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/udp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
114 #include <linux/static_key.h>
115 #include <linux/memcontrol.h>
116 #include <linux/prefetch.h>
117 #include <linux/compat.h>
118 #include <linux/mroute.h>
119 #include <linux/mroute6.h>
120 #include <linux/icmpv6.h>
122 #include <linux/uaccess.h>
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <net/net_namespace.h>
128 #include <net/request_sock.h>
129 #include <net/sock.h>
130 #include <linux/net_tstamp.h>
131 #include <net/xfrm.h>
132 #include <linux/ipsec.h>
133 #include <net/cls_cgroup.h>
134 #include <net/netprio_cgroup.h>
135 #include <linux/sock_diag.h>
137 #include <linux/filter.h>
138 #include <net/sock_reuseport.h>
139 #include <net/bpf_sk_storage.h>
141 #include <trace/events/sock.h>
144 #include <net/busy_poll.h>
145 #include <net/phonet/phonet.h>
147 #include <linux/ethtool.h>
151 static DEFINE_MUTEX(proto_list_mutex);
152 static LIST_HEAD(proto_list);
154 static void sock_def_write_space_wfree(struct sock *sk);
155 static void sock_def_write_space(struct sock *sk);
158 * sk_ns_capable - General socket capability test
159 * @sk: Socket to use a capability on or through
160 * @user_ns: The user namespace of the capability to use
161 * @cap: The capability to use
163 * Test to see if the opener of the socket had when the socket was
164 * created and the current process has the capability @cap in the user
165 * namespace @user_ns.
167 bool sk_ns_capable(const struct sock *sk,
168 struct user_namespace *user_ns, int cap)
170 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
171 ns_capable(user_ns, cap);
173 EXPORT_SYMBOL(sk_ns_capable);
176 * sk_capable - Socket global capability test
177 * @sk: Socket to use a capability on or through
178 * @cap: The global capability to use
180 * Test to see if the opener of the socket had when the socket was
181 * created and the current process has the capability @cap in all user
184 bool sk_capable(const struct sock *sk, int cap)
186 return sk_ns_capable(sk, &init_user_ns, cap);
188 EXPORT_SYMBOL(sk_capable);
191 * sk_net_capable - Network namespace socket capability test
192 * @sk: Socket to use a capability on or through
193 * @cap: The capability to use
195 * Test to see if the opener of the socket had when the socket was created
196 * and the current process has the capability @cap over the network namespace
197 * the socket is a member of.
199 bool sk_net_capable(const struct sock *sk, int cap)
201 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
203 EXPORT_SYMBOL(sk_net_capable);
206 * Each address family might have different locking rules, so we have
207 * one slock key per address family and separate keys for internal and
210 static struct lock_class_key af_family_keys[AF_MAX];
211 static struct lock_class_key af_family_kern_keys[AF_MAX];
212 static struct lock_class_key af_family_slock_keys[AF_MAX];
213 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
216 * Make lock validator output more readable. (we pre-construct these
217 * strings build-time, so that runtime initialization of socket
221 #define _sock_locks(x) \
222 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
223 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
224 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
225 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
226 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
227 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
228 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
229 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
230 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
231 x "27" , x "28" , x "AF_CAN" , \
232 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
233 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
234 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
235 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
236 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
240 static const char *const af_family_key_strings[AF_MAX+1] = {
241 _sock_locks("sk_lock-")
243 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
244 _sock_locks("slock-")
246 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
247 _sock_locks("clock-")
250 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
251 _sock_locks("k-sk_lock-")
253 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
254 _sock_locks("k-slock-")
256 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
257 _sock_locks("k-clock-")
259 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
260 _sock_locks("rlock-")
262 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
263 _sock_locks("wlock-")
265 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
266 _sock_locks("elock-")
270 * sk_callback_lock and sk queues locking rules are per-address-family,
271 * so split the lock classes by using a per-AF key:
273 static struct lock_class_key af_callback_keys[AF_MAX];
274 static struct lock_class_key af_rlock_keys[AF_MAX];
275 static struct lock_class_key af_wlock_keys[AF_MAX];
276 static struct lock_class_key af_elock_keys[AF_MAX];
277 static struct lock_class_key af_kern_callback_keys[AF_MAX];
279 /* Run time adjustable parameters. */
280 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
281 EXPORT_SYMBOL(sysctl_wmem_max);
282 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
283 EXPORT_SYMBOL(sysctl_rmem_max);
284 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
285 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
286 int sysctl_mem_pcpu_rsv __read_mostly = SK_MEMORY_PCPU_RESERVE;
288 int sysctl_tstamp_allow_data __read_mostly = 1;
290 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
291 EXPORT_SYMBOL_GPL(memalloc_socks_key);
294 * sk_set_memalloc - sets %SOCK_MEMALLOC
295 * @sk: socket to set it on
297 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
298 * It's the responsibility of the admin to adjust min_free_kbytes
299 * to meet the requirements
301 void sk_set_memalloc(struct sock *sk)
303 sock_set_flag(sk, SOCK_MEMALLOC);
304 sk->sk_allocation |= __GFP_MEMALLOC;
305 static_branch_inc(&memalloc_socks_key);
307 EXPORT_SYMBOL_GPL(sk_set_memalloc);
309 void sk_clear_memalloc(struct sock *sk)
311 sock_reset_flag(sk, SOCK_MEMALLOC);
312 sk->sk_allocation &= ~__GFP_MEMALLOC;
313 static_branch_dec(&memalloc_socks_key);
316 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
317 * progress of swapping. SOCK_MEMALLOC may be cleared while
318 * it has rmem allocations due to the last swapfile being deactivated
319 * but there is a risk that the socket is unusable due to exceeding
320 * the rmem limits. Reclaim the reserves and obey rmem limits again.
324 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
326 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
329 unsigned int noreclaim_flag;
331 /* these should have been dropped before queueing */
332 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
334 noreclaim_flag = memalloc_noreclaim_save();
335 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
339 memalloc_noreclaim_restore(noreclaim_flag);
343 EXPORT_SYMBOL(__sk_backlog_rcv);
345 void sk_error_report(struct sock *sk)
347 sk->sk_error_report(sk);
349 switch (sk->sk_family) {
353 trace_inet_sk_error_report(sk);
359 EXPORT_SYMBOL(sk_error_report);
361 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
363 struct __kernel_sock_timeval tv;
365 if (timeo == MAX_SCHEDULE_TIMEOUT) {
369 tv.tv_sec = timeo / HZ;
370 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
373 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
374 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
375 *(struct old_timeval32 *)optval = tv32;
380 struct __kernel_old_timeval old_tv;
381 old_tv.tv_sec = tv.tv_sec;
382 old_tv.tv_usec = tv.tv_usec;
383 *(struct __kernel_old_timeval *)optval = old_tv;
384 return sizeof(old_tv);
387 *(struct __kernel_sock_timeval *)optval = tv;
390 EXPORT_SYMBOL(sock_get_timeout);
392 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
393 sockptr_t optval, int optlen, bool old_timeval)
395 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
396 struct old_timeval32 tv32;
398 if (optlen < sizeof(tv32))
401 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
403 tv->tv_sec = tv32.tv_sec;
404 tv->tv_usec = tv32.tv_usec;
405 } else if (old_timeval) {
406 struct __kernel_old_timeval old_tv;
408 if (optlen < sizeof(old_tv))
410 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
412 tv->tv_sec = old_tv.tv_sec;
413 tv->tv_usec = old_tv.tv_usec;
415 if (optlen < sizeof(*tv))
417 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
423 EXPORT_SYMBOL(sock_copy_user_timeval);
425 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
428 struct __kernel_sock_timeval tv;
429 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
435 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
439 static int warned __read_mostly;
441 WRITE_ONCE(*timeo_p, 0);
442 if (warned < 10 && net_ratelimit()) {
444 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
445 __func__, current->comm, task_pid_nr(current));
449 val = MAX_SCHEDULE_TIMEOUT;
450 if ((tv.tv_sec || tv.tv_usec) &&
451 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
452 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
454 WRITE_ONCE(*timeo_p, val);
458 static bool sock_needs_netstamp(const struct sock *sk)
460 switch (sk->sk_family) {
469 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
471 if (sk->sk_flags & flags) {
472 sk->sk_flags &= ~flags;
473 if (sock_needs_netstamp(sk) &&
474 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
475 net_disable_timestamp();
480 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
483 struct sk_buff_head *list = &sk->sk_receive_queue;
485 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
486 atomic_inc(&sk->sk_drops);
487 trace_sock_rcvqueue_full(sk, skb);
491 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
492 atomic_inc(&sk->sk_drops);
497 skb_set_owner_r(skb, sk);
499 /* we escape from rcu protected region, make sure we dont leak
504 spin_lock_irqsave(&list->lock, flags);
505 sock_skb_set_dropcount(sk, skb);
506 __skb_queue_tail(list, skb);
507 spin_unlock_irqrestore(&list->lock, flags);
509 if (!sock_flag(sk, SOCK_DEAD))
510 sk->sk_data_ready(sk);
513 EXPORT_SYMBOL(__sock_queue_rcv_skb);
515 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
516 enum skb_drop_reason *reason)
518 enum skb_drop_reason drop_reason;
521 err = sk_filter(sk, skb);
523 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
526 err = __sock_queue_rcv_skb(sk, skb);
529 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
532 drop_reason = SKB_DROP_REASON_PROTO_MEM;
535 drop_reason = SKB_NOT_DROPPED_YET;
540 *reason = drop_reason;
543 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
545 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
546 const int nested, unsigned int trim_cap, bool refcounted)
548 int rc = NET_RX_SUCCESS;
550 if (sk_filter_trim_cap(sk, skb, trim_cap))
551 goto discard_and_relse;
555 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
556 atomic_inc(&sk->sk_drops);
557 goto discard_and_relse;
560 bh_lock_sock_nested(sk);
563 if (!sock_owned_by_user(sk)) {
565 * trylock + unlock semantics:
567 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
569 rc = sk_backlog_rcv(sk, skb);
571 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
572 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
574 atomic_inc(&sk->sk_drops);
575 goto discard_and_relse;
587 EXPORT_SYMBOL(__sk_receive_skb);
589 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
591 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
593 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
595 struct dst_entry *dst = __sk_dst_get(sk);
597 if (dst && dst->obsolete &&
598 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
599 dst, cookie) == NULL) {
600 sk_tx_queue_clear(sk);
601 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
602 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
609 EXPORT_SYMBOL(__sk_dst_check);
611 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
613 struct dst_entry *dst = sk_dst_get(sk);
615 if (dst && dst->obsolete &&
616 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
617 dst, cookie) == NULL) {
625 EXPORT_SYMBOL(sk_dst_check);
627 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
629 int ret = -ENOPROTOOPT;
630 #ifdef CONFIG_NETDEVICES
631 struct net *net = sock_net(sk);
635 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
642 /* Paired with all READ_ONCE() done locklessly. */
643 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
645 if (sk->sk_prot->rehash)
646 sk->sk_prot->rehash(sk);
657 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
663 ret = sock_bindtoindex_locked(sk, ifindex);
669 EXPORT_SYMBOL(sock_bindtoindex);
671 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
673 int ret = -ENOPROTOOPT;
674 #ifdef CONFIG_NETDEVICES
675 struct net *net = sock_net(sk);
676 char devname[IFNAMSIZ];
683 /* Bind this socket to a particular device like "eth0",
684 * as specified in the passed interface name. If the
685 * name is "" or the option length is zero the socket
688 if (optlen > IFNAMSIZ - 1)
689 optlen = IFNAMSIZ - 1;
690 memset(devname, 0, sizeof(devname));
693 if (copy_from_sockptr(devname, optval, optlen))
697 if (devname[0] != '\0') {
698 struct net_device *dev;
701 dev = dev_get_by_name_rcu(net, devname);
703 index = dev->ifindex;
710 sockopt_lock_sock(sk);
711 ret = sock_bindtoindex_locked(sk, index);
712 sockopt_release_sock(sk);
719 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
720 sockptr_t optlen, int len)
722 int ret = -ENOPROTOOPT;
723 #ifdef CONFIG_NETDEVICES
724 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
725 struct net *net = sock_net(sk);
726 char devname[IFNAMSIZ];
728 if (bound_dev_if == 0) {
737 ret = netdev_get_name(net, devname, bound_dev_if);
741 len = strlen(devname) + 1;
744 if (copy_to_sockptr(optval, devname, len))
749 if (copy_to_sockptr(optlen, &len, sizeof(int)))
760 bool sk_mc_loop(const struct sock *sk)
762 if (dev_recursion_level())
766 /* IPV6_ADDRFORM can change sk->sk_family under us. */
767 switch (READ_ONCE(sk->sk_family)) {
769 return inet_test_bit(MC_LOOP, sk);
770 #if IS_ENABLED(CONFIG_IPV6)
772 return inet6_test_bit(MC6_LOOP, sk);
778 EXPORT_SYMBOL(sk_mc_loop);
780 void sock_set_reuseaddr(struct sock *sk)
783 sk->sk_reuse = SK_CAN_REUSE;
786 EXPORT_SYMBOL(sock_set_reuseaddr);
788 void sock_set_reuseport(struct sock *sk)
791 sk->sk_reuseport = true;
794 EXPORT_SYMBOL(sock_set_reuseport);
796 void sock_no_linger(struct sock *sk)
799 WRITE_ONCE(sk->sk_lingertime, 0);
800 sock_set_flag(sk, SOCK_LINGER);
803 EXPORT_SYMBOL(sock_no_linger);
805 void sock_set_priority(struct sock *sk, u32 priority)
807 WRITE_ONCE(sk->sk_priority, priority);
809 EXPORT_SYMBOL(sock_set_priority);
811 void sock_set_sndtimeo(struct sock *sk, s64 secs)
814 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
815 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
817 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
820 EXPORT_SYMBOL(sock_set_sndtimeo);
822 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
825 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
826 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
827 sock_set_flag(sk, SOCK_RCVTSTAMP);
828 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
830 sock_reset_flag(sk, SOCK_RCVTSTAMP);
831 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
835 void sock_enable_timestamps(struct sock *sk)
838 __sock_set_timestamps(sk, true, false, true);
841 EXPORT_SYMBOL(sock_enable_timestamps);
843 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
846 case SO_TIMESTAMP_OLD:
847 __sock_set_timestamps(sk, valbool, false, false);
849 case SO_TIMESTAMP_NEW:
850 __sock_set_timestamps(sk, valbool, true, false);
852 case SO_TIMESTAMPNS_OLD:
853 __sock_set_timestamps(sk, valbool, false, true);
855 case SO_TIMESTAMPNS_NEW:
856 __sock_set_timestamps(sk, valbool, true, true);
861 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
863 struct net *net = sock_net(sk);
864 struct net_device *dev = NULL;
869 if (sk->sk_bound_dev_if)
870 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
873 pr_err("%s: sock not bind to device\n", __func__);
877 num = ethtool_get_phc_vclocks(dev, &vclock_index);
880 for (i = 0; i < num; i++) {
881 if (*(vclock_index + i) == phc_index) {
893 WRITE_ONCE(sk->sk_bind_phc, phc_index);
898 int sock_set_timestamping(struct sock *sk, int optname,
899 struct so_timestamping timestamping)
901 int val = timestamping.flags;
904 if (val & ~SOF_TIMESTAMPING_MASK)
907 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
908 !(val & SOF_TIMESTAMPING_OPT_ID))
911 if (val & SOF_TIMESTAMPING_OPT_ID &&
912 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
914 if ((1 << sk->sk_state) &
915 (TCPF_CLOSE | TCPF_LISTEN))
917 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
918 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
920 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
922 atomic_set(&sk->sk_tskey, 0);
926 if (val & SOF_TIMESTAMPING_OPT_STATS &&
927 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
930 if (val & SOF_TIMESTAMPING_BIND_PHC) {
931 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
936 WRITE_ONCE(sk->sk_tsflags, val);
937 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
939 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
940 sock_enable_timestamp(sk,
941 SOCK_TIMESTAMPING_RX_SOFTWARE);
943 sock_disable_timestamp(sk,
944 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
948 void sock_set_keepalive(struct sock *sk)
951 if (sk->sk_prot->keepalive)
952 sk->sk_prot->keepalive(sk, true);
953 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
956 EXPORT_SYMBOL(sock_set_keepalive);
958 static void __sock_set_rcvbuf(struct sock *sk, int val)
960 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
961 * as a negative value.
963 val = min_t(int, val, INT_MAX / 2);
964 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
966 /* We double it on the way in to account for "struct sk_buff" etc.
967 * overhead. Applications assume that the SO_RCVBUF setting they make
968 * will allow that much actual data to be received on that socket.
970 * Applications are unaware that "struct sk_buff" and other overheads
971 * allocate from the receive buffer during socket buffer allocation.
973 * And after considering the possible alternatives, returning the value
974 * we actually used in getsockopt is the most desirable behavior.
976 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
979 void sock_set_rcvbuf(struct sock *sk, int val)
982 __sock_set_rcvbuf(sk, val);
985 EXPORT_SYMBOL(sock_set_rcvbuf);
987 static void __sock_set_mark(struct sock *sk, u32 val)
989 if (val != sk->sk_mark) {
990 WRITE_ONCE(sk->sk_mark, val);
995 void sock_set_mark(struct sock *sk, u32 val)
998 __sock_set_mark(sk, val);
1001 EXPORT_SYMBOL(sock_set_mark);
1003 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1005 /* Round down bytes to multiple of pages */
1006 bytes = round_down(bytes, PAGE_SIZE);
1008 WARN_ON(bytes > sk->sk_reserved_mem);
1009 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1013 static int sock_reserve_memory(struct sock *sk, int bytes)
1019 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1025 pages = sk_mem_pages(bytes);
1027 /* pre-charge to memcg */
1028 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1029 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1033 /* pre-charge to forward_alloc */
1034 sk_memory_allocated_add(sk, pages);
1035 allocated = sk_memory_allocated(sk);
1036 /* If the system goes into memory pressure with this
1037 * precharge, give up and return error.
1039 if (allocated > sk_prot_mem_limits(sk, 1)) {
1040 sk_memory_allocated_sub(sk, pages);
1041 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1044 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1046 WRITE_ONCE(sk->sk_reserved_mem,
1047 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1052 void sockopt_lock_sock(struct sock *sk)
1054 /* When current->bpf_ctx is set, the setsockopt is called from
1055 * a bpf prog. bpf has ensured the sk lock has been
1056 * acquired before calling setsockopt().
1058 if (has_current_bpf_ctx())
1063 EXPORT_SYMBOL(sockopt_lock_sock);
1065 void sockopt_release_sock(struct sock *sk)
1067 if (has_current_bpf_ctx())
1072 EXPORT_SYMBOL(sockopt_release_sock);
1074 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1076 return has_current_bpf_ctx() || ns_capable(ns, cap);
1078 EXPORT_SYMBOL(sockopt_ns_capable);
1080 bool sockopt_capable(int cap)
1082 return has_current_bpf_ctx() || capable(cap);
1084 EXPORT_SYMBOL(sockopt_capable);
1087 * This is meant for all protocols to use and covers goings on
1088 * at the socket level. Everything here is generic.
1091 int sk_setsockopt(struct sock *sk, int level, int optname,
1092 sockptr_t optval, unsigned int optlen)
1094 struct so_timestamping timestamping;
1095 struct socket *sock = sk->sk_socket;
1096 struct sock_txtime sk_txtime;
1103 * Options without arguments
1106 if (optname == SO_BINDTODEVICE)
1107 return sock_setbindtodevice(sk, optval, optlen);
1109 if (optlen < sizeof(int))
1112 if (copy_from_sockptr(&val, optval, sizeof(val)))
1115 valbool = val ? 1 : 0;
1117 /* handle options which do not require locking the socket. */
1120 if ((val >= 0 && val <= 6) ||
1121 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1122 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1123 sock_set_priority(sk, val);
1128 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1131 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1134 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1140 return -ENOPROTOOPT;
1141 #ifdef CONFIG_NET_RX_BUSY_POLL
1145 WRITE_ONCE(sk->sk_ll_usec, val);
1147 case SO_PREFER_BUSY_POLL:
1148 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1150 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1152 case SO_BUSY_POLL_BUDGET:
1153 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1154 !sockopt_capable(CAP_NET_ADMIN))
1156 if (val < 0 || val > U16_MAX)
1158 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1161 case SO_MAX_PACING_RATE:
1163 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1164 unsigned long pacing_rate;
1166 if (sizeof(ulval) != sizeof(val) &&
1167 optlen >= sizeof(ulval) &&
1168 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1172 cmpxchg(&sk->sk_pacing_status,
1175 /* Pairs with READ_ONCE() from sk_getsockopt() */
1176 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1177 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1178 if (ulval < pacing_rate)
1179 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1183 if (val < -1 || val > 1)
1185 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1186 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1187 /* Paired with READ_ONCE() in tcp_rtx_synack()
1188 * and sk_getsockopt().
1190 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1194 int (*set_peek_off)(struct sock *sk, int val);
1196 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1198 ret = set_peek_off(sk, val);
1205 sockopt_lock_sock(sk);
1209 if (val && !sockopt_capable(CAP_NET_ADMIN))
1212 sock_valbool_flag(sk, SOCK_DBG, valbool);
1215 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1218 sk->sk_reuseport = valbool;
1221 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1225 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1228 /* Don't error on this BSD doesn't and if you think
1229 * about it this is right. Otherwise apps have to
1230 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1231 * are treated in BSD as hints
1233 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1235 /* Ensure val * 2 fits into an int, to prevent max_t()
1236 * from treating it as a negative value.
1238 val = min_t(int, val, INT_MAX / 2);
1239 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1240 WRITE_ONCE(sk->sk_sndbuf,
1241 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1242 /* Wake up sending tasks if we upped the value. */
1243 sk->sk_write_space(sk);
1246 case SO_SNDBUFFORCE:
1247 if (!sockopt_capable(CAP_NET_ADMIN)) {
1252 /* No negative values (to prevent underflow, as val will be
1260 /* Don't error on this BSD doesn't and if you think
1261 * about it this is right. Otherwise apps have to
1262 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1263 * are treated in BSD as hints
1265 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1268 case SO_RCVBUFFORCE:
1269 if (!sockopt_capable(CAP_NET_ADMIN)) {
1274 /* No negative values (to prevent underflow, as val will be
1277 __sock_set_rcvbuf(sk, max(val, 0));
1281 if (sk->sk_prot->keepalive)
1282 sk->sk_prot->keepalive(sk, valbool);
1283 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1287 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1291 sk->sk_no_check_tx = valbool;
1295 if (optlen < sizeof(ling)) {
1296 ret = -EINVAL; /* 1003.1g */
1299 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1303 if (!ling.l_onoff) {
1304 sock_reset_flag(sk, SOCK_LINGER);
1306 unsigned long t_sec = ling.l_linger;
1308 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1309 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1311 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1312 sock_set_flag(sk, SOCK_LINGER);
1319 case SO_TIMESTAMP_OLD:
1320 case SO_TIMESTAMP_NEW:
1321 case SO_TIMESTAMPNS_OLD:
1322 case SO_TIMESTAMPNS_NEW:
1323 sock_set_timestamp(sk, optname, valbool);
1326 case SO_TIMESTAMPING_NEW:
1327 case SO_TIMESTAMPING_OLD:
1328 if (optlen == sizeof(timestamping)) {
1329 if (copy_from_sockptr(×tamping, optval,
1330 sizeof(timestamping))) {
1335 memset(×tamping, 0, sizeof(timestamping));
1336 timestamping.flags = val;
1338 ret = sock_set_timestamping(sk, optname, timestamping);
1343 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1348 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1350 ret = set_rcvlowat(sk, val);
1352 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1355 case SO_RCVTIMEO_OLD:
1356 case SO_RCVTIMEO_NEW:
1357 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1358 optlen, optname == SO_RCVTIMEO_OLD);
1361 case SO_SNDTIMEO_OLD:
1362 case SO_SNDTIMEO_NEW:
1363 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1364 optlen, optname == SO_SNDTIMEO_OLD);
1367 case SO_ATTACH_FILTER: {
1368 struct sock_fprog fprog;
1370 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1372 ret = sk_attach_filter(&fprog, sk);
1377 if (optlen == sizeof(u32)) {
1381 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1384 ret = sk_attach_bpf(ufd, sk);
1388 case SO_ATTACH_REUSEPORT_CBPF: {
1389 struct sock_fprog fprog;
1391 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1393 ret = sk_reuseport_attach_filter(&fprog, sk);
1396 case SO_ATTACH_REUSEPORT_EBPF:
1398 if (optlen == sizeof(u32)) {
1402 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1405 ret = sk_reuseport_attach_bpf(ufd, sk);
1409 case SO_DETACH_REUSEPORT_BPF:
1410 ret = reuseport_detach_prog(sk);
1413 case SO_DETACH_FILTER:
1414 ret = sk_detach_filter(sk);
1417 case SO_LOCK_FILTER:
1418 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1421 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1425 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1426 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1431 __sock_set_mark(sk, val);
1434 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1438 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1441 case SO_WIFI_STATUS:
1442 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1446 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1449 case SO_SELECT_ERR_QUEUE:
1450 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1454 case SO_INCOMING_CPU:
1455 reuseport_update_incoming_cpu(sk, val);
1460 dst_negative_advice(sk);
1464 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1465 if (!(sk_is_tcp(sk) ||
1466 (sk->sk_type == SOCK_DGRAM &&
1467 sk->sk_protocol == IPPROTO_UDP)))
1469 } else if (sk->sk_family != PF_RDS) {
1473 if (val < 0 || val > 1)
1476 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1481 if (optlen != sizeof(struct sock_txtime)) {
1484 } else if (copy_from_sockptr(&sk_txtime, optval,
1485 sizeof(struct sock_txtime))) {
1488 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1492 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1493 * scheduler has enough safe guards.
1495 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1496 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1500 sock_valbool_flag(sk, SOCK_TXTIME, true);
1501 sk->sk_clockid = sk_txtime.clockid;
1502 sk->sk_txtime_deadline_mode =
1503 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1504 sk->sk_txtime_report_errors =
1505 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1508 case SO_BINDTOIFINDEX:
1509 ret = sock_bindtoindex_locked(sk, val);
1513 if (val & ~SOCK_BUF_LOCK_MASK) {
1517 sk->sk_userlocks = val | (sk->sk_userlocks &
1518 ~SOCK_BUF_LOCK_MASK);
1521 case SO_RESERVE_MEM:
1530 delta = val - sk->sk_reserved_mem;
1532 sock_release_reserved_memory(sk, -delta);
1534 ret = sock_reserve_memory(sk, delta);
1542 sockopt_release_sock(sk);
1546 int sock_setsockopt(struct socket *sock, int level, int optname,
1547 sockptr_t optval, unsigned int optlen)
1549 return sk_setsockopt(sock->sk, level, optname,
1552 EXPORT_SYMBOL(sock_setsockopt);
1554 static const struct cred *sk_get_peer_cred(struct sock *sk)
1556 const struct cred *cred;
1558 spin_lock(&sk->sk_peer_lock);
1559 cred = get_cred(sk->sk_peer_cred);
1560 spin_unlock(&sk->sk_peer_lock);
1565 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1566 struct ucred *ucred)
1568 ucred->pid = pid_vnr(pid);
1569 ucred->uid = ucred->gid = -1;
1571 struct user_namespace *current_ns = current_user_ns();
1573 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1574 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1578 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1580 struct user_namespace *user_ns = current_user_ns();
1583 for (i = 0; i < src->ngroups; i++) {
1584 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1586 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1593 int sk_getsockopt(struct sock *sk, int level, int optname,
1594 sockptr_t optval, sockptr_t optlen)
1596 struct socket *sock = sk->sk_socket;
1601 unsigned long ulval;
1603 struct old_timeval32 tm32;
1604 struct __kernel_old_timeval tm;
1605 struct __kernel_sock_timeval stm;
1606 struct sock_txtime txtime;
1607 struct so_timestamping timestamping;
1610 int lv = sizeof(int);
1613 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1618 memset(&v, 0, sizeof(v));
1622 v.val = sock_flag(sk, SOCK_DBG);
1626 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1630 v.val = sock_flag(sk, SOCK_BROADCAST);
1634 v.val = READ_ONCE(sk->sk_sndbuf);
1638 v.val = READ_ONCE(sk->sk_rcvbuf);
1642 v.val = sk->sk_reuse;
1646 v.val = sk->sk_reuseport;
1650 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1654 v.val = sk->sk_type;
1658 v.val = sk->sk_protocol;
1662 v.val = sk->sk_family;
1666 v.val = -sock_error(sk);
1668 v.val = xchg(&sk->sk_err_soft, 0);
1672 v.val = sock_flag(sk, SOCK_URGINLINE);
1676 v.val = sk->sk_no_check_tx;
1680 v.val = READ_ONCE(sk->sk_priority);
1684 lv = sizeof(v.ling);
1685 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1686 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1692 case SO_TIMESTAMP_OLD:
1693 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1694 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1695 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1698 case SO_TIMESTAMPNS_OLD:
1699 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1702 case SO_TIMESTAMP_NEW:
1703 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1706 case SO_TIMESTAMPNS_NEW:
1707 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1710 case SO_TIMESTAMPING_OLD:
1711 case SO_TIMESTAMPING_NEW:
1712 lv = sizeof(v.timestamping);
1713 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1714 * returning the flags when they were set through the same option.
1715 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1717 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1718 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1719 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1723 case SO_RCVTIMEO_OLD:
1724 case SO_RCVTIMEO_NEW:
1725 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1726 SO_RCVTIMEO_OLD == optname);
1729 case SO_SNDTIMEO_OLD:
1730 case SO_SNDTIMEO_NEW:
1731 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1732 SO_SNDTIMEO_OLD == optname);
1736 v.val = READ_ONCE(sk->sk_rcvlowat);
1744 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1748 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1753 struct ucred peercred;
1754 if (len > sizeof(peercred))
1755 len = sizeof(peercred);
1757 spin_lock(&sk->sk_peer_lock);
1758 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1759 spin_unlock(&sk->sk_peer_lock);
1761 if (copy_to_sockptr(optval, &peercred, len))
1768 struct pid *peer_pid;
1769 struct file *pidfd_file = NULL;
1772 if (len > sizeof(pidfd))
1773 len = sizeof(pidfd);
1775 spin_lock(&sk->sk_peer_lock);
1776 peer_pid = get_pid(sk->sk_peer_pid);
1777 spin_unlock(&sk->sk_peer_lock);
1782 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1787 if (copy_to_sockptr(optval, &pidfd, len) ||
1788 copy_to_sockptr(optlen, &len, sizeof(int))) {
1789 put_unused_fd(pidfd);
1795 fd_install(pidfd, pidfd_file);
1801 const struct cred *cred;
1804 cred = sk_get_peer_cred(sk);
1808 n = cred->group_info->ngroups;
1809 if (len < n * sizeof(gid_t)) {
1810 len = n * sizeof(gid_t);
1812 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1814 len = n * sizeof(gid_t);
1816 ret = groups_to_user(optval, cred->group_info);
1825 struct sockaddr_storage address;
1827 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1832 if (copy_to_sockptr(optval, &address, len))
1837 /* Dubious BSD thing... Probably nobody even uses it, but
1838 * the UNIX standard wants it for whatever reason... -DaveM
1841 v.val = sk->sk_state == TCP_LISTEN;
1845 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1849 return security_socket_getpeersec_stream(sock,
1850 optval, optlen, len);
1853 v.val = READ_ONCE(sk->sk_mark);
1857 v.val = sock_flag(sk, SOCK_RCVMARK);
1861 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1864 case SO_WIFI_STATUS:
1865 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1869 if (!READ_ONCE(sock->ops)->set_peek_off)
1872 v.val = READ_ONCE(sk->sk_peek_off);
1875 v.val = sock_flag(sk, SOCK_NOFCS);
1878 case SO_BINDTODEVICE:
1879 return sock_getbindtodevice(sk, optval, optlen, len);
1882 len = sk_get_filter(sk, optval, len);
1888 case SO_LOCK_FILTER:
1889 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1892 case SO_BPF_EXTENSIONS:
1893 v.val = bpf_tell_extensions();
1896 case SO_SELECT_ERR_QUEUE:
1897 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1900 #ifdef CONFIG_NET_RX_BUSY_POLL
1902 v.val = READ_ONCE(sk->sk_ll_usec);
1904 case SO_PREFER_BUSY_POLL:
1905 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1909 case SO_MAX_PACING_RATE:
1910 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1911 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1912 lv = sizeof(v.ulval);
1913 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1916 v.val = min_t(unsigned long, ~0U,
1917 READ_ONCE(sk->sk_max_pacing_rate));
1921 case SO_INCOMING_CPU:
1922 v.val = READ_ONCE(sk->sk_incoming_cpu);
1927 u32 meminfo[SK_MEMINFO_VARS];
1929 sk_get_meminfo(sk, meminfo);
1931 len = min_t(unsigned int, len, sizeof(meminfo));
1932 if (copy_to_sockptr(optval, &meminfo, len))
1938 #ifdef CONFIG_NET_RX_BUSY_POLL
1939 case SO_INCOMING_NAPI_ID:
1940 v.val = READ_ONCE(sk->sk_napi_id);
1942 /* aggregate non-NAPI IDs down to 0 */
1943 if (v.val < MIN_NAPI_ID)
1953 v.val64 = sock_gen_cookie(sk);
1957 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1961 lv = sizeof(v.txtime);
1962 v.txtime.clockid = sk->sk_clockid;
1963 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1964 SOF_TXTIME_DEADLINE_MODE : 0;
1965 v.txtime.flags |= sk->sk_txtime_report_errors ?
1966 SOF_TXTIME_REPORT_ERRORS : 0;
1969 case SO_BINDTOIFINDEX:
1970 v.val = READ_ONCE(sk->sk_bound_dev_if);
1973 case SO_NETNS_COOKIE:
1977 v.val64 = sock_net(sk)->net_cookie;
1981 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1984 case SO_RESERVE_MEM:
1985 v.val = READ_ONCE(sk->sk_reserved_mem);
1989 /* Paired with WRITE_ONCE() in sk_setsockopt() */
1990 v.val = READ_ONCE(sk->sk_txrehash);
1994 /* We implement the SO_SNDLOWAT etc to not be settable
1997 return -ENOPROTOOPT;
2002 if (copy_to_sockptr(optval, &v, len))
2005 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2011 * Initialize an sk_lock.
2013 * (We also register the sk_lock with the lock validator.)
2015 static inline void sock_lock_init(struct sock *sk)
2017 if (sk->sk_kern_sock)
2018 sock_lock_init_class_and_name(
2020 af_family_kern_slock_key_strings[sk->sk_family],
2021 af_family_kern_slock_keys + sk->sk_family,
2022 af_family_kern_key_strings[sk->sk_family],
2023 af_family_kern_keys + sk->sk_family);
2025 sock_lock_init_class_and_name(
2027 af_family_slock_key_strings[sk->sk_family],
2028 af_family_slock_keys + sk->sk_family,
2029 af_family_key_strings[sk->sk_family],
2030 af_family_keys + sk->sk_family);
2034 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2035 * even temporarly, because of RCU lookups. sk_node should also be left as is.
2036 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2038 static void sock_copy(struct sock *nsk, const struct sock *osk)
2040 const struct proto *prot = READ_ONCE(osk->sk_prot);
2041 #ifdef CONFIG_SECURITY_NETWORK
2042 void *sptr = nsk->sk_security;
2045 /* If we move sk_tx_queue_mapping out of the private section,
2046 * we must check if sk_tx_queue_clear() is called after
2047 * sock_copy() in sk_clone_lock().
2049 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2050 offsetof(struct sock, sk_dontcopy_begin) ||
2051 offsetof(struct sock, sk_tx_queue_mapping) >=
2052 offsetof(struct sock, sk_dontcopy_end));
2054 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2056 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2057 prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2058 /* alloc is larger than struct, see sk_prot_alloc() */);
2060 #ifdef CONFIG_SECURITY_NETWORK
2061 nsk->sk_security = sptr;
2062 security_sk_clone(osk, nsk);
2066 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2070 struct kmem_cache *slab;
2074 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2077 if (want_init_on_alloc(priority))
2078 sk_prot_clear_nulls(sk, prot->obj_size);
2080 sk = kmalloc(prot->obj_size, priority);
2083 if (security_sk_alloc(sk, family, priority))
2086 if (!try_module_get(prot->owner))
2093 security_sk_free(sk);
2096 kmem_cache_free(slab, sk);
2102 static void sk_prot_free(struct proto *prot, struct sock *sk)
2104 struct kmem_cache *slab;
2105 struct module *owner;
2107 owner = prot->owner;
2110 cgroup_sk_free(&sk->sk_cgrp_data);
2111 mem_cgroup_sk_free(sk);
2112 security_sk_free(sk);
2114 kmem_cache_free(slab, sk);
2121 * sk_alloc - All socket objects are allocated here
2122 * @net: the applicable net namespace
2123 * @family: protocol family
2124 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2125 * @prot: struct proto associated with this new sock instance
2126 * @kern: is this to be a kernel socket?
2128 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2129 struct proto *prot, int kern)
2133 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2135 sk->sk_family = family;
2137 * See comment in struct sock definition to understand
2138 * why we need sk_prot_creator -acme
2140 sk->sk_prot = sk->sk_prot_creator = prot;
2141 sk->sk_kern_sock = kern;
2143 sk->sk_net_refcnt = kern ? 0 : 1;
2144 if (likely(sk->sk_net_refcnt)) {
2145 get_net_track(net, &sk->ns_tracker, priority);
2146 sock_inuse_add(net, 1);
2148 __netns_tracker_alloc(net, &sk->ns_tracker,
2152 sock_net_set(sk, net);
2153 refcount_set(&sk->sk_wmem_alloc, 1);
2155 mem_cgroup_sk_alloc(sk);
2156 cgroup_sk_alloc(&sk->sk_cgrp_data);
2157 sock_update_classid(&sk->sk_cgrp_data);
2158 sock_update_netprioidx(&sk->sk_cgrp_data);
2159 sk_tx_queue_clear(sk);
2164 EXPORT_SYMBOL(sk_alloc);
2166 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2167 * grace period. This is the case for UDP sockets and TCP listeners.
2169 static void __sk_destruct(struct rcu_head *head)
2171 struct sock *sk = container_of(head, struct sock, sk_rcu);
2172 struct sk_filter *filter;
2174 if (sk->sk_destruct)
2175 sk->sk_destruct(sk);
2177 filter = rcu_dereference_check(sk->sk_filter,
2178 refcount_read(&sk->sk_wmem_alloc) == 0);
2180 sk_filter_uncharge(sk, filter);
2181 RCU_INIT_POINTER(sk->sk_filter, NULL);
2184 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2186 #ifdef CONFIG_BPF_SYSCALL
2187 bpf_sk_storage_free(sk);
2190 if (atomic_read(&sk->sk_omem_alloc))
2191 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2192 __func__, atomic_read(&sk->sk_omem_alloc));
2194 if (sk->sk_frag.page) {
2195 put_page(sk->sk_frag.page);
2196 sk->sk_frag.page = NULL;
2199 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2200 put_cred(sk->sk_peer_cred);
2201 put_pid(sk->sk_peer_pid);
2203 if (likely(sk->sk_net_refcnt))
2204 put_net_track(sock_net(sk), &sk->ns_tracker);
2206 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2208 sk_prot_free(sk->sk_prot_creator, sk);
2211 void sk_destruct(struct sock *sk)
2213 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2215 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2216 reuseport_detach_sock(sk);
2217 use_call_rcu = true;
2221 call_rcu(&sk->sk_rcu, __sk_destruct);
2223 __sk_destruct(&sk->sk_rcu);
2226 static void __sk_free(struct sock *sk)
2228 if (likely(sk->sk_net_refcnt))
2229 sock_inuse_add(sock_net(sk), -1);
2231 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2232 sock_diag_broadcast_destroy(sk);
2237 void sk_free(struct sock *sk)
2240 * We subtract one from sk_wmem_alloc and can know if
2241 * some packets are still in some tx queue.
2242 * If not null, sock_wfree() will call __sk_free(sk) later
2244 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2247 EXPORT_SYMBOL(sk_free);
2249 static void sk_init_common(struct sock *sk)
2251 skb_queue_head_init(&sk->sk_receive_queue);
2252 skb_queue_head_init(&sk->sk_write_queue);
2253 skb_queue_head_init(&sk->sk_error_queue);
2255 rwlock_init(&sk->sk_callback_lock);
2256 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2257 af_rlock_keys + sk->sk_family,
2258 af_family_rlock_key_strings[sk->sk_family]);
2259 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2260 af_wlock_keys + sk->sk_family,
2261 af_family_wlock_key_strings[sk->sk_family]);
2262 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2263 af_elock_keys + sk->sk_family,
2264 af_family_elock_key_strings[sk->sk_family]);
2265 lockdep_set_class_and_name(&sk->sk_callback_lock,
2266 af_callback_keys + sk->sk_family,
2267 af_family_clock_key_strings[sk->sk_family]);
2271 * sk_clone_lock - clone a socket, and lock its clone
2272 * @sk: the socket to clone
2273 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2275 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2277 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2279 struct proto *prot = READ_ONCE(sk->sk_prot);
2280 struct sk_filter *filter;
2281 bool is_charged = true;
2284 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2288 sock_copy(newsk, sk);
2290 newsk->sk_prot_creator = prot;
2293 if (likely(newsk->sk_net_refcnt)) {
2294 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2295 sock_inuse_add(sock_net(newsk), 1);
2297 /* Kernel sockets are not elevating the struct net refcount.
2298 * Instead, use a tracker to more easily detect if a layer
2299 * is not properly dismantling its kernel sockets at netns
2302 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2305 sk_node_init(&newsk->sk_node);
2306 sock_lock_init(newsk);
2307 bh_lock_sock(newsk);
2308 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2309 newsk->sk_backlog.len = 0;
2311 atomic_set(&newsk->sk_rmem_alloc, 0);
2313 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2314 refcount_set(&newsk->sk_wmem_alloc, 1);
2316 atomic_set(&newsk->sk_omem_alloc, 0);
2317 sk_init_common(newsk);
2319 newsk->sk_dst_cache = NULL;
2320 newsk->sk_dst_pending_confirm = 0;
2321 newsk->sk_wmem_queued = 0;
2322 newsk->sk_forward_alloc = 0;
2323 newsk->sk_reserved_mem = 0;
2324 atomic_set(&newsk->sk_drops, 0);
2325 newsk->sk_send_head = NULL;
2326 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2327 atomic_set(&newsk->sk_zckey, 0);
2329 sock_reset_flag(newsk, SOCK_DONE);
2331 /* sk->sk_memcg will be populated at accept() time */
2332 newsk->sk_memcg = NULL;
2334 cgroup_sk_clone(&newsk->sk_cgrp_data);
2337 filter = rcu_dereference(sk->sk_filter);
2339 /* though it's an empty new sock, the charging may fail
2340 * if sysctl_optmem_max was changed between creation of
2341 * original socket and cloning
2343 is_charged = sk_filter_charge(newsk, filter);
2344 RCU_INIT_POINTER(newsk->sk_filter, filter);
2347 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2348 /* We need to make sure that we don't uncharge the new
2349 * socket if we couldn't charge it in the first place
2350 * as otherwise we uncharge the parent's filter.
2353 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2354 sk_free_unlock_clone(newsk);
2358 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2360 if (bpf_sk_storage_clone(sk, newsk)) {
2361 sk_free_unlock_clone(newsk);
2366 /* Clear sk_user_data if parent had the pointer tagged
2367 * as not suitable for copying when cloning.
2369 if (sk_user_data_is_nocopy(newsk))
2370 newsk->sk_user_data = NULL;
2373 newsk->sk_err_soft = 0;
2374 newsk->sk_priority = 0;
2375 newsk->sk_incoming_cpu = raw_smp_processor_id();
2377 /* Before updating sk_refcnt, we must commit prior changes to memory
2378 * (Documentation/RCU/rculist_nulls.rst for details)
2381 refcount_set(&newsk->sk_refcnt, 2);
2383 sk_set_socket(newsk, NULL);
2384 sk_tx_queue_clear(newsk);
2385 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2387 if (newsk->sk_prot->sockets_allocated)
2388 sk_sockets_allocated_inc(newsk);
2390 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2391 net_enable_timestamp();
2395 EXPORT_SYMBOL_GPL(sk_clone_lock);
2397 void sk_free_unlock_clone(struct sock *sk)
2399 /* It is still raw copy of parent, so invalidate
2400 * destructor and make plain sk_free() */
2401 sk->sk_destruct = NULL;
2405 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2407 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2409 bool is_ipv6 = false;
2412 #if IS_ENABLED(CONFIG_IPV6)
2413 is_ipv6 = (sk->sk_family == AF_INET6 &&
2414 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2416 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2417 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2418 READ_ONCE(dst->dev->gso_ipv4_max_size);
2419 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2420 max_size = GSO_LEGACY_MAX_SIZE;
2422 return max_size - (MAX_TCP_HEADER + 1);
2425 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2429 sk->sk_route_caps = dst->dev->features;
2431 sk->sk_route_caps |= NETIF_F_GSO;
2432 if (sk->sk_route_caps & NETIF_F_GSO)
2433 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2434 if (unlikely(sk->sk_gso_disabled))
2435 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2436 if (sk_can_gso(sk)) {
2437 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2438 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2440 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2441 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2442 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2443 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2446 sk->sk_gso_max_segs = max_segs;
2447 sk_dst_set(sk, dst);
2449 EXPORT_SYMBOL_GPL(sk_setup_caps);
2452 * Simple resource managers for sockets.
2457 * Write buffer destructor automatically called from kfree_skb.
2459 void sock_wfree(struct sk_buff *skb)
2461 struct sock *sk = skb->sk;
2462 unsigned int len = skb->truesize;
2465 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2466 if (sock_flag(sk, SOCK_RCU_FREE) &&
2467 sk->sk_write_space == sock_def_write_space) {
2469 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2470 sock_def_write_space_wfree(sk);
2478 * Keep a reference on sk_wmem_alloc, this will be released
2479 * after sk_write_space() call
2481 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2482 sk->sk_write_space(sk);
2486 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2487 * could not do because of in-flight packets
2489 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2492 EXPORT_SYMBOL(sock_wfree);
2494 /* This variant of sock_wfree() is used by TCP,
2495 * since it sets SOCK_USE_WRITE_QUEUE.
2497 void __sock_wfree(struct sk_buff *skb)
2499 struct sock *sk = skb->sk;
2501 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2505 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2510 if (unlikely(!sk_fullsock(sk))) {
2511 skb->destructor = sock_edemux;
2516 skb->destructor = sock_wfree;
2517 skb_set_hash_from_sk(skb, sk);
2519 * We used to take a refcount on sk, but following operation
2520 * is enough to guarantee sk_free() wont free this sock until
2521 * all in-flight packets are completed
2523 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2525 EXPORT_SYMBOL(skb_set_owner_w);
2527 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2529 #ifdef CONFIG_TLS_DEVICE
2530 /* Drivers depend on in-order delivery for crypto offload,
2531 * partial orphan breaks out-of-order-OK logic.
2536 return (skb->destructor == sock_wfree ||
2537 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2540 /* This helper is used by netem, as it can hold packets in its
2541 * delay queue. We want to allow the owner socket to send more
2542 * packets, as if they were already TX completed by a typical driver.
2543 * But we also want to keep skb->sk set because some packet schedulers
2544 * rely on it (sch_fq for example).
2546 void skb_orphan_partial(struct sk_buff *skb)
2548 if (skb_is_tcp_pure_ack(skb))
2551 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2556 EXPORT_SYMBOL(skb_orphan_partial);
2559 * Read buffer destructor automatically called from kfree_skb.
2561 void sock_rfree(struct sk_buff *skb)
2563 struct sock *sk = skb->sk;
2564 unsigned int len = skb->truesize;
2566 atomic_sub(len, &sk->sk_rmem_alloc);
2567 sk_mem_uncharge(sk, len);
2569 EXPORT_SYMBOL(sock_rfree);
2572 * Buffer destructor for skbs that are not used directly in read or write
2573 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2575 void sock_efree(struct sk_buff *skb)
2579 EXPORT_SYMBOL(sock_efree);
2581 /* Buffer destructor for prefetch/receive path where reference count may
2582 * not be held, e.g. for listen sockets.
2585 void sock_pfree(struct sk_buff *skb)
2587 struct sock *sk = skb->sk;
2589 if (!sk_is_refcounted(sk))
2592 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2593 inet_reqsk(sk)->rsk_listener = NULL;
2594 reqsk_free(inet_reqsk(sk));
2600 EXPORT_SYMBOL(sock_pfree);
2601 #endif /* CONFIG_INET */
2603 kuid_t sock_i_uid(struct sock *sk)
2607 read_lock_bh(&sk->sk_callback_lock);
2608 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2609 read_unlock_bh(&sk->sk_callback_lock);
2612 EXPORT_SYMBOL(sock_i_uid);
2614 unsigned long __sock_i_ino(struct sock *sk)
2618 read_lock(&sk->sk_callback_lock);
2619 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2620 read_unlock(&sk->sk_callback_lock);
2623 EXPORT_SYMBOL(__sock_i_ino);
2625 unsigned long sock_i_ino(struct sock *sk)
2630 ino = __sock_i_ino(sk);
2634 EXPORT_SYMBOL(sock_i_ino);
2637 * Allocate a skb from the socket's send buffer.
2639 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2643 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2644 struct sk_buff *skb = alloc_skb(size, priority);
2647 skb_set_owner_w(skb, sk);
2653 EXPORT_SYMBOL(sock_wmalloc);
2655 static void sock_ofree(struct sk_buff *skb)
2657 struct sock *sk = skb->sk;
2659 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2662 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2665 struct sk_buff *skb;
2667 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2668 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2669 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2672 skb = alloc_skb(size, priority);
2676 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2678 skb->destructor = sock_ofree;
2683 * Allocate a memory block from the socket's option memory buffer.
2685 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2687 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2689 if ((unsigned int)size <= optmem_max &&
2690 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2692 /* First do the add, to avoid the race if kmalloc
2695 atomic_add(size, &sk->sk_omem_alloc);
2696 mem = kmalloc(size, priority);
2699 atomic_sub(size, &sk->sk_omem_alloc);
2703 EXPORT_SYMBOL(sock_kmalloc);
2705 /* Free an option memory block. Note, we actually want the inline
2706 * here as this allows gcc to detect the nullify and fold away the
2707 * condition entirely.
2709 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2712 if (WARN_ON_ONCE(!mem))
2715 kfree_sensitive(mem);
2718 atomic_sub(size, &sk->sk_omem_alloc);
2721 void sock_kfree_s(struct sock *sk, void *mem, int size)
2723 __sock_kfree_s(sk, mem, size, false);
2725 EXPORT_SYMBOL(sock_kfree_s);
2727 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2729 __sock_kfree_s(sk, mem, size, true);
2731 EXPORT_SYMBOL(sock_kzfree_s);
2733 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2734 I think, these locks should be removed for datagram sockets.
2736 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2740 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2744 if (signal_pending(current))
2746 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2747 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2748 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2750 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2752 if (READ_ONCE(sk->sk_err))
2754 timeo = schedule_timeout(timeo);
2756 finish_wait(sk_sleep(sk), &wait);
2762 * Generic send/receive buffer handlers
2765 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2766 unsigned long data_len, int noblock,
2767 int *errcode, int max_page_order)
2769 struct sk_buff *skb;
2773 timeo = sock_sndtimeo(sk, noblock);
2775 err = sock_error(sk);
2780 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2783 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2786 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2787 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2791 if (signal_pending(current))
2793 timeo = sock_wait_for_wmem(sk, timeo);
2795 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2796 errcode, sk->sk_allocation);
2798 skb_set_owner_w(skb, sk);
2802 err = sock_intr_errno(timeo);
2807 EXPORT_SYMBOL(sock_alloc_send_pskb);
2809 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2810 struct sockcm_cookie *sockc)
2814 switch (cmsg->cmsg_type) {
2816 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2817 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2819 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2821 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2823 case SO_TIMESTAMPING_OLD:
2824 case SO_TIMESTAMPING_NEW:
2825 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2828 tsflags = *(u32 *)CMSG_DATA(cmsg);
2829 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2832 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2833 sockc->tsflags |= tsflags;
2836 if (!sock_flag(sk, SOCK_TXTIME))
2838 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2840 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2842 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2844 case SCM_CREDENTIALS:
2851 EXPORT_SYMBOL(__sock_cmsg_send);
2853 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2854 struct sockcm_cookie *sockc)
2856 struct cmsghdr *cmsg;
2859 for_each_cmsghdr(cmsg, msg) {
2860 if (!CMSG_OK(msg, cmsg))
2862 if (cmsg->cmsg_level != SOL_SOCKET)
2864 ret = __sock_cmsg_send(sk, cmsg, sockc);
2870 EXPORT_SYMBOL(sock_cmsg_send);
2872 static void sk_enter_memory_pressure(struct sock *sk)
2874 if (!sk->sk_prot->enter_memory_pressure)
2877 sk->sk_prot->enter_memory_pressure(sk);
2880 static void sk_leave_memory_pressure(struct sock *sk)
2882 if (sk->sk_prot->leave_memory_pressure) {
2883 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2884 tcp_leave_memory_pressure, sk);
2886 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2888 if (memory_pressure && READ_ONCE(*memory_pressure))
2889 WRITE_ONCE(*memory_pressure, 0);
2893 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2896 * skb_page_frag_refill - check that a page_frag contains enough room
2897 * @sz: minimum size of the fragment we want to get
2898 * @pfrag: pointer to page_frag
2899 * @gfp: priority for memory allocation
2901 * Note: While this allocator tries to use high order pages, there is
2902 * no guarantee that allocations succeed. Therefore, @sz MUST be
2903 * less or equal than PAGE_SIZE.
2905 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2908 if (page_ref_count(pfrag->page) == 1) {
2912 if (pfrag->offset + sz <= pfrag->size)
2914 put_page(pfrag->page);
2918 if (SKB_FRAG_PAGE_ORDER &&
2919 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2920 /* Avoid direct reclaim but allow kswapd to wake */
2921 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2922 __GFP_COMP | __GFP_NOWARN |
2924 SKB_FRAG_PAGE_ORDER);
2925 if (likely(pfrag->page)) {
2926 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2930 pfrag->page = alloc_page(gfp);
2931 if (likely(pfrag->page)) {
2932 pfrag->size = PAGE_SIZE;
2937 EXPORT_SYMBOL(skb_page_frag_refill);
2939 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2941 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2944 sk_enter_memory_pressure(sk);
2945 sk_stream_moderate_sndbuf(sk);
2948 EXPORT_SYMBOL(sk_page_frag_refill);
2950 void __lock_sock(struct sock *sk)
2951 __releases(&sk->sk_lock.slock)
2952 __acquires(&sk->sk_lock.slock)
2957 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2958 TASK_UNINTERRUPTIBLE);
2959 spin_unlock_bh(&sk->sk_lock.slock);
2961 spin_lock_bh(&sk->sk_lock.slock);
2962 if (!sock_owned_by_user(sk))
2965 finish_wait(&sk->sk_lock.wq, &wait);
2968 void __release_sock(struct sock *sk)
2969 __releases(&sk->sk_lock.slock)
2970 __acquires(&sk->sk_lock.slock)
2972 struct sk_buff *skb, *next;
2974 while ((skb = sk->sk_backlog.head) != NULL) {
2975 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2977 spin_unlock_bh(&sk->sk_lock.slock);
2982 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2983 skb_mark_not_on_list(skb);
2984 sk_backlog_rcv(sk, skb);
2989 } while (skb != NULL);
2991 spin_lock_bh(&sk->sk_lock.slock);
2995 * Doing the zeroing here guarantee we can not loop forever
2996 * while a wild producer attempts to flood us.
2998 sk->sk_backlog.len = 0;
3001 void __sk_flush_backlog(struct sock *sk)
3003 spin_lock_bh(&sk->sk_lock.slock);
3006 if (sk->sk_prot->release_cb)
3007 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3008 tcp_release_cb, sk);
3010 spin_unlock_bh(&sk->sk_lock.slock);
3012 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3015 * sk_wait_data - wait for data to arrive at sk_receive_queue
3016 * @sk: sock to wait on
3017 * @timeo: for how long
3018 * @skb: last skb seen on sk_receive_queue
3020 * Now socket state including sk->sk_err is changed only under lock,
3021 * hence we may omit checks after joining wait queue.
3022 * We check receive queue before schedule() only as optimization;
3023 * it is very likely that release_sock() added new data.
3025 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3027 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3030 add_wait_queue(sk_sleep(sk), &wait);
3031 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3032 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3033 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3034 remove_wait_queue(sk_sleep(sk), &wait);
3037 EXPORT_SYMBOL(sk_wait_data);
3040 * __sk_mem_raise_allocated - increase memory_allocated
3042 * @size: memory size to allocate
3043 * @amt: pages to allocate
3044 * @kind: allocation type
3046 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3048 * Unlike the globally shared limits among the sockets under same protocol,
3049 * consuming the budget of a memcg won't have direct effect on other ones.
3050 * So be optimistic about memcg's tolerance, and leave the callers to decide
3051 * whether or not to raise allocated through sk_under_memory_pressure() or
3054 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3056 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3057 struct proto *prot = sk->sk_prot;
3058 bool charged = false;
3061 sk_memory_allocated_add(sk, amt);
3062 allocated = sk_memory_allocated(sk);
3065 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3066 goto suppress_allocation;
3071 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3072 sk_leave_memory_pressure(sk);
3076 /* Under pressure. */
3077 if (allocated > sk_prot_mem_limits(sk, 1))
3078 sk_enter_memory_pressure(sk);
3080 /* Over hard limit. */
3081 if (allocated > sk_prot_mem_limits(sk, 2))
3082 goto suppress_allocation;
3084 /* Guarantee minimum buffer size under pressure (either global
3085 * or memcg) to make sure features described in RFC 7323 (TCP
3086 * Extensions for High Performance) work properly.
3088 * This rule does NOT stand when exceeds global or memcg's hard
3089 * limit, or else a DoS attack can be taken place by spawning
3090 * lots of sockets whose usage are under minimum buffer size.
3092 if (kind == SK_MEM_RECV) {
3093 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3096 } else { /* SK_MEM_SEND */
3097 int wmem0 = sk_get_wmem0(sk, prot);
3099 if (sk->sk_type == SOCK_STREAM) {
3100 if (sk->sk_wmem_queued < wmem0)
3102 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3107 if (sk_has_memory_pressure(sk)) {
3110 /* The following 'average' heuristic is within the
3111 * scope of global accounting, so it only makes
3112 * sense for global memory pressure.
3114 if (!sk_under_global_memory_pressure(sk))
3117 /* Try to be fair among all the sockets under global
3118 * pressure by allowing the ones that below average
3121 alloc = sk_sockets_allocated_read_positive(sk);
3122 if (sk_prot_mem_limits(sk, 2) > alloc *
3123 sk_mem_pages(sk->sk_wmem_queued +
3124 atomic_read(&sk->sk_rmem_alloc) +
3125 sk->sk_forward_alloc))
3129 suppress_allocation:
3131 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3132 sk_stream_moderate_sndbuf(sk);
3134 /* Fail only if socket is _under_ its sndbuf.
3135 * In this case we cannot block, so that we have to fail.
3137 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3138 /* Force charge with __GFP_NOFAIL */
3139 if (memcg && !charged) {
3140 mem_cgroup_charge_skmem(memcg, amt,
3141 gfp_memcg_charge() | __GFP_NOFAIL);
3147 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3148 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3150 sk_memory_allocated_sub(sk, amt);
3153 mem_cgroup_uncharge_skmem(memcg, amt);
3159 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3161 * @size: memory size to allocate
3162 * @kind: allocation type
3164 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3165 * rmem allocation. This function assumes that protocols which have
3166 * memory_pressure use sk_wmem_queued as write buffer accounting.
3168 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3170 int ret, amt = sk_mem_pages(size);
3172 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3173 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3175 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3178 EXPORT_SYMBOL(__sk_mem_schedule);
3181 * __sk_mem_reduce_allocated - reclaim memory_allocated
3183 * @amount: number of quanta
3185 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3187 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3189 sk_memory_allocated_sub(sk, amount);
3191 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3192 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3194 if (sk_under_global_memory_pressure(sk) &&
3195 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3196 sk_leave_memory_pressure(sk);
3200 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3202 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3204 void __sk_mem_reclaim(struct sock *sk, int amount)
3206 amount >>= PAGE_SHIFT;
3207 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3208 __sk_mem_reduce_allocated(sk, amount);
3210 EXPORT_SYMBOL(__sk_mem_reclaim);
3212 int sk_set_peek_off(struct sock *sk, int val)
3214 WRITE_ONCE(sk->sk_peek_off, val);
3217 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3220 * Set of default routines for initialising struct proto_ops when
3221 * the protocol does not support a particular function. In certain
3222 * cases where it makes no sense for a protocol to have a "do nothing"
3223 * function, some default processing is provided.
3226 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3230 EXPORT_SYMBOL(sock_no_bind);
3232 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3237 EXPORT_SYMBOL(sock_no_connect);
3239 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3243 EXPORT_SYMBOL(sock_no_socketpair);
3245 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3250 EXPORT_SYMBOL(sock_no_accept);
3252 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3257 EXPORT_SYMBOL(sock_no_getname);
3259 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3263 EXPORT_SYMBOL(sock_no_ioctl);
3265 int sock_no_listen(struct socket *sock, int backlog)
3269 EXPORT_SYMBOL(sock_no_listen);
3271 int sock_no_shutdown(struct socket *sock, int how)
3275 EXPORT_SYMBOL(sock_no_shutdown);
3277 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3281 EXPORT_SYMBOL(sock_no_sendmsg);
3283 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3287 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3289 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3294 EXPORT_SYMBOL(sock_no_recvmsg);
3296 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3298 /* Mirror missing mmap method error code */
3301 EXPORT_SYMBOL(sock_no_mmap);
3304 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3305 * various sock-based usage counts.
3307 void __receive_sock(struct file *file)
3309 struct socket *sock;
3311 sock = sock_from_file(file);
3313 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3314 sock_update_classid(&sock->sk->sk_cgrp_data);
3319 * Default Socket Callbacks
3322 static void sock_def_wakeup(struct sock *sk)
3324 struct socket_wq *wq;
3327 wq = rcu_dereference(sk->sk_wq);
3328 if (skwq_has_sleeper(wq))
3329 wake_up_interruptible_all(&wq->wait);
3333 static void sock_def_error_report(struct sock *sk)
3335 struct socket_wq *wq;
3338 wq = rcu_dereference(sk->sk_wq);
3339 if (skwq_has_sleeper(wq))
3340 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3341 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3345 void sock_def_readable(struct sock *sk)
3347 struct socket_wq *wq;
3349 trace_sk_data_ready(sk);
3352 wq = rcu_dereference(sk->sk_wq);
3353 if (skwq_has_sleeper(wq))
3354 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3355 EPOLLRDNORM | EPOLLRDBAND);
3356 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3360 static void sock_def_write_space(struct sock *sk)
3362 struct socket_wq *wq;
3366 /* Do not wake up a writer until he can make "significant"
3369 if (sock_writeable(sk)) {
3370 wq = rcu_dereference(sk->sk_wq);
3371 if (skwq_has_sleeper(wq))
3372 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3373 EPOLLWRNORM | EPOLLWRBAND);
3375 /* Should agree with poll, otherwise some programs break */
3376 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3382 /* An optimised version of sock_def_write_space(), should only be called
3383 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3386 static void sock_def_write_space_wfree(struct sock *sk)
3388 /* Do not wake up a writer until he can make "significant"
3391 if (sock_writeable(sk)) {
3392 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3394 /* rely on refcount_sub from sock_wfree() */
3395 smp_mb__after_atomic();
3396 if (wq && waitqueue_active(&wq->wait))
3397 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3398 EPOLLWRNORM | EPOLLWRBAND);
3400 /* Should agree with poll, otherwise some programs break */
3401 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3405 static void sock_def_destruct(struct sock *sk)
3409 void sk_send_sigurg(struct sock *sk)
3411 if (sk->sk_socket && sk->sk_socket->file)
3412 if (send_sigurg(&sk->sk_socket->file->f_owner))
3413 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3415 EXPORT_SYMBOL(sk_send_sigurg);
3417 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3418 unsigned long expires)
3420 if (!mod_timer(timer, expires))
3423 EXPORT_SYMBOL(sk_reset_timer);
3425 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3427 if (del_timer(timer))
3430 EXPORT_SYMBOL(sk_stop_timer);
3432 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3434 if (del_timer_sync(timer))
3437 EXPORT_SYMBOL(sk_stop_timer_sync);
3439 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3442 sk->sk_send_head = NULL;
3444 timer_setup(&sk->sk_timer, NULL, 0);
3446 sk->sk_allocation = GFP_KERNEL;
3447 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3448 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3449 sk->sk_state = TCP_CLOSE;
3450 sk->sk_use_task_frag = true;
3451 sk_set_socket(sk, sock);
3453 sock_set_flag(sk, SOCK_ZAPPED);
3456 sk->sk_type = sock->type;
3457 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3460 RCU_INIT_POINTER(sk->sk_wq, NULL);
3464 rwlock_init(&sk->sk_callback_lock);
3465 if (sk->sk_kern_sock)
3466 lockdep_set_class_and_name(
3467 &sk->sk_callback_lock,
3468 af_kern_callback_keys + sk->sk_family,
3469 af_family_kern_clock_key_strings[sk->sk_family]);
3471 lockdep_set_class_and_name(
3472 &sk->sk_callback_lock,
3473 af_callback_keys + sk->sk_family,
3474 af_family_clock_key_strings[sk->sk_family]);
3476 sk->sk_state_change = sock_def_wakeup;
3477 sk->sk_data_ready = sock_def_readable;
3478 sk->sk_write_space = sock_def_write_space;
3479 sk->sk_error_report = sock_def_error_report;
3480 sk->sk_destruct = sock_def_destruct;
3482 sk->sk_frag.page = NULL;
3483 sk->sk_frag.offset = 0;
3484 sk->sk_peek_off = -1;
3486 sk->sk_peer_pid = NULL;
3487 sk->sk_peer_cred = NULL;
3488 spin_lock_init(&sk->sk_peer_lock);
3490 sk->sk_write_pending = 0;
3491 sk->sk_rcvlowat = 1;
3492 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3493 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3495 sk->sk_stamp = SK_DEFAULT_STAMP;
3496 #if BITS_PER_LONG==32
3497 seqlock_init(&sk->sk_stamp_seq);
3499 atomic_set(&sk->sk_zckey, 0);
3501 #ifdef CONFIG_NET_RX_BUSY_POLL
3503 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3506 sk->sk_max_pacing_rate = ~0UL;
3507 sk->sk_pacing_rate = ~0UL;
3508 WRITE_ONCE(sk->sk_pacing_shift, 10);
3509 sk->sk_incoming_cpu = -1;
3511 sk_rx_queue_clear(sk);
3513 * Before updating sk_refcnt, we must commit prior changes to memory
3514 * (Documentation/RCU/rculist_nulls.rst for details)
3517 refcount_set(&sk->sk_refcnt, 1);
3518 atomic_set(&sk->sk_drops, 0);
3520 EXPORT_SYMBOL(sock_init_data_uid);
3522 void sock_init_data(struct socket *sock, struct sock *sk)
3525 SOCK_INODE(sock)->i_uid :
3526 make_kuid(sock_net(sk)->user_ns, 0);
3528 sock_init_data_uid(sock, sk, uid);
3530 EXPORT_SYMBOL(sock_init_data);
3532 void lock_sock_nested(struct sock *sk, int subclass)
3534 /* The sk_lock has mutex_lock() semantics here. */
3535 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3538 spin_lock_bh(&sk->sk_lock.slock);
3539 if (sock_owned_by_user_nocheck(sk))
3541 sk->sk_lock.owned = 1;
3542 spin_unlock_bh(&sk->sk_lock.slock);
3544 EXPORT_SYMBOL(lock_sock_nested);
3546 void release_sock(struct sock *sk)
3548 spin_lock_bh(&sk->sk_lock.slock);
3549 if (sk->sk_backlog.tail)
3552 if (sk->sk_prot->release_cb)
3553 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3554 tcp_release_cb, sk);
3556 sock_release_ownership(sk);
3557 if (waitqueue_active(&sk->sk_lock.wq))
3558 wake_up(&sk->sk_lock.wq);
3559 spin_unlock_bh(&sk->sk_lock.slock);
3561 EXPORT_SYMBOL(release_sock);
3563 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3566 spin_lock_bh(&sk->sk_lock.slock);
3568 if (!sock_owned_by_user_nocheck(sk)) {
3570 * Fast path return with bottom halves disabled and
3571 * sock::sk_lock.slock held.
3573 * The 'mutex' is not contended and holding
3574 * sock::sk_lock.slock prevents all other lockers to
3575 * proceed so the corresponding unlock_sock_fast() can
3576 * avoid the slow path of release_sock() completely and
3577 * just release slock.
3579 * From a semantical POV this is equivalent to 'acquiring'
3580 * the 'mutex', hence the corresponding lockdep
3581 * mutex_release() has to happen in the fast path of
3582 * unlock_sock_fast().
3588 sk->sk_lock.owned = 1;
3589 __acquire(&sk->sk_lock.slock);
3590 spin_unlock_bh(&sk->sk_lock.slock);
3593 EXPORT_SYMBOL(__lock_sock_fast);
3595 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3596 bool timeval, bool time32)
3598 struct sock *sk = sock->sk;
3599 struct timespec64 ts;
3601 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3602 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3603 if (ts.tv_sec == -1)
3605 if (ts.tv_sec == 0) {
3606 ktime_t kt = ktime_get_real();
3607 sock_write_timestamp(sk, kt);
3608 ts = ktime_to_timespec64(kt);
3614 #ifdef CONFIG_COMPAT_32BIT_TIME
3616 return put_old_timespec32(&ts, userstamp);
3618 #ifdef CONFIG_SPARC64
3619 /* beware of padding in sparc64 timeval */
3620 if (timeval && !in_compat_syscall()) {
3621 struct __kernel_old_timeval __user tv = {
3622 .tv_sec = ts.tv_sec,
3623 .tv_usec = ts.tv_nsec,
3625 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3630 return put_timespec64(&ts, userstamp);
3632 EXPORT_SYMBOL(sock_gettstamp);
3634 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3636 if (!sock_flag(sk, flag)) {
3637 unsigned long previous_flags = sk->sk_flags;
3639 sock_set_flag(sk, flag);
3641 * we just set one of the two flags which require net
3642 * time stamping, but time stamping might have been on
3643 * already because of the other one
3645 if (sock_needs_netstamp(sk) &&
3646 !(previous_flags & SK_FLAGS_TIMESTAMP))
3647 net_enable_timestamp();
3651 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3652 int level, int type)
3654 struct sock_exterr_skb *serr;
3655 struct sk_buff *skb;
3659 skb = sock_dequeue_err_skb(sk);
3665 msg->msg_flags |= MSG_TRUNC;
3668 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3672 sock_recv_timestamp(msg, sk, skb);
3674 serr = SKB_EXT_ERR(skb);
3675 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3677 msg->msg_flags |= MSG_ERRQUEUE;
3685 EXPORT_SYMBOL(sock_recv_errqueue);
3688 * Get a socket option on an socket.
3690 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3691 * asynchronous errors should be reported by getsockopt. We assume
3692 * this means if you specify SO_ERROR (otherwise whats the point of it).
3694 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3695 char __user *optval, int __user *optlen)
3697 struct sock *sk = sock->sk;
3699 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3700 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3702 EXPORT_SYMBOL(sock_common_getsockopt);
3704 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3707 struct sock *sk = sock->sk;
3711 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3713 msg->msg_namelen = addr_len;
3716 EXPORT_SYMBOL(sock_common_recvmsg);
3719 * Set socket options on an inet socket.
3721 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3722 sockptr_t optval, unsigned int optlen)
3724 struct sock *sk = sock->sk;
3726 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3727 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3729 EXPORT_SYMBOL(sock_common_setsockopt);
3731 void sk_common_release(struct sock *sk)
3733 if (sk->sk_prot->destroy)
3734 sk->sk_prot->destroy(sk);
3737 * Observation: when sk_common_release is called, processes have
3738 * no access to socket. But net still has.
3739 * Step one, detach it from networking:
3741 * A. Remove from hash tables.
3744 sk->sk_prot->unhash(sk);
3747 * In this point socket cannot receive new packets, but it is possible
3748 * that some packets are in flight because some CPU runs receiver and
3749 * did hash table lookup before we unhashed socket. They will achieve
3750 * receive queue and will be purged by socket destructor.
3752 * Also we still have packets pending on receive queue and probably,
3753 * our own packets waiting in device queues. sock_destroy will drain
3754 * receive queue, but transmitted packets will delay socket destruction
3755 * until the last reference will be released.
3760 xfrm_sk_free_policy(sk);
3764 EXPORT_SYMBOL(sk_common_release);
3766 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3768 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3770 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3771 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3772 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3773 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3774 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3775 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3776 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3777 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3778 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3781 #ifdef CONFIG_PROC_FS
3782 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3784 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3786 int cpu, idx = prot->inuse_idx;
3789 for_each_possible_cpu(cpu)
3790 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3792 return res >= 0 ? res : 0;
3794 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3796 int sock_inuse_get(struct net *net)
3800 for_each_possible_cpu(cpu)
3801 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3806 EXPORT_SYMBOL_GPL(sock_inuse_get);
3808 static int __net_init sock_inuse_init_net(struct net *net)
3810 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3811 if (net->core.prot_inuse == NULL)
3816 static void __net_exit sock_inuse_exit_net(struct net *net)
3818 free_percpu(net->core.prot_inuse);
3821 static struct pernet_operations net_inuse_ops = {
3822 .init = sock_inuse_init_net,
3823 .exit = sock_inuse_exit_net,
3826 static __init int net_inuse_init(void)
3828 if (register_pernet_subsys(&net_inuse_ops))
3829 panic("Cannot initialize net inuse counters");
3834 core_initcall(net_inuse_init);
3836 static int assign_proto_idx(struct proto *prot)
3838 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3840 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3841 pr_err("PROTO_INUSE_NR exhausted\n");
3845 set_bit(prot->inuse_idx, proto_inuse_idx);
3849 static void release_proto_idx(struct proto *prot)
3851 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3852 clear_bit(prot->inuse_idx, proto_inuse_idx);
3855 static inline int assign_proto_idx(struct proto *prot)
3860 static inline void release_proto_idx(struct proto *prot)
3866 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3870 kfree(twsk_prot->twsk_slab_name);
3871 twsk_prot->twsk_slab_name = NULL;
3872 kmem_cache_destroy(twsk_prot->twsk_slab);
3873 twsk_prot->twsk_slab = NULL;
3876 static int tw_prot_init(const struct proto *prot)
3878 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3883 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3885 if (!twsk_prot->twsk_slab_name)
3888 twsk_prot->twsk_slab =
3889 kmem_cache_create(twsk_prot->twsk_slab_name,
3890 twsk_prot->twsk_obj_size, 0,
3891 SLAB_ACCOUNT | prot->slab_flags,
3893 if (!twsk_prot->twsk_slab) {
3894 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3902 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3906 kfree(rsk_prot->slab_name);
3907 rsk_prot->slab_name = NULL;
3908 kmem_cache_destroy(rsk_prot->slab);
3909 rsk_prot->slab = NULL;
3912 static int req_prot_init(const struct proto *prot)
3914 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3919 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3921 if (!rsk_prot->slab_name)
3924 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3925 rsk_prot->obj_size, 0,
3926 SLAB_ACCOUNT | prot->slab_flags,
3929 if (!rsk_prot->slab) {
3930 pr_crit("%s: Can't create request sock SLAB cache!\n",
3937 int proto_register(struct proto *prot, int alloc_slab)
3941 if (prot->memory_allocated && !prot->sysctl_mem) {
3942 pr_err("%s: missing sysctl_mem\n", prot->name);
3945 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3946 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3950 prot->slab = kmem_cache_create_usercopy(prot->name,
3952 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3954 prot->useroffset, prot->usersize,
3957 if (prot->slab == NULL) {
3958 pr_crit("%s: Can't create sock SLAB cache!\n",
3963 if (req_prot_init(prot))
3964 goto out_free_request_sock_slab;
3966 if (tw_prot_init(prot))
3967 goto out_free_timewait_sock_slab;
3970 mutex_lock(&proto_list_mutex);
3971 ret = assign_proto_idx(prot);
3973 mutex_unlock(&proto_list_mutex);
3974 goto out_free_timewait_sock_slab;
3976 list_add(&prot->node, &proto_list);
3977 mutex_unlock(&proto_list_mutex);
3980 out_free_timewait_sock_slab:
3982 tw_prot_cleanup(prot->twsk_prot);
3983 out_free_request_sock_slab:
3985 req_prot_cleanup(prot->rsk_prot);
3987 kmem_cache_destroy(prot->slab);
3993 EXPORT_SYMBOL(proto_register);
3995 void proto_unregister(struct proto *prot)
3997 mutex_lock(&proto_list_mutex);
3998 release_proto_idx(prot);
3999 list_del(&prot->node);
4000 mutex_unlock(&proto_list_mutex);
4002 kmem_cache_destroy(prot->slab);
4005 req_prot_cleanup(prot->rsk_prot);
4006 tw_prot_cleanup(prot->twsk_prot);
4008 EXPORT_SYMBOL(proto_unregister);
4010 int sock_load_diag_module(int family, int protocol)
4013 if (!sock_is_registered(family))
4016 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4017 NETLINK_SOCK_DIAG, family);
4021 if (family == AF_INET &&
4022 protocol != IPPROTO_RAW &&
4023 protocol < MAX_INET_PROTOS &&
4024 !rcu_access_pointer(inet_protos[protocol]))
4028 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4029 NETLINK_SOCK_DIAG, family, protocol);
4031 EXPORT_SYMBOL(sock_load_diag_module);
4033 #ifdef CONFIG_PROC_FS
4034 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4035 __acquires(proto_list_mutex)
4037 mutex_lock(&proto_list_mutex);
4038 return seq_list_start_head(&proto_list, *pos);
4041 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4043 return seq_list_next(v, &proto_list, pos);
4046 static void proto_seq_stop(struct seq_file *seq, void *v)
4047 __releases(proto_list_mutex)
4049 mutex_unlock(&proto_list_mutex);
4052 static char proto_method_implemented(const void *method)
4054 return method == NULL ? 'n' : 'y';
4056 static long sock_prot_memory_allocated(struct proto *proto)
4058 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4061 static const char *sock_prot_memory_pressure(struct proto *proto)
4063 return proto->memory_pressure != NULL ?
4064 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4067 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4070 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4071 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4074 sock_prot_inuse_get(seq_file_net(seq), proto),
4075 sock_prot_memory_allocated(proto),
4076 sock_prot_memory_pressure(proto),
4078 proto->slab == NULL ? "no" : "yes",
4079 module_name(proto->owner),
4080 proto_method_implemented(proto->close),
4081 proto_method_implemented(proto->connect),
4082 proto_method_implemented(proto->disconnect),
4083 proto_method_implemented(proto->accept),
4084 proto_method_implemented(proto->ioctl),
4085 proto_method_implemented(proto->init),
4086 proto_method_implemented(proto->destroy),
4087 proto_method_implemented(proto->shutdown),
4088 proto_method_implemented(proto->setsockopt),
4089 proto_method_implemented(proto->getsockopt),
4090 proto_method_implemented(proto->sendmsg),
4091 proto_method_implemented(proto->recvmsg),
4092 proto_method_implemented(proto->bind),
4093 proto_method_implemented(proto->backlog_rcv),
4094 proto_method_implemented(proto->hash),
4095 proto_method_implemented(proto->unhash),
4096 proto_method_implemented(proto->get_port),
4097 proto_method_implemented(proto->enter_memory_pressure));
4100 static int proto_seq_show(struct seq_file *seq, void *v)
4102 if (v == &proto_list)
4103 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4112 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4114 proto_seq_printf(seq, list_entry(v, struct proto, node));
4118 static const struct seq_operations proto_seq_ops = {
4119 .start = proto_seq_start,
4120 .next = proto_seq_next,
4121 .stop = proto_seq_stop,
4122 .show = proto_seq_show,
4125 static __net_init int proto_init_net(struct net *net)
4127 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4128 sizeof(struct seq_net_private)))
4134 static __net_exit void proto_exit_net(struct net *net)
4136 remove_proc_entry("protocols", net->proc_net);
4140 static __net_initdata struct pernet_operations proto_net_ops = {
4141 .init = proto_init_net,
4142 .exit = proto_exit_net,
4145 static int __init proto_init(void)
4147 return register_pernet_subsys(&proto_net_ops);
4150 subsys_initcall(proto_init);
4152 #endif /* PROC_FS */
4154 #ifdef CONFIG_NET_RX_BUSY_POLL
4155 bool sk_busy_loop_end(void *p, unsigned long start_time)
4157 struct sock *sk = p;
4159 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4162 if (sk_is_udp(sk) &&
4163 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4166 return sk_busy_loop_timeout(sk, start_time);
4168 EXPORT_SYMBOL(sk_busy_loop_end);
4169 #endif /* CONFIG_NET_RX_BUSY_POLL */
4171 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4173 if (!sk->sk_prot->bind_add)
4175 return sk->sk_prot->bind_add(sk, addr, addr_len);
4177 EXPORT_SYMBOL(sock_bind_add);
4179 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4180 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4181 void __user *arg, void *karg, size_t size)
4185 if (copy_from_user(karg, arg, size))
4188 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4192 if (copy_to_user(arg, karg, size))
4197 EXPORT_SYMBOL(sock_ioctl_inout);
4199 /* This is the most common ioctl prep function, where the result (4 bytes) is
4200 * copied back to userspace if the ioctl() returns successfully. No input is
4201 * copied from userspace as input argument.
4203 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4207 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4211 return put_user(karg, (int __user *)arg);
4214 /* A wrapper around sock ioctls, which copies the data from userspace
4215 * (depending on the protocol/ioctl), and copies back the result to userspace.
4216 * The main motivation for this function is to pass kernel memory to the
4217 * protocol ioctl callbacks, instead of userspace memory.
4219 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4223 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4224 rc = ipmr_sk_ioctl(sk, cmd, arg);
4225 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4226 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4227 else if (sk_is_phonet(sk))
4228 rc = phonet_sk_ioctl(sk, cmd, arg);
4230 /* If ioctl was processed, returns its value */
4234 /* Otherwise call the default handler */
4235 return sock_ioctl_out(sk, cmd, arg);
4237 EXPORT_SYMBOL(sk_ioctl);
4239 static int __init sock_struct_check(void)
4241 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4242 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4243 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4244 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4245 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4247 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4248 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4249 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4250 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4251 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4252 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4253 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4254 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4255 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4257 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4258 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4259 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4261 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4262 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4263 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4264 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4266 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4267 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4268 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf);
4269 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4270 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4271 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4272 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4273 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4274 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4275 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm);
4276 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status);
4277 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4278 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4279 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4280 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4281 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4283 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4284 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4285 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4286 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4287 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4288 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4289 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4290 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4291 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4292 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4293 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4294 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4295 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4299 core_initcall(sock_struct_check);