1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/udp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
114 #include <linux/static_key.h>
115 #include <linux/memcontrol.h>
116 #include <linux/prefetch.h>
117 #include <linux/compat.h>
118 #include <linux/mroute.h>
119 #include <linux/mroute6.h>
120 #include <linux/icmpv6.h>
122 #include <linux/uaccess.h>
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <linux/skbuff_ref.h>
128 #include <net/net_namespace.h>
129 #include <net/request_sock.h>
130 #include <net/sock.h>
131 #include <net/proto_memory.h>
132 #include <linux/net_tstamp.h>
133 #include <net/xfrm.h>
134 #include <linux/ipsec.h>
135 #include <net/cls_cgroup.h>
136 #include <net/netprio_cgroup.h>
137 #include <linux/sock_diag.h>
139 #include <linux/filter.h>
140 #include <net/sock_reuseport.h>
141 #include <net/bpf_sk_storage.h>
143 #include <trace/events/sock.h>
146 #include <net/busy_poll.h>
147 #include <net/phonet/phonet.h>
149 #include <linux/ethtool.h>
153 static DEFINE_MUTEX(proto_list_mutex);
154 static LIST_HEAD(proto_list);
156 static void sock_def_write_space_wfree(struct sock *sk);
157 static void sock_def_write_space(struct sock *sk);
160 * sk_ns_capable - General socket capability test
161 * @sk: Socket to use a capability on or through
162 * @user_ns: The user namespace of the capability to use
163 * @cap: The capability to use
165 * Test to see if the opener of the socket had when the socket was
166 * created and the current process has the capability @cap in the user
167 * namespace @user_ns.
169 bool sk_ns_capable(const struct sock *sk,
170 struct user_namespace *user_ns, int cap)
172 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
173 ns_capable(user_ns, cap);
175 EXPORT_SYMBOL(sk_ns_capable);
178 * sk_capable - Socket global capability test
179 * @sk: Socket to use a capability on or through
180 * @cap: The global capability to use
182 * Test to see if the opener of the socket had when the socket was
183 * created and the current process has the capability @cap in all user
186 bool sk_capable(const struct sock *sk, int cap)
188 return sk_ns_capable(sk, &init_user_ns, cap);
190 EXPORT_SYMBOL(sk_capable);
193 * sk_net_capable - Network namespace socket capability test
194 * @sk: Socket to use a capability on or through
195 * @cap: The capability to use
197 * Test to see if the opener of the socket had when the socket was created
198 * and the current process has the capability @cap over the network namespace
199 * the socket is a member of.
201 bool sk_net_capable(const struct sock *sk, int cap)
203 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
205 EXPORT_SYMBOL(sk_net_capable);
208 * Each address family might have different locking rules, so we have
209 * one slock key per address family and separate keys for internal and
212 static struct lock_class_key af_family_keys[AF_MAX];
213 static struct lock_class_key af_family_kern_keys[AF_MAX];
214 static struct lock_class_key af_family_slock_keys[AF_MAX];
215 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
218 * Make lock validator output more readable. (we pre-construct these
219 * strings build-time, so that runtime initialization of socket
223 #define _sock_locks(x) \
224 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
225 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
226 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
227 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
228 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
229 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
230 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
231 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
232 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
233 x "27" , x "28" , x "AF_CAN" , \
234 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
235 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
236 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
237 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
238 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
242 static const char *const af_family_key_strings[AF_MAX+1] = {
243 _sock_locks("sk_lock-")
245 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
246 _sock_locks("slock-")
248 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
249 _sock_locks("clock-")
252 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
253 _sock_locks("k-sk_lock-")
255 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
256 _sock_locks("k-slock-")
258 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
259 _sock_locks("k-clock-")
261 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
262 _sock_locks("rlock-")
264 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
265 _sock_locks("wlock-")
267 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
268 _sock_locks("elock-")
272 * sk_callback_lock and sk queues locking rules are per-address-family,
273 * so split the lock classes by using a per-AF key:
275 static struct lock_class_key af_callback_keys[AF_MAX];
276 static struct lock_class_key af_rlock_keys[AF_MAX];
277 static struct lock_class_key af_wlock_keys[AF_MAX];
278 static struct lock_class_key af_elock_keys[AF_MAX];
279 static struct lock_class_key af_kern_callback_keys[AF_MAX];
281 /* Run time adjustable parameters. */
282 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
283 EXPORT_SYMBOL(sysctl_wmem_max);
284 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
285 EXPORT_SYMBOL(sysctl_rmem_max);
286 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
287 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
289 int sysctl_tstamp_allow_data __read_mostly = 1;
291 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
292 EXPORT_SYMBOL_GPL(memalloc_socks_key);
295 * sk_set_memalloc - sets %SOCK_MEMALLOC
296 * @sk: socket to set it on
298 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
299 * It's the responsibility of the admin to adjust min_free_kbytes
300 * to meet the requirements
302 void sk_set_memalloc(struct sock *sk)
304 sock_set_flag(sk, SOCK_MEMALLOC);
305 sk->sk_allocation |= __GFP_MEMALLOC;
306 static_branch_inc(&memalloc_socks_key);
308 EXPORT_SYMBOL_GPL(sk_set_memalloc);
310 void sk_clear_memalloc(struct sock *sk)
312 sock_reset_flag(sk, SOCK_MEMALLOC);
313 sk->sk_allocation &= ~__GFP_MEMALLOC;
314 static_branch_dec(&memalloc_socks_key);
317 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
318 * progress of swapping. SOCK_MEMALLOC may be cleared while
319 * it has rmem allocations due to the last swapfile being deactivated
320 * but there is a risk that the socket is unusable due to exceeding
321 * the rmem limits. Reclaim the reserves and obey rmem limits again.
325 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
327 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
330 unsigned int noreclaim_flag;
332 /* these should have been dropped before queueing */
333 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
335 noreclaim_flag = memalloc_noreclaim_save();
336 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
340 memalloc_noreclaim_restore(noreclaim_flag);
344 EXPORT_SYMBOL(__sk_backlog_rcv);
346 void sk_error_report(struct sock *sk)
348 sk->sk_error_report(sk);
350 switch (sk->sk_family) {
354 trace_inet_sk_error_report(sk);
360 EXPORT_SYMBOL(sk_error_report);
362 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
364 struct __kernel_sock_timeval tv;
366 if (timeo == MAX_SCHEDULE_TIMEOUT) {
370 tv.tv_sec = timeo / HZ;
371 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
374 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
375 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
376 *(struct old_timeval32 *)optval = tv32;
381 struct __kernel_old_timeval old_tv;
382 old_tv.tv_sec = tv.tv_sec;
383 old_tv.tv_usec = tv.tv_usec;
384 *(struct __kernel_old_timeval *)optval = old_tv;
385 return sizeof(old_tv);
388 *(struct __kernel_sock_timeval *)optval = tv;
391 EXPORT_SYMBOL(sock_get_timeout);
393 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
394 sockptr_t optval, int optlen, bool old_timeval)
396 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
397 struct old_timeval32 tv32;
399 if (optlen < sizeof(tv32))
402 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
404 tv->tv_sec = tv32.tv_sec;
405 tv->tv_usec = tv32.tv_usec;
406 } else if (old_timeval) {
407 struct __kernel_old_timeval old_tv;
409 if (optlen < sizeof(old_tv))
411 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
413 tv->tv_sec = old_tv.tv_sec;
414 tv->tv_usec = old_tv.tv_usec;
416 if (optlen < sizeof(*tv))
418 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
424 EXPORT_SYMBOL(sock_copy_user_timeval);
426 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
429 struct __kernel_sock_timeval tv;
430 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
436 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
440 static int warned __read_mostly;
442 WRITE_ONCE(*timeo_p, 0);
443 if (warned < 10 && net_ratelimit()) {
445 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
446 __func__, current->comm, task_pid_nr(current));
450 val = MAX_SCHEDULE_TIMEOUT;
451 if ((tv.tv_sec || tv.tv_usec) &&
452 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
453 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
455 WRITE_ONCE(*timeo_p, val);
459 static bool sock_needs_netstamp(const struct sock *sk)
461 switch (sk->sk_family) {
470 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
472 if (sk->sk_flags & flags) {
473 sk->sk_flags &= ~flags;
474 if (sock_needs_netstamp(sk) &&
475 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
476 net_disable_timestamp();
481 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
484 struct sk_buff_head *list = &sk->sk_receive_queue;
486 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) {
487 atomic_inc(&sk->sk_drops);
488 trace_sock_rcvqueue_full(sk, skb);
492 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
493 atomic_inc(&sk->sk_drops);
498 skb_set_owner_r(skb, sk);
500 /* we escape from rcu protected region, make sure we dont leak
505 spin_lock_irqsave(&list->lock, flags);
506 sock_skb_set_dropcount(sk, skb);
507 __skb_queue_tail(list, skb);
508 spin_unlock_irqrestore(&list->lock, flags);
510 if (!sock_flag(sk, SOCK_DEAD))
511 sk->sk_data_ready(sk);
514 EXPORT_SYMBOL(__sock_queue_rcv_skb);
516 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
517 enum skb_drop_reason *reason)
519 enum skb_drop_reason drop_reason;
522 err = sk_filter(sk, skb);
524 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
527 err = __sock_queue_rcv_skb(sk, skb);
530 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
533 drop_reason = SKB_DROP_REASON_PROTO_MEM;
536 drop_reason = SKB_NOT_DROPPED_YET;
541 *reason = drop_reason;
544 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
546 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
547 const int nested, unsigned int trim_cap, bool refcounted)
549 int rc = NET_RX_SUCCESS;
551 if (sk_filter_trim_cap(sk, skb, trim_cap))
552 goto discard_and_relse;
556 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) {
557 atomic_inc(&sk->sk_drops);
558 goto discard_and_relse;
561 bh_lock_sock_nested(sk);
564 if (!sock_owned_by_user(sk)) {
566 * trylock + unlock semantics:
568 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
570 rc = sk_backlog_rcv(sk, skb);
572 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
573 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
575 atomic_inc(&sk->sk_drops);
576 goto discard_and_relse;
588 EXPORT_SYMBOL(__sk_receive_skb);
590 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
592 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
594 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
596 struct dst_entry *dst = __sk_dst_get(sk);
598 if (dst && dst->obsolete &&
599 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
600 dst, cookie) == NULL) {
601 sk_tx_queue_clear(sk);
602 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
603 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
610 EXPORT_SYMBOL(__sk_dst_check);
612 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
614 struct dst_entry *dst = sk_dst_get(sk);
616 if (dst && dst->obsolete &&
617 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
618 dst, cookie) == NULL) {
626 EXPORT_SYMBOL(sk_dst_check);
628 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
630 int ret = -ENOPROTOOPT;
631 #ifdef CONFIG_NETDEVICES
632 struct net *net = sock_net(sk);
636 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
643 /* Paired with all READ_ONCE() done locklessly. */
644 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
646 if (sk->sk_prot->rehash)
647 sk->sk_prot->rehash(sk);
658 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
664 ret = sock_bindtoindex_locked(sk, ifindex);
670 EXPORT_SYMBOL(sock_bindtoindex);
672 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
674 int ret = -ENOPROTOOPT;
675 #ifdef CONFIG_NETDEVICES
676 struct net *net = sock_net(sk);
677 char devname[IFNAMSIZ];
684 /* Bind this socket to a particular device like "eth0",
685 * as specified in the passed interface name. If the
686 * name is "" or the option length is zero the socket
689 if (optlen > IFNAMSIZ - 1)
690 optlen = IFNAMSIZ - 1;
691 memset(devname, 0, sizeof(devname));
694 if (copy_from_sockptr(devname, optval, optlen))
698 if (devname[0] != '\0') {
699 struct net_device *dev;
702 dev = dev_get_by_name_rcu(net, devname);
704 index = dev->ifindex;
711 sockopt_lock_sock(sk);
712 ret = sock_bindtoindex_locked(sk, index);
713 sockopt_release_sock(sk);
720 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
721 sockptr_t optlen, int len)
723 int ret = -ENOPROTOOPT;
724 #ifdef CONFIG_NETDEVICES
725 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
726 struct net *net = sock_net(sk);
727 char devname[IFNAMSIZ];
729 if (bound_dev_if == 0) {
738 ret = netdev_get_name(net, devname, bound_dev_if);
742 len = strlen(devname) + 1;
745 if (copy_to_sockptr(optval, devname, len))
750 if (copy_to_sockptr(optlen, &len, sizeof(int)))
761 bool sk_mc_loop(const struct sock *sk)
763 if (dev_recursion_level())
767 /* IPV6_ADDRFORM can change sk->sk_family under us. */
768 switch (READ_ONCE(sk->sk_family)) {
770 return inet_test_bit(MC_LOOP, sk);
771 #if IS_ENABLED(CONFIG_IPV6)
773 return inet6_test_bit(MC6_LOOP, sk);
779 EXPORT_SYMBOL(sk_mc_loop);
781 void sock_set_reuseaddr(struct sock *sk)
784 sk->sk_reuse = SK_CAN_REUSE;
787 EXPORT_SYMBOL(sock_set_reuseaddr);
789 void sock_set_reuseport(struct sock *sk)
792 sk->sk_reuseport = true;
795 EXPORT_SYMBOL(sock_set_reuseport);
797 void sock_no_linger(struct sock *sk)
800 WRITE_ONCE(sk->sk_lingertime, 0);
801 sock_set_flag(sk, SOCK_LINGER);
804 EXPORT_SYMBOL(sock_no_linger);
806 void sock_set_priority(struct sock *sk, u32 priority)
808 WRITE_ONCE(sk->sk_priority, priority);
810 EXPORT_SYMBOL(sock_set_priority);
812 void sock_set_sndtimeo(struct sock *sk, s64 secs)
815 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
816 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
818 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
821 EXPORT_SYMBOL(sock_set_sndtimeo);
823 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
826 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
827 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
828 sock_set_flag(sk, SOCK_RCVTSTAMP);
829 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
831 sock_reset_flag(sk, SOCK_RCVTSTAMP);
832 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
836 void sock_enable_timestamps(struct sock *sk)
839 __sock_set_timestamps(sk, true, false, true);
842 EXPORT_SYMBOL(sock_enable_timestamps);
844 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
847 case SO_TIMESTAMP_OLD:
848 __sock_set_timestamps(sk, valbool, false, false);
850 case SO_TIMESTAMP_NEW:
851 __sock_set_timestamps(sk, valbool, true, false);
853 case SO_TIMESTAMPNS_OLD:
854 __sock_set_timestamps(sk, valbool, false, true);
856 case SO_TIMESTAMPNS_NEW:
857 __sock_set_timestamps(sk, valbool, true, true);
862 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
864 struct net *net = sock_net(sk);
865 struct net_device *dev = NULL;
870 if (sk->sk_bound_dev_if)
871 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
874 pr_err("%s: sock not bind to device\n", __func__);
878 num = ethtool_get_phc_vclocks(dev, &vclock_index);
881 for (i = 0; i < num; i++) {
882 if (*(vclock_index + i) == phc_index) {
894 WRITE_ONCE(sk->sk_bind_phc, phc_index);
899 int sock_set_timestamping(struct sock *sk, int optname,
900 struct so_timestamping timestamping)
902 int val = timestamping.flags;
905 if (val & ~SOF_TIMESTAMPING_MASK)
908 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
909 !(val & SOF_TIMESTAMPING_OPT_ID))
912 if (val & SOF_TIMESTAMPING_OPT_ID &&
913 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
915 if ((1 << sk->sk_state) &
916 (TCPF_CLOSE | TCPF_LISTEN))
918 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
919 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
921 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
923 atomic_set(&sk->sk_tskey, 0);
927 if (val & SOF_TIMESTAMPING_OPT_STATS &&
928 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
931 if (val & SOF_TIMESTAMPING_BIND_PHC) {
932 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
937 WRITE_ONCE(sk->sk_tsflags, val);
938 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
940 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
941 sock_enable_timestamp(sk,
942 SOCK_TIMESTAMPING_RX_SOFTWARE);
944 sock_disable_timestamp(sk,
945 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
949 void sock_set_keepalive(struct sock *sk)
952 if (sk->sk_prot->keepalive)
953 sk->sk_prot->keepalive(sk, true);
954 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
957 EXPORT_SYMBOL(sock_set_keepalive);
959 static void __sock_set_rcvbuf(struct sock *sk, int val)
961 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
962 * as a negative value.
964 val = min_t(int, val, INT_MAX / 2);
965 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
967 /* We double it on the way in to account for "struct sk_buff" etc.
968 * overhead. Applications assume that the SO_RCVBUF setting they make
969 * will allow that much actual data to be received on that socket.
971 * Applications are unaware that "struct sk_buff" and other overheads
972 * allocate from the receive buffer during socket buffer allocation.
974 * And after considering the possible alternatives, returning the value
975 * we actually used in getsockopt is the most desirable behavior.
977 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
980 void sock_set_rcvbuf(struct sock *sk, int val)
983 __sock_set_rcvbuf(sk, val);
986 EXPORT_SYMBOL(sock_set_rcvbuf);
988 static void __sock_set_mark(struct sock *sk, u32 val)
990 if (val != sk->sk_mark) {
991 WRITE_ONCE(sk->sk_mark, val);
996 void sock_set_mark(struct sock *sk, u32 val)
999 __sock_set_mark(sk, val);
1002 EXPORT_SYMBOL(sock_set_mark);
1004 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1006 /* Round down bytes to multiple of pages */
1007 bytes = round_down(bytes, PAGE_SIZE);
1009 WARN_ON(bytes > sk->sk_reserved_mem);
1010 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1014 static int sock_reserve_memory(struct sock *sk, int bytes)
1020 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1026 pages = sk_mem_pages(bytes);
1028 /* pre-charge to memcg */
1029 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1030 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1034 /* pre-charge to forward_alloc */
1035 sk_memory_allocated_add(sk, pages);
1036 allocated = sk_memory_allocated(sk);
1037 /* If the system goes into memory pressure with this
1038 * precharge, give up and return error.
1040 if (allocated > sk_prot_mem_limits(sk, 1)) {
1041 sk_memory_allocated_sub(sk, pages);
1042 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1045 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1047 WRITE_ONCE(sk->sk_reserved_mem,
1048 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1053 #ifdef CONFIG_PAGE_POOL
1055 /* This is the number of tokens that the user can SO_DEVMEM_DONTNEED in
1056 * 1 syscall. The limit exists to limit the amount of memory the kernel
1057 * allocates to copy these tokens.
1059 #define MAX_DONTNEED_TOKENS 128
1061 static noinline_for_stack int
1062 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen)
1064 unsigned int num_tokens, i, j, k, netmem_num = 0;
1065 struct dmabuf_token *tokens;
1066 netmem_ref netmems[16];
1072 if (optlen % sizeof(struct dmabuf_token) ||
1073 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS)
1076 tokens = kvmalloc_array(optlen, sizeof(*tokens), GFP_KERNEL);
1080 num_tokens = optlen / sizeof(struct dmabuf_token);
1081 if (copy_from_sockptr(tokens, optval, optlen)) {
1086 xa_lock_bh(&sk->sk_user_frags);
1087 for (i = 0; i < num_tokens; i++) {
1088 for (j = 0; j < tokens[i].token_count; j++) {
1089 netmem_ref netmem = (__force netmem_ref)__xa_erase(
1090 &sk->sk_user_frags, tokens[i].token_start + j);
1093 !WARN_ON_ONCE(!netmem_is_net_iov(netmem))) {
1094 netmems[netmem_num++] = netmem;
1095 if (netmem_num == ARRAY_SIZE(netmems)) {
1096 xa_unlock_bh(&sk->sk_user_frags);
1097 for (k = 0; k < netmem_num; k++)
1098 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1100 xa_lock_bh(&sk->sk_user_frags);
1107 xa_unlock_bh(&sk->sk_user_frags);
1108 for (k = 0; k < netmem_num; k++)
1109 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1116 void sockopt_lock_sock(struct sock *sk)
1118 /* When current->bpf_ctx is set, the setsockopt is called from
1119 * a bpf prog. bpf has ensured the sk lock has been
1120 * acquired before calling setsockopt().
1122 if (has_current_bpf_ctx())
1127 EXPORT_SYMBOL(sockopt_lock_sock);
1129 void sockopt_release_sock(struct sock *sk)
1131 if (has_current_bpf_ctx())
1136 EXPORT_SYMBOL(sockopt_release_sock);
1138 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1140 return has_current_bpf_ctx() || ns_capable(ns, cap);
1142 EXPORT_SYMBOL(sockopt_ns_capable);
1144 bool sockopt_capable(int cap)
1146 return has_current_bpf_ctx() || capable(cap);
1148 EXPORT_SYMBOL(sockopt_capable);
1150 static int sockopt_validate_clockid(__kernel_clockid_t value)
1153 case CLOCK_REALTIME:
1154 case CLOCK_MONOTONIC:
1162 * This is meant for all protocols to use and covers goings on
1163 * at the socket level. Everything here is generic.
1166 int sk_setsockopt(struct sock *sk, int level, int optname,
1167 sockptr_t optval, unsigned int optlen)
1169 struct so_timestamping timestamping;
1170 struct socket *sock = sk->sk_socket;
1171 struct sock_txtime sk_txtime;
1178 * Options without arguments
1181 if (optname == SO_BINDTODEVICE)
1182 return sock_setbindtodevice(sk, optval, optlen);
1184 if (optlen < sizeof(int))
1187 if (copy_from_sockptr(&val, optval, sizeof(val)))
1190 valbool = val ? 1 : 0;
1192 /* handle options which do not require locking the socket. */
1195 if ((val >= 0 && val <= 6) ||
1196 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1197 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1198 sock_set_priority(sk, val);
1203 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1206 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1209 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1215 return -ENOPROTOOPT;
1216 #ifdef CONFIG_NET_RX_BUSY_POLL
1220 WRITE_ONCE(sk->sk_ll_usec, val);
1222 case SO_PREFER_BUSY_POLL:
1223 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1225 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1227 case SO_BUSY_POLL_BUDGET:
1228 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1229 !sockopt_capable(CAP_NET_ADMIN))
1231 if (val < 0 || val > U16_MAX)
1233 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1236 case SO_MAX_PACING_RATE:
1238 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1239 unsigned long pacing_rate;
1241 if (sizeof(ulval) != sizeof(val) &&
1242 optlen >= sizeof(ulval) &&
1243 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1247 cmpxchg(&sk->sk_pacing_status,
1250 /* Pairs with READ_ONCE() from sk_getsockopt() */
1251 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1252 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1253 if (ulval < pacing_rate)
1254 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1258 if (val < -1 || val > 1)
1260 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1261 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1262 /* Paired with READ_ONCE() in tcp_rtx_synack()
1263 * and sk_getsockopt().
1265 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1269 int (*set_peek_off)(struct sock *sk, int val);
1271 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1273 ret = set_peek_off(sk, val);
1278 #ifdef CONFIG_PAGE_POOL
1279 case SO_DEVMEM_DONTNEED:
1280 return sock_devmem_dontneed(sk, optval, optlen);
1284 sockopt_lock_sock(sk);
1288 if (val && !sockopt_capable(CAP_NET_ADMIN))
1291 sock_valbool_flag(sk, SOCK_DBG, valbool);
1294 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1297 sk->sk_reuseport = valbool;
1300 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1304 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1307 /* Don't error on this BSD doesn't and if you think
1308 * about it this is right. Otherwise apps have to
1309 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1310 * are treated in BSD as hints
1312 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1314 /* Ensure val * 2 fits into an int, to prevent max_t()
1315 * from treating it as a negative value.
1317 val = min_t(int, val, INT_MAX / 2);
1318 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1319 WRITE_ONCE(sk->sk_sndbuf,
1320 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1321 /* Wake up sending tasks if we upped the value. */
1322 sk->sk_write_space(sk);
1325 case SO_SNDBUFFORCE:
1326 if (!sockopt_capable(CAP_NET_ADMIN)) {
1331 /* No negative values (to prevent underflow, as val will be
1339 /* Don't error on this BSD doesn't and if you think
1340 * about it this is right. Otherwise apps have to
1341 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1342 * are treated in BSD as hints
1344 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1347 case SO_RCVBUFFORCE:
1348 if (!sockopt_capable(CAP_NET_ADMIN)) {
1353 /* No negative values (to prevent underflow, as val will be
1356 __sock_set_rcvbuf(sk, max(val, 0));
1360 if (sk->sk_prot->keepalive)
1361 sk->sk_prot->keepalive(sk, valbool);
1362 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1366 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1370 sk->sk_no_check_tx = valbool;
1374 if (optlen < sizeof(ling)) {
1375 ret = -EINVAL; /* 1003.1g */
1378 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1382 if (!ling.l_onoff) {
1383 sock_reset_flag(sk, SOCK_LINGER);
1385 unsigned long t_sec = ling.l_linger;
1387 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1388 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1390 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1391 sock_set_flag(sk, SOCK_LINGER);
1398 case SO_TIMESTAMP_OLD:
1399 case SO_TIMESTAMP_NEW:
1400 case SO_TIMESTAMPNS_OLD:
1401 case SO_TIMESTAMPNS_NEW:
1402 sock_set_timestamp(sk, optname, valbool);
1405 case SO_TIMESTAMPING_NEW:
1406 case SO_TIMESTAMPING_OLD:
1407 if (optlen == sizeof(timestamping)) {
1408 if (copy_from_sockptr(×tamping, optval,
1409 sizeof(timestamping))) {
1414 memset(×tamping, 0, sizeof(timestamping));
1415 timestamping.flags = val;
1417 ret = sock_set_timestamping(sk, optname, timestamping);
1422 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1427 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1429 ret = set_rcvlowat(sk, val);
1431 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1434 case SO_RCVTIMEO_OLD:
1435 case SO_RCVTIMEO_NEW:
1436 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1437 optlen, optname == SO_RCVTIMEO_OLD);
1440 case SO_SNDTIMEO_OLD:
1441 case SO_SNDTIMEO_NEW:
1442 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1443 optlen, optname == SO_SNDTIMEO_OLD);
1446 case SO_ATTACH_FILTER: {
1447 struct sock_fprog fprog;
1449 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1451 ret = sk_attach_filter(&fprog, sk);
1456 if (optlen == sizeof(u32)) {
1460 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1463 ret = sk_attach_bpf(ufd, sk);
1467 case SO_ATTACH_REUSEPORT_CBPF: {
1468 struct sock_fprog fprog;
1470 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1472 ret = sk_reuseport_attach_filter(&fprog, sk);
1475 case SO_ATTACH_REUSEPORT_EBPF:
1477 if (optlen == sizeof(u32)) {
1481 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1484 ret = sk_reuseport_attach_bpf(ufd, sk);
1488 case SO_DETACH_REUSEPORT_BPF:
1489 ret = reuseport_detach_prog(sk);
1492 case SO_DETACH_FILTER:
1493 ret = sk_detach_filter(sk);
1496 case SO_LOCK_FILTER:
1497 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1500 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1504 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1505 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1510 __sock_set_mark(sk, val);
1513 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1517 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1520 case SO_WIFI_STATUS:
1521 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1525 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1528 case SO_SELECT_ERR_QUEUE:
1529 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1533 case SO_INCOMING_CPU:
1534 reuseport_update_incoming_cpu(sk, val);
1539 dst_negative_advice(sk);
1543 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1544 if (!(sk_is_tcp(sk) ||
1545 (sk->sk_type == SOCK_DGRAM &&
1546 sk->sk_protocol == IPPROTO_UDP)))
1548 } else if (sk->sk_family != PF_RDS) {
1552 if (val < 0 || val > 1)
1555 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1560 if (optlen != sizeof(struct sock_txtime)) {
1563 } else if (copy_from_sockptr(&sk_txtime, optval,
1564 sizeof(struct sock_txtime))) {
1567 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1571 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1572 * scheduler has enough safe guards.
1574 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1575 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1580 ret = sockopt_validate_clockid(sk_txtime.clockid);
1584 sock_valbool_flag(sk, SOCK_TXTIME, true);
1585 sk->sk_clockid = sk_txtime.clockid;
1586 sk->sk_txtime_deadline_mode =
1587 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1588 sk->sk_txtime_report_errors =
1589 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1592 case SO_BINDTOIFINDEX:
1593 ret = sock_bindtoindex_locked(sk, val);
1597 if (val & ~SOCK_BUF_LOCK_MASK) {
1601 sk->sk_userlocks = val | (sk->sk_userlocks &
1602 ~SOCK_BUF_LOCK_MASK);
1605 case SO_RESERVE_MEM:
1614 delta = val - sk->sk_reserved_mem;
1616 sock_release_reserved_memory(sk, -delta);
1618 ret = sock_reserve_memory(sk, delta);
1626 sockopt_release_sock(sk);
1630 int sock_setsockopt(struct socket *sock, int level, int optname,
1631 sockptr_t optval, unsigned int optlen)
1633 return sk_setsockopt(sock->sk, level, optname,
1636 EXPORT_SYMBOL(sock_setsockopt);
1638 static const struct cred *sk_get_peer_cred(struct sock *sk)
1640 const struct cred *cred;
1642 spin_lock(&sk->sk_peer_lock);
1643 cred = get_cred(sk->sk_peer_cred);
1644 spin_unlock(&sk->sk_peer_lock);
1649 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1650 struct ucred *ucred)
1652 ucred->pid = pid_vnr(pid);
1653 ucred->uid = ucred->gid = -1;
1655 struct user_namespace *current_ns = current_user_ns();
1657 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1658 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1662 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1664 struct user_namespace *user_ns = current_user_ns();
1667 for (i = 0; i < src->ngroups; i++) {
1668 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1670 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1677 int sk_getsockopt(struct sock *sk, int level, int optname,
1678 sockptr_t optval, sockptr_t optlen)
1680 struct socket *sock = sk->sk_socket;
1685 unsigned long ulval;
1687 struct old_timeval32 tm32;
1688 struct __kernel_old_timeval tm;
1689 struct __kernel_sock_timeval stm;
1690 struct sock_txtime txtime;
1691 struct so_timestamping timestamping;
1694 int lv = sizeof(int);
1697 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1702 memset(&v, 0, sizeof(v));
1706 v.val = sock_flag(sk, SOCK_DBG);
1710 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1714 v.val = sock_flag(sk, SOCK_BROADCAST);
1718 v.val = READ_ONCE(sk->sk_sndbuf);
1722 v.val = READ_ONCE(sk->sk_rcvbuf);
1726 v.val = sk->sk_reuse;
1730 v.val = sk->sk_reuseport;
1734 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1738 v.val = sk->sk_type;
1742 v.val = sk->sk_protocol;
1746 v.val = sk->sk_family;
1750 v.val = -sock_error(sk);
1752 v.val = xchg(&sk->sk_err_soft, 0);
1756 v.val = sock_flag(sk, SOCK_URGINLINE);
1760 v.val = sk->sk_no_check_tx;
1764 v.val = READ_ONCE(sk->sk_priority);
1768 lv = sizeof(v.ling);
1769 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1770 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1776 case SO_TIMESTAMP_OLD:
1777 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1778 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1779 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1782 case SO_TIMESTAMPNS_OLD:
1783 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1786 case SO_TIMESTAMP_NEW:
1787 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1790 case SO_TIMESTAMPNS_NEW:
1791 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1794 case SO_TIMESTAMPING_OLD:
1795 case SO_TIMESTAMPING_NEW:
1796 lv = sizeof(v.timestamping);
1797 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1798 * returning the flags when they were set through the same option.
1799 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1801 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1802 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1803 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1807 case SO_RCVTIMEO_OLD:
1808 case SO_RCVTIMEO_NEW:
1809 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1810 SO_RCVTIMEO_OLD == optname);
1813 case SO_SNDTIMEO_OLD:
1814 case SO_SNDTIMEO_NEW:
1815 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1816 SO_SNDTIMEO_OLD == optname);
1820 v.val = READ_ONCE(sk->sk_rcvlowat);
1828 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1832 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1837 struct ucred peercred;
1838 if (len > sizeof(peercred))
1839 len = sizeof(peercred);
1841 spin_lock(&sk->sk_peer_lock);
1842 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1843 spin_unlock(&sk->sk_peer_lock);
1845 if (copy_to_sockptr(optval, &peercred, len))
1852 struct pid *peer_pid;
1853 struct file *pidfd_file = NULL;
1856 if (len > sizeof(pidfd))
1857 len = sizeof(pidfd);
1859 spin_lock(&sk->sk_peer_lock);
1860 peer_pid = get_pid(sk->sk_peer_pid);
1861 spin_unlock(&sk->sk_peer_lock);
1866 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1871 if (copy_to_sockptr(optval, &pidfd, len) ||
1872 copy_to_sockptr(optlen, &len, sizeof(int))) {
1873 put_unused_fd(pidfd);
1879 fd_install(pidfd, pidfd_file);
1885 const struct cred *cred;
1888 cred = sk_get_peer_cred(sk);
1892 n = cred->group_info->ngroups;
1893 if (len < n * sizeof(gid_t)) {
1894 len = n * sizeof(gid_t);
1896 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1898 len = n * sizeof(gid_t);
1900 ret = groups_to_user(optval, cred->group_info);
1909 struct sockaddr_storage address;
1911 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1916 if (copy_to_sockptr(optval, &address, len))
1921 /* Dubious BSD thing... Probably nobody even uses it, but
1922 * the UNIX standard wants it for whatever reason... -DaveM
1925 v.val = sk->sk_state == TCP_LISTEN;
1929 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1933 return security_socket_getpeersec_stream(sock,
1934 optval, optlen, len);
1937 v.val = READ_ONCE(sk->sk_mark);
1941 v.val = sock_flag(sk, SOCK_RCVMARK);
1945 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1948 case SO_WIFI_STATUS:
1949 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1953 if (!READ_ONCE(sock->ops)->set_peek_off)
1956 v.val = READ_ONCE(sk->sk_peek_off);
1959 v.val = sock_flag(sk, SOCK_NOFCS);
1962 case SO_BINDTODEVICE:
1963 return sock_getbindtodevice(sk, optval, optlen, len);
1966 len = sk_get_filter(sk, optval, len);
1972 case SO_LOCK_FILTER:
1973 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1976 case SO_BPF_EXTENSIONS:
1977 v.val = bpf_tell_extensions();
1980 case SO_SELECT_ERR_QUEUE:
1981 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1984 #ifdef CONFIG_NET_RX_BUSY_POLL
1986 v.val = READ_ONCE(sk->sk_ll_usec);
1988 case SO_PREFER_BUSY_POLL:
1989 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1993 case SO_MAX_PACING_RATE:
1994 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1995 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1996 lv = sizeof(v.ulval);
1997 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
2000 v.val = min_t(unsigned long, ~0U,
2001 READ_ONCE(sk->sk_max_pacing_rate));
2005 case SO_INCOMING_CPU:
2006 v.val = READ_ONCE(sk->sk_incoming_cpu);
2011 u32 meminfo[SK_MEMINFO_VARS];
2013 sk_get_meminfo(sk, meminfo);
2015 len = min_t(unsigned int, len, sizeof(meminfo));
2016 if (copy_to_sockptr(optval, &meminfo, len))
2022 #ifdef CONFIG_NET_RX_BUSY_POLL
2023 case SO_INCOMING_NAPI_ID:
2024 v.val = READ_ONCE(sk->sk_napi_id);
2026 /* aggregate non-NAPI IDs down to 0 */
2027 if (v.val < MIN_NAPI_ID)
2037 v.val64 = sock_gen_cookie(sk);
2041 v.val = sock_flag(sk, SOCK_ZEROCOPY);
2045 lv = sizeof(v.txtime);
2046 v.txtime.clockid = sk->sk_clockid;
2047 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
2048 SOF_TXTIME_DEADLINE_MODE : 0;
2049 v.txtime.flags |= sk->sk_txtime_report_errors ?
2050 SOF_TXTIME_REPORT_ERRORS : 0;
2053 case SO_BINDTOIFINDEX:
2054 v.val = READ_ONCE(sk->sk_bound_dev_if);
2057 case SO_NETNS_COOKIE:
2061 v.val64 = sock_net(sk)->net_cookie;
2065 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
2068 case SO_RESERVE_MEM:
2069 v.val = READ_ONCE(sk->sk_reserved_mem);
2073 /* Paired with WRITE_ONCE() in sk_setsockopt() */
2074 v.val = READ_ONCE(sk->sk_txrehash);
2078 /* We implement the SO_SNDLOWAT etc to not be settable
2081 return -ENOPROTOOPT;
2086 if (copy_to_sockptr(optval, &v, len))
2089 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2095 * Initialize an sk_lock.
2097 * (We also register the sk_lock with the lock validator.)
2099 static inline void sock_lock_init(struct sock *sk)
2101 if (sk->sk_kern_sock)
2102 sock_lock_init_class_and_name(
2104 af_family_kern_slock_key_strings[sk->sk_family],
2105 af_family_kern_slock_keys + sk->sk_family,
2106 af_family_kern_key_strings[sk->sk_family],
2107 af_family_kern_keys + sk->sk_family);
2109 sock_lock_init_class_and_name(
2111 af_family_slock_key_strings[sk->sk_family],
2112 af_family_slock_keys + sk->sk_family,
2113 af_family_key_strings[sk->sk_family],
2114 af_family_keys + sk->sk_family);
2118 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2119 * even temporarily, because of RCU lookups. sk_node should also be left as is.
2120 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2122 static void sock_copy(struct sock *nsk, const struct sock *osk)
2124 const struct proto *prot = READ_ONCE(osk->sk_prot);
2125 #ifdef CONFIG_SECURITY_NETWORK
2126 void *sptr = nsk->sk_security;
2129 /* If we move sk_tx_queue_mapping out of the private section,
2130 * we must check if sk_tx_queue_clear() is called after
2131 * sock_copy() in sk_clone_lock().
2133 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2134 offsetof(struct sock, sk_dontcopy_begin) ||
2135 offsetof(struct sock, sk_tx_queue_mapping) >=
2136 offsetof(struct sock, sk_dontcopy_end));
2138 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2140 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2141 prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2142 /* alloc is larger than struct, see sk_prot_alloc() */);
2144 #ifdef CONFIG_SECURITY_NETWORK
2145 nsk->sk_security = sptr;
2146 security_sk_clone(osk, nsk);
2150 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2154 struct kmem_cache *slab;
2158 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2161 if (want_init_on_alloc(priority))
2162 sk_prot_clear_nulls(sk, prot->obj_size);
2164 sk = kmalloc(prot->obj_size, priority);
2167 if (security_sk_alloc(sk, family, priority))
2170 if (!try_module_get(prot->owner))
2177 security_sk_free(sk);
2180 kmem_cache_free(slab, sk);
2186 static void sk_prot_free(struct proto *prot, struct sock *sk)
2188 struct kmem_cache *slab;
2189 struct module *owner;
2191 owner = prot->owner;
2194 cgroup_sk_free(&sk->sk_cgrp_data);
2195 mem_cgroup_sk_free(sk);
2196 security_sk_free(sk);
2198 kmem_cache_free(slab, sk);
2205 * sk_alloc - All socket objects are allocated here
2206 * @net: the applicable net namespace
2207 * @family: protocol family
2208 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2209 * @prot: struct proto associated with this new sock instance
2210 * @kern: is this to be a kernel socket?
2212 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2213 struct proto *prot, int kern)
2217 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2219 sk->sk_family = family;
2221 * See comment in struct sock definition to understand
2222 * why we need sk_prot_creator -acme
2224 sk->sk_prot = sk->sk_prot_creator = prot;
2225 sk->sk_kern_sock = kern;
2227 sk->sk_net_refcnt = kern ? 0 : 1;
2228 if (likely(sk->sk_net_refcnt)) {
2229 get_net_track(net, &sk->ns_tracker, priority);
2230 sock_inuse_add(net, 1);
2232 __netns_tracker_alloc(net, &sk->ns_tracker,
2236 sock_net_set(sk, net);
2237 refcount_set(&sk->sk_wmem_alloc, 1);
2239 mem_cgroup_sk_alloc(sk);
2240 cgroup_sk_alloc(&sk->sk_cgrp_data);
2241 sock_update_classid(&sk->sk_cgrp_data);
2242 sock_update_netprioidx(&sk->sk_cgrp_data);
2243 sk_tx_queue_clear(sk);
2248 EXPORT_SYMBOL(sk_alloc);
2250 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2251 * grace period. This is the case for UDP sockets and TCP listeners.
2253 static void __sk_destruct(struct rcu_head *head)
2255 struct sock *sk = container_of(head, struct sock, sk_rcu);
2256 struct sk_filter *filter;
2258 if (sk->sk_destruct)
2259 sk->sk_destruct(sk);
2261 filter = rcu_dereference_check(sk->sk_filter,
2262 refcount_read(&sk->sk_wmem_alloc) == 0);
2264 sk_filter_uncharge(sk, filter);
2265 RCU_INIT_POINTER(sk->sk_filter, NULL);
2268 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2270 #ifdef CONFIG_BPF_SYSCALL
2271 bpf_sk_storage_free(sk);
2274 if (atomic_read(&sk->sk_omem_alloc))
2275 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2276 __func__, atomic_read(&sk->sk_omem_alloc));
2278 if (sk->sk_frag.page) {
2279 put_page(sk->sk_frag.page);
2280 sk->sk_frag.page = NULL;
2283 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2284 put_cred(sk->sk_peer_cred);
2285 put_pid(sk->sk_peer_pid);
2287 if (likely(sk->sk_net_refcnt))
2288 put_net_track(sock_net(sk), &sk->ns_tracker);
2290 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2292 sk_prot_free(sk->sk_prot_creator, sk);
2295 void sk_destruct(struct sock *sk)
2297 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2299 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2300 reuseport_detach_sock(sk);
2301 use_call_rcu = true;
2305 call_rcu(&sk->sk_rcu, __sk_destruct);
2307 __sk_destruct(&sk->sk_rcu);
2310 static void __sk_free(struct sock *sk)
2312 if (likely(sk->sk_net_refcnt))
2313 sock_inuse_add(sock_net(sk), -1);
2315 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2316 sock_diag_broadcast_destroy(sk);
2321 void sk_free(struct sock *sk)
2324 * We subtract one from sk_wmem_alloc and can know if
2325 * some packets are still in some tx queue.
2326 * If not null, sock_wfree() will call __sk_free(sk) later
2328 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2331 EXPORT_SYMBOL(sk_free);
2333 static void sk_init_common(struct sock *sk)
2335 skb_queue_head_init(&sk->sk_receive_queue);
2336 skb_queue_head_init(&sk->sk_write_queue);
2337 skb_queue_head_init(&sk->sk_error_queue);
2339 rwlock_init(&sk->sk_callback_lock);
2340 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2341 af_rlock_keys + sk->sk_family,
2342 af_family_rlock_key_strings[sk->sk_family]);
2343 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2344 af_wlock_keys + sk->sk_family,
2345 af_family_wlock_key_strings[sk->sk_family]);
2346 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2347 af_elock_keys + sk->sk_family,
2348 af_family_elock_key_strings[sk->sk_family]);
2349 if (sk->sk_kern_sock)
2350 lockdep_set_class_and_name(&sk->sk_callback_lock,
2351 af_kern_callback_keys + sk->sk_family,
2352 af_family_kern_clock_key_strings[sk->sk_family]);
2354 lockdep_set_class_and_name(&sk->sk_callback_lock,
2355 af_callback_keys + sk->sk_family,
2356 af_family_clock_key_strings[sk->sk_family]);
2360 * sk_clone_lock - clone a socket, and lock its clone
2361 * @sk: the socket to clone
2362 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2364 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2366 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2368 struct proto *prot = READ_ONCE(sk->sk_prot);
2369 struct sk_filter *filter;
2370 bool is_charged = true;
2373 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2377 sock_copy(newsk, sk);
2379 newsk->sk_prot_creator = prot;
2382 if (likely(newsk->sk_net_refcnt)) {
2383 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2384 sock_inuse_add(sock_net(newsk), 1);
2386 /* Kernel sockets are not elevating the struct net refcount.
2387 * Instead, use a tracker to more easily detect if a layer
2388 * is not properly dismantling its kernel sockets at netns
2391 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2394 sk_node_init(&newsk->sk_node);
2395 sock_lock_init(newsk);
2396 bh_lock_sock(newsk);
2397 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2398 newsk->sk_backlog.len = 0;
2400 atomic_set(&newsk->sk_rmem_alloc, 0);
2402 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2403 refcount_set(&newsk->sk_wmem_alloc, 1);
2405 atomic_set(&newsk->sk_omem_alloc, 0);
2406 sk_init_common(newsk);
2408 newsk->sk_dst_cache = NULL;
2409 newsk->sk_dst_pending_confirm = 0;
2410 newsk->sk_wmem_queued = 0;
2411 newsk->sk_forward_alloc = 0;
2412 newsk->sk_reserved_mem = 0;
2413 atomic_set(&newsk->sk_drops, 0);
2414 newsk->sk_send_head = NULL;
2415 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2416 atomic_set(&newsk->sk_zckey, 0);
2418 sock_reset_flag(newsk, SOCK_DONE);
2420 /* sk->sk_memcg will be populated at accept() time */
2421 newsk->sk_memcg = NULL;
2423 cgroup_sk_clone(&newsk->sk_cgrp_data);
2426 filter = rcu_dereference(sk->sk_filter);
2428 /* though it's an empty new sock, the charging may fail
2429 * if sysctl_optmem_max was changed between creation of
2430 * original socket and cloning
2432 is_charged = sk_filter_charge(newsk, filter);
2433 RCU_INIT_POINTER(newsk->sk_filter, filter);
2436 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2437 /* We need to make sure that we don't uncharge the new
2438 * socket if we couldn't charge it in the first place
2439 * as otherwise we uncharge the parent's filter.
2442 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2443 sk_free_unlock_clone(newsk);
2447 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2449 if (bpf_sk_storage_clone(sk, newsk)) {
2450 sk_free_unlock_clone(newsk);
2455 /* Clear sk_user_data if parent had the pointer tagged
2456 * as not suitable for copying when cloning.
2458 if (sk_user_data_is_nocopy(newsk))
2459 newsk->sk_user_data = NULL;
2462 newsk->sk_err_soft = 0;
2463 newsk->sk_priority = 0;
2464 newsk->sk_incoming_cpu = raw_smp_processor_id();
2466 /* Before updating sk_refcnt, we must commit prior changes to memory
2467 * (Documentation/RCU/rculist_nulls.rst for details)
2470 refcount_set(&newsk->sk_refcnt, 2);
2472 sk_set_socket(newsk, NULL);
2473 sk_tx_queue_clear(newsk);
2474 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2476 if (newsk->sk_prot->sockets_allocated)
2477 sk_sockets_allocated_inc(newsk);
2479 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2480 net_enable_timestamp();
2484 EXPORT_SYMBOL_GPL(sk_clone_lock);
2486 void sk_free_unlock_clone(struct sock *sk)
2488 /* It is still raw copy of parent, so invalidate
2489 * destructor and make plain sk_free() */
2490 sk->sk_destruct = NULL;
2494 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2496 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2498 bool is_ipv6 = false;
2501 #if IS_ENABLED(CONFIG_IPV6)
2502 is_ipv6 = (sk->sk_family == AF_INET6 &&
2503 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2505 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2506 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2507 READ_ONCE(dst->dev->gso_ipv4_max_size);
2508 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2509 max_size = GSO_LEGACY_MAX_SIZE;
2511 return max_size - (MAX_TCP_HEADER + 1);
2514 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2518 sk->sk_route_caps = dst->dev->features;
2520 sk->sk_route_caps |= NETIF_F_GSO;
2521 if (sk->sk_route_caps & NETIF_F_GSO)
2522 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2523 if (unlikely(sk->sk_gso_disabled))
2524 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2525 if (sk_can_gso(sk)) {
2526 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2527 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2529 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2530 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2531 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2532 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2535 sk->sk_gso_max_segs = max_segs;
2536 sk_dst_set(sk, dst);
2538 EXPORT_SYMBOL_GPL(sk_setup_caps);
2541 * Simple resource managers for sockets.
2546 * Write buffer destructor automatically called from kfree_skb.
2548 void sock_wfree(struct sk_buff *skb)
2550 struct sock *sk = skb->sk;
2551 unsigned int len = skb->truesize;
2554 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2555 if (sock_flag(sk, SOCK_RCU_FREE) &&
2556 sk->sk_write_space == sock_def_write_space) {
2558 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2559 sock_def_write_space_wfree(sk);
2567 * Keep a reference on sk_wmem_alloc, this will be released
2568 * after sk_write_space() call
2570 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2571 sk->sk_write_space(sk);
2575 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2576 * could not do because of in-flight packets
2578 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2581 EXPORT_SYMBOL(sock_wfree);
2583 /* This variant of sock_wfree() is used by TCP,
2584 * since it sets SOCK_USE_WRITE_QUEUE.
2586 void __sock_wfree(struct sk_buff *skb)
2588 struct sock *sk = skb->sk;
2590 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2594 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2599 if (unlikely(!sk_fullsock(sk))) {
2600 skb->destructor = sock_edemux;
2605 skb->destructor = sock_wfree;
2606 skb_set_hash_from_sk(skb, sk);
2608 * We used to take a refcount on sk, but following operation
2609 * is enough to guarantee sk_free() won't free this sock until
2610 * all in-flight packets are completed
2612 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2614 EXPORT_SYMBOL(skb_set_owner_w);
2616 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2618 /* Drivers depend on in-order delivery for crypto offload,
2619 * partial orphan breaks out-of-order-OK logic.
2621 if (skb_is_decrypted(skb))
2624 return (skb->destructor == sock_wfree ||
2625 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2628 /* This helper is used by netem, as it can hold packets in its
2629 * delay queue. We want to allow the owner socket to send more
2630 * packets, as if they were already TX completed by a typical driver.
2631 * But we also want to keep skb->sk set because some packet schedulers
2632 * rely on it (sch_fq for example).
2634 void skb_orphan_partial(struct sk_buff *skb)
2636 if (skb_is_tcp_pure_ack(skb))
2639 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2644 EXPORT_SYMBOL(skb_orphan_partial);
2647 * Read buffer destructor automatically called from kfree_skb.
2649 void sock_rfree(struct sk_buff *skb)
2651 struct sock *sk = skb->sk;
2652 unsigned int len = skb->truesize;
2654 atomic_sub(len, &sk->sk_rmem_alloc);
2655 sk_mem_uncharge(sk, len);
2657 EXPORT_SYMBOL(sock_rfree);
2660 * Buffer destructor for skbs that are not used directly in read or write
2661 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2663 void sock_efree(struct sk_buff *skb)
2667 EXPORT_SYMBOL(sock_efree);
2669 /* Buffer destructor for prefetch/receive path where reference count may
2670 * not be held, e.g. for listen sockets.
2673 void sock_pfree(struct sk_buff *skb)
2675 struct sock *sk = skb->sk;
2677 if (!sk_is_refcounted(sk))
2680 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2681 inet_reqsk(sk)->rsk_listener = NULL;
2682 reqsk_free(inet_reqsk(sk));
2688 EXPORT_SYMBOL(sock_pfree);
2689 #endif /* CONFIG_INET */
2691 kuid_t sock_i_uid(struct sock *sk)
2695 read_lock_bh(&sk->sk_callback_lock);
2696 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2697 read_unlock_bh(&sk->sk_callback_lock);
2700 EXPORT_SYMBOL(sock_i_uid);
2702 unsigned long __sock_i_ino(struct sock *sk)
2706 read_lock(&sk->sk_callback_lock);
2707 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2708 read_unlock(&sk->sk_callback_lock);
2711 EXPORT_SYMBOL(__sock_i_ino);
2713 unsigned long sock_i_ino(struct sock *sk)
2718 ino = __sock_i_ino(sk);
2722 EXPORT_SYMBOL(sock_i_ino);
2725 * Allocate a skb from the socket's send buffer.
2727 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2731 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2732 struct sk_buff *skb = alloc_skb(size, priority);
2735 skb_set_owner_w(skb, sk);
2741 EXPORT_SYMBOL(sock_wmalloc);
2743 static void sock_ofree(struct sk_buff *skb)
2745 struct sock *sk = skb->sk;
2747 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2750 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2753 struct sk_buff *skb;
2755 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2756 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2757 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2760 skb = alloc_skb(size, priority);
2764 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2766 skb->destructor = sock_ofree;
2771 * Allocate a memory block from the socket's option memory buffer.
2773 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2775 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2777 if ((unsigned int)size <= optmem_max &&
2778 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2780 /* First do the add, to avoid the race if kmalloc
2783 atomic_add(size, &sk->sk_omem_alloc);
2784 mem = kmalloc(size, priority);
2787 atomic_sub(size, &sk->sk_omem_alloc);
2791 EXPORT_SYMBOL(sock_kmalloc);
2793 /* Free an option memory block. Note, we actually want the inline
2794 * here as this allows gcc to detect the nullify and fold away the
2795 * condition entirely.
2797 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2800 if (WARN_ON_ONCE(!mem))
2803 kfree_sensitive(mem);
2806 atomic_sub(size, &sk->sk_omem_alloc);
2809 void sock_kfree_s(struct sock *sk, void *mem, int size)
2811 __sock_kfree_s(sk, mem, size, false);
2813 EXPORT_SYMBOL(sock_kfree_s);
2815 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2817 __sock_kfree_s(sk, mem, size, true);
2819 EXPORT_SYMBOL(sock_kzfree_s);
2821 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2822 I think, these locks should be removed for datagram sockets.
2824 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2828 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2832 if (signal_pending(current))
2834 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2835 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2836 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2838 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2840 if (READ_ONCE(sk->sk_err))
2842 timeo = schedule_timeout(timeo);
2844 finish_wait(sk_sleep(sk), &wait);
2850 * Generic send/receive buffer handlers
2853 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2854 unsigned long data_len, int noblock,
2855 int *errcode, int max_page_order)
2857 struct sk_buff *skb;
2861 timeo = sock_sndtimeo(sk, noblock);
2863 err = sock_error(sk);
2868 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2871 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2874 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2875 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2879 if (signal_pending(current))
2881 timeo = sock_wait_for_wmem(sk, timeo);
2883 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2884 errcode, sk->sk_allocation);
2886 skb_set_owner_w(skb, sk);
2890 err = sock_intr_errno(timeo);
2895 EXPORT_SYMBOL(sock_alloc_send_pskb);
2897 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2898 struct sockcm_cookie *sockc)
2902 switch (cmsg->cmsg_type) {
2904 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2905 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2907 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2909 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2911 case SO_TIMESTAMPING_OLD:
2912 case SO_TIMESTAMPING_NEW:
2913 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2916 tsflags = *(u32 *)CMSG_DATA(cmsg);
2917 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2920 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2921 sockc->tsflags |= tsflags;
2924 if (!sock_flag(sk, SOCK_TXTIME))
2926 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2928 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2930 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2932 case SCM_CREDENTIALS:
2939 EXPORT_SYMBOL(__sock_cmsg_send);
2941 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2942 struct sockcm_cookie *sockc)
2944 struct cmsghdr *cmsg;
2947 for_each_cmsghdr(cmsg, msg) {
2948 if (!CMSG_OK(msg, cmsg))
2950 if (cmsg->cmsg_level != SOL_SOCKET)
2952 ret = __sock_cmsg_send(sk, cmsg, sockc);
2958 EXPORT_SYMBOL(sock_cmsg_send);
2960 static void sk_enter_memory_pressure(struct sock *sk)
2962 if (!sk->sk_prot->enter_memory_pressure)
2965 sk->sk_prot->enter_memory_pressure(sk);
2968 static void sk_leave_memory_pressure(struct sock *sk)
2970 if (sk->sk_prot->leave_memory_pressure) {
2971 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2972 tcp_leave_memory_pressure, sk);
2974 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2976 if (memory_pressure && READ_ONCE(*memory_pressure))
2977 WRITE_ONCE(*memory_pressure, 0);
2981 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2984 * skb_page_frag_refill - check that a page_frag contains enough room
2985 * @sz: minimum size of the fragment we want to get
2986 * @pfrag: pointer to page_frag
2987 * @gfp: priority for memory allocation
2989 * Note: While this allocator tries to use high order pages, there is
2990 * no guarantee that allocations succeed. Therefore, @sz MUST be
2991 * less or equal than PAGE_SIZE.
2993 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2996 if (page_ref_count(pfrag->page) == 1) {
3000 if (pfrag->offset + sz <= pfrag->size)
3002 put_page(pfrag->page);
3006 if (SKB_FRAG_PAGE_ORDER &&
3007 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
3008 /* Avoid direct reclaim but allow kswapd to wake */
3009 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
3010 __GFP_COMP | __GFP_NOWARN |
3012 SKB_FRAG_PAGE_ORDER);
3013 if (likely(pfrag->page)) {
3014 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
3018 pfrag->page = alloc_page(gfp);
3019 if (likely(pfrag->page)) {
3020 pfrag->size = PAGE_SIZE;
3025 EXPORT_SYMBOL(skb_page_frag_refill);
3027 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
3029 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
3032 sk_enter_memory_pressure(sk);
3033 sk_stream_moderate_sndbuf(sk);
3036 EXPORT_SYMBOL(sk_page_frag_refill);
3038 void __lock_sock(struct sock *sk)
3039 __releases(&sk->sk_lock.slock)
3040 __acquires(&sk->sk_lock.slock)
3045 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
3046 TASK_UNINTERRUPTIBLE);
3047 spin_unlock_bh(&sk->sk_lock.slock);
3049 spin_lock_bh(&sk->sk_lock.slock);
3050 if (!sock_owned_by_user(sk))
3053 finish_wait(&sk->sk_lock.wq, &wait);
3056 void __release_sock(struct sock *sk)
3057 __releases(&sk->sk_lock.slock)
3058 __acquires(&sk->sk_lock.slock)
3060 struct sk_buff *skb, *next;
3062 while ((skb = sk->sk_backlog.head) != NULL) {
3063 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
3065 spin_unlock_bh(&sk->sk_lock.slock);
3070 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
3071 skb_mark_not_on_list(skb);
3072 sk_backlog_rcv(sk, skb);
3077 } while (skb != NULL);
3079 spin_lock_bh(&sk->sk_lock.slock);
3083 * Doing the zeroing here guarantee we can not loop forever
3084 * while a wild producer attempts to flood us.
3086 sk->sk_backlog.len = 0;
3089 void __sk_flush_backlog(struct sock *sk)
3091 spin_lock_bh(&sk->sk_lock.slock);
3094 if (sk->sk_prot->release_cb)
3095 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3096 tcp_release_cb, sk);
3098 spin_unlock_bh(&sk->sk_lock.slock);
3100 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3103 * sk_wait_data - wait for data to arrive at sk_receive_queue
3104 * @sk: sock to wait on
3105 * @timeo: for how long
3106 * @skb: last skb seen on sk_receive_queue
3108 * Now socket state including sk->sk_err is changed only under lock,
3109 * hence we may omit checks after joining wait queue.
3110 * We check receive queue before schedule() only as optimization;
3111 * it is very likely that release_sock() added new data.
3113 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3115 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3118 add_wait_queue(sk_sleep(sk), &wait);
3119 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3120 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3121 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3122 remove_wait_queue(sk_sleep(sk), &wait);
3125 EXPORT_SYMBOL(sk_wait_data);
3128 * __sk_mem_raise_allocated - increase memory_allocated
3130 * @size: memory size to allocate
3131 * @amt: pages to allocate
3132 * @kind: allocation type
3134 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3136 * Unlike the globally shared limits among the sockets under same protocol,
3137 * consuming the budget of a memcg won't have direct effect on other ones.
3138 * So be optimistic about memcg's tolerance, and leave the callers to decide
3139 * whether or not to raise allocated through sk_under_memory_pressure() or
3142 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3144 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3145 struct proto *prot = sk->sk_prot;
3146 bool charged = false;
3149 sk_memory_allocated_add(sk, amt);
3150 allocated = sk_memory_allocated(sk);
3153 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3154 goto suppress_allocation;
3159 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3160 sk_leave_memory_pressure(sk);
3164 /* Under pressure. */
3165 if (allocated > sk_prot_mem_limits(sk, 1))
3166 sk_enter_memory_pressure(sk);
3168 /* Over hard limit. */
3169 if (allocated > sk_prot_mem_limits(sk, 2))
3170 goto suppress_allocation;
3172 /* Guarantee minimum buffer size under pressure (either global
3173 * or memcg) to make sure features described in RFC 7323 (TCP
3174 * Extensions for High Performance) work properly.
3176 * This rule does NOT stand when exceeds global or memcg's hard
3177 * limit, or else a DoS attack can be taken place by spawning
3178 * lots of sockets whose usage are under minimum buffer size.
3180 if (kind == SK_MEM_RECV) {
3181 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3184 } else { /* SK_MEM_SEND */
3185 int wmem0 = sk_get_wmem0(sk, prot);
3187 if (sk->sk_type == SOCK_STREAM) {
3188 if (sk->sk_wmem_queued < wmem0)
3190 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3195 if (sk_has_memory_pressure(sk)) {
3198 /* The following 'average' heuristic is within the
3199 * scope of global accounting, so it only makes
3200 * sense for global memory pressure.
3202 if (!sk_under_global_memory_pressure(sk))
3205 /* Try to be fair among all the sockets under global
3206 * pressure by allowing the ones that below average
3209 alloc = sk_sockets_allocated_read_positive(sk);
3210 if (sk_prot_mem_limits(sk, 2) > alloc *
3211 sk_mem_pages(sk->sk_wmem_queued +
3212 atomic_read(&sk->sk_rmem_alloc) +
3213 sk->sk_forward_alloc))
3217 suppress_allocation:
3219 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3220 sk_stream_moderate_sndbuf(sk);
3222 /* Fail only if socket is _under_ its sndbuf.
3223 * In this case we cannot block, so that we have to fail.
3225 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3226 /* Force charge with __GFP_NOFAIL */
3227 if (memcg && !charged) {
3228 mem_cgroup_charge_skmem(memcg, amt,
3229 gfp_memcg_charge() | __GFP_NOFAIL);
3235 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3236 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3238 sk_memory_allocated_sub(sk, amt);
3241 mem_cgroup_uncharge_skmem(memcg, amt);
3247 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3249 * @size: memory size to allocate
3250 * @kind: allocation type
3252 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3253 * rmem allocation. This function assumes that protocols which have
3254 * memory_pressure use sk_wmem_queued as write buffer accounting.
3256 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3258 int ret, amt = sk_mem_pages(size);
3260 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3261 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3263 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3266 EXPORT_SYMBOL(__sk_mem_schedule);
3269 * __sk_mem_reduce_allocated - reclaim memory_allocated
3271 * @amount: number of quanta
3273 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3275 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3277 sk_memory_allocated_sub(sk, amount);
3279 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3280 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3282 if (sk_under_global_memory_pressure(sk) &&
3283 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3284 sk_leave_memory_pressure(sk);
3288 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3290 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3292 void __sk_mem_reclaim(struct sock *sk, int amount)
3294 amount >>= PAGE_SHIFT;
3295 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3296 __sk_mem_reduce_allocated(sk, amount);
3298 EXPORT_SYMBOL(__sk_mem_reclaim);
3300 int sk_set_peek_off(struct sock *sk, int val)
3302 WRITE_ONCE(sk->sk_peek_off, val);
3305 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3308 * Set of default routines for initialising struct proto_ops when
3309 * the protocol does not support a particular function. In certain
3310 * cases where it makes no sense for a protocol to have a "do nothing"
3311 * function, some default processing is provided.
3314 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3318 EXPORT_SYMBOL(sock_no_bind);
3320 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3325 EXPORT_SYMBOL(sock_no_connect);
3327 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3331 EXPORT_SYMBOL(sock_no_socketpair);
3333 int sock_no_accept(struct socket *sock, struct socket *newsock,
3334 struct proto_accept_arg *arg)
3338 EXPORT_SYMBOL(sock_no_accept);
3340 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3345 EXPORT_SYMBOL(sock_no_getname);
3347 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3351 EXPORT_SYMBOL(sock_no_ioctl);
3353 int sock_no_listen(struct socket *sock, int backlog)
3357 EXPORT_SYMBOL(sock_no_listen);
3359 int sock_no_shutdown(struct socket *sock, int how)
3363 EXPORT_SYMBOL(sock_no_shutdown);
3365 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3369 EXPORT_SYMBOL(sock_no_sendmsg);
3371 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3375 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3377 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3382 EXPORT_SYMBOL(sock_no_recvmsg);
3384 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3386 /* Mirror missing mmap method error code */
3389 EXPORT_SYMBOL(sock_no_mmap);
3392 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3393 * various sock-based usage counts.
3395 void __receive_sock(struct file *file)
3397 struct socket *sock;
3399 sock = sock_from_file(file);
3401 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3402 sock_update_classid(&sock->sk->sk_cgrp_data);
3407 * Default Socket Callbacks
3410 static void sock_def_wakeup(struct sock *sk)
3412 struct socket_wq *wq;
3415 wq = rcu_dereference(sk->sk_wq);
3416 if (skwq_has_sleeper(wq))
3417 wake_up_interruptible_all(&wq->wait);
3421 static void sock_def_error_report(struct sock *sk)
3423 struct socket_wq *wq;
3426 wq = rcu_dereference(sk->sk_wq);
3427 if (skwq_has_sleeper(wq))
3428 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3429 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR);
3433 void sock_def_readable(struct sock *sk)
3435 struct socket_wq *wq;
3437 trace_sk_data_ready(sk);
3440 wq = rcu_dereference(sk->sk_wq);
3441 if (skwq_has_sleeper(wq))
3442 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3443 EPOLLRDNORM | EPOLLRDBAND);
3444 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
3448 static void sock_def_write_space(struct sock *sk)
3450 struct socket_wq *wq;
3454 /* Do not wake up a writer until he can make "significant"
3457 if (sock_writeable(sk)) {
3458 wq = rcu_dereference(sk->sk_wq);
3459 if (skwq_has_sleeper(wq))
3460 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3461 EPOLLWRNORM | EPOLLWRBAND);
3463 /* Should agree with poll, otherwise some programs break */
3464 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3470 /* An optimised version of sock_def_write_space(), should only be called
3471 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3474 static void sock_def_write_space_wfree(struct sock *sk)
3476 /* Do not wake up a writer until he can make "significant"
3479 if (sock_writeable(sk)) {
3480 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3482 /* rely on refcount_sub from sock_wfree() */
3483 smp_mb__after_atomic();
3484 if (wq && waitqueue_active(&wq->wait))
3485 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3486 EPOLLWRNORM | EPOLLWRBAND);
3488 /* Should agree with poll, otherwise some programs break */
3489 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3493 static void sock_def_destruct(struct sock *sk)
3497 void sk_send_sigurg(struct sock *sk)
3499 if (sk->sk_socket && sk->sk_socket->file)
3500 if (send_sigurg(sk->sk_socket->file))
3501 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3503 EXPORT_SYMBOL(sk_send_sigurg);
3505 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3506 unsigned long expires)
3508 if (!mod_timer(timer, expires))
3511 EXPORT_SYMBOL(sk_reset_timer);
3513 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3515 if (del_timer(timer))
3518 EXPORT_SYMBOL(sk_stop_timer);
3520 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3522 if (del_timer_sync(timer))
3525 EXPORT_SYMBOL(sk_stop_timer_sync);
3527 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3530 sk->sk_send_head = NULL;
3532 timer_setup(&sk->sk_timer, NULL, 0);
3534 sk->sk_allocation = GFP_KERNEL;
3535 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3536 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3537 sk->sk_state = TCP_CLOSE;
3538 sk->sk_use_task_frag = true;
3539 sk_set_socket(sk, sock);
3541 sock_set_flag(sk, SOCK_ZAPPED);
3544 sk->sk_type = sock->type;
3545 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3548 RCU_INIT_POINTER(sk->sk_wq, NULL);
3552 sk->sk_state_change = sock_def_wakeup;
3553 sk->sk_data_ready = sock_def_readable;
3554 sk->sk_write_space = sock_def_write_space;
3555 sk->sk_error_report = sock_def_error_report;
3556 sk->sk_destruct = sock_def_destruct;
3558 sk->sk_frag.page = NULL;
3559 sk->sk_frag.offset = 0;
3560 sk->sk_peek_off = -1;
3562 sk->sk_peer_pid = NULL;
3563 sk->sk_peer_cred = NULL;
3564 spin_lock_init(&sk->sk_peer_lock);
3566 sk->sk_write_pending = 0;
3567 sk->sk_rcvlowat = 1;
3568 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3569 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3571 sk->sk_stamp = SK_DEFAULT_STAMP;
3572 #if BITS_PER_LONG==32
3573 seqlock_init(&sk->sk_stamp_seq);
3575 atomic_set(&sk->sk_zckey, 0);
3577 #ifdef CONFIG_NET_RX_BUSY_POLL
3579 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3582 sk->sk_max_pacing_rate = ~0UL;
3583 sk->sk_pacing_rate = ~0UL;
3584 WRITE_ONCE(sk->sk_pacing_shift, 10);
3585 sk->sk_incoming_cpu = -1;
3587 sk_rx_queue_clear(sk);
3589 * Before updating sk_refcnt, we must commit prior changes to memory
3590 * (Documentation/RCU/rculist_nulls.rst for details)
3593 refcount_set(&sk->sk_refcnt, 1);
3594 atomic_set(&sk->sk_drops, 0);
3596 EXPORT_SYMBOL(sock_init_data_uid);
3598 void sock_init_data(struct socket *sock, struct sock *sk)
3601 SOCK_INODE(sock)->i_uid :
3602 make_kuid(sock_net(sk)->user_ns, 0);
3604 sock_init_data_uid(sock, sk, uid);
3606 EXPORT_SYMBOL(sock_init_data);
3608 void lock_sock_nested(struct sock *sk, int subclass)
3610 /* The sk_lock has mutex_lock() semantics here. */
3611 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3614 spin_lock_bh(&sk->sk_lock.slock);
3615 if (sock_owned_by_user_nocheck(sk))
3617 sk->sk_lock.owned = 1;
3618 spin_unlock_bh(&sk->sk_lock.slock);
3620 EXPORT_SYMBOL(lock_sock_nested);
3622 void release_sock(struct sock *sk)
3624 spin_lock_bh(&sk->sk_lock.slock);
3625 if (sk->sk_backlog.tail)
3628 if (sk->sk_prot->release_cb)
3629 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3630 tcp_release_cb, sk);
3632 sock_release_ownership(sk);
3633 if (waitqueue_active(&sk->sk_lock.wq))
3634 wake_up(&sk->sk_lock.wq);
3635 spin_unlock_bh(&sk->sk_lock.slock);
3637 EXPORT_SYMBOL(release_sock);
3639 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3642 spin_lock_bh(&sk->sk_lock.slock);
3644 if (!sock_owned_by_user_nocheck(sk)) {
3646 * Fast path return with bottom halves disabled and
3647 * sock::sk_lock.slock held.
3649 * The 'mutex' is not contended and holding
3650 * sock::sk_lock.slock prevents all other lockers to
3651 * proceed so the corresponding unlock_sock_fast() can
3652 * avoid the slow path of release_sock() completely and
3653 * just release slock.
3655 * From a semantical POV this is equivalent to 'acquiring'
3656 * the 'mutex', hence the corresponding lockdep
3657 * mutex_release() has to happen in the fast path of
3658 * unlock_sock_fast().
3664 sk->sk_lock.owned = 1;
3665 __acquire(&sk->sk_lock.slock);
3666 spin_unlock_bh(&sk->sk_lock.slock);
3669 EXPORT_SYMBOL(__lock_sock_fast);
3671 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3672 bool timeval, bool time32)
3674 struct sock *sk = sock->sk;
3675 struct timespec64 ts;
3677 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3678 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3679 if (ts.tv_sec == -1)
3681 if (ts.tv_sec == 0) {
3682 ktime_t kt = ktime_get_real();
3683 sock_write_timestamp(sk, kt);
3684 ts = ktime_to_timespec64(kt);
3690 #ifdef CONFIG_COMPAT_32BIT_TIME
3692 return put_old_timespec32(&ts, userstamp);
3694 #ifdef CONFIG_SPARC64
3695 /* beware of padding in sparc64 timeval */
3696 if (timeval && !in_compat_syscall()) {
3697 struct __kernel_old_timeval __user tv = {
3698 .tv_sec = ts.tv_sec,
3699 .tv_usec = ts.tv_nsec,
3701 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3706 return put_timespec64(&ts, userstamp);
3708 EXPORT_SYMBOL(sock_gettstamp);
3710 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3712 if (!sock_flag(sk, flag)) {
3713 unsigned long previous_flags = sk->sk_flags;
3715 sock_set_flag(sk, flag);
3717 * we just set one of the two flags which require net
3718 * time stamping, but time stamping might have been on
3719 * already because of the other one
3721 if (sock_needs_netstamp(sk) &&
3722 !(previous_flags & SK_FLAGS_TIMESTAMP))
3723 net_enable_timestamp();
3727 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3728 int level, int type)
3730 struct sock_exterr_skb *serr;
3731 struct sk_buff *skb;
3735 skb = sock_dequeue_err_skb(sk);
3741 msg->msg_flags |= MSG_TRUNC;
3744 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3748 sock_recv_timestamp(msg, sk, skb);
3750 serr = SKB_EXT_ERR(skb);
3751 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3753 msg->msg_flags |= MSG_ERRQUEUE;
3761 EXPORT_SYMBOL(sock_recv_errqueue);
3764 * Get a socket option on an socket.
3766 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3767 * asynchronous errors should be reported by getsockopt. We assume
3768 * this means if you specify SO_ERROR (otherwise what is the point of it).
3770 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3771 char __user *optval, int __user *optlen)
3773 struct sock *sk = sock->sk;
3775 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3776 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3778 EXPORT_SYMBOL(sock_common_getsockopt);
3780 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3783 struct sock *sk = sock->sk;
3787 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3789 msg->msg_namelen = addr_len;
3792 EXPORT_SYMBOL(sock_common_recvmsg);
3795 * Set socket options on an inet socket.
3797 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3798 sockptr_t optval, unsigned int optlen)
3800 struct sock *sk = sock->sk;
3802 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3803 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3805 EXPORT_SYMBOL(sock_common_setsockopt);
3807 void sk_common_release(struct sock *sk)
3809 if (sk->sk_prot->destroy)
3810 sk->sk_prot->destroy(sk);
3813 * Observation: when sk_common_release is called, processes have
3814 * no access to socket. But net still has.
3815 * Step one, detach it from networking:
3817 * A. Remove from hash tables.
3820 sk->sk_prot->unhash(sk);
3823 sk->sk_socket->sk = NULL;
3826 * In this point socket cannot receive new packets, but it is possible
3827 * that some packets are in flight because some CPU runs receiver and
3828 * did hash table lookup before we unhashed socket. They will achieve
3829 * receive queue and will be purged by socket destructor.
3831 * Also we still have packets pending on receive queue and probably,
3832 * our own packets waiting in device queues. sock_destroy will drain
3833 * receive queue, but transmitted packets will delay socket destruction
3834 * until the last reference will be released.
3839 xfrm_sk_free_policy(sk);
3843 EXPORT_SYMBOL(sk_common_release);
3845 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3847 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3849 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3850 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3851 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3852 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3853 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3854 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3855 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3856 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3857 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3860 #ifdef CONFIG_PROC_FS
3861 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3863 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3865 int cpu, idx = prot->inuse_idx;
3868 for_each_possible_cpu(cpu)
3869 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3871 return res >= 0 ? res : 0;
3873 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3875 int sock_inuse_get(struct net *net)
3879 for_each_possible_cpu(cpu)
3880 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3885 EXPORT_SYMBOL_GPL(sock_inuse_get);
3887 static int __net_init sock_inuse_init_net(struct net *net)
3889 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3890 if (net->core.prot_inuse == NULL)
3895 static void __net_exit sock_inuse_exit_net(struct net *net)
3897 free_percpu(net->core.prot_inuse);
3900 static struct pernet_operations net_inuse_ops = {
3901 .init = sock_inuse_init_net,
3902 .exit = sock_inuse_exit_net,
3905 static __init int net_inuse_init(void)
3907 if (register_pernet_subsys(&net_inuse_ops))
3908 panic("Cannot initialize net inuse counters");
3913 core_initcall(net_inuse_init);
3915 static int assign_proto_idx(struct proto *prot)
3917 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3919 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3920 pr_err("PROTO_INUSE_NR exhausted\n");
3924 set_bit(prot->inuse_idx, proto_inuse_idx);
3928 static void release_proto_idx(struct proto *prot)
3930 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3931 clear_bit(prot->inuse_idx, proto_inuse_idx);
3934 static inline int assign_proto_idx(struct proto *prot)
3939 static inline void release_proto_idx(struct proto *prot)
3945 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3949 kfree(twsk_prot->twsk_slab_name);
3950 twsk_prot->twsk_slab_name = NULL;
3951 kmem_cache_destroy(twsk_prot->twsk_slab);
3952 twsk_prot->twsk_slab = NULL;
3955 static int tw_prot_init(const struct proto *prot)
3957 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3962 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3964 if (!twsk_prot->twsk_slab_name)
3967 twsk_prot->twsk_slab =
3968 kmem_cache_create(twsk_prot->twsk_slab_name,
3969 twsk_prot->twsk_obj_size, 0,
3970 SLAB_ACCOUNT | prot->slab_flags,
3972 if (!twsk_prot->twsk_slab) {
3973 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3981 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3985 kfree(rsk_prot->slab_name);
3986 rsk_prot->slab_name = NULL;
3987 kmem_cache_destroy(rsk_prot->slab);
3988 rsk_prot->slab = NULL;
3991 static int req_prot_init(const struct proto *prot)
3993 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3998 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
4000 if (!rsk_prot->slab_name)
4003 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
4004 rsk_prot->obj_size, 0,
4005 SLAB_ACCOUNT | prot->slab_flags,
4008 if (!rsk_prot->slab) {
4009 pr_crit("%s: Can't create request sock SLAB cache!\n",
4016 int proto_register(struct proto *prot, int alloc_slab)
4020 if (prot->memory_allocated && !prot->sysctl_mem) {
4021 pr_err("%s: missing sysctl_mem\n", prot->name);
4024 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
4025 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
4029 prot->slab = kmem_cache_create_usercopy(prot->name,
4031 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
4033 prot->useroffset, prot->usersize,
4036 if (prot->slab == NULL) {
4037 pr_crit("%s: Can't create sock SLAB cache!\n",
4042 if (req_prot_init(prot))
4043 goto out_free_request_sock_slab;
4045 if (tw_prot_init(prot))
4046 goto out_free_timewait_sock_slab;
4049 mutex_lock(&proto_list_mutex);
4050 ret = assign_proto_idx(prot);
4052 mutex_unlock(&proto_list_mutex);
4053 goto out_free_timewait_sock_slab;
4055 list_add(&prot->node, &proto_list);
4056 mutex_unlock(&proto_list_mutex);
4059 out_free_timewait_sock_slab:
4061 tw_prot_cleanup(prot->twsk_prot);
4062 out_free_request_sock_slab:
4064 req_prot_cleanup(prot->rsk_prot);
4066 kmem_cache_destroy(prot->slab);
4072 EXPORT_SYMBOL(proto_register);
4074 void proto_unregister(struct proto *prot)
4076 mutex_lock(&proto_list_mutex);
4077 release_proto_idx(prot);
4078 list_del(&prot->node);
4079 mutex_unlock(&proto_list_mutex);
4081 kmem_cache_destroy(prot->slab);
4084 req_prot_cleanup(prot->rsk_prot);
4085 tw_prot_cleanup(prot->twsk_prot);
4087 EXPORT_SYMBOL(proto_unregister);
4089 int sock_load_diag_module(int family, int protocol)
4092 if (!sock_is_registered(family))
4095 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4096 NETLINK_SOCK_DIAG, family);
4100 if (family == AF_INET &&
4101 protocol != IPPROTO_RAW &&
4102 protocol < MAX_INET_PROTOS &&
4103 !rcu_access_pointer(inet_protos[protocol]))
4107 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4108 NETLINK_SOCK_DIAG, family, protocol);
4110 EXPORT_SYMBOL(sock_load_diag_module);
4112 #ifdef CONFIG_PROC_FS
4113 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4114 __acquires(proto_list_mutex)
4116 mutex_lock(&proto_list_mutex);
4117 return seq_list_start_head(&proto_list, *pos);
4120 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4122 return seq_list_next(v, &proto_list, pos);
4125 static void proto_seq_stop(struct seq_file *seq, void *v)
4126 __releases(proto_list_mutex)
4128 mutex_unlock(&proto_list_mutex);
4131 static char proto_method_implemented(const void *method)
4133 return method == NULL ? 'n' : 'y';
4135 static long sock_prot_memory_allocated(struct proto *proto)
4137 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4140 static const char *sock_prot_memory_pressure(struct proto *proto)
4142 return proto->memory_pressure != NULL ?
4143 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4146 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4149 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4150 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4153 sock_prot_inuse_get(seq_file_net(seq), proto),
4154 sock_prot_memory_allocated(proto),
4155 sock_prot_memory_pressure(proto),
4157 proto->slab == NULL ? "no" : "yes",
4158 module_name(proto->owner),
4159 proto_method_implemented(proto->close),
4160 proto_method_implemented(proto->connect),
4161 proto_method_implemented(proto->disconnect),
4162 proto_method_implemented(proto->accept),
4163 proto_method_implemented(proto->ioctl),
4164 proto_method_implemented(proto->init),
4165 proto_method_implemented(proto->destroy),
4166 proto_method_implemented(proto->shutdown),
4167 proto_method_implemented(proto->setsockopt),
4168 proto_method_implemented(proto->getsockopt),
4169 proto_method_implemented(proto->sendmsg),
4170 proto_method_implemented(proto->recvmsg),
4171 proto_method_implemented(proto->bind),
4172 proto_method_implemented(proto->backlog_rcv),
4173 proto_method_implemented(proto->hash),
4174 proto_method_implemented(proto->unhash),
4175 proto_method_implemented(proto->get_port),
4176 proto_method_implemented(proto->enter_memory_pressure));
4179 static int proto_seq_show(struct seq_file *seq, void *v)
4181 if (v == &proto_list)
4182 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4191 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4193 proto_seq_printf(seq, list_entry(v, struct proto, node));
4197 static const struct seq_operations proto_seq_ops = {
4198 .start = proto_seq_start,
4199 .next = proto_seq_next,
4200 .stop = proto_seq_stop,
4201 .show = proto_seq_show,
4204 static __net_init int proto_init_net(struct net *net)
4206 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4207 sizeof(struct seq_net_private)))
4213 static __net_exit void proto_exit_net(struct net *net)
4215 remove_proc_entry("protocols", net->proc_net);
4219 static __net_initdata struct pernet_operations proto_net_ops = {
4220 .init = proto_init_net,
4221 .exit = proto_exit_net,
4224 static int __init proto_init(void)
4226 return register_pernet_subsys(&proto_net_ops);
4229 subsys_initcall(proto_init);
4231 #endif /* PROC_FS */
4233 #ifdef CONFIG_NET_RX_BUSY_POLL
4234 bool sk_busy_loop_end(void *p, unsigned long start_time)
4236 struct sock *sk = p;
4238 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4241 if (sk_is_udp(sk) &&
4242 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4245 return sk_busy_loop_timeout(sk, start_time);
4247 EXPORT_SYMBOL(sk_busy_loop_end);
4248 #endif /* CONFIG_NET_RX_BUSY_POLL */
4250 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4252 if (!sk->sk_prot->bind_add)
4254 return sk->sk_prot->bind_add(sk, addr, addr_len);
4256 EXPORT_SYMBOL(sock_bind_add);
4258 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4259 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4260 void __user *arg, void *karg, size_t size)
4264 if (copy_from_user(karg, arg, size))
4267 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4271 if (copy_to_user(arg, karg, size))
4276 EXPORT_SYMBOL(sock_ioctl_inout);
4278 /* This is the most common ioctl prep function, where the result (4 bytes) is
4279 * copied back to userspace if the ioctl() returns successfully. No input is
4280 * copied from userspace as input argument.
4282 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4286 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4290 return put_user(karg, (int __user *)arg);
4293 /* A wrapper around sock ioctls, which copies the data from userspace
4294 * (depending on the protocol/ioctl), and copies back the result to userspace.
4295 * The main motivation for this function is to pass kernel memory to the
4296 * protocol ioctl callbacks, instead of userspace memory.
4298 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4302 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4303 rc = ipmr_sk_ioctl(sk, cmd, arg);
4304 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4305 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4306 else if (sk_is_phonet(sk))
4307 rc = phonet_sk_ioctl(sk, cmd, arg);
4309 /* If ioctl was processed, returns its value */
4313 /* Otherwise call the default handler */
4314 return sock_ioctl_out(sk, cmd, arg);
4316 EXPORT_SYMBOL(sk_ioctl);
4318 static int __init sock_struct_check(void)
4320 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4321 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4322 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4323 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4324 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4326 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4327 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4328 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4329 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4330 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4331 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4332 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4333 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4334 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4336 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4337 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4338 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4340 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4341 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4342 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4343 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4345 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4346 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4347 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf);
4348 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4349 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4350 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4351 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4352 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4353 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4354 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm);
4355 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status);
4356 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4357 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4358 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4359 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4360 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4362 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4363 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4364 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4365 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4366 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4367 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4368 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4369 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4370 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4371 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4372 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4373 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4374 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4378 core_initcall(sock_struct_check);