struct sk_buff *skb = alloc_skb(size, priority);
2639 skb_set_owner_w(skb, sk);
2645 EXPORT_SYMBOL(sock_wmalloc);
2647 static void sock_ofree(struct sk_buff *skb)
2649 struct sock *sk = skb->sk;
2651 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2654 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2657 struct sk_buff *skb;
2659 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2660 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2661 READ_ONCE(sysctl_optmem_max))
2664 skb = alloc_skb(size, priority);
2668 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2670 skb->destructor = sock_ofree;
2675 * Allocate a memory block from the socket's option memory buffer.
2677 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2679 int optmem_max = READ_ONCE(sysctl_optmem_max);
2681 if ((unsigned int)size <= optmem_max &&
2682 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2684 /* First do the add, to avoid the race if kmalloc
2687 atomic_add(size, &sk->sk_omem_alloc);
2688 mem = kmalloc(size, priority);
2691 atomic_sub(size, &sk->sk_omem_alloc);
2695 EXPORT_SYMBOL(sock_kmalloc);
2697 /* Free an option memory block. Note, we actually want the inline
2698 * here as this allows gcc to detect the nullify and fold away the
2699 * condition entirely.
2701 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2704 if (WARN_ON_ONCE(!mem))
2707 kfree_sensitive(mem);
2710 atomic_sub(size, &sk->sk_omem_alloc);
2713 void sock_kfree_s(struct sock *sk, void *mem, int size)
2715 __sock_kfree_s(sk, mem, size, false);
2717 EXPORT_SYMBOL(sock_kfree_s);
2719 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2721 __sock_kfree_s(sk, mem, size, true);
2723 EXPORT_SYMBOL(sock_kzfree_s);
2725 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2726 I think, these locks should be removed for datagram sockets.
2728 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2732 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2736 if (signal_pending(current))
2738 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2739 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2740 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2742 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2744 if (READ_ONCE(sk->sk_err))
2746 timeo = schedule_timeout(timeo);
2748 finish_wait(sk_sleep(sk), &wait);
2754 * Generic send/receive buffer handlers
2757 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2758 unsigned long data_len, int noblock,
2759 int *errcode, int max_page_order)
2761 struct sk_buff *skb;
2765 timeo = sock_sndtimeo(sk, noblock);
2767 err = sock_error(sk);
2772 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2775 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2778 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2779 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2783 if (signal_pending(current))
2785 timeo = sock_wait_for_wmem(sk, timeo);
2787 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2788 errcode, sk->sk_allocation);
2790 skb_set_owner_w(skb, sk);
2794 err = sock_intr_errno(timeo);
2799 EXPORT_SYMBOL(sock_alloc_send_pskb);
2801 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2802 struct sockcm_cookie *sockc)
2806 switch (cmsg->cmsg_type) {
2808 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2809 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2811 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2813 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2815 case SO_TIMESTAMPING_OLD:
2816 case SO_TIMESTAMPING_NEW:
2817 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2820 tsflags = *(u32 *)CMSG_DATA(cmsg);
2821 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2824 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2825 sockc->tsflags |= tsflags;
2828 if (!sock_flag(sk, SOCK_TXTIME))
2830 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2832 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2834 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2836 case SCM_CREDENTIALS:
2843 EXPORT_SYMBOL(__sock_cmsg_send);
2845 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2846 struct sockcm_cookie *sockc)
2848 struct cmsghdr *cmsg;
2851 for_each_cmsghdr(cmsg, msg) {
2852 if (!CMSG_OK(msg, cmsg))
2854 if (cmsg->cmsg_level != SOL_SOCKET)
2856 ret = __sock_cmsg_send(sk, cmsg, sockc);
2862 EXPORT_SYMBOL(sock_cmsg_send);
2864 static void sk_enter_memory_pressure(struct sock *sk)
2866 if (!sk->sk_prot->enter_memory_pressure)
2869 sk->sk_prot->enter_memory_pressure(sk);
2872 static void sk_leave_memory_pressure(struct sock *sk)
2874 if (sk->sk_prot->leave_memory_pressure) {
2875 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2876 tcp_leave_memory_pressure, sk);
2878 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2880 if (memory_pressure && READ_ONCE(*memory_pressure))
2881 WRITE_ONCE(*memory_pressure, 0);
2885 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2888 * skb_page_frag_refill - check that a page_frag contains enough room
2889 * @sz: minimum size of the fragment we want to get
2890 * @pfrag: pointer to page_frag
2891 * @gfp: priority for memory allocation
2893 * Note: While this allocator tries to use high order pages, there is
2894 * no guarantee that allocations succeed. Therefore, @sz MUST be
2895 * less or equal than PAGE_SIZE.
2897 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2900 if (page_ref_count(pfrag->page) == 1) {
2904 if (pfrag->offset + sz <= pfrag->size)
2906 put_page(pfrag->page);
2910 if (SKB_FRAG_PAGE_ORDER &&
2911 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2912 /* Avoid direct reclaim but allow kswapd to wake */
2913 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2914 __GFP_COMP | __GFP_NOWARN |
2916 SKB_FRAG_PAGE_ORDER);
2917 if (likely(pfrag->page)) {
2918 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2922 pfrag->page = alloc_page(gfp);
2923 if (likely(pfrag->page)) {
2924 pfrag->size = PAGE_SIZE;
2929 EXPORT_SYMBOL(skb_page_frag_refill);
2931 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2933 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2936 sk_enter_memory_pressure(sk);
2937 sk_stream_moderate_sndbuf(sk);
2940 EXPORT_SYMBOL(sk_page_frag_refill);
2942 void __lock_sock(struct sock *sk)
2943 __releases(&sk->sk_lock.slock)
2944 __acquires(&sk->sk_lock.slock)
2949 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2950 TASK_UNINTERRUPTIBLE);
2951 spin_unlock_bh(&sk->sk_lock.slock);
2953 spin_lock_bh(&sk->sk_lock.slock);
2954 if (!sock_owned_by_user(sk))
2957 finish_wait(&sk->sk_lock.wq, &wait);
2960 void __release_sock(struct sock *sk)
2961 __releases(&sk->sk_lock.slock)
2962 __acquires(&sk->sk_lock.slock)
2964 struct sk_buff *skb, *next;
2966 while ((skb = sk->sk_backlog.head) != NULL) {
2967 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2969 spin_unlock_bh(&sk->sk_lock.slock);
2974 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2975 skb_mark_not_on_list(skb);
2976 sk_backlog_rcv(sk, skb);
2981 } while (skb != NULL);
2983 spin_lock_bh(&sk->sk_lock.slock);
2987 * Doing the zeroing here guarantee we can not loop forever
2988 * while a wild producer attempts to flood us.
2990 sk->sk_backlog.len = 0;
2993 void __sk_flush_backlog(struct sock *sk)
2995 spin_lock_bh(&sk->sk_lock.slock);
2998 if (sk->sk_prot->release_cb)
2999 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3000 tcp_release_cb, sk);
3002 spin_unlock_bh(&sk->sk_lock.slock);
3004 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3007 * sk_wait_data - wait for data to arrive at sk_receive_queue
3008 * @sk: sock to wait on
3009 * @timeo: for how long
3010 * @skb: last skb seen on sk_receive_queue
3012 * Now socket state including sk->sk_err is changed only under lock,
3013 * hence we may omit checks after joining wait queue.
3014 * We check receive queue before schedule() only as optimization;
3015 * it is very likely that release_sock() added new data.
3017 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3019 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3022 add_wait_queue(sk_sleep(sk), &wait);
3023 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3024 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3025 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3026 remove_wait_queue(sk_sleep(sk), &wait);
3029 EXPORT_SYMBOL(sk_wait_data);
3032 * __sk_mem_raise_allocated - increase memory_allocated
3034 * @size: memory size to allocate
3035 * @amt: pages to allocate
3036 * @kind: allocation type
3038 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3040 * Unlike the globally shared limits among the sockets under same protocol,
3041 * consuming the budget of a memcg won't have direct effect on other ones.
3042 * So be optimistic about memcg's tolerance, and leave the callers to decide
3043 * whether or not to raise allocated through sk_under_memory_pressure() or
3046 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3048 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3049 struct proto *prot = sk->sk_prot;
3050 bool charged = false;
3053 sk_memory_allocated_add(sk, amt);
3054 allocated = sk_memory_allocated(sk);
3057 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3058 goto suppress_allocation;
3063 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3064 sk_leave_memory_pressure(sk);
3068 /* Under pressure. */
3069 if (allocated > sk_prot_mem_limits(sk, 1))
3070 sk_enter_memory_pressure(sk);
3072 /* Over hard limit. */
3073 if (allocated > sk_prot_mem_limits(sk, 2))
3074 goto suppress_allocation;
3076 /* Guarantee minimum buffer size under pressure (either global
3077 * or memcg) to make sure features described in RFC 7323 (TCP
3078 * Extensions for High Performance) work properly.
3080 * This rule does NOT stand when exceeds global or memcg's hard
3081 * limit, or else a DoS attack can be taken place by spawning
3082 * lots of sockets whose usage are under minimum buffer size.
3084 if (kind == SK_MEM_RECV) {
3085 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3088 } else { /* SK_MEM_SEND */
3089 int wmem0 = sk_get_wmem0(sk, prot);
3091 if (sk->sk_type == SOCK_STREAM) {
3092 if (sk->sk_wmem_queued < wmem0)
3094 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3099 if (sk_has_memory_pressure(sk)) {
3102 /* The following 'average' heuristic is within the
3103 * scope of global accounting, so it only makes
3104 * sense for global memory pressure.
3106 if (!sk_under_global_memory_pressure(sk))
3109 /* Try to be fair among all the sockets under global
3110 * pressure by allowing the ones that below average
3113 alloc = sk_sockets_allocated_read_positive(sk);
3114 if (sk_prot_mem_limits(sk, 2) > alloc *
3115 sk_mem_pages(sk->sk_wmem_queued +
3116 atomic_read(&sk->sk_rmem_alloc) +
3117 sk->sk_forward_alloc))
3121 suppress_allocation:
3123 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3124 sk_stream_moderate_sndbuf(sk);
3126 /* Fail only if socket is _under_ its sndbuf.
3127 * In this case we cannot block, so that we have to fail.
3129 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3130 /* Force charge with __GFP_NOFAIL */
3131 if (memcg && !charged) {
3132 mem_cgroup_charge_skmem(memcg, amt,
3133 gfp_memcg_charge() | __GFP_NOFAIL);
3139 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3140 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3142 sk_memory_allocated_sub(sk, amt);
3145 mem_cgroup_uncharge_skmem(memcg, amt);
3151 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3153 * @size: memory size to allocate
3154 * @kind: allocation type
3156 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3157 * rmem allocation. This function assumes that protocols which have
3158 * memory_pressure use sk_wmem_queued as write buffer accounting.
3160 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3162 int ret, amt = sk_mem_pages(size);
3164 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3165 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3167 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3170 EXPORT_SYMBOL(__sk_mem_schedule);
3173 * __sk_mem_reduce_allocated - reclaim memory_allocated
3175 * @amount: number of quanta
3177 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3179 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3181 sk_memory_allocated_sub(sk, amount);
3183 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3184 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3186 if (sk_under_global_memory_pressure(sk) &&
3187 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3188 sk_leave_memory_pressure(sk);
3192 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3194 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3196 void __sk_mem_reclaim(struct sock *sk, int amount)
3198 amount >>= PAGE_SHIFT;
3199 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3200 __sk_mem_reduce_allocated(sk, amount);
3202 EXPORT_SYMBOL(__sk_mem_reclaim);
3204 int sk_set_peek_off(struct sock *sk, int val)
3206 WRITE_ONCE(sk->sk_peek_off, val);
3209 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3212 * Set of default routines for initialising struct proto_ops when
3213 * the protocol does not support a particular function. In certain
3214 * cases where it makes no sense for a protocol to have a "do nothing"
3215 * function, some default processing is provided.
3218 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3222 EXPORT_SYMBOL(sock_no_bind);
3224 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3229 EXPORT_SYMBOL(sock_no_connect);
3231 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3235 EXPORT_SYMBOL(sock_no_socketpair);
3237 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3242 EXPORT_SYMBOL(sock_no_accept);
3244 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3249 EXPORT_SYMBOL(sock_no_getname);
3251 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3255 EXPORT_SYMBOL(sock_no_ioctl);
3257 int sock_no_listen(struct socket *sock, int backlog)
3261 EXPORT_SYMBOL(sock_no_listen);
3263 int sock_no_shutdown(struct socket *sock, int how)
3267 EXPORT_SYMBOL(sock_no_shutdown);
3269 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3273 EXPORT_SYMBOL(sock_no_sendmsg);
3275 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3279 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3281 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3286 EXPORT_SYMBOL(sock_no_recvmsg);
3288 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3290 /* Mirror missing mmap method error code */
3293 EXPORT_SYMBOL(sock_no_mmap);
3296 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3297 * various sock-based usage counts.
3299 void __receive_sock(struct file *file)
3301 struct socket *sock;
3303 sock = sock_from_file(file);
3305 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3306 sock_update_classid(&sock->sk->sk_cgrp_data);
3311 * Default Socket Callbacks
3314 static void sock_def_wakeup(struct sock *sk)
3316 struct socket_wq *wq;
3319 wq = rcu_dereference(sk->sk_wq);
3320 if (skwq_has_sleeper(wq))
3321 wake_up_interruptible_all(&wq->wait);
3325 static void sock_def_error_report(struct sock *sk)
3327 struct socket_wq *wq;
3330 wq = rcu_dereference(sk->sk_wq);
3331 if (skwq_has_sleeper(wq))
3332 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3333 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3337 void sock_def_readable(struct sock *sk)
3339 struct socket_wq *wq;
3341 trace_sk_data_ready(sk);
3344 wq = rcu_dereference(sk->sk_wq);
3345 if (skwq_has_sleeper(wq))
3346 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3347 EPOLLRDNORM | EPOLLRDBAND);
3348 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3352 static void sock_def_write_space(struct sock *sk)
3354 struct socket_wq *wq;
3358 /* Do not wake up a writer until he can make "significant"
3361 if (sock_writeable(sk)) {
3362 wq = rcu_dereference(sk->sk_wq);
3363 if (skwq_has_sleeper(wq))
3364 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3365 EPOLLWRNORM | EPOLLWRBAND);
3367 /* Should agree with poll, otherwise some programs break */
3368 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3374 /* An optimised version of sock_def_write_space(), should only be called
3375 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3378 static void sock_def_write_space_wfree(struct sock *sk)
3380 /* Do not wake up a writer until he can make "significant"
3383 if (sock_writeable(sk)) {
3384 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3386 /* rely on refcount_sub from sock_wfree() */
3387 smp_mb__after_atomic();
3388 if (wq && waitqueue_active(&wq->wait))
3389 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3390 EPOLLWRNORM | EPOLLWRBAND);
3392 /* Should agree with poll, otherwise some programs break */
3393 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3397 static void sock_def_destruct(struct sock *sk)
3401 void sk_send_sigurg(struct sock *sk)
3403 if (sk->sk_socket && sk->sk_socket->file)
3404 if (send_sigurg(&sk->sk_socket->file->f_owner))
3405 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3407 EXPORT_SYMBOL(sk_send_sigurg);
3409 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3410 unsigned long expires)
3412 if (!mod_timer(timer, expires))
3415 EXPORT_SYMBOL(sk_reset_timer);
3417 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3419 if (del_timer(timer))
3422 EXPORT_SYMBOL(sk_stop_timer);
3424 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3426 if (del_timer_sync(timer))
3429 EXPORT_SYMBOL(sk_stop_timer_sync);
3431 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3434 sk->sk_send_head = NULL;
3436 timer_setup(&sk->sk_timer, NULL, 0);
3438 sk->sk_allocation = GFP_KERNEL;
3439 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3440 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3441 sk->sk_state = TCP_CLOSE;
3442 sk->sk_use_task_frag = true;
3443 sk_set_socket(sk, sock);
3445 sock_set_flag(sk, SOCK_ZAPPED);
3448 sk->sk_type = sock->type;
3449 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3452 RCU_INIT_POINTER(sk->sk_wq, NULL);
3456 rwlock_init(&sk->sk_callback_lock);
3457 if (sk->sk_kern_sock)
3458 lockdep_set_class_and_name(
3459 &sk->sk_callback_lock,
3460 af_kern_callback_keys + sk->sk_family,
3461 af_family_kern_clock_key_strings[sk->sk_family]);
3463 lockdep_set_class_and_name(
3464 &sk->sk_callback_lock,
3465 af_callback_keys + sk->sk_family,
3466 af_family_clock_key_strings[sk->sk_family]);
3468 sk->sk_state_change = sock_def_wakeup;
3469 sk->sk_data_ready = sock_def_readable;
3470 sk->sk_write_space = sock_def_write_space;
3471 sk->sk_error_report = sock_def_error_report;
3472 sk->sk_destruct = sock_def_destruct;
3474 sk->sk_frag.page = NULL;
3475 sk->sk_frag.offset = 0;
3476 sk->sk_peek_off = -1;
3478 sk->sk_peer_pid = NULL;
3479 sk->sk_peer_cred = NULL;
3480 spin_lock_init(&sk->sk_peer_lock);
3482 sk->sk_write_pending = 0;
3483 sk->sk_rcvlowat = 1;
3484 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3485 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3487 sk->sk_stamp = SK_DEFAULT_STAMP;
3488 #if BITS_PER_LONG==32
3489 seqlock_init(&sk->sk_stamp_seq);
3491 atomic_set(&sk->sk_zckey, 0);
3493 #ifdef CONFIG_NET_RX_BUSY_POLL
3495 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3498 sk->sk_max_pacing_rate = ~0UL;
3499 sk->sk_pacing_rate = ~0UL;
3500 WRITE_ONCE(sk->sk_pacing_shift, 10);
3501 sk->sk_incoming_cpu = -1;
3503 sk_rx_queue_clear(sk);
3505 * Before updating sk_refcnt, we must commit prior changes to memory
3506 * (Documentation/RCU/rculist_nulls.rst for details)
3509 refcount_set(&sk->sk_refcnt, 1);
3510 atomic_set(&sk->sk_drops, 0);
3512 EXPORT_SYMBOL(sock_init_data_uid);
3514 void sock_init_data(struct socket *sock, struct sock *sk)
3517 SOCK_INODE(sock)->i_uid :
3518 make_kuid(sock_net(sk)->user_ns, 0);
3520 sock_init_data_uid(sock, sk, uid);
3522 EXPORT_SYMBOL(sock_init_data);
3524 void lock_sock_nested(struct sock *sk, int subclass)
3526 /* The sk_lock has mutex_lock() semantics here. */
3527 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3530 spin_lock_bh(&sk->sk_lock.slock);
3531 if (sock_owned_by_user_nocheck(sk))
3533 sk->sk_lock.owned = 1;
3534 spin_unlock_bh(&sk->sk_lock.slock);
3536 EXPORT_SYMBOL(lock_sock_nested);
3538 void release_sock(struct sock *sk)
3540 spin_lock_bh(&sk->sk_lock.slock);
3541 if (sk->sk_backlog.tail)
3544 if (sk->sk_prot->release_cb)
3545 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3546 tcp_release_cb, sk);
3548 sock_release_ownership(sk);
3549 if (waitqueue_active(&sk->sk_lock.wq))
3550 wake_up(&sk->sk_lock.wq);
3551 spin_unlock_bh(&sk->sk_lock.slock);
3553 EXPORT_SYMBOL(release_sock);
3555 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3558 spin_lock_bh(&sk->sk_lock.slock);
3560 if (!sock_owned_by_user_nocheck(sk)) {
3562 * Fast path return with bottom halves disabled and
3563 * sock::sk_lock.slock held.
3565 * The 'mutex' is not contended and holding
3566 * sock::sk_lock.slock prevents all other lockers to
3567 * proceed so the corresponding unlock_sock_fast() can
3568 * avoid the slow path of release_sock() completely and
3569 * just release slock.
3571 * From a semantical POV this is equivalent to 'acquiring'
3572 * the 'mutex', hence the corresponding lockdep
3573 * mutex_release() has to happen in the fast path of
3574 * unlock_sock_fast().
3580 sk->sk_lock.owned = 1;
3581 __acquire(&sk->sk_lock.slock);
3582 spin_unlock_bh(&sk->sk_lock.slock);
3585 EXPORT_SYMBOL(__lock_sock_fast);
3587 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3588 bool timeval, bool time32)
3590 struct sock *sk = sock->sk;
3591 struct timespec64 ts;
3593 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3594 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3595 if (ts.tv_sec == -1)
3597 if (ts.tv_sec == 0) {
3598 ktime_t kt = ktime_get_real();
3599 sock_write_timestamp(sk, kt);
3600 ts = ktime_to_timespec64(kt);
3606 #ifdef CONFIG_COMPAT_32BIT_TIME
3608 return put_old_timespec32(&ts, userstamp);
3610 #ifdef CONFIG_SPARC64
3611 /* beware of padding in sparc64 timeval */
3612 if (timeval && !in_compat_syscall()) {
3613 struct __kernel_old_timeval __user tv = {
3614 .tv_sec = ts.tv_sec,
3615 .tv_usec = ts.tv_nsec,
3617 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3622 return put_timespec64(&ts, userstamp);
3624 EXPORT_SYMBOL(sock_gettstamp);
3626 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3628 if (!sock_flag(sk, flag)) {
3629 unsigned long previous_flags = sk->sk_flags;
3631 sock_set_flag(sk, flag);
3633 * we just set one of the two flags which require net
3634 * time stamping, but time stamping might have been on
3635 * already because of the other one
3637 if (sock_needs_netstamp(sk) &&
3638 !(previous_flags & SK_FLAGS_TIMESTAMP))
3639 net_enable_timestamp();
3643 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3644 int level, int type)
3646 struct sock_exterr_skb *serr;
3647 struct sk_buff *skb;
3651 skb = sock_dequeue_err_skb(sk);
3657 msg->msg_flags |= MSG_TRUNC;
3660 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3664 sock_recv_timestamp(msg, sk, skb);
3666 serr = SKB_EXT_ERR(skb);
3667 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3669 msg->msg_flags |= MSG_ERRQUEUE;
3677 EXPORT_SYMBOL(sock_recv_errqueue);
3680 * Get a socket option on an socket.
3682 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3683 * asynchronous errors should be reported by getsockopt. We assume
3684 * this means if you specify SO_ERROR (otherwise whats the point of it).
3686 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3687 char __user *optval, int __user *optlen)
3689 struct sock *sk = sock->sk;
3691 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3692 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3694 EXPORT_SYMBOL(sock_common_getsockopt);
3696 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3699 struct sock *sk = sock->sk;
3703 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3705 msg->msg_namelen = addr_len;
3708 EXPORT_SYMBOL(sock_common_recvmsg);
3711 * Set socket options on an inet socket.
3713 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3714 sockptr_t optval, unsigned int optlen)
3716 struct sock *sk = sock->sk;
3718 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3719 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3721 EXPORT_SYMBOL(sock_common_setsockopt);
3723 void sk_common_release(struct sock *sk)
3725 if (sk->sk_prot->destroy)
3726 sk->sk_prot->destroy(sk);
3729 * Observation: when sk_common_release is called, processes have
3730 * no access to socket. But net still has.
3731 * Step one, detach it from networking:
3733 * A. Remove from hash tables.
3736 sk->sk_prot->unhash(sk);
3739 * In this point socket cannot receive new packets, but it is possible
3740 * that some packets are in flight because some CPU runs receiver and
3741 * did hash table lookup before we unhashed socket. They will achieve
3742 * receive queue and will be purged by socket destructor.
3744 * Also we still have packets pending on receive queue and probably,
3745 * our own packets waiting in device queues. sock_destroy will drain
3746 * receive queue, but transmitted packets will delay socket destruction
3747 * until the last reference will be released.
3752 xfrm_sk_free_policy(sk);
3756 EXPORT_SYMBOL(sk_common_release);
3758 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3760 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3762 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3763 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3764 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3765 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3766 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3767 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3768 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3769 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3770 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3773 #ifdef CONFIG_PROC_FS
3774 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3776 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3778 int cpu, idx = prot->inuse_idx;
3781 for_each_possible_cpu(cpu)
3782 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3784 return res >= 0 ? res : 0;
3786 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3788 int sock_inuse_get(struct net *net)
3792 for_each_possible_cpu(cpu)
3793 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3798 EXPORT_SYMBOL_GPL(sock_inuse_get);
3800 static int __net_init sock_inuse_init_net(struct net *net)
3802 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3803 if (net->core.prot_inuse == NULL)
3808 static void __net_exit sock_inuse_exit_net(struct net *net)
3810 free_percpu(net->core.prot_inuse);
3813 static struct pernet_operations net_inuse_ops = {
3814 .init = sock_inuse_init_net,
3815 .exit = sock_inuse_exit_net,
3818 static __init int net_inuse_init(void)
3820 if (register_pernet_subsys(&net_inuse_ops))
3821 panic("Cannot initialize net inuse counters");
3826 core_initcall(net_inuse_init);
3828 static int assign_proto_idx(struct proto *prot)
3830 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3832 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3833 pr_err("PROTO_INUSE_NR exhausted\n");
3837 set_bit(prot->inuse_idx, proto_inuse_idx);
3841 static void release_proto_idx(struct proto *prot)
3843 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3844 clear_bit(prot->inuse_idx, proto_inuse_idx);
3847 static inline int assign_proto_idx(struct proto *prot)
3852 static inline void release_proto_idx(struct proto *prot)
3858 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3862 kfree(twsk_prot->twsk_slab_name);
3863 twsk_prot->twsk_slab_name = NULL;
3864 kmem_cache_destroy(twsk_prot->twsk_slab);
3865 twsk_prot->twsk_slab = NULL;
3868 static int tw_prot_init(const struct proto *prot)
3870 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3875 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3877 if (!twsk_prot->twsk_slab_name)
3880 twsk_prot->twsk_slab =
3881 kmem_cache_create(twsk_prot->twsk_slab_name,
3882 twsk_prot->twsk_obj_size, 0,
3883 SLAB_ACCOUNT | prot->slab_flags,
3885 if (!twsk_prot->twsk_slab) {
3886 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3894 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3898 kfree(rsk_prot->slab_name);
3899 rsk_prot->slab_name = NULL;
3900 kmem_cache_destroy(rsk_prot->slab);
3901 rsk_prot->slab = NULL;
3904 static int req_prot_init(const struct proto *prot)
3906 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3911 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3913 if (!rsk_prot->slab_name)
3916 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3917 rsk_prot->obj_size, 0,
3918 SLAB_ACCOUNT | prot->slab_flags,
3921 if (!rsk_prot->slab) {
3922 pr_crit("%s: Can't create request sock SLAB cache!\n",
3929 int proto_register(struct proto *prot, int alloc_slab)
3933 if (prot->memory_allocated && !prot->sysctl_mem) {
3934 pr_err("%s: missing sysctl_mem\n", prot->name);
3937 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3938 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3942 prot->slab = kmem_cache_create_usercopy(prot->name,
3944 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3946 prot->useroffset, prot->usersize,
3949 if (prot->slab == NULL) {
3950 pr_crit("%s: Can't create sock SLAB cache!\n",
3955 if (req_prot_init(prot))
3956 goto out_free_request_sock_slab;
3958 if (tw_prot_init(prot))
3959 goto out_free_timewait_sock_slab;
3962 mutex_lock(&proto_list_mutex);
3963 ret = assign_proto_idx(prot);
3965 mutex_unlock(&proto_list_mutex);
3966 goto out_free_timewait_sock_slab;
3968 list_add(&prot->node, &proto_list);
3969 mutex_unlock(&proto_list_mutex);
3972 out_free_timewait_sock_slab:
3974 tw_prot_cleanup(prot->twsk_prot);
3975 out_free_request_sock_slab:
3977 req_prot_cleanup(prot->rsk_prot);
3979 kmem_cache_destroy(prot->slab);
3985 EXPORT_SYMBOL(proto_register);
3987 void proto_unregister(struct proto *prot)
3989 mutex_lock(&proto_list_mutex);
3990 release_proto_idx(prot);
3991 list_del(&prot->node);
3992 mutex_unlock(&proto_list_mutex);
3994 kmem_cache_destroy(prot->slab);
3997 req_prot_cleanup(prot->rsk_prot);
3998 tw_prot_cleanup(prot->twsk_prot);
4000 EXPORT_SYMBOL(proto_unregister);
4002 int sock_load_diag_module(int family, int protocol)
4005 if (!sock_is_registered(family))
4008 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4009 NETLINK_SOCK_DIAG, family);
4013 if (family == AF_INET &&
4014 protocol != IPPROTO_RAW &&
4015 protocol < MAX_INET_PROTOS &&
4016 !rcu_access_pointer(inet_protos[protocol]))
4020 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4021 NETLINK_SOCK_DIAG, family, protocol);
4023 EXPORT_SYMBOL(sock_load_diag_module);
4025 #ifdef CONFIG_PROC_FS
4026 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4027 __acquires(proto_list_mutex)
4029 mutex_lock(&proto_list_mutex);
4030 return seq_list_start_head(&proto_list, *pos);
4033 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4035 return seq_list_next(v, &proto_list, pos);
4038 static void proto_seq_stop(struct seq_file *seq, void *v)
4039 __releases(proto_list_mutex)
4041 mutex_unlock(&proto_list_mutex);
4044 static char proto_method_implemented(const void *method)
4046 return method == NULL ? 'n' : 'y';
4048 static long sock_prot_memory_allocated(struct proto *proto)
4050 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4053 static const char *sock_prot_memory_pressure(struct proto *proto)
4055 return proto->memory_pressure != NULL ?
4056 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4059 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4062 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4063 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4066 sock_prot_inuse_get(seq_file_net(seq), proto),
4067 sock_prot_memory_allocated(proto),
4068 sock_prot_memory_pressure(proto),
4070 proto->slab == NULL ? "no" : "yes",
4071 module_name(proto->owner),
4072 proto_method_implemented(proto->close),
4073 proto_method_implemented(proto->connect),
4074 proto_method_implemented(proto->disconnect),
4075 proto_method_implemented(proto->accept),
4076 proto_method_implemented(proto->ioctl),
4077 proto_method_implemented(proto->init),
4078 proto_method_implemented(proto->destroy),
4079 proto_method_implemented(proto->shutdown),
4080 proto_method_implemented(proto->setsockopt),
4081 proto_method_implemented(proto->getsockopt),
4082 proto_method_implemented(proto->sendmsg),
4083 proto_method_implemented(proto->recvmsg),
4084 proto_method_implemented(proto->bind),
4085 proto_method_implemented(proto->backlog_rcv),
4086 proto_method_implemented(proto->hash),
4087 proto_method_implemented(proto->unhash),
4088 proto_method_implemented(proto->get_port),
4089 proto_method_implemented(proto->enter_memory_pressure));
4092 static int proto_seq_show(struct seq_file *seq, void *v)
4094 if (v == &proto_list)
4095 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4104 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4106 proto_seq_printf(seq, list_entry(v, struct proto, node));
4110 static const struct seq_operations proto_seq_ops = {
4111 .start = proto_seq_start,
4112 .next = proto_seq_next,
4113 .stop = proto_seq_stop,
4114 .show = proto_seq_show,
4117 static __net_init int proto_init_net(struct net *net)
4119 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4120 sizeof(struct seq_net_private)))
4126 static __net_exit void proto_exit_net(struct net *net)
4128 remove_proc_entry("protocols", net->proc_net);
4132 static __net_initdata struct pernet_operations proto_net_ops = {
4133 .init = proto_init_net,
4134 .exit = proto_exit_net,
4137 static int __init proto_init(void)
4139 return register_pernet_subsys(&proto_net_ops);
4142 subsys_initcall(proto_init);
4144 #endif /* PROC_FS */
4146 #ifdef CONFIG_NET_RX_BUSY_POLL
4147 bool sk_busy_loop_end(void *p, unsigned long start_time)
4149 struct sock *sk = p;
4151 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
4152 sk_busy_loop_timeout(sk, start_time);
4154 EXPORT_SYMBOL(sk_busy_loop_end);
4155 #endif /* CONFIG_NET_RX_BUSY_POLL */
4157 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4159 if (!sk->sk_prot->bind_add)
4161 return sk->sk_prot->bind_add(sk, addr, addr_len);
4163 EXPORT_SYMBOL(sock_bind_add);
4165 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4166 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4167 void __user *arg, void *karg, size_t size)
4171 if (copy_from_user(karg, arg, size))
4174 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4178 if (copy_to_user(arg, karg, size))
4183 EXPORT_SYMBOL(sock_ioctl_inout);
4185 /* This is the most common ioctl prep function, where the result (4 bytes) is
4186 * copied back to userspace if the ioctl() returns successfully. No input is
4187 * copied from userspace as input argument.
4189 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4193 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4197 return put_user(karg, (int __user *)arg);
4200 /* A wrapper around sock ioctls, which copies the data from userspace
4201 * (depending on the protocol/ioctl), and copies back the result to userspace.
4202 * The main motivation for this function is to pass kernel memory to the
4203 * protocol ioctl callbacks, instead of userspace memory.
4205 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4209 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4210 rc = ipmr_sk_ioctl(sk, cmd, arg);
4211 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4212 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4213 else if (sk_is_phonet(sk))
4214 rc = phonet_sk_ioctl(sk, cmd, arg);
4216 /* If ioctl was processed, returns its value */
4220 /* Otherwise call the default handler */
4221 return sock_ioctl_out(sk, cmd, arg);
4223 EXPORT_SYMBOL(sk_ioctl);