1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
117 #include <linux/mroute.h>
118 #include <linux/mroute6.h>
119 #include <linux/icmpv6.h>
121 #include <linux/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
138 #include <net/bpf_sk_storage.h>
140 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
144 #include <net/phonet/phonet.h>
146 #include <linux/ethtool.h>
150 static DEFINE_MUTEX(proto_list_mutex);
151 static LIST_HEAD(proto_list);
153 static void sock_def_write_space_wfree(struct sock *sk);
154 static void sock_def_write_space(struct sock *sk);
157 * sk_ns_capable - General socket capability test
158 * @sk: Socket to use a capability on or through
159 * @user_ns: The user namespace of the capability to use
160 * @cap: The capability to use
162 * Test to see if the opener of the socket had when the socket was
163 * created and the current process has the capability @cap in the user
164 * namespace @user_ns.
166 bool sk_ns_capable(const struct sock *sk,
167 struct user_namespace *user_ns, int cap)
169 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
170 ns_capable(user_ns, cap);
172 EXPORT_SYMBOL(sk_ns_capable);
175 * sk_capable - Socket global capability test
176 * @sk: Socket to use a capability on or through
177 * @cap: The global capability to use
179 * Test to see if the opener of the socket had when the socket was
180 * created and the current process has the capability @cap in all user
183 bool sk_capable(const struct sock *sk, int cap)
185 return sk_ns_capable(sk, &init_user_ns, cap);
187 EXPORT_SYMBOL(sk_capable);
190 * sk_net_capable - Network namespace socket capability test
191 * @sk: Socket to use a capability on or through
192 * @cap: The capability to use
194 * Test to see if the opener of the socket had when the socket was created
195 * and the current process has the capability @cap over the network namespace
196 * the socket is a member of.
198 bool sk_net_capable(const struct sock *sk, int cap)
200 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
202 EXPORT_SYMBOL(sk_net_capable);
205 * Each address family might have different locking rules, so we have
206 * one slock key per address family and separate keys for internal and
209 static struct lock_class_key af_family_keys[AF_MAX];
210 static struct lock_class_key af_family_kern_keys[AF_MAX];
211 static struct lock_class_key af_family_slock_keys[AF_MAX];
212 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
215 * Make lock validator output more readable. (we pre-construct these
216 * strings build-time, so that runtime initialization of socket
220 #define _sock_locks(x) \
221 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
222 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
223 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
224 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
225 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
226 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
227 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
228 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
229 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
230 x "27" , x "28" , x "AF_CAN" , \
231 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
232 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
233 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
234 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
235 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
239 static const char *const af_family_key_strings[AF_MAX+1] = {
240 _sock_locks("sk_lock-")
242 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("slock-")
245 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("clock-")
249 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
250 _sock_locks("k-sk_lock-")
252 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
253 _sock_locks("k-slock-")
255 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
256 _sock_locks("k-clock-")
258 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
259 _sock_locks("rlock-")
261 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
262 _sock_locks("wlock-")
264 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
265 _sock_locks("elock-")
269 * sk_callback_lock and sk queues locking rules are per-address-family,
270 * so split the lock classes by using a per-AF key:
272 static struct lock_class_key af_callback_keys[AF_MAX];
273 static struct lock_class_key af_rlock_keys[AF_MAX];
274 static struct lock_class_key af_wlock_keys[AF_MAX];
275 static struct lock_class_key af_elock_keys[AF_MAX];
276 static struct lock_class_key af_kern_callback_keys[AF_MAX];
278 /* Run time adjustable parameters. */
279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
280 EXPORT_SYMBOL(sysctl_wmem_max);
281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
282 EXPORT_SYMBOL(sysctl_rmem_max);
283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
286 /* Maximal space eaten by iovec or ancillary data plus some space */
287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
288 EXPORT_SYMBOL(sysctl_optmem_max);
290 int sysctl_tstamp_allow_data __read_mostly = 1;
292 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
293 EXPORT_SYMBOL_GPL(memalloc_socks_key);
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
303 void sk_set_memalloc(struct sock *sk)
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
307 static_branch_inc(&memalloc_socks_key);
309 EXPORT_SYMBOL_GPL(sk_set_memalloc);
311 void sk_clear_memalloc(struct sock *sk)
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
315 static_branch_dec(&memalloc_socks_key);
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
326 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
331 unsigned int noreclaim_flag;
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
336 noreclaim_flag = memalloc_noreclaim_save();
337 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
341 memalloc_noreclaim_restore(noreclaim_flag);
345 EXPORT_SYMBOL(__sk_backlog_rcv);
347 void sk_error_report(struct sock *sk)
349 sk->sk_error_report(sk);
351 switch (sk->sk_family) {
355 trace_inet_sk_error_report(sk);
361 EXPORT_SYMBOL(sk_error_report);
363 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
365 struct __kernel_sock_timeval tv;
367 if (timeo == MAX_SCHEDULE_TIMEOUT) {
371 tv.tv_sec = timeo / HZ;
372 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
375 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
376 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
377 *(struct old_timeval32 *)optval = tv32;
382 struct __kernel_old_timeval old_tv;
383 old_tv.tv_sec = tv.tv_sec;
384 old_tv.tv_usec = tv.tv_usec;
385 *(struct __kernel_old_timeval *)optval = old_tv;
386 return sizeof(old_tv);
389 *(struct __kernel_sock_timeval *)optval = tv;
392 EXPORT_SYMBOL(sock_get_timeout);
394 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
395 sockptr_t optval, int optlen, bool old_timeval)
397 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
398 struct old_timeval32 tv32;
400 if (optlen < sizeof(tv32))
403 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
405 tv->tv_sec = tv32.tv_sec;
406 tv->tv_usec = tv32.tv_usec;
407 } else if (old_timeval) {
408 struct __kernel_old_timeval old_tv;
410 if (optlen < sizeof(old_tv))
412 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
414 tv->tv_sec = old_tv.tv_sec;
415 tv->tv_usec = old_tv.tv_usec;
417 if (optlen < sizeof(*tv))
419 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
425 EXPORT_SYMBOL(sock_copy_user_timeval);
427 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
430 struct __kernel_sock_timeval tv;
431 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
436 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
440 static int warned __read_mostly;
443 if (warned < 10 && net_ratelimit()) {
445 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
446 __func__, current->comm, task_pid_nr(current));
450 *timeo_p = MAX_SCHEDULE_TIMEOUT;
451 if (tv.tv_sec == 0 && tv.tv_usec == 0)
453 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
454 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
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 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(struct sock *sk)
762 if (dev_recursion_level())
766 switch (sk->sk_family) {
768 return inet_sk(sk)->mc_loop;
769 #if IS_ENABLED(CONFIG_IPV6)
771 return inet6_sk(sk)->mc_loop;
777 EXPORT_SYMBOL(sk_mc_loop);
779 void sock_set_reuseaddr(struct sock *sk)
782 sk->sk_reuse = SK_CAN_REUSE;
785 EXPORT_SYMBOL(sock_set_reuseaddr);
787 void sock_set_reuseport(struct sock *sk)
790 sk->sk_reuseport = true;
793 EXPORT_SYMBOL(sock_set_reuseport);
795 void sock_no_linger(struct sock *sk)
798 sk->sk_lingertime = 0;
799 sock_set_flag(sk, SOCK_LINGER);
802 EXPORT_SYMBOL(sock_no_linger);
804 void sock_set_priority(struct sock *sk, u32 priority)
807 sk->sk_priority = priority;
810 EXPORT_SYMBOL(sock_set_priority);
812 void sock_set_sndtimeo(struct sock *sk, s64 secs)
815 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
816 sk->sk_sndtimeo = secs * HZ;
818 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
821 EXPORT_SYMBOL(sock_set_sndtimeo);
823 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
826 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
827 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
828 sock_set_flag(sk, SOCK_RCVTSTAMP);
829 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
831 sock_reset_flag(sk, SOCK_RCVTSTAMP);
832 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
836 void sock_enable_timestamps(struct sock *sk)
839 __sock_set_timestamps(sk, true, false, true);
842 EXPORT_SYMBOL(sock_enable_timestamps);
844 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
847 case SO_TIMESTAMP_OLD:
848 __sock_set_timestamps(sk, valbool, false, false);
850 case SO_TIMESTAMP_NEW:
851 __sock_set_timestamps(sk, valbool, true, false);
853 case SO_TIMESTAMPNS_OLD:
854 __sock_set_timestamps(sk, valbool, false, true);
856 case SO_TIMESTAMPNS_NEW:
857 __sock_set_timestamps(sk, valbool, true, true);
862 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
864 struct net *net = sock_net(sk);
865 struct net_device *dev = NULL;
870 if (sk->sk_bound_dev_if)
871 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
874 pr_err("%s: sock not bind to device\n", __func__);
878 num = ethtool_get_phc_vclocks(dev, &vclock_index);
881 for (i = 0; i < num; i++) {
882 if (*(vclock_index + i) == phc_index) {
894 sk->sk_bind_phc = phc_index;
899 int sock_set_timestamping(struct sock *sk, int optname,
900 struct so_timestamping timestamping)
902 int val = timestamping.flags;
905 if (val & ~SOF_TIMESTAMPING_MASK)
908 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
909 !(val & SOF_TIMESTAMPING_OPT_ID))
912 if (val & SOF_TIMESTAMPING_OPT_ID &&
913 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
915 if ((1 << sk->sk_state) &
916 (TCPF_CLOSE | TCPF_LISTEN))
918 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
919 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
921 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
923 atomic_set(&sk->sk_tskey, 0);
927 if (val & SOF_TIMESTAMPING_OPT_STATS &&
928 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
931 if (val & SOF_TIMESTAMPING_BIND_PHC) {
932 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
937 sk->sk_tsflags = val;
938 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
940 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
941 sock_enable_timestamp(sk,
942 SOCK_TIMESTAMPING_RX_SOFTWARE);
944 sock_disable_timestamp(sk,
945 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
949 void sock_set_keepalive(struct sock *sk)
952 if (sk->sk_prot->keepalive)
953 sk->sk_prot->keepalive(sk, true);
954 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
957 EXPORT_SYMBOL(sock_set_keepalive);
959 static void __sock_set_rcvbuf(struct sock *sk, int val)
961 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
962 * as a negative value.
964 val = min_t(int, val, INT_MAX / 2);
965 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
967 /* We double it on the way in to account for "struct sk_buff" etc.
968 * overhead. Applications assume that the SO_RCVBUF setting they make
969 * will allow that much actual data to be received on that socket.
971 * Applications are unaware that "struct sk_buff" and other overheads
972 * allocate from the receive buffer during socket buffer allocation.
974 * And after considering the possible alternatives, returning the value
975 * we actually used in getsockopt is the most desirable behavior.
977 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
980 void sock_set_rcvbuf(struct sock *sk, int val)
983 __sock_set_rcvbuf(sk, val);
986 EXPORT_SYMBOL(sock_set_rcvbuf);
988 static void __sock_set_mark(struct sock *sk, u32 val)
990 if (val != sk->sk_mark) {
996 void sock_set_mark(struct sock *sk, u32 val)
999 __sock_set_mark(sk, val);
1002 EXPORT_SYMBOL(sock_set_mark);
1004 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1006 /* Round down bytes to multiple of pages */
1007 bytes = round_down(bytes, PAGE_SIZE);
1009 WARN_ON(bytes > sk->sk_reserved_mem);
1010 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1014 static int sock_reserve_memory(struct sock *sk, int bytes)
1020 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1026 pages = sk_mem_pages(bytes);
1028 /* pre-charge to memcg */
1029 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1030 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1034 /* pre-charge to forward_alloc */
1035 sk_memory_allocated_add(sk, pages);
1036 allocated = sk_memory_allocated(sk);
1037 /* If the system goes into memory pressure with this
1038 * precharge, give up and return error.
1040 if (allocated > sk_prot_mem_limits(sk, 1)) {
1041 sk_memory_allocated_sub(sk, pages);
1042 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1045 sk->sk_forward_alloc += pages << PAGE_SHIFT;
1047 WRITE_ONCE(sk->sk_reserved_mem,
1048 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1053 void sockopt_lock_sock(struct sock *sk)
1055 /* When current->bpf_ctx is set, the setsockopt is called from
1056 * a bpf prog. bpf has ensured the sk lock has been
1057 * acquired before calling setsockopt().
1059 if (has_current_bpf_ctx())
1064 EXPORT_SYMBOL(sockopt_lock_sock);
1066 void sockopt_release_sock(struct sock *sk)
1068 if (has_current_bpf_ctx())
1073 EXPORT_SYMBOL(sockopt_release_sock);
1075 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1077 return has_current_bpf_ctx() || ns_capable(ns, cap);
1079 EXPORT_SYMBOL(sockopt_ns_capable);
1081 bool sockopt_capable(int cap)
1083 return has_current_bpf_ctx() || capable(cap);
1085 EXPORT_SYMBOL(sockopt_capable);
1088 * This is meant for all protocols to use and covers goings on
1089 * at the socket level. Everything here is generic.
1092 int sk_setsockopt(struct sock *sk, int level, int optname,
1093 sockptr_t optval, unsigned int optlen)
1095 struct so_timestamping timestamping;
1096 struct socket *sock = sk->sk_socket;
1097 struct sock_txtime sk_txtime;
1104 * Options without arguments
1107 if (optname == SO_BINDTODEVICE)
1108 return sock_setbindtodevice(sk, optval, optlen);
1110 if (optlen < sizeof(int))
1113 if (copy_from_sockptr(&val, optval, sizeof(val)))
1116 valbool = val ? 1 : 0;
1118 sockopt_lock_sock(sk);
1122 if (val && !sockopt_capable(CAP_NET_ADMIN))
1125 sock_valbool_flag(sk, SOCK_DBG, valbool);
1128 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1131 sk->sk_reuseport = valbool;
1140 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1144 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1147 /* Don't error on this BSD doesn't and if you think
1148 * about it this is right. Otherwise apps have to
1149 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1150 * are treated in BSD as hints
1152 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1154 /* Ensure val * 2 fits into an int, to prevent max_t()
1155 * from treating it as a negative value.
1157 val = min_t(int, val, INT_MAX / 2);
1158 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1159 WRITE_ONCE(sk->sk_sndbuf,
1160 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1161 /* Wake up sending tasks if we upped the value. */
1162 sk->sk_write_space(sk);
1165 case SO_SNDBUFFORCE:
1166 if (!sockopt_capable(CAP_NET_ADMIN)) {
1171 /* No negative values (to prevent underflow, as val will be
1179 /* Don't error on this BSD doesn't and if you think
1180 * about it this is right. Otherwise apps have to
1181 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1182 * are treated in BSD as hints
1184 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1187 case SO_RCVBUFFORCE:
1188 if (!sockopt_capable(CAP_NET_ADMIN)) {
1193 /* No negative values (to prevent underflow, as val will be
1196 __sock_set_rcvbuf(sk, max(val, 0));
1200 if (sk->sk_prot->keepalive)
1201 sk->sk_prot->keepalive(sk, valbool);
1202 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1206 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1210 sk->sk_no_check_tx = valbool;
1214 if ((val >= 0 && val <= 6) ||
1215 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1216 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1217 sk->sk_priority = val;
1223 if (optlen < sizeof(ling)) {
1224 ret = -EINVAL; /* 1003.1g */
1227 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1232 sock_reset_flag(sk, SOCK_LINGER);
1234 #if (BITS_PER_LONG == 32)
1235 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1236 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1239 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1240 sock_set_flag(sk, SOCK_LINGER);
1249 set_bit(SOCK_PASSCRED, &sock->flags);
1251 clear_bit(SOCK_PASSCRED, &sock->flags);
1256 set_bit(SOCK_PASSPIDFD, &sock->flags);
1258 clear_bit(SOCK_PASSPIDFD, &sock->flags);
1261 case SO_TIMESTAMP_OLD:
1262 case SO_TIMESTAMP_NEW:
1263 case SO_TIMESTAMPNS_OLD:
1264 case SO_TIMESTAMPNS_NEW:
1265 sock_set_timestamp(sk, optname, valbool);
1268 case SO_TIMESTAMPING_NEW:
1269 case SO_TIMESTAMPING_OLD:
1270 if (optlen == sizeof(timestamping)) {
1271 if (copy_from_sockptr(×tamping, optval,
1272 sizeof(timestamping))) {
1277 memset(×tamping, 0, sizeof(timestamping));
1278 timestamping.flags = val;
1280 ret = sock_set_timestamping(sk, optname, timestamping);
1286 if (sock && sock->ops->set_rcvlowat)
1287 ret = sock->ops->set_rcvlowat(sk, val);
1289 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1292 case SO_RCVTIMEO_OLD:
1293 case SO_RCVTIMEO_NEW:
1294 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1295 optlen, optname == SO_RCVTIMEO_OLD);
1298 case SO_SNDTIMEO_OLD:
1299 case SO_SNDTIMEO_NEW:
1300 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1301 optlen, optname == SO_SNDTIMEO_OLD);
1304 case SO_ATTACH_FILTER: {
1305 struct sock_fprog fprog;
1307 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1309 ret = sk_attach_filter(&fprog, sk);
1314 if (optlen == sizeof(u32)) {
1318 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1321 ret = sk_attach_bpf(ufd, sk);
1325 case SO_ATTACH_REUSEPORT_CBPF: {
1326 struct sock_fprog fprog;
1328 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1330 ret = sk_reuseport_attach_filter(&fprog, sk);
1333 case SO_ATTACH_REUSEPORT_EBPF:
1335 if (optlen == sizeof(u32)) {
1339 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1342 ret = sk_reuseport_attach_bpf(ufd, sk);
1346 case SO_DETACH_REUSEPORT_BPF:
1347 ret = reuseport_detach_prog(sk);
1350 case SO_DETACH_FILTER:
1351 ret = sk_detach_filter(sk);
1354 case SO_LOCK_FILTER:
1355 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1358 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1363 set_bit(SOCK_PASSSEC, &sock->flags);
1365 clear_bit(SOCK_PASSSEC, &sock->flags);
1368 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1369 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1374 __sock_set_mark(sk, val);
1377 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1381 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1384 case SO_WIFI_STATUS:
1385 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1389 if (sock->ops->set_peek_off)
1390 ret = sock->ops->set_peek_off(sk, val);
1396 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1399 case SO_SELECT_ERR_QUEUE:
1400 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1403 #ifdef CONFIG_NET_RX_BUSY_POLL
1408 WRITE_ONCE(sk->sk_ll_usec, val);
1410 case SO_PREFER_BUSY_POLL:
1411 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1414 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1416 case SO_BUSY_POLL_BUDGET:
1417 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) {
1420 if (val < 0 || val > U16_MAX)
1423 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1428 case SO_MAX_PACING_RATE:
1430 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1432 if (sizeof(ulval) != sizeof(val) &&
1433 optlen >= sizeof(ulval) &&
1434 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1439 cmpxchg(&sk->sk_pacing_status,
1442 /* Pairs with READ_ONCE() from sk_getsockopt() */
1443 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1444 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1447 case SO_INCOMING_CPU:
1448 reuseport_update_incoming_cpu(sk, val);
1453 dst_negative_advice(sk);
1457 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1458 if (!(sk_is_tcp(sk) ||
1459 (sk->sk_type == SOCK_DGRAM &&
1460 sk->sk_protocol == IPPROTO_UDP)))
1462 } else if (sk->sk_family != PF_RDS) {
1466 if (val < 0 || val > 1)
1469 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1474 if (optlen != sizeof(struct sock_txtime)) {
1477 } else if (copy_from_sockptr(&sk_txtime, optval,
1478 sizeof(struct sock_txtime))) {
1481 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1485 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1486 * scheduler has enough safe guards.
1488 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1489 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1493 sock_valbool_flag(sk, SOCK_TXTIME, true);
1494 sk->sk_clockid = sk_txtime.clockid;
1495 sk->sk_txtime_deadline_mode =
1496 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1497 sk->sk_txtime_report_errors =
1498 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1501 case SO_BINDTOIFINDEX:
1502 ret = sock_bindtoindex_locked(sk, val);
1506 if (val & ~SOCK_BUF_LOCK_MASK) {
1510 sk->sk_userlocks = val | (sk->sk_userlocks &
1511 ~SOCK_BUF_LOCK_MASK);
1514 case SO_RESERVE_MEM:
1523 delta = val - sk->sk_reserved_mem;
1525 sock_release_reserved_memory(sk, -delta);
1527 ret = sock_reserve_memory(sk, delta);
1532 if (val < -1 || val > 1) {
1536 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1537 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1538 /* Paired with READ_ONCE() in tcp_rtx_synack()
1539 * and sk_getsockopt().
1541 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1548 sockopt_release_sock(sk);
1552 int sock_setsockopt(struct socket *sock, int level, int optname,
1553 sockptr_t optval, unsigned int optlen)
1555 return sk_setsockopt(sock->sk, level, optname,
1558 EXPORT_SYMBOL(sock_setsockopt);
1560 static const struct cred *sk_get_peer_cred(struct sock *sk)
1562 const struct cred *cred;
1564 spin_lock(&sk->sk_peer_lock);
1565 cred = get_cred(sk->sk_peer_cred);
1566 spin_unlock(&sk->sk_peer_lock);
1571 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1572 struct ucred *ucred)
1574 ucred->pid = pid_vnr(pid);
1575 ucred->uid = ucred->gid = -1;
1577 struct user_namespace *current_ns = current_user_ns();
1579 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1580 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1584 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1586 struct user_namespace *user_ns = current_user_ns();
1589 for (i = 0; i < src->ngroups; i++) {
1590 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1592 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1599 int sk_getsockopt(struct sock *sk, int level, int optname,
1600 sockptr_t optval, sockptr_t optlen)
1602 struct socket *sock = sk->sk_socket;
1607 unsigned long ulval;
1609 struct old_timeval32 tm32;
1610 struct __kernel_old_timeval tm;
1611 struct __kernel_sock_timeval stm;
1612 struct sock_txtime txtime;
1613 struct so_timestamping timestamping;
1616 int lv = sizeof(int);
1619 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1624 memset(&v, 0, sizeof(v));
1628 v.val = sock_flag(sk, SOCK_DBG);
1632 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1636 v.val = sock_flag(sk, SOCK_BROADCAST);
1640 v.val = sk->sk_sndbuf;
1644 v.val = sk->sk_rcvbuf;
1648 v.val = sk->sk_reuse;
1652 v.val = sk->sk_reuseport;
1656 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1660 v.val = sk->sk_type;
1664 v.val = sk->sk_protocol;
1668 v.val = sk->sk_family;
1672 v.val = -sock_error(sk);
1674 v.val = xchg(&sk->sk_err_soft, 0);
1678 v.val = sock_flag(sk, SOCK_URGINLINE);
1682 v.val = sk->sk_no_check_tx;
1686 v.val = sk->sk_priority;
1690 lv = sizeof(v.ling);
1691 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1692 v.ling.l_linger = sk->sk_lingertime / HZ;
1698 case SO_TIMESTAMP_OLD:
1699 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1700 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1701 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1704 case SO_TIMESTAMPNS_OLD:
1705 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1708 case SO_TIMESTAMP_NEW:
1709 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1712 case SO_TIMESTAMPNS_NEW:
1713 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1716 case SO_TIMESTAMPING_OLD:
1717 lv = sizeof(v.timestamping);
1718 v.timestamping.flags = sk->sk_tsflags;
1719 v.timestamping.bind_phc = sk->sk_bind_phc;
1722 case SO_RCVTIMEO_OLD:
1723 case SO_RCVTIMEO_NEW:
1724 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1727 case SO_SNDTIMEO_OLD:
1728 case SO_SNDTIMEO_NEW:
1729 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1733 v.val = READ_ONCE(sk->sk_rcvlowat);
1741 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1745 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1750 struct ucred peercred;
1751 if (len > sizeof(peercred))
1752 len = sizeof(peercred);
1754 spin_lock(&sk->sk_peer_lock);
1755 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1756 spin_unlock(&sk->sk_peer_lock);
1758 if (copy_to_sockptr(optval, &peercred, len))
1765 struct pid *peer_pid;
1766 struct file *pidfd_file = NULL;
1769 if (len > sizeof(pidfd))
1770 len = sizeof(pidfd);
1772 spin_lock(&sk->sk_peer_lock);
1773 peer_pid = get_pid(sk->sk_peer_pid);
1774 spin_unlock(&sk->sk_peer_lock);
1779 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1784 if (copy_to_sockptr(optval, &pidfd, len) ||
1785 copy_to_sockptr(optlen, &len, sizeof(int))) {
1786 put_unused_fd(pidfd);
1792 fd_install(pidfd, pidfd_file);
1798 const struct cred *cred;
1801 cred = sk_get_peer_cred(sk);
1805 n = cred->group_info->ngroups;
1806 if (len < n * sizeof(gid_t)) {
1807 len = n * sizeof(gid_t);
1809 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1811 len = n * sizeof(gid_t);
1813 ret = groups_to_user(optval, cred->group_info);
1824 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1829 if (copy_to_sockptr(optval, address, len))
1834 /* Dubious BSD thing... Probably nobody even uses it, but
1835 * the UNIX standard wants it for whatever reason... -DaveM
1838 v.val = sk->sk_state == TCP_LISTEN;
1842 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1846 return security_socket_getpeersec_stream(sock,
1847 optval, optlen, len);
1850 v.val = sk->sk_mark;
1854 v.val = sock_flag(sk, SOCK_RCVMARK);
1858 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1861 case SO_WIFI_STATUS:
1862 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1866 if (!sock->ops->set_peek_off)
1869 v.val = sk->sk_peek_off;
1872 v.val = sock_flag(sk, SOCK_NOFCS);
1875 case SO_BINDTODEVICE:
1876 return sock_getbindtodevice(sk, optval, optlen, len);
1879 len = sk_get_filter(sk, optval, len);
1885 case SO_LOCK_FILTER:
1886 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1889 case SO_BPF_EXTENSIONS:
1890 v.val = bpf_tell_extensions();
1893 case SO_SELECT_ERR_QUEUE:
1894 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1897 #ifdef CONFIG_NET_RX_BUSY_POLL
1899 v.val = sk->sk_ll_usec;
1901 case SO_PREFER_BUSY_POLL:
1902 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1906 case SO_MAX_PACING_RATE:
1907 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1908 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1909 lv = sizeof(v.ulval);
1910 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1913 v.val = min_t(unsigned long, ~0U,
1914 READ_ONCE(sk->sk_max_pacing_rate));
1918 case SO_INCOMING_CPU:
1919 v.val = READ_ONCE(sk->sk_incoming_cpu);
1924 u32 meminfo[SK_MEMINFO_VARS];
1926 sk_get_meminfo(sk, meminfo);
1928 len = min_t(unsigned int, len, sizeof(meminfo));
1929 if (copy_to_sockptr(optval, &meminfo, len))
1935 #ifdef CONFIG_NET_RX_BUSY_POLL
1936 case SO_INCOMING_NAPI_ID:
1937 v.val = READ_ONCE(sk->sk_napi_id);
1939 /* aggregate non-NAPI IDs down to 0 */
1940 if (v.val < MIN_NAPI_ID)
1950 v.val64 = sock_gen_cookie(sk);
1954 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1958 lv = sizeof(v.txtime);
1959 v.txtime.clockid = sk->sk_clockid;
1960 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1961 SOF_TXTIME_DEADLINE_MODE : 0;
1962 v.txtime.flags |= sk->sk_txtime_report_errors ?
1963 SOF_TXTIME_REPORT_ERRORS : 0;
1966 case SO_BINDTOIFINDEX:
1967 v.val = READ_ONCE(sk->sk_bound_dev_if);
1970 case SO_NETNS_COOKIE:
1974 v.val64 = sock_net(sk)->net_cookie;
1978 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1981 case SO_RESERVE_MEM:
1982 v.val = READ_ONCE(sk->sk_reserved_mem);
1986 /* Paired with WRITE_ONCE() in sk_setsockopt() */
1987 v.val = READ_ONCE(sk->sk_txrehash);
1991 /* We implement the SO_SNDLOWAT etc to not be settable
1994 return -ENOPROTOOPT;
1999 if (copy_to_sockptr(optval, &v, len))
2002 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2007 int sock_getsockopt(struct socket *sock, int level, int optname,
2008 char __user *optval, int __user *optlen)
2010 return sk_getsockopt(sock->sk, level, optname,
2011 USER_SOCKPTR(optval),
2012 USER_SOCKPTR(optlen));
2016 * Initialize an sk_lock.
2018 * (We also register the sk_lock with the lock validator.)
2020 static inline void sock_lock_init(struct sock *sk)
2022 if (sk->sk_kern_sock)
2023 sock_lock_init_class_and_name(
2025 af_family_kern_slock_key_strings[sk->sk_family],
2026 af_family_kern_slock_keys + sk->sk_family,
2027 af_family_kern_key_strings[sk->sk_family],
2028 af_family_kern_keys + sk->sk_family);
2030 sock_lock_init_class_and_name(
2032 af_family_slock_key_strings[sk->sk_family],
2033 af_family_slock_keys + sk->sk_family,
2034 af_family_key_strings[sk->sk_family],
2035 af_family_keys + sk->sk_family);
2039 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2040 * even temporarly, because of RCU lookups. sk_node should also be left as is.
2041 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2043 static void sock_copy(struct sock *nsk, const struct sock *osk)
2045 const struct proto *prot = READ_ONCE(osk->sk_prot);
2046 #ifdef CONFIG_SECURITY_NETWORK
2047 void *sptr = nsk->sk_security;
2050 /* If we move sk_tx_queue_mapping out of the private section,
2051 * we must check if sk_tx_queue_clear() is called after
2052 * sock_copy() in sk_clone_lock().
2054 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2055 offsetof(struct sock, sk_dontcopy_begin) ||
2056 offsetof(struct sock, sk_tx_queue_mapping) >=
2057 offsetof(struct sock, sk_dontcopy_end));
2059 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2061 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2062 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2064 #ifdef CONFIG_SECURITY_NETWORK
2065 nsk->sk_security = sptr;
2066 security_sk_clone(osk, nsk);
2070 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2074 struct kmem_cache *slab;
2078 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2081 if (want_init_on_alloc(priority))
2082 sk_prot_clear_nulls(sk, prot->obj_size);
2084 sk = kmalloc(prot->obj_size, priority);
2087 if (security_sk_alloc(sk, family, priority))
2090 if (!try_module_get(prot->owner))
2097 security_sk_free(sk);
2100 kmem_cache_free(slab, sk);
2106 static void sk_prot_free(struct proto *prot, struct sock *sk)
2108 struct kmem_cache *slab;
2109 struct module *owner;
2111 owner = prot->owner;
2114 cgroup_sk_free(&sk->sk_cgrp_data);
2115 mem_cgroup_sk_free(sk);
2116 security_sk_free(sk);
2118 kmem_cache_free(slab, sk);
2125 * sk_alloc - All socket objects are allocated here
2126 * @net: the applicable net namespace
2127 * @family: protocol family
2128 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2129 * @prot: struct proto associated with this new sock instance
2130 * @kern: is this to be a kernel socket?
2132 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2133 struct proto *prot, int kern)
2137 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2139 sk->sk_family = family;
2141 * See comment in struct sock definition to understand
2142 * why we need sk_prot_creator -acme
2144 sk->sk_prot = sk->sk_prot_creator = prot;
2145 sk->sk_kern_sock = kern;
2147 sk->sk_net_refcnt = kern ? 0 : 1;
2148 if (likely(sk->sk_net_refcnt)) {
2149 get_net_track(net, &sk->ns_tracker, priority);
2150 sock_inuse_add(net, 1);
2152 __netns_tracker_alloc(net, &sk->ns_tracker,
2156 sock_net_set(sk, net);
2157 refcount_set(&sk->sk_wmem_alloc, 1);
2159 mem_cgroup_sk_alloc(sk);
2160 cgroup_sk_alloc(&sk->sk_cgrp_data);
2161 sock_update_classid(&sk->sk_cgrp_data);
2162 sock_update_netprioidx(&sk->sk_cgrp_data);
2163 sk_tx_queue_clear(sk);
2168 EXPORT_SYMBOL(sk_alloc);
2170 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2171 * grace period. This is the case for UDP sockets and TCP listeners.
2173 static void __sk_destruct(struct rcu_head *head)
2175 struct sock *sk = container_of(head, struct sock, sk_rcu);
2176 struct sk_filter *filter;
2178 if (sk->sk_destruct)
2179 sk->sk_destruct(sk);
2181 filter = rcu_dereference_check(sk->sk_filter,
2182 refcount_read(&sk->sk_wmem_alloc) == 0);
2184 sk_filter_uncharge(sk, filter);
2185 RCU_INIT_POINTER(sk->sk_filter, NULL);
2188 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2190 #ifdef CONFIG_BPF_SYSCALL
2191 bpf_sk_storage_free(sk);
2194 if (atomic_read(&sk->sk_omem_alloc))
2195 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2196 __func__, atomic_read(&sk->sk_omem_alloc));
2198 if (sk->sk_frag.page) {
2199 put_page(sk->sk_frag.page);
2200 sk->sk_frag.page = NULL;
2203 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2204 put_cred(sk->sk_peer_cred);
2205 put_pid(sk->sk_peer_pid);
2207 if (likely(sk->sk_net_refcnt))
2208 put_net_track(sock_net(sk), &sk->ns_tracker);
2210 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2212 sk_prot_free(sk->sk_prot_creator, sk);
2215 void sk_destruct(struct sock *sk)
2217 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2219 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2220 reuseport_detach_sock(sk);
2221 use_call_rcu = true;
2225 call_rcu(&sk->sk_rcu, __sk_destruct);
2227 __sk_destruct(&sk->sk_rcu);
2230 static void __sk_free(struct sock *sk)
2232 if (likely(sk->sk_net_refcnt))
2233 sock_inuse_add(sock_net(sk), -1);
2235 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2236 sock_diag_broadcast_destroy(sk);
2241 void sk_free(struct sock *sk)
2244 * We subtract one from sk_wmem_alloc and can know if
2245 * some packets are still in some tx queue.
2246 * If not null, sock_wfree() will call __sk_free(sk) later
2248 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2251 EXPORT_SYMBOL(sk_free);
2253 static void sk_init_common(struct sock *sk)
2255 skb_queue_head_init(&sk->sk_receive_queue);
2256 skb_queue_head_init(&sk->sk_write_queue);
2257 skb_queue_head_init(&sk->sk_error_queue);
2259 rwlock_init(&sk->sk_callback_lock);
2260 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2261 af_rlock_keys + sk->sk_family,
2262 af_family_rlock_key_strings[sk->sk_family]);
2263 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2264 af_wlock_keys + sk->sk_family,
2265 af_family_wlock_key_strings[sk->sk_family]);
2266 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2267 af_elock_keys + sk->sk_family,
2268 af_family_elock_key_strings[sk->sk_family]);
2269 lockdep_set_class_and_name(&sk->sk_callback_lock,
2270 af_callback_keys + sk->sk_family,
2271 af_family_clock_key_strings[sk->sk_family]);
2275 * sk_clone_lock - clone a socket, and lock its clone
2276 * @sk: the socket to clone
2277 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2279 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2281 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2283 struct proto *prot = READ_ONCE(sk->sk_prot);
2284 struct sk_filter *filter;
2285 bool is_charged = true;
2288 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2292 sock_copy(newsk, sk);
2294 newsk->sk_prot_creator = prot;
2297 if (likely(newsk->sk_net_refcnt)) {
2298 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2299 sock_inuse_add(sock_net(newsk), 1);
2301 /* Kernel sockets are not elevating the struct net refcount.
2302 * Instead, use a tracker to more easily detect if a layer
2303 * is not properly dismantling its kernel sockets at netns
2306 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2309 sk_node_init(&newsk->sk_node);
2310 sock_lock_init(newsk);
2311 bh_lock_sock(newsk);
2312 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2313 newsk->sk_backlog.len = 0;
2315 atomic_set(&newsk->sk_rmem_alloc, 0);
2317 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2318 refcount_set(&newsk->sk_wmem_alloc, 1);
2320 atomic_set(&newsk->sk_omem_alloc, 0);
2321 sk_init_common(newsk);
2323 newsk->sk_dst_cache = NULL;
2324 newsk->sk_dst_pending_confirm = 0;
2325 newsk->sk_wmem_queued = 0;
2326 newsk->sk_forward_alloc = 0;
2327 newsk->sk_reserved_mem = 0;
2328 atomic_set(&newsk->sk_drops, 0);
2329 newsk->sk_send_head = NULL;
2330 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2331 atomic_set(&newsk->sk_zckey, 0);
2333 sock_reset_flag(newsk, SOCK_DONE);
2335 /* sk->sk_memcg will be populated at accept() time */
2336 newsk->sk_memcg = NULL;
2338 cgroup_sk_clone(&newsk->sk_cgrp_data);
2341 filter = rcu_dereference(sk->sk_filter);
2343 /* though it's an empty new sock, the charging may fail
2344 * if sysctl_optmem_max was changed between creation of
2345 * original socket and cloning
2347 is_charged = sk_filter_charge(newsk, filter);
2348 RCU_INIT_POINTER(newsk->sk_filter, filter);
2351 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2352 /* We need to make sure that we don't uncharge the new
2353 * socket if we couldn't charge it in the first place
2354 * as otherwise we uncharge the parent's filter.
2357 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2358 sk_free_unlock_clone(newsk);
2362 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2364 if (bpf_sk_storage_clone(sk, newsk)) {
2365 sk_free_unlock_clone(newsk);
2370 /* Clear sk_user_data if parent had the pointer tagged
2371 * as not suitable for copying when cloning.
2373 if (sk_user_data_is_nocopy(newsk))
2374 newsk->sk_user_data = NULL;
2377 newsk->sk_err_soft = 0;
2378 newsk->sk_priority = 0;
2379 newsk->sk_incoming_cpu = raw_smp_processor_id();
2381 /* Before updating sk_refcnt, we must commit prior changes to memory
2382 * (Documentation/RCU/rculist_nulls.rst for details)
2385 refcount_set(&newsk->sk_refcnt, 2);
2387 sk_set_socket(newsk, NULL);
2388 sk_tx_queue_clear(newsk);
2389 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2391 if (newsk->sk_prot->sockets_allocated)
2392 sk_sockets_allocated_inc(newsk);
2394 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2395 net_enable_timestamp();
2399 EXPORT_SYMBOL_GPL(sk_clone_lock);
2401 void sk_free_unlock_clone(struct sock *sk)
2403 /* It is still raw copy of parent, so invalidate
2404 * destructor and make plain sk_free() */
2405 sk->sk_destruct = NULL;
2409 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2411 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2413 bool is_ipv6 = false;
2416 #if IS_ENABLED(CONFIG_IPV6)
2417 is_ipv6 = (sk->sk_family == AF_INET6 &&
2418 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2420 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2421 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2422 READ_ONCE(dst->dev->gso_ipv4_max_size);
2423 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2424 max_size = GSO_LEGACY_MAX_SIZE;
2426 return max_size - (MAX_TCP_HEADER + 1);
2429 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2433 sk->sk_route_caps = dst->dev->features;
2435 sk->sk_route_caps |= NETIF_F_GSO;
2436 if (sk->sk_route_caps & NETIF_F_GSO)
2437 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2438 if (unlikely(sk->sk_gso_disabled))
2439 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2440 if (sk_can_gso(sk)) {
2441 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2442 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2444 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2445 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2446 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2447 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2450 sk->sk_gso_max_segs = max_segs;
2451 sk_dst_set(sk, dst);
2453 EXPORT_SYMBOL_GPL(sk_setup_caps);
2456 * Simple resource managers for sockets.
2461 * Write buffer destructor automatically called from kfree_skb.
2463 void sock_wfree(struct sk_buff *skb)
2465 struct sock *sk = skb->sk;
2466 unsigned int len = skb->truesize;
2469 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2470 if (sock_flag(sk, SOCK_RCU_FREE) &&
2471 sk->sk_write_space == sock_def_write_space) {
2473 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2474 sock_def_write_space_wfree(sk);
2482 * Keep a reference on sk_wmem_alloc, this will be released
2483 * after sk_write_space() call
2485 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2486 sk->sk_write_space(sk);
2490 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2491 * could not do because of in-flight packets
2493 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2496 EXPORT_SYMBOL(sock_wfree);
2498 /* This variant of sock_wfree() is used by TCP,
2499 * since it sets SOCK_USE_WRITE_QUEUE.
2501 void __sock_wfree(struct sk_buff *skb)
2503 struct sock *sk = skb->sk;
2505 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2509 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2514 if (unlikely(!sk_fullsock(sk))) {
2515 skb->destructor = sock_edemux;
2520 skb->destructor = sock_wfree;
2521 skb_set_hash_from_sk(skb, sk);
2523 * We used to take a refcount on sk, but following operation
2524 * is enough to guarantee sk_free() wont free this sock until
2525 * all in-flight packets are completed
2527 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2529 EXPORT_SYMBOL(skb_set_owner_w);
2531 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2533 #ifdef CONFIG_TLS_DEVICE
2534 /* Drivers depend on in-order delivery for crypto offload,
2535 * partial orphan breaks out-of-order-OK logic.
2540 return (skb->destructor == sock_wfree ||
2541 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2544 /* This helper is used by netem, as it can hold packets in its
2545 * delay queue. We want to allow the owner socket to send more
2546 * packets, as if they were already TX completed by a typical driver.
2547 * But we also want to keep skb->sk set because some packet schedulers
2548 * rely on it (sch_fq for example).
2550 void skb_orphan_partial(struct sk_buff *skb)
2552 if (skb_is_tcp_pure_ack(skb))
2555 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2560 EXPORT_SYMBOL(skb_orphan_partial);
2563 * Read buffer destructor automatically called from kfree_skb.
2565 void sock_rfree(struct sk_buff *skb)
2567 struct sock *sk = skb->sk;
2568 unsigned int len = skb->truesize;
2570 atomic_sub(len, &sk->sk_rmem_alloc);
2571 sk_mem_uncharge(sk, len);
2573 EXPORT_SYMBOL(sock_rfree);
2576 * Buffer destructor for skbs that are not used directly in read or write
2577 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2579 void sock_efree(struct sk_buff *skb)
2583 EXPORT_SYMBOL(sock_efree);
2585 /* Buffer destructor for prefetch/receive path where reference count may
2586 * not be held, e.g. for listen sockets.
2589 void sock_pfree(struct sk_buff *skb)
2591 if (sk_is_refcounted(skb->sk))
2592 sock_gen_put(skb->sk);
2594 EXPORT_SYMBOL(sock_pfree);
2595 #endif /* CONFIG_INET */
2597 kuid_t sock_i_uid(struct sock *sk)
2601 read_lock_bh(&sk->sk_callback_lock);
2602 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2603 read_unlock_bh(&sk->sk_callback_lock);
2606 EXPORT_SYMBOL(sock_i_uid);
2608 unsigned long __sock_i_ino(struct sock *sk)
2612 read_lock(&sk->sk_callback_lock);
2613 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2614 read_unlock(&sk->sk_callback_lock);
2617 EXPORT_SYMBOL(__sock_i_ino);
2619 unsigned long sock_i_ino(struct sock *sk)
2624 ino = __sock_i_ino(sk);
2628 EXPORT_SYMBOL(sock_i_ino);
2631 * Allocate a skb from the socket's send buffer.
2633 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2637 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2638 struct sk_buff *skb = alloc_skb(size, priority);
2641 skb_set_owner_w(skb, sk);
2647 EXPORT_SYMBOL(sock_wmalloc);
2649 static void sock_ofree(struct sk_buff *skb)
2651 struct sock *sk = skb->sk;
2653 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2656 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2659 struct sk_buff *skb;
2661 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2662 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2663 READ_ONCE(sysctl_optmem_max))
2666 skb = alloc_skb(size, priority);
2670 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2672 skb->destructor = sock_ofree;
2677 * Allocate a memory block from the socket's option memory buffer.
2679 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2681 int optmem_max = READ_ONCE(sysctl_optmem_max);
2683 if ((unsigned int)size <= optmem_max &&
2684 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2686 /* First do the add, to avoid the race if kmalloc
2689 atomic_add(size, &sk->sk_omem_alloc);
2690 mem = kmalloc(size, priority);
2693 atomic_sub(size, &sk->sk_omem_alloc);
2697 EXPORT_SYMBOL(sock_kmalloc);
2699 /* Free an option memory block. Note, we actually want the inline
2700 * here as this allows gcc to detect the nullify and fold away the
2701 * condition entirely.
2703 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2706 if (WARN_ON_ONCE(!mem))
2709 kfree_sensitive(mem);
2712 atomic_sub(size, &sk->sk_omem_alloc);
2715 void sock_kfree_s(struct sock *sk, void *mem, int size)
2717 __sock_kfree_s(sk, mem, size, false);
2719 EXPORT_SYMBOL(sock_kfree_s);
2721 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2723 __sock_kfree_s(sk, mem, size, true);
2725 EXPORT_SYMBOL(sock_kzfree_s);
2727 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2728 I think, these locks should be removed for datagram sockets.
2730 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2734 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2738 if (signal_pending(current))
2740 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2741 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2742 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2744 if (sk->sk_shutdown & SEND_SHUTDOWN)
2748 timeo = schedule_timeout(timeo);
2750 finish_wait(sk_sleep(sk), &wait);
2756 * Generic send/receive buffer handlers
2759 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2760 unsigned long data_len, int noblock,
2761 int *errcode, int max_page_order)
2763 struct sk_buff *skb;
2767 timeo = sock_sndtimeo(sk, noblock);
2769 err = sock_error(sk);
2774 if (sk->sk_shutdown & SEND_SHUTDOWN)
2777 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2780 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2781 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2785 if (signal_pending(current))
2787 timeo = sock_wait_for_wmem(sk, timeo);
2789 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2790 errcode, sk->sk_allocation);
2792 skb_set_owner_w(skb, sk);
2796 err = sock_intr_errno(timeo);
2801 EXPORT_SYMBOL(sock_alloc_send_pskb);
2803 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2804 struct sockcm_cookie *sockc)
2808 switch (cmsg->cmsg_type) {
2810 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2811 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2813 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2815 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2817 case SO_TIMESTAMPING_OLD:
2818 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2821 tsflags = *(u32 *)CMSG_DATA(cmsg);
2822 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2825 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2826 sockc->tsflags |= tsflags;
2829 if (!sock_flag(sk, SOCK_TXTIME))
2831 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2833 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2835 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2837 case SCM_CREDENTIALS:
2844 EXPORT_SYMBOL(__sock_cmsg_send);
2846 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2847 struct sockcm_cookie *sockc)
2849 struct cmsghdr *cmsg;
2852 for_each_cmsghdr(cmsg, msg) {
2853 if (!CMSG_OK(msg, cmsg))
2855 if (cmsg->cmsg_level != SOL_SOCKET)
2857 ret = __sock_cmsg_send(sk, cmsg, sockc);
2863 EXPORT_SYMBOL(sock_cmsg_send);
2865 static void sk_enter_memory_pressure(struct sock *sk)
2867 if (!sk->sk_prot->enter_memory_pressure)
2870 sk->sk_prot->enter_memory_pressure(sk);
2873 static void sk_leave_memory_pressure(struct sock *sk)
2875 if (sk->sk_prot->leave_memory_pressure) {
2876 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2877 tcp_leave_memory_pressure, sk);
2879 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2881 if (memory_pressure && READ_ONCE(*memory_pressure))
2882 WRITE_ONCE(*memory_pressure, 0);
2886 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2889 * skb_page_frag_refill - check that a page_frag contains enough room
2890 * @sz: minimum size of the fragment we want to get
2891 * @pfrag: pointer to page_frag
2892 * @gfp: priority for memory allocation
2894 * Note: While this allocator tries to use high order pages, there is
2895 * no guarantee that allocations succeed. Therefore, @sz MUST be
2896 * less or equal than PAGE_SIZE.
2898 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2901 if (page_ref_count(pfrag->page) == 1) {
2905 if (pfrag->offset + sz <= pfrag->size)
2907 put_page(pfrag->page);
2911 if (SKB_FRAG_PAGE_ORDER &&
2912 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2913 /* Avoid direct reclaim but allow kswapd to wake */
2914 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2915 __GFP_COMP | __GFP_NOWARN |
2917 SKB_FRAG_PAGE_ORDER);
2918 if (likely(pfrag->page)) {
2919 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2923 pfrag->page = alloc_page(gfp);
2924 if (likely(pfrag->page)) {
2925 pfrag->size = PAGE_SIZE;
2930 EXPORT_SYMBOL(skb_page_frag_refill);
2932 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2934 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2937 sk_enter_memory_pressure(sk);
2938 sk_stream_moderate_sndbuf(sk);
2941 EXPORT_SYMBOL(sk_page_frag_refill);
2943 void __lock_sock(struct sock *sk)
2944 __releases(&sk->sk_lock.slock)
2945 __acquires(&sk->sk_lock.slock)
2950 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2951 TASK_UNINTERRUPTIBLE);
2952 spin_unlock_bh(&sk->sk_lock.slock);
2954 spin_lock_bh(&sk->sk_lock.slock);
2955 if (!sock_owned_by_user(sk))
2958 finish_wait(&sk->sk_lock.wq, &wait);
2961 void __release_sock(struct sock *sk)
2962 __releases(&sk->sk_lock.slock)
2963 __acquires(&sk->sk_lock.slock)
2965 struct sk_buff *skb, *next;
2967 while ((skb = sk->sk_backlog.head) != NULL) {
2968 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2970 spin_unlock_bh(&sk->sk_lock.slock);
2975 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2976 skb_mark_not_on_list(skb);
2977 sk_backlog_rcv(sk, skb);
2982 } while (skb != NULL);
2984 spin_lock_bh(&sk->sk_lock.slock);
2988 * Doing the zeroing here guarantee we can not loop forever
2989 * while a wild producer attempts to flood us.
2991 sk->sk_backlog.len = 0;
2994 void __sk_flush_backlog(struct sock *sk)
2996 spin_lock_bh(&sk->sk_lock.slock);
2998 spin_unlock_bh(&sk->sk_lock.slock);
3000 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3003 * sk_wait_data - wait for data to arrive at sk_receive_queue
3004 * @sk: sock to wait on
3005 * @timeo: for how long
3006 * @skb: last skb seen on sk_receive_queue
3008 * Now socket state including sk->sk_err is changed only under lock,
3009 * hence we may omit checks after joining wait queue.
3010 * We check receive queue before schedule() only as optimization;
3011 * it is very likely that release_sock() added new data.
3013 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3015 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3018 add_wait_queue(sk_sleep(sk), &wait);
3019 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3020 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3021 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3022 remove_wait_queue(sk_sleep(sk), &wait);
3025 EXPORT_SYMBOL(sk_wait_data);
3028 * __sk_mem_raise_allocated - increase memory_allocated
3030 * @size: memory size to allocate
3031 * @amt: pages to allocate
3032 * @kind: allocation type
3034 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
3036 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3038 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
3039 struct proto *prot = sk->sk_prot;
3040 bool charged = true;
3043 sk_memory_allocated_add(sk, amt);
3044 allocated = sk_memory_allocated(sk);
3046 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3047 gfp_memcg_charge())))
3048 goto suppress_allocation;
3051 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3052 sk_leave_memory_pressure(sk);
3056 /* Under pressure. */
3057 if (allocated > sk_prot_mem_limits(sk, 1))
3058 sk_enter_memory_pressure(sk);
3060 /* Over hard limit. */
3061 if (allocated > sk_prot_mem_limits(sk, 2))
3062 goto suppress_allocation;
3064 /* guarantee minimum buffer size under pressure */
3065 if (kind == SK_MEM_RECV) {
3066 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3069 } else { /* SK_MEM_SEND */
3070 int wmem0 = sk_get_wmem0(sk, prot);
3072 if (sk->sk_type == SOCK_STREAM) {
3073 if (sk->sk_wmem_queued < wmem0)
3075 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3080 if (sk_has_memory_pressure(sk)) {
3083 if (!sk_under_memory_pressure(sk))
3085 alloc = sk_sockets_allocated_read_positive(sk);
3086 if (sk_prot_mem_limits(sk, 2) > alloc *
3087 sk_mem_pages(sk->sk_wmem_queued +
3088 atomic_read(&sk->sk_rmem_alloc) +
3089 sk->sk_forward_alloc))
3093 suppress_allocation:
3095 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3096 sk_stream_moderate_sndbuf(sk);
3098 /* Fail only if socket is _under_ its sndbuf.
3099 * In this case we cannot block, so that we have to fail.
3101 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3102 /* Force charge with __GFP_NOFAIL */
3103 if (memcg_charge && !charged) {
3104 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3105 gfp_memcg_charge() | __GFP_NOFAIL);
3111 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3112 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3114 sk_memory_allocated_sub(sk, amt);
3116 if (memcg_charge && charged)
3117 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
3123 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3125 * @size: memory size to allocate
3126 * @kind: allocation type
3128 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3129 * rmem allocation. This function assumes that protocols which have
3130 * memory_pressure use sk_wmem_queued as write buffer accounting.
3132 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3134 int ret, amt = sk_mem_pages(size);
3136 sk->sk_forward_alloc += amt << PAGE_SHIFT;
3137 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3139 sk->sk_forward_alloc -= amt << PAGE_SHIFT;
3142 EXPORT_SYMBOL(__sk_mem_schedule);
3145 * __sk_mem_reduce_allocated - reclaim memory_allocated
3147 * @amount: number of quanta
3149 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3151 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3153 sk_memory_allocated_sub(sk, amount);
3155 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3156 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3158 if (sk_under_memory_pressure(sk) &&
3159 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3160 sk_leave_memory_pressure(sk);
3164 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3166 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3168 void __sk_mem_reclaim(struct sock *sk, int amount)
3170 amount >>= PAGE_SHIFT;
3171 sk->sk_forward_alloc -= amount << PAGE_SHIFT;
3172 __sk_mem_reduce_allocated(sk, amount);
3174 EXPORT_SYMBOL(__sk_mem_reclaim);
3176 int sk_set_peek_off(struct sock *sk, int val)
3178 sk->sk_peek_off = val;
3181 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3184 * Set of default routines for initialising struct proto_ops when
3185 * the protocol does not support a particular function. In certain
3186 * cases where it makes no sense for a protocol to have a "do nothing"
3187 * function, some default processing is provided.
3190 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3194 EXPORT_SYMBOL(sock_no_bind);
3196 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3201 EXPORT_SYMBOL(sock_no_connect);
3203 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3207 EXPORT_SYMBOL(sock_no_socketpair);
3209 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3214 EXPORT_SYMBOL(sock_no_accept);
3216 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3221 EXPORT_SYMBOL(sock_no_getname);
3223 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3227 EXPORT_SYMBOL(sock_no_ioctl);
3229 int sock_no_listen(struct socket *sock, int backlog)
3233 EXPORT_SYMBOL(sock_no_listen);
3235 int sock_no_shutdown(struct socket *sock, int how)
3239 EXPORT_SYMBOL(sock_no_shutdown);
3241 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3245 EXPORT_SYMBOL(sock_no_sendmsg);
3247 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3251 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3253 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3258 EXPORT_SYMBOL(sock_no_recvmsg);
3260 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3262 /* Mirror missing mmap method error code */
3265 EXPORT_SYMBOL(sock_no_mmap);
3268 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3269 * various sock-based usage counts.
3271 void __receive_sock(struct file *file)
3273 struct socket *sock;
3275 sock = sock_from_file(file);
3277 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3278 sock_update_classid(&sock->sk->sk_cgrp_data);
3283 * Default Socket Callbacks
3286 static void sock_def_wakeup(struct sock *sk)
3288 struct socket_wq *wq;
3291 wq = rcu_dereference(sk->sk_wq);
3292 if (skwq_has_sleeper(wq))
3293 wake_up_interruptible_all(&wq->wait);
3297 static void sock_def_error_report(struct sock *sk)
3299 struct socket_wq *wq;
3302 wq = rcu_dereference(sk->sk_wq);
3303 if (skwq_has_sleeper(wq))
3304 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3305 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3309 void sock_def_readable(struct sock *sk)
3311 struct socket_wq *wq;
3313 trace_sk_data_ready(sk);
3316 wq = rcu_dereference(sk->sk_wq);
3317 if (skwq_has_sleeper(wq))
3318 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3319 EPOLLRDNORM | EPOLLRDBAND);
3320 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3324 static void sock_def_write_space(struct sock *sk)
3326 struct socket_wq *wq;
3330 /* Do not wake up a writer until he can make "significant"
3333 if (sock_writeable(sk)) {
3334 wq = rcu_dereference(sk->sk_wq);
3335 if (skwq_has_sleeper(wq))
3336 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3337 EPOLLWRNORM | EPOLLWRBAND);
3339 /* Should agree with poll, otherwise some programs break */
3340 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3346 /* An optimised version of sock_def_write_space(), should only be called
3347 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3350 static void sock_def_write_space_wfree(struct sock *sk)
3352 /* Do not wake up a writer until he can make "significant"
3355 if (sock_writeable(sk)) {
3356 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3358 /* rely on refcount_sub from sock_wfree() */
3359 smp_mb__after_atomic();
3360 if (wq && waitqueue_active(&wq->wait))
3361 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3362 EPOLLWRNORM | EPOLLWRBAND);
3364 /* Should agree with poll, otherwise some programs break */
3365 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3369 static void sock_def_destruct(struct sock *sk)
3373 void sk_send_sigurg(struct sock *sk)
3375 if (sk->sk_socket && sk->sk_socket->file)
3376 if (send_sigurg(&sk->sk_socket->file->f_owner))
3377 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3379 EXPORT_SYMBOL(sk_send_sigurg);
3381 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3382 unsigned long expires)
3384 if (!mod_timer(timer, expires))
3387 EXPORT_SYMBOL(sk_reset_timer);
3389 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3391 if (del_timer(timer))
3394 EXPORT_SYMBOL(sk_stop_timer);
3396 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3398 if (del_timer_sync(timer))
3401 EXPORT_SYMBOL(sk_stop_timer_sync);
3403 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3406 sk->sk_send_head = NULL;
3408 timer_setup(&sk->sk_timer, NULL, 0);
3410 sk->sk_allocation = GFP_KERNEL;
3411 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3412 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3413 sk->sk_state = TCP_CLOSE;
3414 sk->sk_use_task_frag = true;
3415 sk_set_socket(sk, sock);
3417 sock_set_flag(sk, SOCK_ZAPPED);
3420 sk->sk_type = sock->type;
3421 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3424 RCU_INIT_POINTER(sk->sk_wq, NULL);
3428 rwlock_init(&sk->sk_callback_lock);
3429 if (sk->sk_kern_sock)
3430 lockdep_set_class_and_name(
3431 &sk->sk_callback_lock,
3432 af_kern_callback_keys + sk->sk_family,
3433 af_family_kern_clock_key_strings[sk->sk_family]);
3435 lockdep_set_class_and_name(
3436 &sk->sk_callback_lock,
3437 af_callback_keys + sk->sk_family,
3438 af_family_clock_key_strings[sk->sk_family]);
3440 sk->sk_state_change = sock_def_wakeup;
3441 sk->sk_data_ready = sock_def_readable;
3442 sk->sk_write_space = sock_def_write_space;
3443 sk->sk_error_report = sock_def_error_report;
3444 sk->sk_destruct = sock_def_destruct;
3446 sk->sk_frag.page = NULL;
3447 sk->sk_frag.offset = 0;
3448 sk->sk_peek_off = -1;
3450 sk->sk_peer_pid = NULL;
3451 sk->sk_peer_cred = NULL;
3452 spin_lock_init(&sk->sk_peer_lock);
3454 sk->sk_write_pending = 0;
3455 sk->sk_rcvlowat = 1;
3456 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3457 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3459 sk->sk_stamp = SK_DEFAULT_STAMP;
3460 #if BITS_PER_LONG==32
3461 seqlock_init(&sk->sk_stamp_seq);
3463 atomic_set(&sk->sk_zckey, 0);
3465 #ifdef CONFIG_NET_RX_BUSY_POLL
3467 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3470 sk->sk_max_pacing_rate = ~0UL;
3471 sk->sk_pacing_rate = ~0UL;
3472 WRITE_ONCE(sk->sk_pacing_shift, 10);
3473 sk->sk_incoming_cpu = -1;
3475 sk_rx_queue_clear(sk);
3477 * Before updating sk_refcnt, we must commit prior changes to memory
3478 * (Documentation/RCU/rculist_nulls.rst for details)
3481 refcount_set(&sk->sk_refcnt, 1);
3482 atomic_set(&sk->sk_drops, 0);
3484 EXPORT_SYMBOL(sock_init_data_uid);
3486 void sock_init_data(struct socket *sock, struct sock *sk)
3489 SOCK_INODE(sock)->i_uid :
3490 make_kuid(sock_net(sk)->user_ns, 0);
3492 sock_init_data_uid(sock, sk, uid);
3494 EXPORT_SYMBOL(sock_init_data);
3496 void lock_sock_nested(struct sock *sk, int subclass)
3498 /* The sk_lock has mutex_lock() semantics here. */
3499 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3502 spin_lock_bh(&sk->sk_lock.slock);
3503 if (sock_owned_by_user_nocheck(sk))
3505 sk->sk_lock.owned = 1;
3506 spin_unlock_bh(&sk->sk_lock.slock);
3508 EXPORT_SYMBOL(lock_sock_nested);
3510 void release_sock(struct sock *sk)
3512 spin_lock_bh(&sk->sk_lock.slock);
3513 if (sk->sk_backlog.tail)
3516 /* Warning : release_cb() might need to release sk ownership,
3517 * ie call sock_release_ownership(sk) before us.
3519 if (sk->sk_prot->release_cb)
3520 sk->sk_prot->release_cb(sk);
3522 sock_release_ownership(sk);
3523 if (waitqueue_active(&sk->sk_lock.wq))
3524 wake_up(&sk->sk_lock.wq);
3525 spin_unlock_bh(&sk->sk_lock.slock);
3527 EXPORT_SYMBOL(release_sock);
3529 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3532 spin_lock_bh(&sk->sk_lock.slock);
3534 if (!sock_owned_by_user_nocheck(sk)) {
3536 * Fast path return with bottom halves disabled and
3537 * sock::sk_lock.slock held.
3539 * The 'mutex' is not contended and holding
3540 * sock::sk_lock.slock prevents all other lockers to
3541 * proceed so the corresponding unlock_sock_fast() can
3542 * avoid the slow path of release_sock() completely and
3543 * just release slock.
3545 * From a semantical POV this is equivalent to 'acquiring'
3546 * the 'mutex', hence the corresponding lockdep
3547 * mutex_release() has to happen in the fast path of
3548 * unlock_sock_fast().
3554 sk->sk_lock.owned = 1;
3555 __acquire(&sk->sk_lock.slock);
3556 spin_unlock_bh(&sk->sk_lock.slock);
3559 EXPORT_SYMBOL(__lock_sock_fast);
3561 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3562 bool timeval, bool time32)
3564 struct sock *sk = sock->sk;
3565 struct timespec64 ts;
3567 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3568 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3569 if (ts.tv_sec == -1)
3571 if (ts.tv_sec == 0) {
3572 ktime_t kt = ktime_get_real();
3573 sock_write_timestamp(sk, kt);
3574 ts = ktime_to_timespec64(kt);
3580 #ifdef CONFIG_COMPAT_32BIT_TIME
3582 return put_old_timespec32(&ts, userstamp);
3584 #ifdef CONFIG_SPARC64
3585 /* beware of padding in sparc64 timeval */
3586 if (timeval && !in_compat_syscall()) {
3587 struct __kernel_old_timeval __user tv = {
3588 .tv_sec = ts.tv_sec,
3589 .tv_usec = ts.tv_nsec,
3591 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3596 return put_timespec64(&ts, userstamp);
3598 EXPORT_SYMBOL(sock_gettstamp);
3600 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3602 if (!sock_flag(sk, flag)) {
3603 unsigned long previous_flags = sk->sk_flags;
3605 sock_set_flag(sk, flag);
3607 * we just set one of the two flags which require net
3608 * time stamping, but time stamping might have been on
3609 * already because of the other one
3611 if (sock_needs_netstamp(sk) &&
3612 !(previous_flags & SK_FLAGS_TIMESTAMP))
3613 net_enable_timestamp();
3617 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3618 int level, int type)
3620 struct sock_exterr_skb *serr;
3621 struct sk_buff *skb;
3625 skb = sock_dequeue_err_skb(sk);
3631 msg->msg_flags |= MSG_TRUNC;
3634 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3638 sock_recv_timestamp(msg, sk, skb);
3640 serr = SKB_EXT_ERR(skb);
3641 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3643 msg->msg_flags |= MSG_ERRQUEUE;
3651 EXPORT_SYMBOL(sock_recv_errqueue);
3654 * Get a socket option on an socket.
3656 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3657 * asynchronous errors should be reported by getsockopt. We assume
3658 * this means if you specify SO_ERROR (otherwise whats the point of it).
3660 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3661 char __user *optval, int __user *optlen)
3663 struct sock *sk = sock->sk;
3665 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3666 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3668 EXPORT_SYMBOL(sock_common_getsockopt);
3670 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3673 struct sock *sk = sock->sk;
3677 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3679 msg->msg_namelen = addr_len;
3682 EXPORT_SYMBOL(sock_common_recvmsg);
3685 * Set socket options on an inet socket.
3687 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3688 sockptr_t optval, unsigned int optlen)
3690 struct sock *sk = sock->sk;
3692 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3693 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3695 EXPORT_SYMBOL(sock_common_setsockopt);
3697 void sk_common_release(struct sock *sk)
3699 if (sk->sk_prot->destroy)
3700 sk->sk_prot->destroy(sk);
3703 * Observation: when sk_common_release is called, processes have
3704 * no access to socket. But net still has.
3705 * Step one, detach it from networking:
3707 * A. Remove from hash tables.
3710 sk->sk_prot->unhash(sk);
3713 * In this point socket cannot receive new packets, but it is possible
3714 * that some packets are in flight because some CPU runs receiver and
3715 * did hash table lookup before we unhashed socket. They will achieve
3716 * receive queue and will be purged by socket destructor.
3718 * Also we still have packets pending on receive queue and probably,
3719 * our own packets waiting in device queues. sock_destroy will drain
3720 * receive queue, but transmitted packets will delay socket destruction
3721 * until the last reference will be released.
3726 xfrm_sk_free_policy(sk);
3730 EXPORT_SYMBOL(sk_common_release);
3732 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3734 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3736 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3737 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3738 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3739 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3740 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3741 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3742 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3743 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3744 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3747 #ifdef CONFIG_PROC_FS
3748 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3750 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3752 int cpu, idx = prot->inuse_idx;
3755 for_each_possible_cpu(cpu)
3756 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3758 return res >= 0 ? res : 0;
3760 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3762 int sock_inuse_get(struct net *net)
3766 for_each_possible_cpu(cpu)
3767 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3772 EXPORT_SYMBOL_GPL(sock_inuse_get);
3774 static int __net_init sock_inuse_init_net(struct net *net)
3776 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3777 if (net->core.prot_inuse == NULL)
3782 static void __net_exit sock_inuse_exit_net(struct net *net)
3784 free_percpu(net->core.prot_inuse);
3787 static struct pernet_operations net_inuse_ops = {
3788 .init = sock_inuse_init_net,
3789 .exit = sock_inuse_exit_net,
3792 static __init int net_inuse_init(void)
3794 if (register_pernet_subsys(&net_inuse_ops))
3795 panic("Cannot initialize net inuse counters");
3800 core_initcall(net_inuse_init);
3802 static int assign_proto_idx(struct proto *prot)
3804 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3806 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3807 pr_err("PROTO_INUSE_NR exhausted\n");
3811 set_bit(prot->inuse_idx, proto_inuse_idx);
3815 static void release_proto_idx(struct proto *prot)
3817 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3818 clear_bit(prot->inuse_idx, proto_inuse_idx);
3821 static inline int assign_proto_idx(struct proto *prot)
3826 static inline void release_proto_idx(struct proto *prot)
3832 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3836 kfree(twsk_prot->twsk_slab_name);
3837 twsk_prot->twsk_slab_name = NULL;
3838 kmem_cache_destroy(twsk_prot->twsk_slab);
3839 twsk_prot->twsk_slab = NULL;
3842 static int tw_prot_init(const struct proto *prot)
3844 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3849 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3851 if (!twsk_prot->twsk_slab_name)
3854 twsk_prot->twsk_slab =
3855 kmem_cache_create(twsk_prot->twsk_slab_name,
3856 twsk_prot->twsk_obj_size, 0,
3857 SLAB_ACCOUNT | prot->slab_flags,
3859 if (!twsk_prot->twsk_slab) {
3860 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3868 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3872 kfree(rsk_prot->slab_name);
3873 rsk_prot->slab_name = NULL;
3874 kmem_cache_destroy(rsk_prot->slab);
3875 rsk_prot->slab = NULL;
3878 static int req_prot_init(const struct proto *prot)
3880 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3885 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3887 if (!rsk_prot->slab_name)
3890 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3891 rsk_prot->obj_size, 0,
3892 SLAB_ACCOUNT | prot->slab_flags,
3895 if (!rsk_prot->slab) {
3896 pr_crit("%s: Can't create request sock SLAB cache!\n",
3903 int proto_register(struct proto *prot, int alloc_slab)
3907 if (prot->memory_allocated && !prot->sysctl_mem) {
3908 pr_err("%s: missing sysctl_mem\n", prot->name);
3911 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3912 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3916 prot->slab = kmem_cache_create_usercopy(prot->name,
3918 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3920 prot->useroffset, prot->usersize,
3923 if (prot->slab == NULL) {
3924 pr_crit("%s: Can't create sock SLAB cache!\n",
3929 if (req_prot_init(prot))
3930 goto out_free_request_sock_slab;
3932 if (tw_prot_init(prot))
3933 goto out_free_timewait_sock_slab;
3936 mutex_lock(&proto_list_mutex);
3937 ret = assign_proto_idx(prot);
3939 mutex_unlock(&proto_list_mutex);
3940 goto out_free_timewait_sock_slab;
3942 list_add(&prot->node, &proto_list);
3943 mutex_unlock(&proto_list_mutex);
3946 out_free_timewait_sock_slab:
3948 tw_prot_cleanup(prot->twsk_prot);
3949 out_free_request_sock_slab:
3951 req_prot_cleanup(prot->rsk_prot);
3953 kmem_cache_destroy(prot->slab);
3959 EXPORT_SYMBOL(proto_register);
3961 void proto_unregister(struct proto *prot)
3963 mutex_lock(&proto_list_mutex);
3964 release_proto_idx(prot);
3965 list_del(&prot->node);
3966 mutex_unlock(&proto_list_mutex);
3968 kmem_cache_destroy(prot->slab);
3971 req_prot_cleanup(prot->rsk_prot);
3972 tw_prot_cleanup(prot->twsk_prot);
3974 EXPORT_SYMBOL(proto_unregister);
3976 int sock_load_diag_module(int family, int protocol)
3979 if (!sock_is_registered(family))
3982 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3983 NETLINK_SOCK_DIAG, family);
3987 if (family == AF_INET &&
3988 protocol != IPPROTO_RAW &&
3989 protocol < MAX_INET_PROTOS &&
3990 !rcu_access_pointer(inet_protos[protocol]))
3994 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3995 NETLINK_SOCK_DIAG, family, protocol);
3997 EXPORT_SYMBOL(sock_load_diag_module);
3999 #ifdef CONFIG_PROC_FS
4000 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4001 __acquires(proto_list_mutex)
4003 mutex_lock(&proto_list_mutex);
4004 return seq_list_start_head(&proto_list, *pos);
4007 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4009 return seq_list_next(v, &proto_list, pos);
4012 static void proto_seq_stop(struct seq_file *seq, void *v)
4013 __releases(proto_list_mutex)
4015 mutex_unlock(&proto_list_mutex);
4018 static char proto_method_implemented(const void *method)
4020 return method == NULL ? 'n' : 'y';
4022 static long sock_prot_memory_allocated(struct proto *proto)
4024 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4027 static const char *sock_prot_memory_pressure(struct proto *proto)
4029 return proto->memory_pressure != NULL ?
4030 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4033 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4036 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4037 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4040 sock_prot_inuse_get(seq_file_net(seq), proto),
4041 sock_prot_memory_allocated(proto),
4042 sock_prot_memory_pressure(proto),
4044 proto->slab == NULL ? "no" : "yes",
4045 module_name(proto->owner),
4046 proto_method_implemented(proto->close),
4047 proto_method_implemented(proto->connect),
4048 proto_method_implemented(proto->disconnect),
4049 proto_method_implemented(proto->accept),
4050 proto_method_implemented(proto->ioctl),
4051 proto_method_implemented(proto->init),
4052 proto_method_implemented(proto->destroy),
4053 proto_method_implemented(proto->shutdown),
4054 proto_method_implemented(proto->setsockopt),
4055 proto_method_implemented(proto->getsockopt),
4056 proto_method_implemented(proto->sendmsg),
4057 proto_method_implemented(proto->recvmsg),
4058 proto_method_implemented(proto->bind),
4059 proto_method_implemented(proto->backlog_rcv),
4060 proto_method_implemented(proto->hash),
4061 proto_method_implemented(proto->unhash),
4062 proto_method_implemented(proto->get_port),
4063 proto_method_implemented(proto->enter_memory_pressure));
4066 static int proto_seq_show(struct seq_file *seq, void *v)
4068 if (v == &proto_list)
4069 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4078 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4080 proto_seq_printf(seq, list_entry(v, struct proto, node));
4084 static const struct seq_operations proto_seq_ops = {
4085 .start = proto_seq_start,
4086 .next = proto_seq_next,
4087 .stop = proto_seq_stop,
4088 .show = proto_seq_show,
4091 static __net_init int proto_init_net(struct net *net)
4093 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4094 sizeof(struct seq_net_private)))
4100 static __net_exit void proto_exit_net(struct net *net)
4102 remove_proc_entry("protocols", net->proc_net);
4106 static __net_initdata struct pernet_operations proto_net_ops = {
4107 .init = proto_init_net,
4108 .exit = proto_exit_net,
4111 static int __init proto_init(void)
4113 return register_pernet_subsys(&proto_net_ops);
4116 subsys_initcall(proto_init);
4118 #endif /* PROC_FS */
4120 #ifdef CONFIG_NET_RX_BUSY_POLL
4121 bool sk_busy_loop_end(void *p, unsigned long start_time)
4123 struct sock *sk = p;
4125 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
4126 sk_busy_loop_timeout(sk, start_time);
4128 EXPORT_SYMBOL(sk_busy_loop_end);
4129 #endif /* CONFIG_NET_RX_BUSY_POLL */
4131 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4133 if (!sk->sk_prot->bind_add)
4135 return sk->sk_prot->bind_add(sk, addr, addr_len);
4137 EXPORT_SYMBOL(sock_bind_add);
4139 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4140 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4141 void __user *arg, void *karg, size_t size)
4145 if (copy_from_user(karg, arg, size))
4148 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4152 if (copy_to_user(arg, karg, size))
4157 EXPORT_SYMBOL(sock_ioctl_inout);
4159 /* This is the most common ioctl prep function, where the result (4 bytes) is
4160 * copied back to userspace if the ioctl() returns successfully. No input is
4161 * copied from userspace as input argument.
4163 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4167 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4171 return put_user(karg, (int __user *)arg);
4174 /* A wrapper around sock ioctls, which copies the data from userspace
4175 * (depending on the protocol/ioctl), and copies back the result to userspace.
4176 * The main motivation for this function is to pass kernel memory to the
4177 * protocol ioctl callbacks, instead of userspace memory.
4179 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4183 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4184 rc = ipmr_sk_ioctl(sk, cmd, arg);
4185 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4186 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4187 else if (sk_is_phonet(sk))
4188 rc = phonet_sk_ioctl(sk, cmd, arg);
4190 /* If ioctl was processed, returns its value */
4194 /* Otherwise call the default handler */
4195 return sock_ioctl_out(sk, cmd, arg);
4197 EXPORT_SYMBOL(sk_ioctl);