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>
118 #include <linux/uaccess.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
137 #include <trace/events/sock.h>
140 #include <net/busy_poll.h>
142 #include <linux/ethtool.h>
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
149 static void sock_def_write_space_wfree(struct sock *sk);
150 static void sock_def_write_space(struct sock *sk);
153 * sk_ns_capable - General socket capability test
154 * @sk: Socket to use a capability on or through
155 * @user_ns: The user namespace of the capability to use
156 * @cap: The capability to use
158 * Test to see if the opener of the socket had when the socket was
159 * created and the current process has the capability @cap in the user
160 * namespace @user_ns.
162 bool sk_ns_capable(const struct sock *sk,
163 struct user_namespace *user_ns, int cap)
165 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
166 ns_capable(user_ns, cap);
168 EXPORT_SYMBOL(sk_ns_capable);
171 * sk_capable - Socket global capability test
172 * @sk: Socket to use a capability on or through
173 * @cap: The global capability to use
175 * Test to see if the opener of the socket had when the socket was
176 * created and the current process has the capability @cap in all user
179 bool sk_capable(const struct sock *sk, int cap)
181 return sk_ns_capable(sk, &init_user_ns, cap);
183 EXPORT_SYMBOL(sk_capable);
186 * sk_net_capable - Network namespace socket capability test
187 * @sk: Socket to use a capability on or through
188 * @cap: The capability to use
190 * Test to see if the opener of the socket had when the socket was created
191 * and the current process has the capability @cap over the network namespace
192 * the socket is a member of.
194 bool sk_net_capable(const struct sock *sk, int cap)
196 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
198 EXPORT_SYMBOL(sk_net_capable);
201 * Each address family might have different locking rules, so we have
202 * one slock key per address family and separate keys for internal and
205 static struct lock_class_key af_family_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_keys[AF_MAX];
207 static struct lock_class_key af_family_slock_keys[AF_MAX];
208 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
211 * Make lock validator output more readable. (we pre-construct these
212 * strings build-time, so that runtime initialization of socket
216 #define _sock_locks(x) \
217 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
218 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
219 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
220 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
221 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
222 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
223 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
224 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
225 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
226 x "27" , x "28" , x "AF_CAN" , \
227 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
228 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
229 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
230 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
231 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
235 static const char *const af_family_key_strings[AF_MAX+1] = {
236 _sock_locks("sk_lock-")
238 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
239 _sock_locks("slock-")
241 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
242 _sock_locks("clock-")
245 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
246 _sock_locks("k-sk_lock-")
248 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
249 _sock_locks("k-slock-")
251 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
252 _sock_locks("k-clock-")
254 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
255 _sock_locks("rlock-")
257 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
258 _sock_locks("wlock-")
260 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
261 _sock_locks("elock-")
265 * sk_callback_lock and sk queues locking rules are per-address-family,
266 * so split the lock classes by using a per-AF key:
268 static struct lock_class_key af_callback_keys[AF_MAX];
269 static struct lock_class_key af_rlock_keys[AF_MAX];
270 static struct lock_class_key af_wlock_keys[AF_MAX];
271 static struct lock_class_key af_elock_keys[AF_MAX];
272 static struct lock_class_key af_kern_callback_keys[AF_MAX];
274 /* Run time adjustable parameters. */
275 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
276 EXPORT_SYMBOL(sysctl_wmem_max);
277 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
278 EXPORT_SYMBOL(sysctl_rmem_max);
279 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
280 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
282 /* Maximal space eaten by iovec or ancillary data plus some space */
283 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
284 EXPORT_SYMBOL(sysctl_optmem_max);
286 int sysctl_tstamp_allow_data __read_mostly = 1;
288 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
289 EXPORT_SYMBOL_GPL(memalloc_socks_key);
292 * sk_set_memalloc - sets %SOCK_MEMALLOC
293 * @sk: socket to set it on
295 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
296 * It's the responsibility of the admin to adjust min_free_kbytes
297 * to meet the requirements
299 void sk_set_memalloc(struct sock *sk)
301 sock_set_flag(sk, SOCK_MEMALLOC);
302 sk->sk_allocation |= __GFP_MEMALLOC;
303 static_branch_inc(&memalloc_socks_key);
305 EXPORT_SYMBOL_GPL(sk_set_memalloc);
307 void sk_clear_memalloc(struct sock *sk)
309 sock_reset_flag(sk, SOCK_MEMALLOC);
310 sk->sk_allocation &= ~__GFP_MEMALLOC;
311 static_branch_dec(&memalloc_socks_key);
314 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
315 * progress of swapping. SOCK_MEMALLOC may be cleared while
316 * it has rmem allocations due to the last swapfile being deactivated
317 * but there is a risk that the socket is unusable due to exceeding
318 * the rmem limits. Reclaim the reserves and obey rmem limits again.
322 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
324 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
327 unsigned int noreclaim_flag;
329 /* these should have been dropped before queueing */
330 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
332 noreclaim_flag = memalloc_noreclaim_save();
333 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
337 memalloc_noreclaim_restore(noreclaim_flag);
341 EXPORT_SYMBOL(__sk_backlog_rcv);
343 void sk_error_report(struct sock *sk)
345 sk->sk_error_report(sk);
347 switch (sk->sk_family) {
351 trace_inet_sk_error_report(sk);
357 EXPORT_SYMBOL(sk_error_report);
359 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
361 struct __kernel_sock_timeval tv;
363 if (timeo == MAX_SCHEDULE_TIMEOUT) {
367 tv.tv_sec = timeo / HZ;
368 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
371 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
372 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
373 *(struct old_timeval32 *)optval = tv32;
378 struct __kernel_old_timeval old_tv;
379 old_tv.tv_sec = tv.tv_sec;
380 old_tv.tv_usec = tv.tv_usec;
381 *(struct __kernel_old_timeval *)optval = old_tv;
382 return sizeof(old_tv);
385 *(struct __kernel_sock_timeval *)optval = tv;
388 EXPORT_SYMBOL(sock_get_timeout);
390 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
391 sockptr_t optval, int optlen, bool old_timeval)
393 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
394 struct old_timeval32 tv32;
396 if (optlen < sizeof(tv32))
399 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
401 tv->tv_sec = tv32.tv_sec;
402 tv->tv_usec = tv32.tv_usec;
403 } else if (old_timeval) {
404 struct __kernel_old_timeval old_tv;
406 if (optlen < sizeof(old_tv))
408 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
410 tv->tv_sec = old_tv.tv_sec;
411 tv->tv_usec = old_tv.tv_usec;
413 if (optlen < sizeof(*tv))
415 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
421 EXPORT_SYMBOL(sock_copy_user_timeval);
423 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
426 struct __kernel_sock_timeval tv;
427 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
432 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
436 static int warned __read_mostly;
439 if (warned < 10 && net_ratelimit()) {
441 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
442 __func__, current->comm, task_pid_nr(current));
446 *timeo_p = MAX_SCHEDULE_TIMEOUT;
447 if (tv.tv_sec == 0 && tv.tv_usec == 0)
449 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
450 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
454 static bool sock_needs_netstamp(const struct sock *sk)
456 switch (sk->sk_family) {
465 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
467 if (sk->sk_flags & flags) {
468 sk->sk_flags &= ~flags;
469 if (sock_needs_netstamp(sk) &&
470 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
471 net_disable_timestamp();
476 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
479 struct sk_buff_head *list = &sk->sk_receive_queue;
481 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
482 atomic_inc(&sk->sk_drops);
483 trace_sock_rcvqueue_full(sk, skb);
487 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
488 atomic_inc(&sk->sk_drops);
493 skb_set_owner_r(skb, sk);
495 /* we escape from rcu protected region, make sure we dont leak
500 spin_lock_irqsave(&list->lock, flags);
501 sock_skb_set_dropcount(sk, skb);
502 __skb_queue_tail(list, skb);
503 spin_unlock_irqrestore(&list->lock, flags);
505 if (!sock_flag(sk, SOCK_DEAD))
506 sk->sk_data_ready(sk);
509 EXPORT_SYMBOL(__sock_queue_rcv_skb);
511 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
512 enum skb_drop_reason *reason)
514 enum skb_drop_reason drop_reason;
517 err = sk_filter(sk, skb);
519 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
522 err = __sock_queue_rcv_skb(sk, skb);
525 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
528 drop_reason = SKB_DROP_REASON_PROTO_MEM;
531 drop_reason = SKB_NOT_DROPPED_YET;
536 *reason = drop_reason;
539 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
541 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
542 const int nested, unsigned int trim_cap, bool refcounted)
544 int rc = NET_RX_SUCCESS;
546 if (sk_filter_trim_cap(sk, skb, trim_cap))
547 goto discard_and_relse;
551 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
552 atomic_inc(&sk->sk_drops);
553 goto discard_and_relse;
556 bh_lock_sock_nested(sk);
559 if (!sock_owned_by_user(sk)) {
561 * trylock + unlock semantics:
563 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
565 rc = sk_backlog_rcv(sk, skb);
567 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
568 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
570 atomic_inc(&sk->sk_drops);
571 goto discard_and_relse;
583 EXPORT_SYMBOL(__sk_receive_skb);
585 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
587 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
589 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
591 struct dst_entry *dst = __sk_dst_get(sk);
593 if (dst && dst->obsolete &&
594 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
595 dst, cookie) == NULL) {
596 sk_tx_queue_clear(sk);
597 sk->sk_dst_pending_confirm = 0;
598 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
605 EXPORT_SYMBOL(__sk_dst_check);
607 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
609 struct dst_entry *dst = sk_dst_get(sk);
611 if (dst && dst->obsolete &&
612 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
613 dst, cookie) == NULL) {
621 EXPORT_SYMBOL(sk_dst_check);
623 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
625 int ret = -ENOPROTOOPT;
626 #ifdef CONFIG_NETDEVICES
627 struct net *net = sock_net(sk);
631 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
638 /* Paired with all READ_ONCE() done locklessly. */
639 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
641 if (sk->sk_prot->rehash)
642 sk->sk_prot->rehash(sk);
653 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
659 ret = sock_bindtoindex_locked(sk, ifindex);
665 EXPORT_SYMBOL(sock_bindtoindex);
667 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
669 int ret = -ENOPROTOOPT;
670 #ifdef CONFIG_NETDEVICES
671 struct net *net = sock_net(sk);
672 char devname[IFNAMSIZ];
679 /* Bind this socket to a particular device like "eth0",
680 * as specified in the passed interface name. If the
681 * name is "" or the option length is zero the socket
684 if (optlen > IFNAMSIZ - 1)
685 optlen = IFNAMSIZ - 1;
686 memset(devname, 0, sizeof(devname));
689 if (copy_from_sockptr(devname, optval, optlen))
693 if (devname[0] != '\0') {
694 struct net_device *dev;
697 dev = dev_get_by_name_rcu(net, devname);
699 index = dev->ifindex;
706 sockopt_lock_sock(sk);
707 ret = sock_bindtoindex_locked(sk, index);
708 sockopt_release_sock(sk);
715 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
716 sockptr_t optlen, int len)
718 int ret = -ENOPROTOOPT;
719 #ifdef CONFIG_NETDEVICES
720 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
721 struct net *net = sock_net(sk);
722 char devname[IFNAMSIZ];
724 if (bound_dev_if == 0) {
733 ret = netdev_get_name(net, devname, bound_dev_if);
737 len = strlen(devname) + 1;
740 if (copy_to_sockptr(optval, devname, len))
745 if (copy_to_sockptr(optlen, &len, sizeof(int)))
756 bool sk_mc_loop(struct sock *sk)
758 if (dev_recursion_level())
762 switch (sk->sk_family) {
764 return inet_sk(sk)->mc_loop;
765 #if IS_ENABLED(CONFIG_IPV6)
767 return inet6_sk(sk)->mc_loop;
773 EXPORT_SYMBOL(sk_mc_loop);
775 void sock_set_reuseaddr(struct sock *sk)
778 sk->sk_reuse = SK_CAN_REUSE;
781 EXPORT_SYMBOL(sock_set_reuseaddr);
783 void sock_set_reuseport(struct sock *sk)
786 sk->sk_reuseport = true;
789 EXPORT_SYMBOL(sock_set_reuseport);
791 void sock_no_linger(struct sock *sk)
794 sk->sk_lingertime = 0;
795 sock_set_flag(sk, SOCK_LINGER);
798 EXPORT_SYMBOL(sock_no_linger);
800 void sock_set_priority(struct sock *sk, u32 priority)
803 sk->sk_priority = priority;
806 EXPORT_SYMBOL(sock_set_priority);
808 void sock_set_sndtimeo(struct sock *sk, s64 secs)
811 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
812 sk->sk_sndtimeo = secs * HZ;
814 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
817 EXPORT_SYMBOL(sock_set_sndtimeo);
819 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
822 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
823 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
824 sock_set_flag(sk, SOCK_RCVTSTAMP);
825 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
827 sock_reset_flag(sk, SOCK_RCVTSTAMP);
828 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
832 void sock_enable_timestamps(struct sock *sk)
835 __sock_set_timestamps(sk, true, false, true);
838 EXPORT_SYMBOL(sock_enable_timestamps);
840 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
843 case SO_TIMESTAMP_OLD:
844 __sock_set_timestamps(sk, valbool, false, false);
846 case SO_TIMESTAMP_NEW:
847 __sock_set_timestamps(sk, valbool, true, false);
849 case SO_TIMESTAMPNS_OLD:
850 __sock_set_timestamps(sk, valbool, false, true);
852 case SO_TIMESTAMPNS_NEW:
853 __sock_set_timestamps(sk, valbool, true, true);
858 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
860 struct net *net = sock_net(sk);
861 struct net_device *dev = NULL;
866 if (sk->sk_bound_dev_if)
867 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
870 pr_err("%s: sock not bind to device\n", __func__);
874 num = ethtool_get_phc_vclocks(dev, &vclock_index);
877 for (i = 0; i < num; i++) {
878 if (*(vclock_index + i) == phc_index) {
890 sk->sk_bind_phc = phc_index;
895 int sock_set_timestamping(struct sock *sk, int optname,
896 struct so_timestamping timestamping)
898 int val = timestamping.flags;
901 if (val & ~SOF_TIMESTAMPING_MASK)
904 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
905 !(val & SOF_TIMESTAMPING_OPT_ID))
908 if (val & SOF_TIMESTAMPING_OPT_ID &&
909 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
911 if ((1 << sk->sk_state) &
912 (TCPF_CLOSE | TCPF_LISTEN))
914 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
915 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
917 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
919 atomic_set(&sk->sk_tskey, 0);
923 if (val & SOF_TIMESTAMPING_OPT_STATS &&
924 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
927 if (val & SOF_TIMESTAMPING_BIND_PHC) {
928 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
933 sk->sk_tsflags = val;
934 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
936 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
937 sock_enable_timestamp(sk,
938 SOCK_TIMESTAMPING_RX_SOFTWARE);
940 sock_disable_timestamp(sk,
941 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
945 void sock_set_keepalive(struct sock *sk)
948 if (sk->sk_prot->keepalive)
949 sk->sk_prot->keepalive(sk, true);
950 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
953 EXPORT_SYMBOL(sock_set_keepalive);
955 static void __sock_set_rcvbuf(struct sock *sk, int val)
957 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
958 * as a negative value.
960 val = min_t(int, val, INT_MAX / 2);
961 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
963 /* We double it on the way in to account for "struct sk_buff" etc.
964 * overhead. Applications assume that the SO_RCVBUF setting they make
965 * will allow that much actual data to be received on that socket.
967 * Applications are unaware that "struct sk_buff" and other overheads
968 * allocate from the receive buffer during socket buffer allocation.
970 * And after considering the possible alternatives, returning the value
971 * we actually used in getsockopt is the most desirable behavior.
973 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
976 void sock_set_rcvbuf(struct sock *sk, int val)
979 __sock_set_rcvbuf(sk, val);
982 EXPORT_SYMBOL(sock_set_rcvbuf);
984 static void __sock_set_mark(struct sock *sk, u32 val)
986 if (val != sk->sk_mark) {
992 void sock_set_mark(struct sock *sk, u32 val)
995 __sock_set_mark(sk, val);
998 EXPORT_SYMBOL(sock_set_mark);
1000 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1002 /* Round down bytes to multiple of pages */
1003 bytes = round_down(bytes, PAGE_SIZE);
1005 WARN_ON(bytes > sk->sk_reserved_mem);
1006 sk->sk_reserved_mem -= bytes;
1010 static int sock_reserve_memory(struct sock *sk, int bytes)
1016 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1022 pages = sk_mem_pages(bytes);
1024 /* pre-charge to memcg */
1025 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1026 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1030 /* pre-charge to forward_alloc */
1031 sk_memory_allocated_add(sk, pages);
1032 allocated = sk_memory_allocated(sk);
1033 /* If the system goes into memory pressure with this
1034 * precharge, give up and return error.
1036 if (allocated > sk_prot_mem_limits(sk, 1)) {
1037 sk_memory_allocated_sub(sk, pages);
1038 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1041 sk->sk_forward_alloc += pages << PAGE_SHIFT;
1043 sk->sk_reserved_mem += pages << PAGE_SHIFT;
1048 void sockopt_lock_sock(struct sock *sk)
1050 /* When current->bpf_ctx is set, the setsockopt is called from
1051 * a bpf prog. bpf has ensured the sk lock has been
1052 * acquired before calling setsockopt().
1054 if (has_current_bpf_ctx())
1059 EXPORT_SYMBOL(sockopt_lock_sock);
1061 void sockopt_release_sock(struct sock *sk)
1063 if (has_current_bpf_ctx())
1068 EXPORT_SYMBOL(sockopt_release_sock);
1070 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1072 return has_current_bpf_ctx() || ns_capable(ns, cap);
1074 EXPORT_SYMBOL(sockopt_ns_capable);
1076 bool sockopt_capable(int cap)
1078 return has_current_bpf_ctx() || capable(cap);
1080 EXPORT_SYMBOL(sockopt_capable);
1083 * This is meant for all protocols to use and covers goings on
1084 * at the socket level. Everything here is generic.
1087 int sk_setsockopt(struct sock *sk, int level, int optname,
1088 sockptr_t optval, unsigned int optlen)
1090 struct so_timestamping timestamping;
1091 struct socket *sock = sk->sk_socket;
1092 struct sock_txtime sk_txtime;
1099 * Options without arguments
1102 if (optname == SO_BINDTODEVICE)
1103 return sock_setbindtodevice(sk, optval, optlen);
1105 if (optlen < sizeof(int))
1108 if (copy_from_sockptr(&val, optval, sizeof(val)))
1111 valbool = val ? 1 : 0;
1113 sockopt_lock_sock(sk);
1117 if (val && !sockopt_capable(CAP_NET_ADMIN))
1120 sock_valbool_flag(sk, SOCK_DBG, valbool);
1123 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1126 sk->sk_reuseport = valbool;
1135 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1139 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1142 /* Don't error on this BSD doesn't and if you think
1143 * about it this is right. Otherwise apps have to
1144 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1145 * are treated in BSD as hints
1147 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1149 /* Ensure val * 2 fits into an int, to prevent max_t()
1150 * from treating it as a negative value.
1152 val = min_t(int, val, INT_MAX / 2);
1153 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1154 WRITE_ONCE(sk->sk_sndbuf,
1155 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1156 /* Wake up sending tasks if we upped the value. */
1157 sk->sk_write_space(sk);
1160 case SO_SNDBUFFORCE:
1161 if (!sockopt_capable(CAP_NET_ADMIN)) {
1166 /* No negative values (to prevent underflow, as val will be
1174 /* Don't error on this BSD doesn't and if you think
1175 * about it this is right. Otherwise apps have to
1176 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1177 * are treated in BSD as hints
1179 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1182 case SO_RCVBUFFORCE:
1183 if (!sockopt_capable(CAP_NET_ADMIN)) {
1188 /* No negative values (to prevent underflow, as val will be
1191 __sock_set_rcvbuf(sk, max(val, 0));
1195 if (sk->sk_prot->keepalive)
1196 sk->sk_prot->keepalive(sk, valbool);
1197 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1201 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1205 sk->sk_no_check_tx = valbool;
1209 if ((val >= 0 && val <= 6) ||
1210 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1211 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1212 sk->sk_priority = val;
1218 if (optlen < sizeof(ling)) {
1219 ret = -EINVAL; /* 1003.1g */
1222 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1227 sock_reset_flag(sk, SOCK_LINGER);
1229 #if (BITS_PER_LONG == 32)
1230 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1231 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1234 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1235 sock_set_flag(sk, SOCK_LINGER);
1244 set_bit(SOCK_PASSCRED, &sock->flags);
1246 clear_bit(SOCK_PASSCRED, &sock->flags);
1249 case SO_TIMESTAMP_OLD:
1250 case SO_TIMESTAMP_NEW:
1251 case SO_TIMESTAMPNS_OLD:
1252 case SO_TIMESTAMPNS_NEW:
1253 sock_set_timestamp(sk, optname, valbool);
1256 case SO_TIMESTAMPING_NEW:
1257 case SO_TIMESTAMPING_OLD:
1258 if (optlen == sizeof(timestamping)) {
1259 if (copy_from_sockptr(×tamping, optval,
1260 sizeof(timestamping))) {
1265 memset(×tamping, 0, sizeof(timestamping));
1266 timestamping.flags = val;
1268 ret = sock_set_timestamping(sk, optname, timestamping);
1274 if (sock && sock->ops->set_rcvlowat)
1275 ret = sock->ops->set_rcvlowat(sk, val);
1277 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1280 case SO_RCVTIMEO_OLD:
1281 case SO_RCVTIMEO_NEW:
1282 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1283 optlen, optname == SO_RCVTIMEO_OLD);
1286 case SO_SNDTIMEO_OLD:
1287 case SO_SNDTIMEO_NEW:
1288 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1289 optlen, optname == SO_SNDTIMEO_OLD);
1292 case SO_ATTACH_FILTER: {
1293 struct sock_fprog fprog;
1295 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1297 ret = sk_attach_filter(&fprog, sk);
1302 if (optlen == sizeof(u32)) {
1306 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1309 ret = sk_attach_bpf(ufd, sk);
1313 case SO_ATTACH_REUSEPORT_CBPF: {
1314 struct sock_fprog fprog;
1316 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1318 ret = sk_reuseport_attach_filter(&fprog, sk);
1321 case SO_ATTACH_REUSEPORT_EBPF:
1323 if (optlen == sizeof(u32)) {
1327 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1330 ret = sk_reuseport_attach_bpf(ufd, sk);
1334 case SO_DETACH_REUSEPORT_BPF:
1335 ret = reuseport_detach_prog(sk);
1338 case SO_DETACH_FILTER:
1339 ret = sk_detach_filter(sk);
1342 case SO_LOCK_FILTER:
1343 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1346 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1351 set_bit(SOCK_PASSSEC, &sock->flags);
1353 clear_bit(SOCK_PASSSEC, &sock->flags);
1356 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1357 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1362 __sock_set_mark(sk, val);
1365 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1366 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1371 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1375 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1378 case SO_WIFI_STATUS:
1379 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1383 if (sock->ops->set_peek_off)
1384 ret = sock->ops->set_peek_off(sk, val);
1390 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1393 case SO_SELECT_ERR_QUEUE:
1394 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1397 #ifdef CONFIG_NET_RX_BUSY_POLL
1399 /* allow unprivileged users to decrease the value */
1400 if ((val > sk->sk_ll_usec) && !sockopt_capable(CAP_NET_ADMIN))
1406 WRITE_ONCE(sk->sk_ll_usec, val);
1409 case SO_PREFER_BUSY_POLL:
1410 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1413 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1415 case SO_BUSY_POLL_BUDGET:
1416 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) {
1419 if (val < 0 || val > U16_MAX)
1422 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1427 case SO_MAX_PACING_RATE:
1429 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1431 if (sizeof(ulval) != sizeof(val) &&
1432 optlen >= sizeof(ulval) &&
1433 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1438 cmpxchg(&sk->sk_pacing_status,
1441 sk->sk_max_pacing_rate = ulval;
1442 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1445 case SO_INCOMING_CPU:
1446 reuseport_update_incoming_cpu(sk, val);
1451 dst_negative_advice(sk);
1455 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1456 if (!(sk_is_tcp(sk) ||
1457 (sk->sk_type == SOCK_DGRAM &&
1458 sk->sk_protocol == IPPROTO_UDP)))
1460 } else if (sk->sk_family != PF_RDS) {
1464 if (val < 0 || val > 1)
1467 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1472 if (optlen != sizeof(struct sock_txtime)) {
1475 } else if (copy_from_sockptr(&sk_txtime, optval,
1476 sizeof(struct sock_txtime))) {
1479 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1483 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1484 * scheduler has enough safe guards.
1486 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1487 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1491 sock_valbool_flag(sk, SOCK_TXTIME, true);
1492 sk->sk_clockid = sk_txtime.clockid;
1493 sk->sk_txtime_deadline_mode =
1494 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1495 sk->sk_txtime_report_errors =
1496 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1499 case SO_BINDTOIFINDEX:
1500 ret = sock_bindtoindex_locked(sk, val);
1504 if (val & ~SOCK_BUF_LOCK_MASK) {
1508 sk->sk_userlocks = val | (sk->sk_userlocks &
1509 ~SOCK_BUF_LOCK_MASK);
1512 case SO_RESERVE_MEM:
1521 delta = val - sk->sk_reserved_mem;
1523 sock_release_reserved_memory(sk, -delta);
1525 ret = sock_reserve_memory(sk, delta);
1530 if (val < -1 || val > 1) {
1534 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1535 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1536 /* Paired with READ_ONCE() in tcp_rtx_synack() */
1537 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1544 sockopt_release_sock(sk);
1548 int sock_setsockopt(struct socket *sock, int level, int optname,
1549 sockptr_t optval, unsigned int optlen)
1551 return sk_setsockopt(sock->sk, level, optname,
1554 EXPORT_SYMBOL(sock_setsockopt);
1556 static const struct cred *sk_get_peer_cred(struct sock *sk)
1558 const struct cred *cred;
1560 spin_lock(&sk->sk_peer_lock);
1561 cred = get_cred(sk->sk_peer_cred);
1562 spin_unlock(&sk->sk_peer_lock);
1567 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1568 struct ucred *ucred)
1570 ucred->pid = pid_vnr(pid);
1571 ucred->uid = ucred->gid = -1;
1573 struct user_namespace *current_ns = current_user_ns();
1575 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1576 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1580 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1582 struct user_namespace *user_ns = current_user_ns();
1585 for (i = 0; i < src->ngroups; i++) {
1586 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1588 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1595 int sk_getsockopt(struct sock *sk, int level, int optname,
1596 sockptr_t optval, sockptr_t optlen)
1598 struct socket *sock = sk->sk_socket;
1603 unsigned long ulval;
1605 struct old_timeval32 tm32;
1606 struct __kernel_old_timeval tm;
1607 struct __kernel_sock_timeval stm;
1608 struct sock_txtime txtime;
1609 struct so_timestamping timestamping;
1612 int lv = sizeof(int);
1615 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1620 memset(&v, 0, sizeof(v));
1624 v.val = sock_flag(sk, SOCK_DBG);
1628 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1632 v.val = sock_flag(sk, SOCK_BROADCAST);
1636 v.val = sk->sk_sndbuf;
1640 v.val = sk->sk_rcvbuf;
1644 v.val = sk->sk_reuse;
1648 v.val = sk->sk_reuseport;
1652 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1656 v.val = sk->sk_type;
1660 v.val = sk->sk_protocol;
1664 v.val = sk->sk_family;
1668 v.val = -sock_error(sk);
1670 v.val = xchg(&sk->sk_err_soft, 0);
1674 v.val = sock_flag(sk, SOCK_URGINLINE);
1678 v.val = sk->sk_no_check_tx;
1682 v.val = sk->sk_priority;
1686 lv = sizeof(v.ling);
1687 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1688 v.ling.l_linger = sk->sk_lingertime / HZ;
1694 case SO_TIMESTAMP_OLD:
1695 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1696 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1697 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1700 case SO_TIMESTAMPNS_OLD:
1701 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1704 case SO_TIMESTAMP_NEW:
1705 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1708 case SO_TIMESTAMPNS_NEW:
1709 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1712 case SO_TIMESTAMPING_OLD:
1713 lv = sizeof(v.timestamping);
1714 v.timestamping.flags = sk->sk_tsflags;
1715 v.timestamping.bind_phc = sk->sk_bind_phc;
1718 case SO_RCVTIMEO_OLD:
1719 case SO_RCVTIMEO_NEW:
1720 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1723 case SO_SNDTIMEO_OLD:
1724 case SO_SNDTIMEO_NEW:
1725 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1729 v.val = sk->sk_rcvlowat;
1737 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1742 struct ucred peercred;
1743 if (len > sizeof(peercred))
1744 len = sizeof(peercred);
1746 spin_lock(&sk->sk_peer_lock);
1747 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1748 spin_unlock(&sk->sk_peer_lock);
1750 if (copy_to_sockptr(optval, &peercred, len))
1757 const struct cred *cred;
1760 cred = sk_get_peer_cred(sk);
1764 n = cred->group_info->ngroups;
1765 if (len < n * sizeof(gid_t)) {
1766 len = n * sizeof(gid_t);
1768 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1770 len = n * sizeof(gid_t);
1772 ret = groups_to_user(optval, cred->group_info);
1783 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1788 if (copy_to_sockptr(optval, address, len))
1793 /* Dubious BSD thing... Probably nobody even uses it, but
1794 * the UNIX standard wants it for whatever reason... -DaveM
1797 v.val = sk->sk_state == TCP_LISTEN;
1801 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1805 return security_socket_getpeersec_stream(sock,
1806 optval, optlen, len);
1809 v.val = sk->sk_mark;
1813 v.val = sock_flag(sk, SOCK_RCVMARK);
1817 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1820 case SO_WIFI_STATUS:
1821 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1825 if (!sock->ops->set_peek_off)
1828 v.val = sk->sk_peek_off;
1831 v.val = sock_flag(sk, SOCK_NOFCS);
1834 case SO_BINDTODEVICE:
1835 return sock_getbindtodevice(sk, optval, optlen, len);
1838 len = sk_get_filter(sk, optval, len);
1844 case SO_LOCK_FILTER:
1845 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1848 case SO_BPF_EXTENSIONS:
1849 v.val = bpf_tell_extensions();
1852 case SO_SELECT_ERR_QUEUE:
1853 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1856 #ifdef CONFIG_NET_RX_BUSY_POLL
1858 v.val = sk->sk_ll_usec;
1860 case SO_PREFER_BUSY_POLL:
1861 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1865 case SO_MAX_PACING_RATE:
1866 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1867 lv = sizeof(v.ulval);
1868 v.ulval = sk->sk_max_pacing_rate;
1871 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1875 case SO_INCOMING_CPU:
1876 v.val = READ_ONCE(sk->sk_incoming_cpu);
1881 u32 meminfo[SK_MEMINFO_VARS];
1883 sk_get_meminfo(sk, meminfo);
1885 len = min_t(unsigned int, len, sizeof(meminfo));
1886 if (copy_to_sockptr(optval, &meminfo, len))
1892 #ifdef CONFIG_NET_RX_BUSY_POLL
1893 case SO_INCOMING_NAPI_ID:
1894 v.val = READ_ONCE(sk->sk_napi_id);
1896 /* aggregate non-NAPI IDs down to 0 */
1897 if (v.val < MIN_NAPI_ID)
1907 v.val64 = sock_gen_cookie(sk);
1911 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1915 lv = sizeof(v.txtime);
1916 v.txtime.clockid = sk->sk_clockid;
1917 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1918 SOF_TXTIME_DEADLINE_MODE : 0;
1919 v.txtime.flags |= sk->sk_txtime_report_errors ?
1920 SOF_TXTIME_REPORT_ERRORS : 0;
1923 case SO_BINDTOIFINDEX:
1924 v.val = READ_ONCE(sk->sk_bound_dev_if);
1927 case SO_NETNS_COOKIE:
1931 v.val64 = sock_net(sk)->net_cookie;
1935 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1938 case SO_RESERVE_MEM:
1939 v.val = sk->sk_reserved_mem;
1943 v.val = sk->sk_txrehash;
1947 /* We implement the SO_SNDLOWAT etc to not be settable
1950 return -ENOPROTOOPT;
1955 if (copy_to_sockptr(optval, &v, len))
1958 if (copy_to_sockptr(optlen, &len, sizeof(int)))
1963 int sock_getsockopt(struct socket *sock, int level, int optname,
1964 char __user *optval, int __user *optlen)
1966 return sk_getsockopt(sock->sk, level, optname,
1967 USER_SOCKPTR(optval),
1968 USER_SOCKPTR(optlen));
1972 * Initialize an sk_lock.
1974 * (We also register the sk_lock with the lock validator.)
1976 static inline void sock_lock_init(struct sock *sk)
1978 if (sk->sk_kern_sock)
1979 sock_lock_init_class_and_name(
1981 af_family_kern_slock_key_strings[sk->sk_family],
1982 af_family_kern_slock_keys + sk->sk_family,
1983 af_family_kern_key_strings[sk->sk_family],
1984 af_family_kern_keys + sk->sk_family);
1986 sock_lock_init_class_and_name(
1988 af_family_slock_key_strings[sk->sk_family],
1989 af_family_slock_keys + sk->sk_family,
1990 af_family_key_strings[sk->sk_family],
1991 af_family_keys + sk->sk_family);
1995 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1996 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1997 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1999 static void sock_copy(struct sock *nsk, const struct sock *osk)
2001 const struct proto *prot = READ_ONCE(osk->sk_prot);
2002 #ifdef CONFIG_SECURITY_NETWORK
2003 void *sptr = nsk->sk_security;
2006 /* If we move sk_tx_queue_mapping out of the private section,
2007 * we must check if sk_tx_queue_clear() is called after
2008 * sock_copy() in sk_clone_lock().
2010 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2011 offsetof(struct sock, sk_dontcopy_begin) ||
2012 offsetof(struct sock, sk_tx_queue_mapping) >=
2013 offsetof(struct sock, sk_dontcopy_end));
2015 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2017 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2018 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2020 #ifdef CONFIG_SECURITY_NETWORK
2021 nsk->sk_security = sptr;
2022 security_sk_clone(osk, nsk);
2026 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2030 struct kmem_cache *slab;
2034 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2037 if (want_init_on_alloc(priority))
2038 sk_prot_clear_nulls(sk, prot->obj_size);
2040 sk = kmalloc(prot->obj_size, priority);
2043 if (security_sk_alloc(sk, family, priority))
2046 if (!try_module_get(prot->owner))
2053 security_sk_free(sk);
2056 kmem_cache_free(slab, sk);
2062 static void sk_prot_free(struct proto *prot, struct sock *sk)
2064 struct kmem_cache *slab;
2065 struct module *owner;
2067 owner = prot->owner;
2070 cgroup_sk_free(&sk->sk_cgrp_data);
2071 mem_cgroup_sk_free(sk);
2072 security_sk_free(sk);
2074 kmem_cache_free(slab, sk);
2081 * sk_alloc - All socket objects are allocated here
2082 * @net: the applicable net namespace
2083 * @family: protocol family
2084 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2085 * @prot: struct proto associated with this new sock instance
2086 * @kern: is this to be a kernel socket?
2088 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2089 struct proto *prot, int kern)
2093 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2095 sk->sk_family = family;
2097 * See comment in struct sock definition to understand
2098 * why we need sk_prot_creator -acme
2100 sk->sk_prot = sk->sk_prot_creator = prot;
2101 sk->sk_kern_sock = kern;
2103 sk->sk_net_refcnt = kern ? 0 : 1;
2104 if (likely(sk->sk_net_refcnt)) {
2105 get_net_track(net, &sk->ns_tracker, priority);
2106 sock_inuse_add(net, 1);
2108 __netns_tracker_alloc(net, &sk->ns_tracker,
2112 sock_net_set(sk, net);
2113 refcount_set(&sk->sk_wmem_alloc, 1);
2115 mem_cgroup_sk_alloc(sk);
2116 cgroup_sk_alloc(&sk->sk_cgrp_data);
2117 sock_update_classid(&sk->sk_cgrp_data);
2118 sock_update_netprioidx(&sk->sk_cgrp_data);
2119 sk_tx_queue_clear(sk);
2124 EXPORT_SYMBOL(sk_alloc);
2126 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2127 * grace period. This is the case for UDP sockets and TCP listeners.
2129 static void __sk_destruct(struct rcu_head *head)
2131 struct sock *sk = container_of(head, struct sock, sk_rcu);
2132 struct sk_filter *filter;
2134 if (sk->sk_destruct)
2135 sk->sk_destruct(sk);
2137 filter = rcu_dereference_check(sk->sk_filter,
2138 refcount_read(&sk->sk_wmem_alloc) == 0);
2140 sk_filter_uncharge(sk, filter);
2141 RCU_INIT_POINTER(sk->sk_filter, NULL);
2144 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2146 #ifdef CONFIG_BPF_SYSCALL
2147 bpf_sk_storage_free(sk);
2150 if (atomic_read(&sk->sk_omem_alloc))
2151 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2152 __func__, atomic_read(&sk->sk_omem_alloc));
2154 if (sk->sk_frag.page) {
2155 put_page(sk->sk_frag.page);
2156 sk->sk_frag.page = NULL;
2159 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2160 put_cred(sk->sk_peer_cred);
2161 put_pid(sk->sk_peer_pid);
2163 if (likely(sk->sk_net_refcnt))
2164 put_net_track(sock_net(sk), &sk->ns_tracker);
2166 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2168 sk_prot_free(sk->sk_prot_creator, sk);
2171 void sk_destruct(struct sock *sk)
2173 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2175 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2176 reuseport_detach_sock(sk);
2177 use_call_rcu = true;
2181 call_rcu(&sk->sk_rcu, __sk_destruct);
2183 __sk_destruct(&sk->sk_rcu);
2186 static void __sk_free(struct sock *sk)
2188 if (likely(sk->sk_net_refcnt))
2189 sock_inuse_add(sock_net(sk), -1);
2191 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2192 sock_diag_broadcast_destroy(sk);
2197 void sk_free(struct sock *sk)
2200 * We subtract one from sk_wmem_alloc and can know if
2201 * some packets are still in some tx queue.
2202 * If not null, sock_wfree() will call __sk_free(sk) later
2204 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2207 EXPORT_SYMBOL(sk_free);
2209 static void sk_init_common(struct sock *sk)
2211 skb_queue_head_init(&sk->sk_receive_queue);
2212 skb_queue_head_init(&sk->sk_write_queue);
2213 skb_queue_head_init(&sk->sk_error_queue);
2215 rwlock_init(&sk->sk_callback_lock);
2216 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2217 af_rlock_keys + sk->sk_family,
2218 af_family_rlock_key_strings[sk->sk_family]);
2219 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2220 af_wlock_keys + sk->sk_family,
2221 af_family_wlock_key_strings[sk->sk_family]);
2222 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2223 af_elock_keys + sk->sk_family,
2224 af_family_elock_key_strings[sk->sk_family]);
2225 lockdep_set_class_and_name(&sk->sk_callback_lock,
2226 af_callback_keys + sk->sk_family,
2227 af_family_clock_key_strings[sk->sk_family]);
2231 * sk_clone_lock - clone a socket, and lock its clone
2232 * @sk: the socket to clone
2233 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2235 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2237 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2239 struct proto *prot = READ_ONCE(sk->sk_prot);
2240 struct sk_filter *filter;
2241 bool is_charged = true;
2244 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2248 sock_copy(newsk, sk);
2250 newsk->sk_prot_creator = prot;
2253 if (likely(newsk->sk_net_refcnt)) {
2254 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2255 sock_inuse_add(sock_net(newsk), 1);
2257 /* Kernel sockets are not elevating the struct net refcount.
2258 * Instead, use a tracker to more easily detect if a layer
2259 * is not properly dismantling its kernel sockets at netns
2262 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2265 sk_node_init(&newsk->sk_node);
2266 sock_lock_init(newsk);
2267 bh_lock_sock(newsk);
2268 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2269 newsk->sk_backlog.len = 0;
2271 atomic_set(&newsk->sk_rmem_alloc, 0);
2273 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2274 refcount_set(&newsk->sk_wmem_alloc, 1);
2276 atomic_set(&newsk->sk_omem_alloc, 0);
2277 sk_init_common(newsk);
2279 newsk->sk_dst_cache = NULL;
2280 newsk->sk_dst_pending_confirm = 0;
2281 newsk->sk_wmem_queued = 0;
2282 newsk->sk_forward_alloc = 0;
2283 newsk->sk_reserved_mem = 0;
2284 atomic_set(&newsk->sk_drops, 0);
2285 newsk->sk_send_head = NULL;
2286 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2287 atomic_set(&newsk->sk_zckey, 0);
2289 sock_reset_flag(newsk, SOCK_DONE);
2291 /* sk->sk_memcg will be populated at accept() time */
2292 newsk->sk_memcg = NULL;
2294 cgroup_sk_clone(&newsk->sk_cgrp_data);
2297 filter = rcu_dereference(sk->sk_filter);
2299 /* though it's an empty new sock, the charging may fail
2300 * if sysctl_optmem_max was changed between creation of
2301 * original socket and cloning
2303 is_charged = sk_filter_charge(newsk, filter);
2304 RCU_INIT_POINTER(newsk->sk_filter, filter);
2307 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2308 /* We need to make sure that we don't uncharge the new
2309 * socket if we couldn't charge it in the first place
2310 * as otherwise we uncharge the parent's filter.
2313 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2314 sk_free_unlock_clone(newsk);
2318 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2320 if (bpf_sk_storage_clone(sk, newsk)) {
2321 sk_free_unlock_clone(newsk);
2326 /* Clear sk_user_data if parent had the pointer tagged
2327 * as not suitable for copying when cloning.
2329 if (sk_user_data_is_nocopy(newsk))
2330 newsk->sk_user_data = NULL;
2333 newsk->sk_err_soft = 0;
2334 newsk->sk_priority = 0;
2335 newsk->sk_incoming_cpu = raw_smp_processor_id();
2337 /* Before updating sk_refcnt, we must commit prior changes to memory
2338 * (Documentation/RCU/rculist_nulls.rst for details)
2341 refcount_set(&newsk->sk_refcnt, 2);
2343 sk_set_socket(newsk, NULL);
2344 sk_tx_queue_clear(newsk);
2345 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2347 if (newsk->sk_prot->sockets_allocated)
2348 sk_sockets_allocated_inc(newsk);
2350 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2351 net_enable_timestamp();
2355 EXPORT_SYMBOL_GPL(sk_clone_lock);
2357 void sk_free_unlock_clone(struct sock *sk)
2359 /* It is still raw copy of parent, so invalidate
2360 * destructor and make plain sk_free() */
2361 sk->sk_destruct = NULL;
2365 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2367 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2369 bool is_ipv6 = false;
2372 #if IS_ENABLED(CONFIG_IPV6)
2373 is_ipv6 = (sk->sk_family == AF_INET6 &&
2374 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2376 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2377 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2378 READ_ONCE(dst->dev->gso_ipv4_max_size);
2379 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2380 max_size = GSO_LEGACY_MAX_SIZE;
2382 return max_size - (MAX_TCP_HEADER + 1);
2385 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2389 sk_dst_set(sk, dst);
2390 sk->sk_route_caps = dst->dev->features;
2392 sk->sk_route_caps |= NETIF_F_GSO;
2393 if (sk->sk_route_caps & NETIF_F_GSO)
2394 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2395 if (unlikely(sk->sk_gso_disabled))
2396 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2397 if (sk_can_gso(sk)) {
2398 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2399 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2401 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2402 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2403 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2404 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2407 sk->sk_gso_max_segs = max_segs;
2409 EXPORT_SYMBOL_GPL(sk_setup_caps);
2412 * Simple resource managers for sockets.
2417 * Write buffer destructor automatically called from kfree_skb.
2419 void sock_wfree(struct sk_buff *skb)
2421 struct sock *sk = skb->sk;
2422 unsigned int len = skb->truesize;
2425 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2426 if (sock_flag(sk, SOCK_RCU_FREE) &&
2427 sk->sk_write_space == sock_def_write_space) {
2429 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2430 sock_def_write_space_wfree(sk);
2438 * Keep a reference on sk_wmem_alloc, this will be released
2439 * after sk_write_space() call
2441 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2442 sk->sk_write_space(sk);
2446 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2447 * could not do because of in-flight packets
2449 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2452 EXPORT_SYMBOL(sock_wfree);
2454 /* This variant of sock_wfree() is used by TCP,
2455 * since it sets SOCK_USE_WRITE_QUEUE.
2457 void __sock_wfree(struct sk_buff *skb)
2459 struct sock *sk = skb->sk;
2461 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2465 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2470 if (unlikely(!sk_fullsock(sk))) {
2471 skb->destructor = sock_edemux;
2476 skb->destructor = sock_wfree;
2477 skb_set_hash_from_sk(skb, sk);
2479 * We used to take a refcount on sk, but following operation
2480 * is enough to guarantee sk_free() wont free this sock until
2481 * all in-flight packets are completed
2483 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2485 EXPORT_SYMBOL(skb_set_owner_w);
2487 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2489 #ifdef CONFIG_TLS_DEVICE
2490 /* Drivers depend on in-order delivery for crypto offload,
2491 * partial orphan breaks out-of-order-OK logic.
2496 return (skb->destructor == sock_wfree ||
2497 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2500 /* This helper is used by netem, as it can hold packets in its
2501 * delay queue. We want to allow the owner socket to send more
2502 * packets, as if they were already TX completed by a typical driver.
2503 * But we also want to keep skb->sk set because some packet schedulers
2504 * rely on it (sch_fq for example).
2506 void skb_orphan_partial(struct sk_buff *skb)
2508 if (skb_is_tcp_pure_ack(skb))
2511 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2516 EXPORT_SYMBOL(skb_orphan_partial);
2519 * Read buffer destructor automatically called from kfree_skb.
2521 void sock_rfree(struct sk_buff *skb)
2523 struct sock *sk = skb->sk;
2524 unsigned int len = skb->truesize;
2526 atomic_sub(len, &sk->sk_rmem_alloc);
2527 sk_mem_uncharge(sk, len);
2529 EXPORT_SYMBOL(sock_rfree);
2532 * Buffer destructor for skbs that are not used directly in read or write
2533 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2535 void sock_efree(struct sk_buff *skb)
2539 EXPORT_SYMBOL(sock_efree);
2541 /* Buffer destructor for prefetch/receive path where reference count may
2542 * not be held, e.g. for listen sockets.
2545 void sock_pfree(struct sk_buff *skb)
2547 if (sk_is_refcounted(skb->sk))
2548 sock_gen_put(skb->sk);
2550 EXPORT_SYMBOL(sock_pfree);
2551 #endif /* CONFIG_INET */
2553 kuid_t sock_i_uid(struct sock *sk)
2557 read_lock_bh(&sk->sk_callback_lock);
2558 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2559 read_unlock_bh(&sk->sk_callback_lock);
2562 EXPORT_SYMBOL(sock_i_uid);
2564 unsigned long sock_i_ino(struct sock *sk)
2568 read_lock_bh(&sk->sk_callback_lock);
2569 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2570 read_unlock_bh(&sk->sk_callback_lock);
2573 EXPORT_SYMBOL(sock_i_ino);
2576 * Allocate a skb from the socket's send buffer.
2578 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2582 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2583 struct sk_buff *skb = alloc_skb(size, priority);
2586 skb_set_owner_w(skb, sk);
2592 EXPORT_SYMBOL(sock_wmalloc);
2594 static void sock_ofree(struct sk_buff *skb)
2596 struct sock *sk = skb->sk;
2598 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2601 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2604 struct sk_buff *skb;
2606 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2607 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2608 READ_ONCE(sysctl_optmem_max))
2611 skb = alloc_skb(size, priority);
2615 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2617 skb->destructor = sock_ofree;
2622 * Allocate a memory block from the socket's option memory buffer.
2624 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2626 int optmem_max = READ_ONCE(sysctl_optmem_max);
2628 if ((unsigned int)size <= optmem_max &&
2629 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2631 /* First do the add, to avoid the race if kmalloc
2634 atomic_add(size, &sk->sk_omem_alloc);
2635 mem = kmalloc(size, priority);
2638 atomic_sub(size, &sk->sk_omem_alloc);
2642 EXPORT_SYMBOL(sock_kmalloc);
2644 /* Free an option memory block. Note, we actually want the inline
2645 * here as this allows gcc to detect the nullify and fold away the
2646 * condition entirely.
2648 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2651 if (WARN_ON_ONCE(!mem))
2654 kfree_sensitive(mem);
2657 atomic_sub(size, &sk->sk_omem_alloc);
2660 void sock_kfree_s(struct sock *sk, void *mem, int size)
2662 __sock_kfree_s(sk, mem, size, false);
2664 EXPORT_SYMBOL(sock_kfree_s);
2666 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2668 __sock_kfree_s(sk, mem, size, true);
2670 EXPORT_SYMBOL(sock_kzfree_s);
2672 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2673 I think, these locks should be removed for datagram sockets.
2675 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2679 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2683 if (signal_pending(current))
2685 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2686 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2687 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2689 if (sk->sk_shutdown & SEND_SHUTDOWN)
2693 timeo = schedule_timeout(timeo);
2695 finish_wait(sk_sleep(sk), &wait);
2701 * Generic send/receive buffer handlers
2704 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2705 unsigned long data_len, int noblock,
2706 int *errcode, int max_page_order)
2708 struct sk_buff *skb;
2712 timeo = sock_sndtimeo(sk, noblock);
2714 err = sock_error(sk);
2719 if (sk->sk_shutdown & SEND_SHUTDOWN)
2722 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2725 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2726 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2730 if (signal_pending(current))
2732 timeo = sock_wait_for_wmem(sk, timeo);
2734 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2735 errcode, sk->sk_allocation);
2737 skb_set_owner_w(skb, sk);
2741 err = sock_intr_errno(timeo);
2746 EXPORT_SYMBOL(sock_alloc_send_pskb);
2748 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2749 struct sockcm_cookie *sockc)
2753 switch (cmsg->cmsg_type) {
2755 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2756 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2758 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2760 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2762 case SO_TIMESTAMPING_OLD:
2763 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2766 tsflags = *(u32 *)CMSG_DATA(cmsg);
2767 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2770 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2771 sockc->tsflags |= tsflags;
2774 if (!sock_flag(sk, SOCK_TXTIME))
2776 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2778 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2780 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2782 case SCM_CREDENTIALS:
2789 EXPORT_SYMBOL(__sock_cmsg_send);
2791 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2792 struct sockcm_cookie *sockc)
2794 struct cmsghdr *cmsg;
2797 for_each_cmsghdr(cmsg, msg) {
2798 if (!CMSG_OK(msg, cmsg))
2800 if (cmsg->cmsg_level != SOL_SOCKET)
2802 ret = __sock_cmsg_send(sk, cmsg, sockc);
2808 EXPORT_SYMBOL(sock_cmsg_send);
2810 static void sk_enter_memory_pressure(struct sock *sk)
2812 if (!sk->sk_prot->enter_memory_pressure)
2815 sk->sk_prot->enter_memory_pressure(sk);
2818 static void sk_leave_memory_pressure(struct sock *sk)
2820 if (sk->sk_prot->leave_memory_pressure) {
2821 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2822 tcp_leave_memory_pressure, sk);
2824 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2826 if (memory_pressure && READ_ONCE(*memory_pressure))
2827 WRITE_ONCE(*memory_pressure, 0);
2831 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2834 * skb_page_frag_refill - check that a page_frag contains enough room
2835 * @sz: minimum size of the fragment we want to get
2836 * @pfrag: pointer to page_frag
2837 * @gfp: priority for memory allocation
2839 * Note: While this allocator tries to use high order pages, there is
2840 * no guarantee that allocations succeed. Therefore, @sz MUST be
2841 * less or equal than PAGE_SIZE.
2843 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2846 if (page_ref_count(pfrag->page) == 1) {
2850 if (pfrag->offset + sz <= pfrag->size)
2852 put_page(pfrag->page);
2856 if (SKB_FRAG_PAGE_ORDER &&
2857 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2858 /* Avoid direct reclaim but allow kswapd to wake */
2859 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2860 __GFP_COMP | __GFP_NOWARN |
2862 SKB_FRAG_PAGE_ORDER);
2863 if (likely(pfrag->page)) {
2864 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2868 pfrag->page = alloc_page(gfp);
2869 if (likely(pfrag->page)) {
2870 pfrag->size = PAGE_SIZE;
2875 EXPORT_SYMBOL(skb_page_frag_refill);
2877 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2879 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2882 sk_enter_memory_pressure(sk);
2883 sk_stream_moderate_sndbuf(sk);
2886 EXPORT_SYMBOL(sk_page_frag_refill);
2888 void __lock_sock(struct sock *sk)
2889 __releases(&sk->sk_lock.slock)
2890 __acquires(&sk->sk_lock.slock)
2895 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2896 TASK_UNINTERRUPTIBLE);
2897 spin_unlock_bh(&sk->sk_lock.slock);
2899 spin_lock_bh(&sk->sk_lock.slock);
2900 if (!sock_owned_by_user(sk))
2903 finish_wait(&sk->sk_lock.wq, &wait);
2906 void __release_sock(struct sock *sk)
2907 __releases(&sk->sk_lock.slock)
2908 __acquires(&sk->sk_lock.slock)
2910 struct sk_buff *skb, *next;
2912 while ((skb = sk->sk_backlog.head) != NULL) {
2913 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2915 spin_unlock_bh(&sk->sk_lock.slock);
2920 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2921 skb_mark_not_on_list(skb);
2922 sk_backlog_rcv(sk, skb);
2927 } while (skb != NULL);
2929 spin_lock_bh(&sk->sk_lock.slock);
2933 * Doing the zeroing here guarantee we can not loop forever
2934 * while a wild producer attempts to flood us.
2936 sk->sk_backlog.len = 0;
2939 void __sk_flush_backlog(struct sock *sk)
2941 spin_lock_bh(&sk->sk_lock.slock);
2943 spin_unlock_bh(&sk->sk_lock.slock);
2945 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
2948 * sk_wait_data - wait for data to arrive at sk_receive_queue
2949 * @sk: sock to wait on
2950 * @timeo: for how long
2951 * @skb: last skb seen on sk_receive_queue
2953 * Now socket state including sk->sk_err is changed only under lock,
2954 * hence we may omit checks after joining wait queue.
2955 * We check receive queue before schedule() only as optimization;
2956 * it is very likely that release_sock() added new data.
2958 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2960 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2963 add_wait_queue(sk_sleep(sk), &wait);
2964 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2965 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2966 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2967 remove_wait_queue(sk_sleep(sk), &wait);
2970 EXPORT_SYMBOL(sk_wait_data);
2973 * __sk_mem_raise_allocated - increase memory_allocated
2975 * @size: memory size to allocate
2976 * @amt: pages to allocate
2977 * @kind: allocation type
2979 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2981 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2983 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2984 struct proto *prot = sk->sk_prot;
2985 bool charged = true;
2988 sk_memory_allocated_add(sk, amt);
2989 allocated = sk_memory_allocated(sk);
2991 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2992 gfp_memcg_charge())))
2993 goto suppress_allocation;
2996 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2997 sk_leave_memory_pressure(sk);
3001 /* Under pressure. */
3002 if (allocated > sk_prot_mem_limits(sk, 1))
3003 sk_enter_memory_pressure(sk);
3005 /* Over hard limit. */
3006 if (allocated > sk_prot_mem_limits(sk, 2))
3007 goto suppress_allocation;
3009 /* guarantee minimum buffer size under pressure */
3010 if (kind == SK_MEM_RECV) {
3011 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3014 } else { /* SK_MEM_SEND */
3015 int wmem0 = sk_get_wmem0(sk, prot);
3017 if (sk->sk_type == SOCK_STREAM) {
3018 if (sk->sk_wmem_queued < wmem0)
3020 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3025 if (sk_has_memory_pressure(sk)) {
3028 if (!sk_under_memory_pressure(sk))
3030 alloc = sk_sockets_allocated_read_positive(sk);
3031 if (sk_prot_mem_limits(sk, 2) > alloc *
3032 sk_mem_pages(sk->sk_wmem_queued +
3033 atomic_read(&sk->sk_rmem_alloc) +
3034 sk->sk_forward_alloc))
3038 suppress_allocation:
3040 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3041 sk_stream_moderate_sndbuf(sk);
3043 /* Fail only if socket is _under_ its sndbuf.
3044 * In this case we cannot block, so that we have to fail.
3046 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3047 /* Force charge with __GFP_NOFAIL */
3048 if (memcg_charge && !charged) {
3049 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3050 gfp_memcg_charge() | __GFP_NOFAIL);
3056 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3057 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3059 sk_memory_allocated_sub(sk, amt);
3061 if (memcg_charge && charged)
3062 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
3068 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3070 * @size: memory size to allocate
3071 * @kind: allocation type
3073 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3074 * rmem allocation. This function assumes that protocols which have
3075 * memory_pressure use sk_wmem_queued as write buffer accounting.
3077 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3079 int ret, amt = sk_mem_pages(size);
3081 sk->sk_forward_alloc += amt << PAGE_SHIFT;
3082 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3084 sk->sk_forward_alloc -= amt << PAGE_SHIFT;
3087 EXPORT_SYMBOL(__sk_mem_schedule);
3090 * __sk_mem_reduce_allocated - reclaim memory_allocated
3092 * @amount: number of quanta
3094 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3096 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3098 sk_memory_allocated_sub(sk, amount);
3100 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3101 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3103 if (sk_under_memory_pressure(sk) &&
3104 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3105 sk_leave_memory_pressure(sk);
3109 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3111 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3113 void __sk_mem_reclaim(struct sock *sk, int amount)
3115 amount >>= PAGE_SHIFT;
3116 sk->sk_forward_alloc -= amount << PAGE_SHIFT;
3117 __sk_mem_reduce_allocated(sk, amount);
3119 EXPORT_SYMBOL(__sk_mem_reclaim);
3121 int sk_set_peek_off(struct sock *sk, int val)
3123 sk->sk_peek_off = val;
3126 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3129 * Set of default routines for initialising struct proto_ops when
3130 * the protocol does not support a particular function. In certain
3131 * cases where it makes no sense for a protocol to have a "do nothing"
3132 * function, some default processing is provided.
3135 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3139 EXPORT_SYMBOL(sock_no_bind);
3141 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3146 EXPORT_SYMBOL(sock_no_connect);
3148 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3152 EXPORT_SYMBOL(sock_no_socketpair);
3154 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3159 EXPORT_SYMBOL(sock_no_accept);
3161 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3166 EXPORT_SYMBOL(sock_no_getname);
3168 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3172 EXPORT_SYMBOL(sock_no_ioctl);
3174 int sock_no_listen(struct socket *sock, int backlog)
3178 EXPORT_SYMBOL(sock_no_listen);
3180 int sock_no_shutdown(struct socket *sock, int how)
3184 EXPORT_SYMBOL(sock_no_shutdown);
3186 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3190 EXPORT_SYMBOL(sock_no_sendmsg);
3192 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3196 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3198 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3203 EXPORT_SYMBOL(sock_no_recvmsg);
3205 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3207 /* Mirror missing mmap method error code */
3210 EXPORT_SYMBOL(sock_no_mmap);
3213 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3214 * various sock-based usage counts.
3216 void __receive_sock(struct file *file)
3218 struct socket *sock;
3220 sock = sock_from_file(file);
3222 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3223 sock_update_classid(&sock->sk->sk_cgrp_data);
3227 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3230 struct msghdr msg = {.msg_flags = flags};
3232 char *kaddr = kmap(page);
3233 iov.iov_base = kaddr + offset;
3235 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3239 EXPORT_SYMBOL(sock_no_sendpage);
3241 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3242 int offset, size_t size, int flags)
3245 struct msghdr msg = {.msg_flags = flags};
3247 char *kaddr = kmap(page);
3249 iov.iov_base = kaddr + offset;
3251 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3255 EXPORT_SYMBOL(sock_no_sendpage_locked);
3258 * Default Socket Callbacks
3261 static void sock_def_wakeup(struct sock *sk)
3263 struct socket_wq *wq;
3266 wq = rcu_dereference(sk->sk_wq);
3267 if (skwq_has_sleeper(wq))
3268 wake_up_interruptible_all(&wq->wait);
3272 static void sock_def_error_report(struct sock *sk)
3274 struct socket_wq *wq;
3277 wq = rcu_dereference(sk->sk_wq);
3278 if (skwq_has_sleeper(wq))
3279 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3280 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3284 void sock_def_readable(struct sock *sk)
3286 struct socket_wq *wq;
3288 trace_sk_data_ready(sk);
3291 wq = rcu_dereference(sk->sk_wq);
3292 if (skwq_has_sleeper(wq))
3293 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3294 EPOLLRDNORM | EPOLLRDBAND);
3295 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3299 static void sock_def_write_space(struct sock *sk)
3301 struct socket_wq *wq;
3305 /* Do not wake up a writer until he can make "significant"
3308 if (sock_writeable(sk)) {
3309 wq = rcu_dereference(sk->sk_wq);
3310 if (skwq_has_sleeper(wq))
3311 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3312 EPOLLWRNORM | EPOLLWRBAND);
3314 /* Should agree with poll, otherwise some programs break */
3315 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3321 /* An optimised version of sock_def_write_space(), should only be called
3322 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3325 static void sock_def_write_space_wfree(struct sock *sk)
3327 /* Do not wake up a writer until he can make "significant"
3330 if (sock_writeable(sk)) {
3331 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3333 /* rely on refcount_sub from sock_wfree() */
3334 smp_mb__after_atomic();
3335 if (wq && waitqueue_active(&wq->wait))
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);
3344 static void sock_def_destruct(struct sock *sk)
3348 void sk_send_sigurg(struct sock *sk)
3350 if (sk->sk_socket && sk->sk_socket->file)
3351 if (send_sigurg(&sk->sk_socket->file->f_owner))
3352 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3354 EXPORT_SYMBOL(sk_send_sigurg);
3356 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3357 unsigned long expires)
3359 if (!mod_timer(timer, expires))
3362 EXPORT_SYMBOL(sk_reset_timer);
3364 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3366 if (del_timer(timer))
3369 EXPORT_SYMBOL(sk_stop_timer);
3371 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3373 if (del_timer_sync(timer))
3376 EXPORT_SYMBOL(sk_stop_timer_sync);
3378 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3381 sk->sk_send_head = NULL;
3383 timer_setup(&sk->sk_timer, NULL, 0);
3385 sk->sk_allocation = GFP_KERNEL;
3386 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3387 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3388 sk->sk_state = TCP_CLOSE;
3389 sk->sk_use_task_frag = true;
3390 sk_set_socket(sk, sock);
3392 sock_set_flag(sk, SOCK_ZAPPED);
3395 sk->sk_type = sock->type;
3396 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3399 RCU_INIT_POINTER(sk->sk_wq, NULL);
3403 rwlock_init(&sk->sk_callback_lock);
3404 if (sk->sk_kern_sock)
3405 lockdep_set_class_and_name(
3406 &sk->sk_callback_lock,
3407 af_kern_callback_keys + sk->sk_family,
3408 af_family_kern_clock_key_strings[sk->sk_family]);
3410 lockdep_set_class_and_name(
3411 &sk->sk_callback_lock,
3412 af_callback_keys + sk->sk_family,
3413 af_family_clock_key_strings[sk->sk_family]);
3415 sk->sk_state_change = sock_def_wakeup;
3416 sk->sk_data_ready = sock_def_readable;
3417 sk->sk_write_space = sock_def_write_space;
3418 sk->sk_error_report = sock_def_error_report;
3419 sk->sk_destruct = sock_def_destruct;
3421 sk->sk_frag.page = NULL;
3422 sk->sk_frag.offset = 0;
3423 sk->sk_peek_off = -1;
3425 sk->sk_peer_pid = NULL;
3426 sk->sk_peer_cred = NULL;
3427 spin_lock_init(&sk->sk_peer_lock);
3429 sk->sk_write_pending = 0;
3430 sk->sk_rcvlowat = 1;
3431 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3432 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3434 sk->sk_stamp = SK_DEFAULT_STAMP;
3435 #if BITS_PER_LONG==32
3436 seqlock_init(&sk->sk_stamp_seq);
3438 atomic_set(&sk->sk_zckey, 0);
3440 #ifdef CONFIG_NET_RX_BUSY_POLL
3442 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3445 sk->sk_max_pacing_rate = ~0UL;
3446 sk->sk_pacing_rate = ~0UL;
3447 WRITE_ONCE(sk->sk_pacing_shift, 10);
3448 sk->sk_incoming_cpu = -1;
3450 sk_rx_queue_clear(sk);
3452 * Before updating sk_refcnt, we must commit prior changes to memory
3453 * (Documentation/RCU/rculist_nulls.rst for details)
3456 refcount_set(&sk->sk_refcnt, 1);
3457 atomic_set(&sk->sk_drops, 0);
3459 EXPORT_SYMBOL(sock_init_data_uid);
3461 void sock_init_data(struct socket *sock, struct sock *sk)
3464 SOCK_INODE(sock)->i_uid :
3465 make_kuid(sock_net(sk)->user_ns, 0);
3467 sock_init_data_uid(sock, sk, uid);
3469 EXPORT_SYMBOL(sock_init_data);
3471 void lock_sock_nested(struct sock *sk, int subclass)
3473 /* The sk_lock has mutex_lock() semantics here. */
3474 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3477 spin_lock_bh(&sk->sk_lock.slock);
3478 if (sock_owned_by_user_nocheck(sk))
3480 sk->sk_lock.owned = 1;
3481 spin_unlock_bh(&sk->sk_lock.slock);
3483 EXPORT_SYMBOL(lock_sock_nested);
3485 void release_sock(struct sock *sk)
3487 spin_lock_bh(&sk->sk_lock.slock);
3488 if (sk->sk_backlog.tail)
3491 /* Warning : release_cb() might need to release sk ownership,
3492 * ie call sock_release_ownership(sk) before us.
3494 if (sk->sk_prot->release_cb)
3495 sk->sk_prot->release_cb(sk);
3497 sock_release_ownership(sk);
3498 if (waitqueue_active(&sk->sk_lock.wq))
3499 wake_up(&sk->sk_lock.wq);
3500 spin_unlock_bh(&sk->sk_lock.slock);
3502 EXPORT_SYMBOL(release_sock);
3504 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3507 spin_lock_bh(&sk->sk_lock.slock);
3509 if (!sock_owned_by_user_nocheck(sk)) {
3511 * Fast path return with bottom halves disabled and
3512 * sock::sk_lock.slock held.
3514 * The 'mutex' is not contended and holding
3515 * sock::sk_lock.slock prevents all other lockers to
3516 * proceed so the corresponding unlock_sock_fast() can
3517 * avoid the slow path of release_sock() completely and
3518 * just release slock.
3520 * From a semantical POV this is equivalent to 'acquiring'
3521 * the 'mutex', hence the corresponding lockdep
3522 * mutex_release() has to happen in the fast path of
3523 * unlock_sock_fast().
3529 sk->sk_lock.owned = 1;
3530 __acquire(&sk->sk_lock.slock);
3531 spin_unlock_bh(&sk->sk_lock.slock);
3534 EXPORT_SYMBOL(__lock_sock_fast);
3536 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3537 bool timeval, bool time32)
3539 struct sock *sk = sock->sk;
3540 struct timespec64 ts;
3542 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3543 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3544 if (ts.tv_sec == -1)
3546 if (ts.tv_sec == 0) {
3547 ktime_t kt = ktime_get_real();
3548 sock_write_timestamp(sk, kt);
3549 ts = ktime_to_timespec64(kt);
3555 #ifdef CONFIG_COMPAT_32BIT_TIME
3557 return put_old_timespec32(&ts, userstamp);
3559 #ifdef CONFIG_SPARC64
3560 /* beware of padding in sparc64 timeval */
3561 if (timeval && !in_compat_syscall()) {
3562 struct __kernel_old_timeval __user tv = {
3563 .tv_sec = ts.tv_sec,
3564 .tv_usec = ts.tv_nsec,
3566 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3571 return put_timespec64(&ts, userstamp);
3573 EXPORT_SYMBOL(sock_gettstamp);
3575 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3577 if (!sock_flag(sk, flag)) {
3578 unsigned long previous_flags = sk->sk_flags;
3580 sock_set_flag(sk, flag);
3582 * we just set one of the two flags which require net
3583 * time stamping, but time stamping might have been on
3584 * already because of the other one
3586 if (sock_needs_netstamp(sk) &&
3587 !(previous_flags & SK_FLAGS_TIMESTAMP))
3588 net_enable_timestamp();
3592 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3593 int level, int type)
3595 struct sock_exterr_skb *serr;
3596 struct sk_buff *skb;
3600 skb = sock_dequeue_err_skb(sk);
3606 msg->msg_flags |= MSG_TRUNC;
3609 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3613 sock_recv_timestamp(msg, sk, skb);
3615 serr = SKB_EXT_ERR(skb);
3616 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3618 msg->msg_flags |= MSG_ERRQUEUE;
3626 EXPORT_SYMBOL(sock_recv_errqueue);
3629 * Get a socket option on an socket.
3631 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3632 * asynchronous errors should be reported by getsockopt. We assume
3633 * this means if you specify SO_ERROR (otherwise whats the point of it).
3635 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3636 char __user *optval, int __user *optlen)
3638 struct sock *sk = sock->sk;
3640 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3641 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3643 EXPORT_SYMBOL(sock_common_getsockopt);
3645 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3648 struct sock *sk = sock->sk;
3652 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3654 msg->msg_namelen = addr_len;
3657 EXPORT_SYMBOL(sock_common_recvmsg);
3660 * Set socket options on an inet socket.
3662 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3663 sockptr_t optval, unsigned int optlen)
3665 struct sock *sk = sock->sk;
3667 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3668 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3670 EXPORT_SYMBOL(sock_common_setsockopt);
3672 void sk_common_release(struct sock *sk)
3674 if (sk->sk_prot->destroy)
3675 sk->sk_prot->destroy(sk);
3678 * Observation: when sk_common_release is called, processes have
3679 * no access to socket. But net still has.
3680 * Step one, detach it from networking:
3682 * A. Remove from hash tables.
3685 sk->sk_prot->unhash(sk);
3688 * In this point socket cannot receive new packets, but it is possible
3689 * that some packets are in flight because some CPU runs receiver and
3690 * did hash table lookup before we unhashed socket. They will achieve
3691 * receive queue and will be purged by socket destructor.
3693 * Also we still have packets pending on receive queue and probably,
3694 * our own packets waiting in device queues. sock_destroy will drain
3695 * receive queue, but transmitted packets will delay socket destruction
3696 * until the last reference will be released.
3701 xfrm_sk_free_policy(sk);
3705 EXPORT_SYMBOL(sk_common_release);
3707 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3709 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3711 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3712 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3713 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3714 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3715 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3716 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3717 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3718 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3719 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3722 #ifdef CONFIG_PROC_FS
3723 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3725 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3727 int cpu, idx = prot->inuse_idx;
3730 for_each_possible_cpu(cpu)
3731 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3733 return res >= 0 ? res : 0;
3735 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3737 int sock_inuse_get(struct net *net)
3741 for_each_possible_cpu(cpu)
3742 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3747 EXPORT_SYMBOL_GPL(sock_inuse_get);
3749 static int __net_init sock_inuse_init_net(struct net *net)
3751 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3752 if (net->core.prot_inuse == NULL)
3757 static void __net_exit sock_inuse_exit_net(struct net *net)
3759 free_percpu(net->core.prot_inuse);
3762 static struct pernet_operations net_inuse_ops = {
3763 .init = sock_inuse_init_net,
3764 .exit = sock_inuse_exit_net,
3767 static __init int net_inuse_init(void)
3769 if (register_pernet_subsys(&net_inuse_ops))
3770 panic("Cannot initialize net inuse counters");
3775 core_initcall(net_inuse_init);
3777 static int assign_proto_idx(struct proto *prot)
3779 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3781 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3782 pr_err("PROTO_INUSE_NR exhausted\n");
3786 set_bit(prot->inuse_idx, proto_inuse_idx);
3790 static void release_proto_idx(struct proto *prot)
3792 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3793 clear_bit(prot->inuse_idx, proto_inuse_idx);
3796 static inline int assign_proto_idx(struct proto *prot)
3801 static inline void release_proto_idx(struct proto *prot)
3807 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3811 kfree(twsk_prot->twsk_slab_name);
3812 twsk_prot->twsk_slab_name = NULL;
3813 kmem_cache_destroy(twsk_prot->twsk_slab);
3814 twsk_prot->twsk_slab = NULL;
3817 static int tw_prot_init(const struct proto *prot)
3819 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3824 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3826 if (!twsk_prot->twsk_slab_name)
3829 twsk_prot->twsk_slab =
3830 kmem_cache_create(twsk_prot->twsk_slab_name,
3831 twsk_prot->twsk_obj_size, 0,
3832 SLAB_ACCOUNT | prot->slab_flags,
3834 if (!twsk_prot->twsk_slab) {
3835 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3843 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3847 kfree(rsk_prot->slab_name);
3848 rsk_prot->slab_name = NULL;
3849 kmem_cache_destroy(rsk_prot->slab);
3850 rsk_prot->slab = NULL;
3853 static int req_prot_init(const struct proto *prot)
3855 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3860 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3862 if (!rsk_prot->slab_name)
3865 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3866 rsk_prot->obj_size, 0,
3867 SLAB_ACCOUNT | prot->slab_flags,
3870 if (!rsk_prot->slab) {
3871 pr_crit("%s: Can't create request sock SLAB cache!\n",
3878 int proto_register(struct proto *prot, int alloc_slab)
3882 if (prot->memory_allocated && !prot->sysctl_mem) {
3883 pr_err("%s: missing sysctl_mem\n", prot->name);
3886 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3887 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3891 prot->slab = kmem_cache_create_usercopy(prot->name,
3893 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3895 prot->useroffset, prot->usersize,
3898 if (prot->slab == NULL) {
3899 pr_crit("%s: Can't create sock SLAB cache!\n",
3904 if (req_prot_init(prot))
3905 goto out_free_request_sock_slab;
3907 if (tw_prot_init(prot))
3908 goto out_free_timewait_sock_slab;
3911 mutex_lock(&proto_list_mutex);
3912 ret = assign_proto_idx(prot);
3914 mutex_unlock(&proto_list_mutex);
3915 goto out_free_timewait_sock_slab;
3917 list_add(&prot->node, &proto_list);
3918 mutex_unlock(&proto_list_mutex);
3921 out_free_timewait_sock_slab:
3923 tw_prot_cleanup(prot->twsk_prot);
3924 out_free_request_sock_slab:
3926 req_prot_cleanup(prot->rsk_prot);
3928 kmem_cache_destroy(prot->slab);
3934 EXPORT_SYMBOL(proto_register);
3936 void proto_unregister(struct proto *prot)
3938 mutex_lock(&proto_list_mutex);
3939 release_proto_idx(prot);
3940 list_del(&prot->node);
3941 mutex_unlock(&proto_list_mutex);
3943 kmem_cache_destroy(prot->slab);
3946 req_prot_cleanup(prot->rsk_prot);
3947 tw_prot_cleanup(prot->twsk_prot);
3949 EXPORT_SYMBOL(proto_unregister);
3951 int sock_load_diag_module(int family, int protocol)
3954 if (!sock_is_registered(family))
3957 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3958 NETLINK_SOCK_DIAG, family);
3962 if (family == AF_INET &&
3963 protocol != IPPROTO_RAW &&
3964 protocol < MAX_INET_PROTOS &&
3965 !rcu_access_pointer(inet_protos[protocol]))
3969 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3970 NETLINK_SOCK_DIAG, family, protocol);
3972 EXPORT_SYMBOL(sock_load_diag_module);
3974 #ifdef CONFIG_PROC_FS
3975 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3976 __acquires(proto_list_mutex)
3978 mutex_lock(&proto_list_mutex);
3979 return seq_list_start_head(&proto_list, *pos);
3982 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3984 return seq_list_next(v, &proto_list, pos);
3987 static void proto_seq_stop(struct seq_file *seq, void *v)
3988 __releases(proto_list_mutex)
3990 mutex_unlock(&proto_list_mutex);
3993 static char proto_method_implemented(const void *method)
3995 return method == NULL ? 'n' : 'y';
3997 static long sock_prot_memory_allocated(struct proto *proto)
3999 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4002 static const char *sock_prot_memory_pressure(struct proto *proto)
4004 return proto->memory_pressure != NULL ?
4005 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4008 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4011 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4012 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4015 sock_prot_inuse_get(seq_file_net(seq), proto),
4016 sock_prot_memory_allocated(proto),
4017 sock_prot_memory_pressure(proto),
4019 proto->slab == NULL ? "no" : "yes",
4020 module_name(proto->owner),
4021 proto_method_implemented(proto->close),
4022 proto_method_implemented(proto->connect),
4023 proto_method_implemented(proto->disconnect),
4024 proto_method_implemented(proto->accept),
4025 proto_method_implemented(proto->ioctl),
4026 proto_method_implemented(proto->init),
4027 proto_method_implemented(proto->destroy),
4028 proto_method_implemented(proto->shutdown),
4029 proto_method_implemented(proto->setsockopt),
4030 proto_method_implemented(proto->getsockopt),
4031 proto_method_implemented(proto->sendmsg),
4032 proto_method_implemented(proto->recvmsg),
4033 proto_method_implemented(proto->sendpage),
4034 proto_method_implemented(proto->bind),
4035 proto_method_implemented(proto->backlog_rcv),
4036 proto_method_implemented(proto->hash),
4037 proto_method_implemented(proto->unhash),
4038 proto_method_implemented(proto->get_port),
4039 proto_method_implemented(proto->enter_memory_pressure));
4042 static int proto_seq_show(struct seq_file *seq, void *v)
4044 if (v == &proto_list)
4045 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4054 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
4056 proto_seq_printf(seq, list_entry(v, struct proto, node));
4060 static const struct seq_operations proto_seq_ops = {
4061 .start = proto_seq_start,
4062 .next = proto_seq_next,
4063 .stop = proto_seq_stop,
4064 .show = proto_seq_show,
4067 static __net_init int proto_init_net(struct net *net)
4069 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4070 sizeof(struct seq_net_private)))
4076 static __net_exit void proto_exit_net(struct net *net)
4078 remove_proc_entry("protocols", net->proc_net);
4082 static __net_initdata struct pernet_operations proto_net_ops = {
4083 .init = proto_init_net,
4084 .exit = proto_exit_net,
4087 static int __init proto_init(void)
4089 return register_pernet_subsys(&proto_net_ops);
4092 subsys_initcall(proto_init);
4094 #endif /* PROC_FS */
4096 #ifdef CONFIG_NET_RX_BUSY_POLL
4097 bool sk_busy_loop_end(void *p, unsigned long start_time)
4099 struct sock *sk = p;
4101 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
4102 sk_busy_loop_timeout(sk, start_time);
4104 EXPORT_SYMBOL(sk_busy_loop_end);
4105 #endif /* CONFIG_NET_RX_BUSY_POLL */
4107 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4109 if (!sk->sk_prot->bind_add)
4111 return sk->sk_prot->bind_add(sk, addr, addr_len);
4113 EXPORT_SYMBOL(sock_bind_add);