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 &&
905 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
907 if ((1 << sk->sk_state) &
908 (TCPF_CLOSE | TCPF_LISTEN))
910 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
912 atomic_set(&sk->sk_tskey, 0);
916 if (val & SOF_TIMESTAMPING_OPT_STATS &&
917 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
920 if (val & SOF_TIMESTAMPING_BIND_PHC) {
921 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
926 sk->sk_tsflags = val;
927 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
929 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
930 sock_enable_timestamp(sk,
931 SOCK_TIMESTAMPING_RX_SOFTWARE);
933 sock_disable_timestamp(sk,
934 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
938 void sock_set_keepalive(struct sock *sk)
941 if (sk->sk_prot->keepalive)
942 sk->sk_prot->keepalive(sk, true);
943 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
946 EXPORT_SYMBOL(sock_set_keepalive);
948 static void __sock_set_rcvbuf(struct sock *sk, int val)
950 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
951 * as a negative value.
953 val = min_t(int, val, INT_MAX / 2);
954 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
956 /* We double it on the way in to account for "struct sk_buff" etc.
957 * overhead. Applications assume that the SO_RCVBUF setting they make
958 * will allow that much actual data to be received on that socket.
960 * Applications are unaware that "struct sk_buff" and other overheads
961 * allocate from the receive buffer during socket buffer allocation.
963 * And after considering the possible alternatives, returning the value
964 * we actually used in getsockopt is the most desirable behavior.
966 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
969 void sock_set_rcvbuf(struct sock *sk, int val)
972 __sock_set_rcvbuf(sk, val);
975 EXPORT_SYMBOL(sock_set_rcvbuf);
977 static void __sock_set_mark(struct sock *sk, u32 val)
979 if (val != sk->sk_mark) {
985 void sock_set_mark(struct sock *sk, u32 val)
988 __sock_set_mark(sk, val);
991 EXPORT_SYMBOL(sock_set_mark);
993 static void sock_release_reserved_memory(struct sock *sk, int bytes)
995 /* Round down bytes to multiple of pages */
996 bytes = round_down(bytes, PAGE_SIZE);
998 WARN_ON(bytes > sk->sk_reserved_mem);
999 sk->sk_reserved_mem -= bytes;
1003 static int sock_reserve_memory(struct sock *sk, int bytes)
1009 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1015 pages = sk_mem_pages(bytes);
1017 /* pre-charge to memcg */
1018 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1019 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1023 /* pre-charge to forward_alloc */
1024 sk_memory_allocated_add(sk, pages);
1025 allocated = sk_memory_allocated(sk);
1026 /* If the system goes into memory pressure with this
1027 * precharge, give up and return error.
1029 if (allocated > sk_prot_mem_limits(sk, 1)) {
1030 sk_memory_allocated_sub(sk, pages);
1031 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1034 sk->sk_forward_alloc += pages << PAGE_SHIFT;
1036 sk->sk_reserved_mem += pages << PAGE_SHIFT;
1041 void sockopt_lock_sock(struct sock *sk)
1043 /* When current->bpf_ctx is set, the setsockopt is called from
1044 * a bpf prog. bpf has ensured the sk lock has been
1045 * acquired before calling setsockopt().
1047 if (has_current_bpf_ctx())
1052 EXPORT_SYMBOL(sockopt_lock_sock);
1054 void sockopt_release_sock(struct sock *sk)
1056 if (has_current_bpf_ctx())
1061 EXPORT_SYMBOL(sockopt_release_sock);
1063 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1065 return has_current_bpf_ctx() || ns_capable(ns, cap);
1067 EXPORT_SYMBOL(sockopt_ns_capable);
1069 bool sockopt_capable(int cap)
1071 return has_current_bpf_ctx() || capable(cap);
1073 EXPORT_SYMBOL(sockopt_capable);
1076 * This is meant for all protocols to use and covers goings on
1077 * at the socket level. Everything here is generic.
1080 int sk_setsockopt(struct sock *sk, int level, int optname,
1081 sockptr_t optval, unsigned int optlen)
1083 struct so_timestamping timestamping;
1084 struct socket *sock = sk->sk_socket;
1085 struct sock_txtime sk_txtime;
1092 * Options without arguments
1095 if (optname == SO_BINDTODEVICE)
1096 return sock_setbindtodevice(sk, optval, optlen);
1098 if (optlen < sizeof(int))
1101 if (copy_from_sockptr(&val, optval, sizeof(val)))
1104 valbool = val ? 1 : 0;
1106 sockopt_lock_sock(sk);
1110 if (val && !sockopt_capable(CAP_NET_ADMIN))
1113 sock_valbool_flag(sk, SOCK_DBG, valbool);
1116 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1119 sk->sk_reuseport = valbool;
1128 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1132 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1135 /* Don't error on this BSD doesn't and if you think
1136 * about it this is right. Otherwise apps have to
1137 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1138 * are treated in BSD as hints
1140 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1142 /* Ensure val * 2 fits into an int, to prevent max_t()
1143 * from treating it as a negative value.
1145 val = min_t(int, val, INT_MAX / 2);
1146 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1147 WRITE_ONCE(sk->sk_sndbuf,
1148 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1149 /* Wake up sending tasks if we upped the value. */
1150 sk->sk_write_space(sk);
1153 case SO_SNDBUFFORCE:
1154 if (!sockopt_capable(CAP_NET_ADMIN)) {
1159 /* No negative values (to prevent underflow, as val will be
1167 /* Don't error on this BSD doesn't and if you think
1168 * about it this is right. Otherwise apps have to
1169 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1170 * are treated in BSD as hints
1172 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1175 case SO_RCVBUFFORCE:
1176 if (!sockopt_capable(CAP_NET_ADMIN)) {
1181 /* No negative values (to prevent underflow, as val will be
1184 __sock_set_rcvbuf(sk, max(val, 0));
1188 if (sk->sk_prot->keepalive)
1189 sk->sk_prot->keepalive(sk, valbool);
1190 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1194 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1198 sk->sk_no_check_tx = valbool;
1202 if ((val >= 0 && val <= 6) ||
1203 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1204 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1205 sk->sk_priority = val;
1211 if (optlen < sizeof(ling)) {
1212 ret = -EINVAL; /* 1003.1g */
1215 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1220 sock_reset_flag(sk, SOCK_LINGER);
1222 #if (BITS_PER_LONG == 32)
1223 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1224 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1227 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1228 sock_set_flag(sk, SOCK_LINGER);
1237 set_bit(SOCK_PASSCRED, &sock->flags);
1239 clear_bit(SOCK_PASSCRED, &sock->flags);
1242 case SO_TIMESTAMP_OLD:
1243 case SO_TIMESTAMP_NEW:
1244 case SO_TIMESTAMPNS_OLD:
1245 case SO_TIMESTAMPNS_NEW:
1246 sock_set_timestamp(sk, optname, valbool);
1249 case SO_TIMESTAMPING_NEW:
1250 case SO_TIMESTAMPING_OLD:
1251 if (optlen == sizeof(timestamping)) {
1252 if (copy_from_sockptr(×tamping, optval,
1253 sizeof(timestamping))) {
1258 memset(×tamping, 0, sizeof(timestamping));
1259 timestamping.flags = val;
1261 ret = sock_set_timestamping(sk, optname, timestamping);
1267 if (sock && sock->ops->set_rcvlowat)
1268 ret = sock->ops->set_rcvlowat(sk, val);
1270 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1273 case SO_RCVTIMEO_OLD:
1274 case SO_RCVTIMEO_NEW:
1275 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1276 optlen, optname == SO_RCVTIMEO_OLD);
1279 case SO_SNDTIMEO_OLD:
1280 case SO_SNDTIMEO_NEW:
1281 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1282 optlen, optname == SO_SNDTIMEO_OLD);
1285 case SO_ATTACH_FILTER: {
1286 struct sock_fprog fprog;
1288 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1290 ret = sk_attach_filter(&fprog, sk);
1295 if (optlen == sizeof(u32)) {
1299 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1302 ret = sk_attach_bpf(ufd, sk);
1306 case SO_ATTACH_REUSEPORT_CBPF: {
1307 struct sock_fprog fprog;
1309 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1311 ret = sk_reuseport_attach_filter(&fprog, sk);
1314 case SO_ATTACH_REUSEPORT_EBPF:
1316 if (optlen == sizeof(u32)) {
1320 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1323 ret = sk_reuseport_attach_bpf(ufd, sk);
1327 case SO_DETACH_REUSEPORT_BPF:
1328 ret = reuseport_detach_prog(sk);
1331 case SO_DETACH_FILTER:
1332 ret = sk_detach_filter(sk);
1335 case SO_LOCK_FILTER:
1336 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1339 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1344 set_bit(SOCK_PASSSEC, &sock->flags);
1346 clear_bit(SOCK_PASSSEC, &sock->flags);
1349 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1350 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1355 __sock_set_mark(sk, val);
1358 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1359 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1364 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1368 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1371 case SO_WIFI_STATUS:
1372 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1376 if (sock->ops->set_peek_off)
1377 ret = sock->ops->set_peek_off(sk, val);
1383 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1386 case SO_SELECT_ERR_QUEUE:
1387 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1390 #ifdef CONFIG_NET_RX_BUSY_POLL
1392 /* allow unprivileged users to decrease the value */
1393 if ((val > sk->sk_ll_usec) && !sockopt_capable(CAP_NET_ADMIN))
1399 WRITE_ONCE(sk->sk_ll_usec, val);
1402 case SO_PREFER_BUSY_POLL:
1403 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1406 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1408 case SO_BUSY_POLL_BUDGET:
1409 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) {
1412 if (val < 0 || val > U16_MAX)
1415 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1420 case SO_MAX_PACING_RATE:
1422 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1424 if (sizeof(ulval) != sizeof(val) &&
1425 optlen >= sizeof(ulval) &&
1426 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1431 cmpxchg(&sk->sk_pacing_status,
1434 sk->sk_max_pacing_rate = ulval;
1435 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1438 case SO_INCOMING_CPU:
1439 reuseport_update_incoming_cpu(sk, val);
1444 dst_negative_advice(sk);
1448 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1449 if (!(sk_is_tcp(sk) ||
1450 (sk->sk_type == SOCK_DGRAM &&
1451 sk->sk_protocol == IPPROTO_UDP)))
1453 } else if (sk->sk_family != PF_RDS) {
1457 if (val < 0 || val > 1)
1460 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1465 if (optlen != sizeof(struct sock_txtime)) {
1468 } else if (copy_from_sockptr(&sk_txtime, optval,
1469 sizeof(struct sock_txtime))) {
1472 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1476 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1477 * scheduler has enough safe guards.
1479 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1480 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1484 sock_valbool_flag(sk, SOCK_TXTIME, true);
1485 sk->sk_clockid = sk_txtime.clockid;
1486 sk->sk_txtime_deadline_mode =
1487 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1488 sk->sk_txtime_report_errors =
1489 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1492 case SO_BINDTOIFINDEX:
1493 ret = sock_bindtoindex_locked(sk, val);
1497 if (val & ~SOCK_BUF_LOCK_MASK) {
1501 sk->sk_userlocks = val | (sk->sk_userlocks &
1502 ~SOCK_BUF_LOCK_MASK);
1505 case SO_RESERVE_MEM:
1514 delta = val - sk->sk_reserved_mem;
1516 sock_release_reserved_memory(sk, -delta);
1518 ret = sock_reserve_memory(sk, delta);
1523 if (val < -1 || val > 1) {
1527 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1528 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1529 /* Paired with READ_ONCE() in tcp_rtx_synack() */
1530 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1537 sockopt_release_sock(sk);
1541 int sock_setsockopt(struct socket *sock, int level, int optname,
1542 sockptr_t optval, unsigned int optlen)
1544 return sk_setsockopt(sock->sk, level, optname,
1547 EXPORT_SYMBOL(sock_setsockopt);
1549 static const struct cred *sk_get_peer_cred(struct sock *sk)
1551 const struct cred *cred;
1553 spin_lock(&sk->sk_peer_lock);
1554 cred = get_cred(sk->sk_peer_cred);
1555 spin_unlock(&sk->sk_peer_lock);
1560 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1561 struct ucred *ucred)
1563 ucred->pid = pid_vnr(pid);
1564 ucred->uid = ucred->gid = -1;
1566 struct user_namespace *current_ns = current_user_ns();
1568 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1569 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1573 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1575 struct user_namespace *user_ns = current_user_ns();
1578 for (i = 0; i < src->ngroups; i++) {
1579 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1581 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1588 int sk_getsockopt(struct sock *sk, int level, int optname,
1589 sockptr_t optval, sockptr_t optlen)
1591 struct socket *sock = sk->sk_socket;
1596 unsigned long ulval;
1598 struct old_timeval32 tm32;
1599 struct __kernel_old_timeval tm;
1600 struct __kernel_sock_timeval stm;
1601 struct sock_txtime txtime;
1602 struct so_timestamping timestamping;
1605 int lv = sizeof(int);
1608 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1613 memset(&v, 0, sizeof(v));
1617 v.val = sock_flag(sk, SOCK_DBG);
1621 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1625 v.val = sock_flag(sk, SOCK_BROADCAST);
1629 v.val = sk->sk_sndbuf;
1633 v.val = sk->sk_rcvbuf;
1637 v.val = sk->sk_reuse;
1641 v.val = sk->sk_reuseport;
1645 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1649 v.val = sk->sk_type;
1653 v.val = sk->sk_protocol;
1657 v.val = sk->sk_family;
1661 v.val = -sock_error(sk);
1663 v.val = xchg(&sk->sk_err_soft, 0);
1667 v.val = sock_flag(sk, SOCK_URGINLINE);
1671 v.val = sk->sk_no_check_tx;
1675 v.val = sk->sk_priority;
1679 lv = sizeof(v.ling);
1680 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1681 v.ling.l_linger = sk->sk_lingertime / HZ;
1687 case SO_TIMESTAMP_OLD:
1688 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1689 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1690 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1693 case SO_TIMESTAMPNS_OLD:
1694 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1697 case SO_TIMESTAMP_NEW:
1698 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1701 case SO_TIMESTAMPNS_NEW:
1702 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1705 case SO_TIMESTAMPING_OLD:
1706 lv = sizeof(v.timestamping);
1707 v.timestamping.flags = sk->sk_tsflags;
1708 v.timestamping.bind_phc = sk->sk_bind_phc;
1711 case SO_RCVTIMEO_OLD:
1712 case SO_RCVTIMEO_NEW:
1713 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1716 case SO_SNDTIMEO_OLD:
1717 case SO_SNDTIMEO_NEW:
1718 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1722 v.val = sk->sk_rcvlowat;
1730 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1735 struct ucred peercred;
1736 if (len > sizeof(peercred))
1737 len = sizeof(peercred);
1739 spin_lock(&sk->sk_peer_lock);
1740 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1741 spin_unlock(&sk->sk_peer_lock);
1743 if (copy_to_sockptr(optval, &peercred, len))
1750 const struct cred *cred;
1753 cred = sk_get_peer_cred(sk);
1757 n = cred->group_info->ngroups;
1758 if (len < n * sizeof(gid_t)) {
1759 len = n * sizeof(gid_t);
1761 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1763 len = n * sizeof(gid_t);
1765 ret = groups_to_user(optval, cred->group_info);
1776 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1781 if (copy_to_sockptr(optval, address, len))
1786 /* Dubious BSD thing... Probably nobody even uses it, but
1787 * the UNIX standard wants it for whatever reason... -DaveM
1790 v.val = sk->sk_state == TCP_LISTEN;
1794 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1798 return security_socket_getpeersec_stream(sock, optval.user, optlen.user, len);
1801 v.val = sk->sk_mark;
1805 v.val = sock_flag(sk, SOCK_RCVMARK);
1809 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1812 case SO_WIFI_STATUS:
1813 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1817 if (!sock->ops->set_peek_off)
1820 v.val = sk->sk_peek_off;
1823 v.val = sock_flag(sk, SOCK_NOFCS);
1826 case SO_BINDTODEVICE:
1827 return sock_getbindtodevice(sk, optval, optlen, len);
1830 len = sk_get_filter(sk, optval, len);
1836 case SO_LOCK_FILTER:
1837 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1840 case SO_BPF_EXTENSIONS:
1841 v.val = bpf_tell_extensions();
1844 case SO_SELECT_ERR_QUEUE:
1845 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1848 #ifdef CONFIG_NET_RX_BUSY_POLL
1850 v.val = sk->sk_ll_usec;
1852 case SO_PREFER_BUSY_POLL:
1853 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1857 case SO_MAX_PACING_RATE:
1858 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1859 lv = sizeof(v.ulval);
1860 v.ulval = sk->sk_max_pacing_rate;
1863 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1867 case SO_INCOMING_CPU:
1868 v.val = READ_ONCE(sk->sk_incoming_cpu);
1873 u32 meminfo[SK_MEMINFO_VARS];
1875 sk_get_meminfo(sk, meminfo);
1877 len = min_t(unsigned int, len, sizeof(meminfo));
1878 if (copy_to_sockptr(optval, &meminfo, len))
1884 #ifdef CONFIG_NET_RX_BUSY_POLL
1885 case SO_INCOMING_NAPI_ID:
1886 v.val = READ_ONCE(sk->sk_napi_id);
1888 /* aggregate non-NAPI IDs down to 0 */
1889 if (v.val < MIN_NAPI_ID)
1899 v.val64 = sock_gen_cookie(sk);
1903 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1907 lv = sizeof(v.txtime);
1908 v.txtime.clockid = sk->sk_clockid;
1909 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1910 SOF_TXTIME_DEADLINE_MODE : 0;
1911 v.txtime.flags |= sk->sk_txtime_report_errors ?
1912 SOF_TXTIME_REPORT_ERRORS : 0;
1915 case SO_BINDTOIFINDEX:
1916 v.val = READ_ONCE(sk->sk_bound_dev_if);
1919 case SO_NETNS_COOKIE:
1923 v.val64 = sock_net(sk)->net_cookie;
1927 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1930 case SO_RESERVE_MEM:
1931 v.val = sk->sk_reserved_mem;
1935 v.val = sk->sk_txrehash;
1939 /* We implement the SO_SNDLOWAT etc to not be settable
1942 return -ENOPROTOOPT;
1947 if (copy_to_sockptr(optval, &v, len))
1950 if (copy_to_sockptr(optlen, &len, sizeof(int)))
1955 int sock_getsockopt(struct socket *sock, int level, int optname,
1956 char __user *optval, int __user *optlen)
1958 return sk_getsockopt(sock->sk, level, optname,
1959 USER_SOCKPTR(optval),
1960 USER_SOCKPTR(optlen));
1964 * Initialize an sk_lock.
1966 * (We also register the sk_lock with the lock validator.)
1968 static inline void sock_lock_init(struct sock *sk)
1970 if (sk->sk_kern_sock)
1971 sock_lock_init_class_and_name(
1973 af_family_kern_slock_key_strings[sk->sk_family],
1974 af_family_kern_slock_keys + sk->sk_family,
1975 af_family_kern_key_strings[sk->sk_family],
1976 af_family_kern_keys + sk->sk_family);
1978 sock_lock_init_class_and_name(
1980 af_family_slock_key_strings[sk->sk_family],
1981 af_family_slock_keys + sk->sk_family,
1982 af_family_key_strings[sk->sk_family],
1983 af_family_keys + sk->sk_family);
1987 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1988 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1989 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1991 static void sock_copy(struct sock *nsk, const struct sock *osk)
1993 const struct proto *prot = READ_ONCE(osk->sk_prot);
1994 #ifdef CONFIG_SECURITY_NETWORK
1995 void *sptr = nsk->sk_security;
1998 /* If we move sk_tx_queue_mapping out of the private section,
1999 * we must check if sk_tx_queue_clear() is called after
2000 * sock_copy() in sk_clone_lock().
2002 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2003 offsetof(struct sock, sk_dontcopy_begin) ||
2004 offsetof(struct sock, sk_tx_queue_mapping) >=
2005 offsetof(struct sock, sk_dontcopy_end));
2007 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2009 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2010 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2012 #ifdef CONFIG_SECURITY_NETWORK
2013 nsk->sk_security = sptr;
2014 security_sk_clone(osk, nsk);
2018 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2022 struct kmem_cache *slab;
2026 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2029 if (want_init_on_alloc(priority))
2030 sk_prot_clear_nulls(sk, prot->obj_size);
2032 sk = kmalloc(prot->obj_size, priority);
2035 if (security_sk_alloc(sk, family, priority))
2038 if (!try_module_get(prot->owner))
2045 security_sk_free(sk);
2048 kmem_cache_free(slab, sk);
2054 static void sk_prot_free(struct proto *prot, struct sock *sk)
2056 struct kmem_cache *slab;
2057 struct module *owner;
2059 owner = prot->owner;
2062 cgroup_sk_free(&sk->sk_cgrp_data);
2063 mem_cgroup_sk_free(sk);
2064 security_sk_free(sk);
2066 kmem_cache_free(slab, sk);
2073 * sk_alloc - All socket objects are allocated here
2074 * @net: the applicable net namespace
2075 * @family: protocol family
2076 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2077 * @prot: struct proto associated with this new sock instance
2078 * @kern: is this to be a kernel socket?
2080 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2081 struct proto *prot, int kern)
2085 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2087 sk->sk_family = family;
2089 * See comment in struct sock definition to understand
2090 * why we need sk_prot_creator -acme
2092 sk->sk_prot = sk->sk_prot_creator = prot;
2093 sk->sk_kern_sock = kern;
2095 sk->sk_net_refcnt = kern ? 0 : 1;
2096 if (likely(sk->sk_net_refcnt)) {
2097 get_net_track(net, &sk->ns_tracker, priority);
2098 sock_inuse_add(net, 1);
2101 sock_net_set(sk, net);
2102 refcount_set(&sk->sk_wmem_alloc, 1);
2104 mem_cgroup_sk_alloc(sk);
2105 cgroup_sk_alloc(&sk->sk_cgrp_data);
2106 sock_update_classid(&sk->sk_cgrp_data);
2107 sock_update_netprioidx(&sk->sk_cgrp_data);
2108 sk_tx_queue_clear(sk);
2113 EXPORT_SYMBOL(sk_alloc);
2115 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2116 * grace period. This is the case for UDP sockets and TCP listeners.
2118 static void __sk_destruct(struct rcu_head *head)
2120 struct sock *sk = container_of(head, struct sock, sk_rcu);
2121 struct sk_filter *filter;
2123 if (sk->sk_destruct)
2124 sk->sk_destruct(sk);
2126 filter = rcu_dereference_check(sk->sk_filter,
2127 refcount_read(&sk->sk_wmem_alloc) == 0);
2129 sk_filter_uncharge(sk, filter);
2130 RCU_INIT_POINTER(sk->sk_filter, NULL);
2133 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2135 #ifdef CONFIG_BPF_SYSCALL
2136 bpf_sk_storage_free(sk);
2139 if (atomic_read(&sk->sk_omem_alloc))
2140 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2141 __func__, atomic_read(&sk->sk_omem_alloc));
2143 if (sk->sk_frag.page) {
2144 put_page(sk->sk_frag.page);
2145 sk->sk_frag.page = NULL;
2148 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2149 put_cred(sk->sk_peer_cred);
2150 put_pid(sk->sk_peer_pid);
2152 if (likely(sk->sk_net_refcnt))
2153 put_net_track(sock_net(sk), &sk->ns_tracker);
2154 sk_prot_free(sk->sk_prot_creator, sk);
2157 void sk_destruct(struct sock *sk)
2159 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2161 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2162 reuseport_detach_sock(sk);
2163 use_call_rcu = true;
2167 call_rcu(&sk->sk_rcu, __sk_destruct);
2169 __sk_destruct(&sk->sk_rcu);
2172 static void __sk_free(struct sock *sk)
2174 if (likely(sk->sk_net_refcnt))
2175 sock_inuse_add(sock_net(sk), -1);
2177 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2178 sock_diag_broadcast_destroy(sk);
2183 void sk_free(struct sock *sk)
2186 * We subtract one from sk_wmem_alloc and can know if
2187 * some packets are still in some tx queue.
2188 * If not null, sock_wfree() will call __sk_free(sk) later
2190 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2193 EXPORT_SYMBOL(sk_free);
2195 static void sk_init_common(struct sock *sk)
2197 skb_queue_head_init(&sk->sk_receive_queue);
2198 skb_queue_head_init(&sk->sk_write_queue);
2199 skb_queue_head_init(&sk->sk_error_queue);
2201 rwlock_init(&sk->sk_callback_lock);
2202 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2203 af_rlock_keys + sk->sk_family,
2204 af_family_rlock_key_strings[sk->sk_family]);
2205 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2206 af_wlock_keys + sk->sk_family,
2207 af_family_wlock_key_strings[sk->sk_family]);
2208 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2209 af_elock_keys + sk->sk_family,
2210 af_family_elock_key_strings[sk->sk_family]);
2211 lockdep_set_class_and_name(&sk->sk_callback_lock,
2212 af_callback_keys + sk->sk_family,
2213 af_family_clock_key_strings[sk->sk_family]);
2217 * sk_clone_lock - clone a socket, and lock its clone
2218 * @sk: the socket to clone
2219 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2221 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2223 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2225 struct proto *prot = READ_ONCE(sk->sk_prot);
2226 struct sk_filter *filter;
2227 bool is_charged = true;
2230 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2234 sock_copy(newsk, sk);
2236 newsk->sk_prot_creator = prot;
2239 if (likely(newsk->sk_net_refcnt)) {
2240 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2241 sock_inuse_add(sock_net(newsk), 1);
2243 sk_node_init(&newsk->sk_node);
2244 sock_lock_init(newsk);
2245 bh_lock_sock(newsk);
2246 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2247 newsk->sk_backlog.len = 0;
2249 atomic_set(&newsk->sk_rmem_alloc, 0);
2251 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2252 refcount_set(&newsk->sk_wmem_alloc, 1);
2254 atomic_set(&newsk->sk_omem_alloc, 0);
2255 sk_init_common(newsk);
2257 newsk->sk_dst_cache = NULL;
2258 newsk->sk_dst_pending_confirm = 0;
2259 newsk->sk_wmem_queued = 0;
2260 newsk->sk_forward_alloc = 0;
2261 newsk->sk_reserved_mem = 0;
2262 atomic_set(&newsk->sk_drops, 0);
2263 newsk->sk_send_head = NULL;
2264 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2265 atomic_set(&newsk->sk_zckey, 0);
2267 sock_reset_flag(newsk, SOCK_DONE);
2269 /* sk->sk_memcg will be populated at accept() time */
2270 newsk->sk_memcg = NULL;
2272 cgroup_sk_clone(&newsk->sk_cgrp_data);
2275 filter = rcu_dereference(sk->sk_filter);
2277 /* though it's an empty new sock, the charging may fail
2278 * if sysctl_optmem_max was changed between creation of
2279 * original socket and cloning
2281 is_charged = sk_filter_charge(newsk, filter);
2282 RCU_INIT_POINTER(newsk->sk_filter, filter);
2285 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2286 /* We need to make sure that we don't uncharge the new
2287 * socket if we couldn't charge it in the first place
2288 * as otherwise we uncharge the parent's filter.
2291 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2292 sk_free_unlock_clone(newsk);
2296 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2298 if (bpf_sk_storage_clone(sk, newsk)) {
2299 sk_free_unlock_clone(newsk);
2304 /* Clear sk_user_data if parent had the pointer tagged
2305 * as not suitable for copying when cloning.
2307 if (sk_user_data_is_nocopy(newsk))
2308 newsk->sk_user_data = NULL;
2311 newsk->sk_err_soft = 0;
2312 newsk->sk_priority = 0;
2313 newsk->sk_incoming_cpu = raw_smp_processor_id();
2315 /* Before updating sk_refcnt, we must commit prior changes to memory
2316 * (Documentation/RCU/rculist_nulls.rst for details)
2319 refcount_set(&newsk->sk_refcnt, 2);
2321 /* Increment the counter in the same struct proto as the master
2322 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2323 * is the same as sk->sk_prot->socks, as this field was copied
2326 * This _changes_ the previous behaviour, where
2327 * tcp_create_openreq_child always was incrementing the
2328 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2329 * to be taken into account in all callers. -acme
2331 sk_refcnt_debug_inc(newsk);
2332 sk_set_socket(newsk, NULL);
2333 sk_tx_queue_clear(newsk);
2334 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2336 if (newsk->sk_prot->sockets_allocated)
2337 sk_sockets_allocated_inc(newsk);
2339 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2340 net_enable_timestamp();
2344 EXPORT_SYMBOL_GPL(sk_clone_lock);
2346 void sk_free_unlock_clone(struct sock *sk)
2348 /* It is still raw copy of parent, so invalidate
2349 * destructor and make plain sk_free() */
2350 sk->sk_destruct = NULL;
2354 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2356 static void sk_trim_gso_size(struct sock *sk)
2358 if (sk->sk_gso_max_size <= GSO_LEGACY_MAX_SIZE)
2360 #if IS_ENABLED(CONFIG_IPV6)
2361 if (sk->sk_family == AF_INET6 &&
2363 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
2366 sk->sk_gso_max_size = GSO_LEGACY_MAX_SIZE;
2369 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2373 sk_dst_set(sk, dst);
2374 sk->sk_route_caps = dst->dev->features;
2376 sk->sk_route_caps |= NETIF_F_GSO;
2377 if (sk->sk_route_caps & NETIF_F_GSO)
2378 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2379 if (unlikely(sk->sk_gso_disabled))
2380 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2381 if (sk_can_gso(sk)) {
2382 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2383 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2385 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2386 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */
2387 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size);
2388 sk_trim_gso_size(sk);
2389 sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1);
2390 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2391 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2394 sk->sk_gso_max_segs = max_segs;
2396 EXPORT_SYMBOL_GPL(sk_setup_caps);
2399 * Simple resource managers for sockets.
2404 * Write buffer destructor automatically called from kfree_skb.
2406 void sock_wfree(struct sk_buff *skb)
2408 struct sock *sk = skb->sk;
2409 unsigned int len = skb->truesize;
2412 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2413 if (sock_flag(sk, SOCK_RCU_FREE) &&
2414 sk->sk_write_space == sock_def_write_space) {
2416 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2417 sock_def_write_space_wfree(sk);
2425 * Keep a reference on sk_wmem_alloc, this will be released
2426 * after sk_write_space() call
2428 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2429 sk->sk_write_space(sk);
2433 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2434 * could not do because of in-flight packets
2436 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2439 EXPORT_SYMBOL(sock_wfree);
2441 /* This variant of sock_wfree() is used by TCP,
2442 * since it sets SOCK_USE_WRITE_QUEUE.
2444 void __sock_wfree(struct sk_buff *skb)
2446 struct sock *sk = skb->sk;
2448 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2452 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2457 if (unlikely(!sk_fullsock(sk))) {
2458 skb->destructor = sock_edemux;
2463 skb->destructor = sock_wfree;
2464 skb_set_hash_from_sk(skb, sk);
2466 * We used to take a refcount on sk, but following operation
2467 * is enough to guarantee sk_free() wont free this sock until
2468 * all in-flight packets are completed
2470 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2472 EXPORT_SYMBOL(skb_set_owner_w);
2474 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2476 #ifdef CONFIG_TLS_DEVICE
2477 /* Drivers depend on in-order delivery for crypto offload,
2478 * partial orphan breaks out-of-order-OK logic.
2483 return (skb->destructor == sock_wfree ||
2484 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2487 /* This helper is used by netem, as it can hold packets in its
2488 * delay queue. We want to allow the owner socket to send more
2489 * packets, as if they were already TX completed by a typical driver.
2490 * But we also want to keep skb->sk set because some packet schedulers
2491 * rely on it (sch_fq for example).
2493 void skb_orphan_partial(struct sk_buff *skb)
2495 if (skb_is_tcp_pure_ack(skb))
2498 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2503 EXPORT_SYMBOL(skb_orphan_partial);
2506 * Read buffer destructor automatically called from kfree_skb.
2508 void sock_rfree(struct sk_buff *skb)
2510 struct sock *sk = skb->sk;
2511 unsigned int len = skb->truesize;
2513 atomic_sub(len, &sk->sk_rmem_alloc);
2514 sk_mem_uncharge(sk, len);
2516 EXPORT_SYMBOL(sock_rfree);
2519 * Buffer destructor for skbs that are not used directly in read or write
2520 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2522 void sock_efree(struct sk_buff *skb)
2526 EXPORT_SYMBOL(sock_efree);
2528 /* Buffer destructor for prefetch/receive path where reference count may
2529 * not be held, e.g. for listen sockets.
2532 void sock_pfree(struct sk_buff *skb)
2534 if (sk_is_refcounted(skb->sk))
2535 sock_gen_put(skb->sk);
2537 EXPORT_SYMBOL(sock_pfree);
2538 #endif /* CONFIG_INET */
2540 kuid_t sock_i_uid(struct sock *sk)
2544 read_lock_bh(&sk->sk_callback_lock);
2545 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2546 read_unlock_bh(&sk->sk_callback_lock);
2549 EXPORT_SYMBOL(sock_i_uid);
2551 unsigned long sock_i_ino(struct sock *sk)
2555 read_lock_bh(&sk->sk_callback_lock);
2556 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2557 read_unlock_bh(&sk->sk_callback_lock);
2560 EXPORT_SYMBOL(sock_i_ino);
2563 * Allocate a skb from the socket's send buffer.
2565 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2569 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2570 struct sk_buff *skb = alloc_skb(size, priority);
2573 skb_set_owner_w(skb, sk);
2579 EXPORT_SYMBOL(sock_wmalloc);
2581 static void sock_ofree(struct sk_buff *skb)
2583 struct sock *sk = skb->sk;
2585 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2588 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2591 struct sk_buff *skb;
2593 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2594 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2595 READ_ONCE(sysctl_optmem_max))
2598 skb = alloc_skb(size, priority);
2602 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2604 skb->destructor = sock_ofree;
2609 * Allocate a memory block from the socket's option memory buffer.
2611 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2613 int optmem_max = READ_ONCE(sysctl_optmem_max);
2615 if ((unsigned int)size <= optmem_max &&
2616 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2618 /* First do the add, to avoid the race if kmalloc
2621 atomic_add(size, &sk->sk_omem_alloc);
2622 mem = kmalloc(size, priority);
2625 atomic_sub(size, &sk->sk_omem_alloc);
2629 EXPORT_SYMBOL(sock_kmalloc);
2631 /* Free an option memory block. Note, we actually want the inline
2632 * here as this allows gcc to detect the nullify and fold away the
2633 * condition entirely.
2635 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2638 if (WARN_ON_ONCE(!mem))
2641 kfree_sensitive(mem);
2644 atomic_sub(size, &sk->sk_omem_alloc);
2647 void sock_kfree_s(struct sock *sk, void *mem, int size)
2649 __sock_kfree_s(sk, mem, size, false);
2651 EXPORT_SYMBOL(sock_kfree_s);
2653 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2655 __sock_kfree_s(sk, mem, size, true);
2657 EXPORT_SYMBOL(sock_kzfree_s);
2659 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2660 I think, these locks should be removed for datagram sockets.
2662 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2666 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2670 if (signal_pending(current))
2672 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2673 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2674 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2676 if (sk->sk_shutdown & SEND_SHUTDOWN)
2680 timeo = schedule_timeout(timeo);
2682 finish_wait(sk_sleep(sk), &wait);
2688 * Generic send/receive buffer handlers
2691 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2692 unsigned long data_len, int noblock,
2693 int *errcode, int max_page_order)
2695 struct sk_buff *skb;
2699 timeo = sock_sndtimeo(sk, noblock);
2701 err = sock_error(sk);
2706 if (sk->sk_shutdown & SEND_SHUTDOWN)
2709 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2712 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2713 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2717 if (signal_pending(current))
2719 timeo = sock_wait_for_wmem(sk, timeo);
2721 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2722 errcode, sk->sk_allocation);
2724 skb_set_owner_w(skb, sk);
2728 err = sock_intr_errno(timeo);
2733 EXPORT_SYMBOL(sock_alloc_send_pskb);
2735 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2736 struct sockcm_cookie *sockc)
2740 switch (cmsg->cmsg_type) {
2742 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2743 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2745 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2747 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2749 case SO_TIMESTAMPING_OLD:
2750 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2753 tsflags = *(u32 *)CMSG_DATA(cmsg);
2754 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2757 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2758 sockc->tsflags |= tsflags;
2761 if (!sock_flag(sk, SOCK_TXTIME))
2763 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2765 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2767 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2769 case SCM_CREDENTIALS:
2776 EXPORT_SYMBOL(__sock_cmsg_send);
2778 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2779 struct sockcm_cookie *sockc)
2781 struct cmsghdr *cmsg;
2784 for_each_cmsghdr(cmsg, msg) {
2785 if (!CMSG_OK(msg, cmsg))
2787 if (cmsg->cmsg_level != SOL_SOCKET)
2789 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2795 EXPORT_SYMBOL(sock_cmsg_send);
2797 static void sk_enter_memory_pressure(struct sock *sk)
2799 if (!sk->sk_prot->enter_memory_pressure)
2802 sk->sk_prot->enter_memory_pressure(sk);
2805 static void sk_leave_memory_pressure(struct sock *sk)
2807 if (sk->sk_prot->leave_memory_pressure) {
2808 sk->sk_prot->leave_memory_pressure(sk);
2810 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2812 if (memory_pressure && READ_ONCE(*memory_pressure))
2813 WRITE_ONCE(*memory_pressure, 0);
2817 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2820 * skb_page_frag_refill - check that a page_frag contains enough room
2821 * @sz: minimum size of the fragment we want to get
2822 * @pfrag: pointer to page_frag
2823 * @gfp: priority for memory allocation
2825 * Note: While this allocator tries to use high order pages, there is
2826 * no guarantee that allocations succeed. Therefore, @sz MUST be
2827 * less or equal than PAGE_SIZE.
2829 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2832 if (page_ref_count(pfrag->page) == 1) {
2836 if (pfrag->offset + sz <= pfrag->size)
2838 put_page(pfrag->page);
2842 if (SKB_FRAG_PAGE_ORDER &&
2843 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2844 /* Avoid direct reclaim but allow kswapd to wake */
2845 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2846 __GFP_COMP | __GFP_NOWARN |
2848 SKB_FRAG_PAGE_ORDER);
2849 if (likely(pfrag->page)) {
2850 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2854 pfrag->page = alloc_page(gfp);
2855 if (likely(pfrag->page)) {
2856 pfrag->size = PAGE_SIZE;
2861 EXPORT_SYMBOL(skb_page_frag_refill);
2863 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2865 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2868 sk_enter_memory_pressure(sk);
2869 sk_stream_moderate_sndbuf(sk);
2872 EXPORT_SYMBOL(sk_page_frag_refill);
2874 void __lock_sock(struct sock *sk)
2875 __releases(&sk->sk_lock.slock)
2876 __acquires(&sk->sk_lock.slock)
2881 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2882 TASK_UNINTERRUPTIBLE);
2883 spin_unlock_bh(&sk->sk_lock.slock);
2885 spin_lock_bh(&sk->sk_lock.slock);
2886 if (!sock_owned_by_user(sk))
2889 finish_wait(&sk->sk_lock.wq, &wait);
2892 void __release_sock(struct sock *sk)
2893 __releases(&sk->sk_lock.slock)
2894 __acquires(&sk->sk_lock.slock)
2896 struct sk_buff *skb, *next;
2898 while ((skb = sk->sk_backlog.head) != NULL) {
2899 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2901 spin_unlock_bh(&sk->sk_lock.slock);
2906 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2907 skb_mark_not_on_list(skb);
2908 sk_backlog_rcv(sk, skb);
2913 } while (skb != NULL);
2915 spin_lock_bh(&sk->sk_lock.slock);
2919 * Doing the zeroing here guarantee we can not loop forever
2920 * while a wild producer attempts to flood us.
2922 sk->sk_backlog.len = 0;
2925 void __sk_flush_backlog(struct sock *sk)
2927 spin_lock_bh(&sk->sk_lock.slock);
2929 spin_unlock_bh(&sk->sk_lock.slock);
2931 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
2934 * sk_wait_data - wait for data to arrive at sk_receive_queue
2935 * @sk: sock to wait on
2936 * @timeo: for how long
2937 * @skb: last skb seen on sk_receive_queue
2939 * Now socket state including sk->sk_err is changed only under lock,
2940 * hence we may omit checks after joining wait queue.
2941 * We check receive queue before schedule() only as optimization;
2942 * it is very likely that release_sock() added new data.
2944 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2946 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2949 add_wait_queue(sk_sleep(sk), &wait);
2950 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2951 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2952 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2953 remove_wait_queue(sk_sleep(sk), &wait);
2956 EXPORT_SYMBOL(sk_wait_data);
2959 * __sk_mem_raise_allocated - increase memory_allocated
2961 * @size: memory size to allocate
2962 * @amt: pages to allocate
2963 * @kind: allocation type
2965 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2967 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2969 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2970 struct proto *prot = sk->sk_prot;
2971 bool charged = true;
2974 sk_memory_allocated_add(sk, amt);
2975 allocated = sk_memory_allocated(sk);
2977 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2978 gfp_memcg_charge())))
2979 goto suppress_allocation;
2982 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2983 sk_leave_memory_pressure(sk);
2987 /* Under pressure. */
2988 if (allocated > sk_prot_mem_limits(sk, 1))
2989 sk_enter_memory_pressure(sk);
2991 /* Over hard limit. */
2992 if (allocated > sk_prot_mem_limits(sk, 2))
2993 goto suppress_allocation;
2995 /* guarantee minimum buffer size under pressure */
2996 if (kind == SK_MEM_RECV) {
2997 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3000 } else { /* SK_MEM_SEND */
3001 int wmem0 = sk_get_wmem0(sk, prot);
3003 if (sk->sk_type == SOCK_STREAM) {
3004 if (sk->sk_wmem_queued < wmem0)
3006 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3011 if (sk_has_memory_pressure(sk)) {
3014 if (!sk_under_memory_pressure(sk))
3016 alloc = sk_sockets_allocated_read_positive(sk);
3017 if (sk_prot_mem_limits(sk, 2) > alloc *
3018 sk_mem_pages(sk->sk_wmem_queued +
3019 atomic_read(&sk->sk_rmem_alloc) +
3020 sk->sk_forward_alloc))
3024 suppress_allocation:
3026 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3027 sk_stream_moderate_sndbuf(sk);
3029 /* Fail only if socket is _under_ its sndbuf.
3030 * In this case we cannot block, so that we have to fail.
3032 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3033 /* Force charge with __GFP_NOFAIL */
3034 if (memcg_charge && !charged) {
3035 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3036 gfp_memcg_charge() | __GFP_NOFAIL);
3042 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3043 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3045 sk_memory_allocated_sub(sk, amt);
3047 if (memcg_charge && charged)
3048 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
3054 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3056 * @size: memory size to allocate
3057 * @kind: allocation type
3059 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3060 * rmem allocation. This function assumes that protocols which have
3061 * memory_pressure use sk_wmem_queued as write buffer accounting.
3063 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3065 int ret, amt = sk_mem_pages(size);
3067 sk->sk_forward_alloc += amt << PAGE_SHIFT;
3068 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3070 sk->sk_forward_alloc -= amt << PAGE_SHIFT;
3073 EXPORT_SYMBOL(__sk_mem_schedule);
3076 * __sk_mem_reduce_allocated - reclaim memory_allocated
3078 * @amount: number of quanta
3080 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3082 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3084 sk_memory_allocated_sub(sk, amount);
3086 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3087 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3089 if (sk_under_memory_pressure(sk) &&
3090 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3091 sk_leave_memory_pressure(sk);
3095 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3097 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3099 void __sk_mem_reclaim(struct sock *sk, int amount)
3101 amount >>= PAGE_SHIFT;
3102 sk->sk_forward_alloc -= amount << PAGE_SHIFT;
3103 __sk_mem_reduce_allocated(sk, amount);
3105 EXPORT_SYMBOL(__sk_mem_reclaim);
3107 int sk_set_peek_off(struct sock *sk, int val)
3109 sk->sk_peek_off = val;
3112 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3115 * Set of default routines for initialising struct proto_ops when
3116 * the protocol does not support a particular function. In certain
3117 * cases where it makes no sense for a protocol to have a "do nothing"
3118 * function, some default processing is provided.
3121 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3125 EXPORT_SYMBOL(sock_no_bind);
3127 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3132 EXPORT_SYMBOL(sock_no_connect);
3134 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3138 EXPORT_SYMBOL(sock_no_socketpair);
3140 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3145 EXPORT_SYMBOL(sock_no_accept);
3147 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3152 EXPORT_SYMBOL(sock_no_getname);
3154 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3158 EXPORT_SYMBOL(sock_no_ioctl);
3160 int sock_no_listen(struct socket *sock, int backlog)
3164 EXPORT_SYMBOL(sock_no_listen);
3166 int sock_no_shutdown(struct socket *sock, int how)
3170 EXPORT_SYMBOL(sock_no_shutdown);
3172 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3176 EXPORT_SYMBOL(sock_no_sendmsg);
3178 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3182 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3184 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3189 EXPORT_SYMBOL(sock_no_recvmsg);
3191 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3193 /* Mirror missing mmap method error code */
3196 EXPORT_SYMBOL(sock_no_mmap);
3199 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3200 * various sock-based usage counts.
3202 void __receive_sock(struct file *file)
3204 struct socket *sock;
3206 sock = sock_from_file(file);
3208 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3209 sock_update_classid(&sock->sk->sk_cgrp_data);
3213 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3216 struct msghdr msg = {.msg_flags = flags};
3218 char *kaddr = kmap(page);
3219 iov.iov_base = kaddr + offset;
3221 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3225 EXPORT_SYMBOL(sock_no_sendpage);
3227 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3228 int offset, size_t size, int flags)
3231 struct msghdr msg = {.msg_flags = flags};
3233 char *kaddr = kmap(page);
3235 iov.iov_base = kaddr + offset;
3237 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3241 EXPORT_SYMBOL(sock_no_sendpage_locked);
3244 * Default Socket Callbacks
3247 static void sock_def_wakeup(struct sock *sk)
3249 struct socket_wq *wq;
3252 wq = rcu_dereference(sk->sk_wq);
3253 if (skwq_has_sleeper(wq))
3254 wake_up_interruptible_all(&wq->wait);
3258 static void sock_def_error_report(struct sock *sk)
3260 struct socket_wq *wq;
3263 wq = rcu_dereference(sk->sk_wq);
3264 if (skwq_has_sleeper(wq))
3265 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3266 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3270 void sock_def_readable(struct sock *sk)
3272 struct socket_wq *wq;
3275 wq = rcu_dereference(sk->sk_wq);
3276 if (skwq_has_sleeper(wq))
3277 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3278 EPOLLRDNORM | EPOLLRDBAND);
3279 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3283 static void sock_def_write_space(struct sock *sk)
3285 struct socket_wq *wq;
3289 /* Do not wake up a writer until he can make "significant"
3292 if (sock_writeable(sk)) {
3293 wq = rcu_dereference(sk->sk_wq);
3294 if (skwq_has_sleeper(wq))
3295 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3296 EPOLLWRNORM | EPOLLWRBAND);
3298 /* Should agree with poll, otherwise some programs break */
3299 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3305 /* An optimised version of sock_def_write_space(), should only be called
3306 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3309 static void sock_def_write_space_wfree(struct sock *sk)
3311 /* Do not wake up a writer until he can make "significant"
3314 if (sock_writeable(sk)) {
3315 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3317 /* rely on refcount_sub from sock_wfree() */
3318 smp_mb__after_atomic();
3319 if (wq && waitqueue_active(&wq->wait))
3320 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3321 EPOLLWRNORM | EPOLLWRBAND);
3323 /* Should agree with poll, otherwise some programs break */
3324 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3328 static void sock_def_destruct(struct sock *sk)
3332 void sk_send_sigurg(struct sock *sk)
3334 if (sk->sk_socket && sk->sk_socket->file)
3335 if (send_sigurg(&sk->sk_socket->file->f_owner))
3336 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3338 EXPORT_SYMBOL(sk_send_sigurg);
3340 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3341 unsigned long expires)
3343 if (!mod_timer(timer, expires))
3346 EXPORT_SYMBOL(sk_reset_timer);
3348 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3350 if (del_timer(timer))
3353 EXPORT_SYMBOL(sk_stop_timer);
3355 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3357 if (del_timer_sync(timer))
3360 EXPORT_SYMBOL(sk_stop_timer_sync);
3362 void sock_init_data(struct socket *sock, struct sock *sk)
3365 sk->sk_send_head = NULL;
3367 timer_setup(&sk->sk_timer, NULL, 0);
3369 sk->sk_allocation = GFP_KERNEL;
3370 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3371 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3372 sk->sk_state = TCP_CLOSE;
3373 sk_set_socket(sk, sock);
3375 sock_set_flag(sk, SOCK_ZAPPED);
3378 sk->sk_type = sock->type;
3379 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3381 sk->sk_uid = SOCK_INODE(sock)->i_uid;
3383 RCU_INIT_POINTER(sk->sk_wq, NULL);
3384 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3387 rwlock_init(&sk->sk_callback_lock);
3388 if (sk->sk_kern_sock)
3389 lockdep_set_class_and_name(
3390 &sk->sk_callback_lock,
3391 af_kern_callback_keys + sk->sk_family,
3392 af_family_kern_clock_key_strings[sk->sk_family]);
3394 lockdep_set_class_and_name(
3395 &sk->sk_callback_lock,
3396 af_callback_keys + sk->sk_family,
3397 af_family_clock_key_strings[sk->sk_family]);
3399 sk->sk_state_change = sock_def_wakeup;
3400 sk->sk_data_ready = sock_def_readable;
3401 sk->sk_write_space = sock_def_write_space;
3402 sk->sk_error_report = sock_def_error_report;
3403 sk->sk_destruct = sock_def_destruct;
3405 sk->sk_frag.page = NULL;
3406 sk->sk_frag.offset = 0;
3407 sk->sk_peek_off = -1;
3409 sk->sk_peer_pid = NULL;
3410 sk->sk_peer_cred = NULL;
3411 spin_lock_init(&sk->sk_peer_lock);
3413 sk->sk_write_pending = 0;
3414 sk->sk_rcvlowat = 1;
3415 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3416 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3418 sk->sk_stamp = SK_DEFAULT_STAMP;
3419 #if BITS_PER_LONG==32
3420 seqlock_init(&sk->sk_stamp_seq);
3422 atomic_set(&sk->sk_zckey, 0);
3424 #ifdef CONFIG_NET_RX_BUSY_POLL
3426 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3429 sk->sk_max_pacing_rate = ~0UL;
3430 sk->sk_pacing_rate = ~0UL;
3431 WRITE_ONCE(sk->sk_pacing_shift, 10);
3432 sk->sk_incoming_cpu = -1;
3434 sk_rx_queue_clear(sk);
3436 * Before updating sk_refcnt, we must commit prior changes to memory
3437 * (Documentation/RCU/rculist_nulls.rst for details)
3440 refcount_set(&sk->sk_refcnt, 1);
3441 atomic_set(&sk->sk_drops, 0);
3443 EXPORT_SYMBOL(sock_init_data);
3445 void lock_sock_nested(struct sock *sk, int subclass)
3447 /* The sk_lock has mutex_lock() semantics here. */
3448 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3451 spin_lock_bh(&sk->sk_lock.slock);
3452 if (sock_owned_by_user_nocheck(sk))
3454 sk->sk_lock.owned = 1;
3455 spin_unlock_bh(&sk->sk_lock.slock);
3457 EXPORT_SYMBOL(lock_sock_nested);
3459 void release_sock(struct sock *sk)
3461 spin_lock_bh(&sk->sk_lock.slock);
3462 if (sk->sk_backlog.tail)
3465 /* Warning : release_cb() might need to release sk ownership,
3466 * ie call sock_release_ownership(sk) before us.
3468 if (sk->sk_prot->release_cb)
3469 sk->sk_prot->release_cb(sk);
3471 sock_release_ownership(sk);
3472 if (waitqueue_active(&sk->sk_lock.wq))
3473 wake_up(&sk->sk_lock.wq);
3474 spin_unlock_bh(&sk->sk_lock.slock);
3476 EXPORT_SYMBOL(release_sock);
3478 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3481 spin_lock_bh(&sk->sk_lock.slock);
3483 if (!sock_owned_by_user_nocheck(sk)) {
3485 * Fast path return with bottom halves disabled and
3486 * sock::sk_lock.slock held.
3488 * The 'mutex' is not contended and holding
3489 * sock::sk_lock.slock prevents all other lockers to
3490 * proceed so the corresponding unlock_sock_fast() can
3491 * avoid the slow path of release_sock() completely and
3492 * just release slock.
3494 * From a semantical POV this is equivalent to 'acquiring'
3495 * the 'mutex', hence the corresponding lockdep
3496 * mutex_release() has to happen in the fast path of
3497 * unlock_sock_fast().
3503 sk->sk_lock.owned = 1;
3504 __acquire(&sk->sk_lock.slock);
3505 spin_unlock_bh(&sk->sk_lock.slock);
3508 EXPORT_SYMBOL(__lock_sock_fast);
3510 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3511 bool timeval, bool time32)
3513 struct sock *sk = sock->sk;
3514 struct timespec64 ts;
3516 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3517 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3518 if (ts.tv_sec == -1)
3520 if (ts.tv_sec == 0) {
3521 ktime_t kt = ktime_get_real();
3522 sock_write_timestamp(sk, kt);
3523 ts = ktime_to_timespec64(kt);
3529 #ifdef CONFIG_COMPAT_32BIT_TIME
3531 return put_old_timespec32(&ts, userstamp);
3533 #ifdef CONFIG_SPARC64
3534 /* beware of padding in sparc64 timeval */
3535 if (timeval && !in_compat_syscall()) {
3536 struct __kernel_old_timeval __user tv = {
3537 .tv_sec = ts.tv_sec,
3538 .tv_usec = ts.tv_nsec,
3540 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3545 return put_timespec64(&ts, userstamp);
3547 EXPORT_SYMBOL(sock_gettstamp);
3549 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3551 if (!sock_flag(sk, flag)) {
3552 unsigned long previous_flags = sk->sk_flags;
3554 sock_set_flag(sk, flag);
3556 * we just set one of the two flags which require net
3557 * time stamping, but time stamping might have been on
3558 * already because of the other one
3560 if (sock_needs_netstamp(sk) &&
3561 !(previous_flags & SK_FLAGS_TIMESTAMP))
3562 net_enable_timestamp();
3566 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3567 int level, int type)
3569 struct sock_exterr_skb *serr;
3570 struct sk_buff *skb;
3574 skb = sock_dequeue_err_skb(sk);
3580 msg->msg_flags |= MSG_TRUNC;
3583 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3587 sock_recv_timestamp(msg, sk, skb);
3589 serr = SKB_EXT_ERR(skb);
3590 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3592 msg->msg_flags |= MSG_ERRQUEUE;
3600 EXPORT_SYMBOL(sock_recv_errqueue);
3603 * Get a socket option on an socket.
3605 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3606 * asynchronous errors should be reported by getsockopt. We assume
3607 * this means if you specify SO_ERROR (otherwise whats the point of it).
3609 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3610 char __user *optval, int __user *optlen)
3612 struct sock *sk = sock->sk;
3614 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3615 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3617 EXPORT_SYMBOL(sock_common_getsockopt);
3619 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3622 struct sock *sk = sock->sk;
3626 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3628 msg->msg_namelen = addr_len;
3631 EXPORT_SYMBOL(sock_common_recvmsg);
3634 * Set socket options on an inet socket.
3636 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3637 sockptr_t optval, unsigned int optlen)
3639 struct sock *sk = sock->sk;
3641 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3642 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3644 EXPORT_SYMBOL(sock_common_setsockopt);
3646 void sk_common_release(struct sock *sk)
3648 if (sk->sk_prot->destroy)
3649 sk->sk_prot->destroy(sk);
3652 * Observation: when sk_common_release is called, processes have
3653 * no access to socket. But net still has.
3654 * Step one, detach it from networking:
3656 * A. Remove from hash tables.
3659 sk->sk_prot->unhash(sk);
3662 * In this point socket cannot receive new packets, but it is possible
3663 * that some packets are in flight because some CPU runs receiver and
3664 * did hash table lookup before we unhashed socket. They will achieve
3665 * receive queue and will be purged by socket destructor.
3667 * Also we still have packets pending on receive queue and probably,
3668 * our own packets waiting in device queues. sock_destroy will drain
3669 * receive queue, but transmitted packets will delay socket destruction
3670 * until the last reference will be released.
3675 xfrm_sk_free_policy(sk);
3677 sk_refcnt_debug_release(sk);
3681 EXPORT_SYMBOL(sk_common_release);
3683 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3685 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3687 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3688 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3689 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3690 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3691 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3692 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3693 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3694 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3695 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3698 #ifdef CONFIG_PROC_FS
3699 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3701 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3703 int cpu, idx = prot->inuse_idx;
3706 for_each_possible_cpu(cpu)
3707 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3709 return res >= 0 ? res : 0;
3711 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3713 int sock_inuse_get(struct net *net)
3717 for_each_possible_cpu(cpu)
3718 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3723 EXPORT_SYMBOL_GPL(sock_inuse_get);
3725 static int __net_init sock_inuse_init_net(struct net *net)
3727 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3728 if (net->core.prot_inuse == NULL)
3733 static void __net_exit sock_inuse_exit_net(struct net *net)
3735 free_percpu(net->core.prot_inuse);
3738 static struct pernet_operations net_inuse_ops = {
3739 .init = sock_inuse_init_net,
3740 .exit = sock_inuse_exit_net,
3743 static __init int net_inuse_init(void)
3745 if (register_pernet_subsys(&net_inuse_ops))
3746 panic("Cannot initialize net inuse counters");
3751 core_initcall(net_inuse_init);
3753 static int assign_proto_idx(struct proto *prot)
3755 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3757 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3758 pr_err("PROTO_INUSE_NR exhausted\n");
3762 set_bit(prot->inuse_idx, proto_inuse_idx);
3766 static void release_proto_idx(struct proto *prot)
3768 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3769 clear_bit(prot->inuse_idx, proto_inuse_idx);
3772 static inline int assign_proto_idx(struct proto *prot)
3777 static inline void release_proto_idx(struct proto *prot)
3783 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3787 kfree(twsk_prot->twsk_slab_name);
3788 twsk_prot->twsk_slab_name = NULL;
3789 kmem_cache_destroy(twsk_prot->twsk_slab);
3790 twsk_prot->twsk_slab = NULL;
3793 static int tw_prot_init(const struct proto *prot)
3795 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3800 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3802 if (!twsk_prot->twsk_slab_name)
3805 twsk_prot->twsk_slab =
3806 kmem_cache_create(twsk_prot->twsk_slab_name,
3807 twsk_prot->twsk_obj_size, 0,
3808 SLAB_ACCOUNT | prot->slab_flags,
3810 if (!twsk_prot->twsk_slab) {
3811 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3819 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3823 kfree(rsk_prot->slab_name);
3824 rsk_prot->slab_name = NULL;
3825 kmem_cache_destroy(rsk_prot->slab);
3826 rsk_prot->slab = NULL;
3829 static int req_prot_init(const struct proto *prot)
3831 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3836 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3838 if (!rsk_prot->slab_name)
3841 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3842 rsk_prot->obj_size, 0,
3843 SLAB_ACCOUNT | prot->slab_flags,
3846 if (!rsk_prot->slab) {
3847 pr_crit("%s: Can't create request sock SLAB cache!\n",
3854 int proto_register(struct proto *prot, int alloc_slab)
3858 if (prot->memory_allocated && !prot->sysctl_mem) {
3859 pr_err("%s: missing sysctl_mem\n", prot->name);
3862 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3863 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3867 prot->slab = kmem_cache_create_usercopy(prot->name,
3869 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3871 prot->useroffset, prot->usersize,
3874 if (prot->slab == NULL) {
3875 pr_crit("%s: Can't create sock SLAB cache!\n",
3880 if (req_prot_init(prot))
3881 goto out_free_request_sock_slab;
3883 if (tw_prot_init(prot))
3884 goto out_free_timewait_sock_slab;
3887 mutex_lock(&proto_list_mutex);
3888 ret = assign_proto_idx(prot);
3890 mutex_unlock(&proto_list_mutex);
3891 goto out_free_timewait_sock_slab;
3893 list_add(&prot->node, &proto_list);
3894 mutex_unlock(&proto_list_mutex);
3897 out_free_timewait_sock_slab:
3899 tw_prot_cleanup(prot->twsk_prot);
3900 out_free_request_sock_slab:
3902 req_prot_cleanup(prot->rsk_prot);
3904 kmem_cache_destroy(prot->slab);
3910 EXPORT_SYMBOL(proto_register);
3912 void proto_unregister(struct proto *prot)
3914 mutex_lock(&proto_list_mutex);
3915 release_proto_idx(prot);
3916 list_del(&prot->node);
3917 mutex_unlock(&proto_list_mutex);
3919 kmem_cache_destroy(prot->slab);
3922 req_prot_cleanup(prot->rsk_prot);
3923 tw_prot_cleanup(prot->twsk_prot);
3925 EXPORT_SYMBOL(proto_unregister);
3927 int sock_load_diag_module(int family, int protocol)
3930 if (!sock_is_registered(family))
3933 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3934 NETLINK_SOCK_DIAG, family);
3938 if (family == AF_INET &&
3939 protocol != IPPROTO_RAW &&
3940 protocol < MAX_INET_PROTOS &&
3941 !rcu_access_pointer(inet_protos[protocol]))
3945 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3946 NETLINK_SOCK_DIAG, family, protocol);
3948 EXPORT_SYMBOL(sock_load_diag_module);
3950 #ifdef CONFIG_PROC_FS
3951 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3952 __acquires(proto_list_mutex)
3954 mutex_lock(&proto_list_mutex);
3955 return seq_list_start_head(&proto_list, *pos);
3958 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3960 return seq_list_next(v, &proto_list, pos);
3963 static void proto_seq_stop(struct seq_file *seq, void *v)
3964 __releases(proto_list_mutex)
3966 mutex_unlock(&proto_list_mutex);
3969 static char proto_method_implemented(const void *method)
3971 return method == NULL ? 'n' : 'y';
3973 static long sock_prot_memory_allocated(struct proto *proto)
3975 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3978 static const char *sock_prot_memory_pressure(struct proto *proto)
3980 return proto->memory_pressure != NULL ?
3981 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3984 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3987 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3988 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3991 sock_prot_inuse_get(seq_file_net(seq), proto),
3992 sock_prot_memory_allocated(proto),
3993 sock_prot_memory_pressure(proto),
3995 proto->slab == NULL ? "no" : "yes",
3996 module_name(proto->owner),
3997 proto_method_implemented(proto->close),
3998 proto_method_implemented(proto->connect),
3999 proto_method_implemented(proto->disconnect),
4000 proto_method_implemented(proto->accept),
4001 proto_method_implemented(proto->ioctl),
4002 proto_method_implemented(proto->init),
4003 proto_method_implemented(proto->destroy),
4004 proto_method_implemented(proto->shutdown),
4005 proto_method_implemented(proto->setsockopt),
4006 proto_method_implemented(proto->getsockopt),
4007 proto_method_implemented(proto->sendmsg),
4008 proto_method_implemented(proto->recvmsg),
4009 proto_method_implemented(proto->sendpage),
4010 proto_method_implemented(proto->bind),
4011 proto_method_implemented(proto->backlog_rcv),
4012 proto_method_implemented(proto->hash),
4013 proto_method_implemented(proto->unhash),
4014 proto_method_implemented(proto->get_port),
4015 proto_method_implemented(proto->enter_memory_pressure));
4018 static int proto_seq_show(struct seq_file *seq, void *v)
4020 if (v == &proto_list)
4021 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4030 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
4032 proto_seq_printf(seq, list_entry(v, struct proto, node));
4036 static const struct seq_operations proto_seq_ops = {
4037 .start = proto_seq_start,
4038 .next = proto_seq_next,
4039 .stop = proto_seq_stop,
4040 .show = proto_seq_show,
4043 static __net_init int proto_init_net(struct net *net)
4045 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4046 sizeof(struct seq_net_private)))
4052 static __net_exit void proto_exit_net(struct net *net)
4054 remove_proc_entry("protocols", net->proc_net);
4058 static __net_initdata struct pernet_operations proto_net_ops = {
4059 .init = proto_init_net,
4060 .exit = proto_exit_net,
4063 static int __init proto_init(void)
4065 return register_pernet_subsys(&proto_net_ops);
4068 subsys_initcall(proto_init);
4070 #endif /* PROC_FS */
4072 #ifdef CONFIG_NET_RX_BUSY_POLL
4073 bool sk_busy_loop_end(void *p, unsigned long start_time)
4075 struct sock *sk = p;
4077 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
4078 sk_busy_loop_timeout(sk, start_time);
4080 EXPORT_SYMBOL(sk_busy_loop_end);
4081 #endif /* CONFIG_NET_RX_BUSY_POLL */
4083 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4085 if (!sk->sk_prot->bind_add)
4087 return sk->sk_prot->bind_add(sk, addr, addr_len);
4089 EXPORT_SYMBOL(sock_bind_add);