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.
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
118 #include <linux/uaccess.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
137 #include <trace/events/sock.h>
140 #include <net/busy_poll.h>
142 #include <linux/ethtool.h>
144 static DEFINE_MUTEX(proto_list_mutex);
145 static LIST_HEAD(proto_list);
148 * sk_ns_capable - General socket capability test
149 * @sk: Socket to use a capability on or through
150 * @user_ns: The user namespace of the capability to use
151 * @cap: The capability to use
153 * Test to see if the opener of the socket had when the socket was
154 * created and the current process has the capability @cap in the user
155 * namespace @user_ns.
157 bool sk_ns_capable(const struct sock *sk,
158 struct user_namespace *user_ns, int cap)
160 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
161 ns_capable(user_ns, cap);
163 EXPORT_SYMBOL(sk_ns_capable);
166 * sk_capable - Socket global capability test
167 * @sk: Socket to use a capability on or through
168 * @cap: The global capability to use
170 * Test to see if the opener of the socket had when the socket was
171 * created and the current process has the capability @cap in all user
174 bool sk_capable(const struct sock *sk, int cap)
176 return sk_ns_capable(sk, &init_user_ns, cap);
178 EXPORT_SYMBOL(sk_capable);
181 * sk_net_capable - Network namespace socket capability test
182 * @sk: Socket to use a capability on or through
183 * @cap: The capability to use
185 * Test to see if the opener of the socket had when the socket was created
186 * and the current process has the capability @cap over the network namespace
187 * the socket is a member of.
189 bool sk_net_capable(const struct sock *sk, int cap)
191 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193 EXPORT_SYMBOL(sk_net_capable);
196 * Each address family might have different locking rules, so we have
197 * one slock key per address family and separate keys for internal and
200 static struct lock_class_key af_family_keys[AF_MAX];
201 static struct lock_class_key af_family_kern_keys[AF_MAX];
202 static struct lock_class_key af_family_slock_keys[AF_MAX];
203 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
211 #define _sock_locks(x) \
212 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
213 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
214 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
215 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
216 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
217 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
218 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
219 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
220 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
221 x "27" , x "28" , x "AF_CAN" , \
222 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
223 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
224 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
225 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
226 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
230 static const char *const af_family_key_strings[AF_MAX+1] = {
231 _sock_locks("sk_lock-")
233 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
234 _sock_locks("slock-")
236 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
237 _sock_locks("clock-")
240 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
241 _sock_locks("k-sk_lock-")
243 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
244 _sock_locks("k-slock-")
246 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
247 _sock_locks("k-clock-")
249 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
250 _sock_locks("rlock-")
252 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
253 _sock_locks("wlock-")
255 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
256 _sock_locks("elock-")
260 * sk_callback_lock and sk queues locking rules are per-address-family,
261 * so split the lock classes by using a per-AF key:
263 static struct lock_class_key af_callback_keys[AF_MAX];
264 static struct lock_class_key af_rlock_keys[AF_MAX];
265 static struct lock_class_key af_wlock_keys[AF_MAX];
266 static struct lock_class_key af_elock_keys[AF_MAX];
267 static struct lock_class_key af_kern_callback_keys[AF_MAX];
269 /* Run time adjustable parameters. */
270 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
271 EXPORT_SYMBOL(sysctl_wmem_max);
272 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
273 EXPORT_SYMBOL(sysctl_rmem_max);
274 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
275 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
277 /* Maximal space eaten by iovec or ancillary data plus some space */
278 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
279 EXPORT_SYMBOL(sysctl_optmem_max);
281 int sysctl_tstamp_allow_data __read_mostly = 1;
283 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
284 EXPORT_SYMBOL_GPL(memalloc_socks_key);
287 * sk_set_memalloc - sets %SOCK_MEMALLOC
288 * @sk: socket to set it on
290 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
291 * It's the responsibility of the admin to adjust min_free_kbytes
292 * to meet the requirements
294 void sk_set_memalloc(struct sock *sk)
296 sock_set_flag(sk, SOCK_MEMALLOC);
297 sk->sk_allocation |= __GFP_MEMALLOC;
298 static_branch_inc(&memalloc_socks_key);
300 EXPORT_SYMBOL_GPL(sk_set_memalloc);
302 void sk_clear_memalloc(struct sock *sk)
304 sock_reset_flag(sk, SOCK_MEMALLOC);
305 sk->sk_allocation &= ~__GFP_MEMALLOC;
306 static_branch_dec(&memalloc_socks_key);
309 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
310 * progress of swapping. SOCK_MEMALLOC may be cleared while
311 * it has rmem allocations due to the last swapfile being deactivated
312 * but there is a risk that the socket is unusable due to exceeding
313 * the rmem limits. Reclaim the reserves and obey rmem limits again.
317 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
319 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
322 unsigned int noreclaim_flag;
324 /* these should have been dropped before queueing */
325 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
327 noreclaim_flag = memalloc_noreclaim_save();
328 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
332 memalloc_noreclaim_restore(noreclaim_flag);
336 EXPORT_SYMBOL(__sk_backlog_rcv);
338 void sk_error_report(struct sock *sk)
340 sk->sk_error_report(sk);
342 switch (sk->sk_family) {
346 trace_inet_sk_error_report(sk);
352 EXPORT_SYMBOL(sk_error_report);
354 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
356 struct __kernel_sock_timeval tv;
358 if (timeo == MAX_SCHEDULE_TIMEOUT) {
362 tv.tv_sec = timeo / HZ;
363 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
366 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
367 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
368 *(struct old_timeval32 *)optval = tv32;
373 struct __kernel_old_timeval old_tv;
374 old_tv.tv_sec = tv.tv_sec;
375 old_tv.tv_usec = tv.tv_usec;
376 *(struct __kernel_old_timeval *)optval = old_tv;
377 return sizeof(old_tv);
380 *(struct __kernel_sock_timeval *)optval = tv;
383 EXPORT_SYMBOL(sock_get_timeout);
385 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
386 sockptr_t optval, int optlen, bool old_timeval)
388 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
389 struct old_timeval32 tv32;
391 if (optlen < sizeof(tv32))
394 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
396 tv->tv_sec = tv32.tv_sec;
397 tv->tv_usec = tv32.tv_usec;
398 } else if (old_timeval) {
399 struct __kernel_old_timeval old_tv;
401 if (optlen < sizeof(old_tv))
403 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
405 tv->tv_sec = old_tv.tv_sec;
406 tv->tv_usec = old_tv.tv_usec;
408 if (optlen < sizeof(*tv))
410 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
416 EXPORT_SYMBOL(sock_copy_user_timeval);
418 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
421 struct __kernel_sock_timeval tv;
422 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
427 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
431 static int warned __read_mostly;
434 if (warned < 10 && net_ratelimit()) {
436 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
437 __func__, current->comm, task_pid_nr(current));
441 *timeo_p = MAX_SCHEDULE_TIMEOUT;
442 if (tv.tv_sec == 0 && tv.tv_usec == 0)
444 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
445 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
449 static bool sock_needs_netstamp(const struct sock *sk)
451 switch (sk->sk_family) {
460 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
462 if (sk->sk_flags & flags) {
463 sk->sk_flags &= ~flags;
464 if (sock_needs_netstamp(sk) &&
465 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
466 net_disable_timestamp();
471 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
474 struct sk_buff_head *list = &sk->sk_receive_queue;
476 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
477 atomic_inc(&sk->sk_drops);
478 trace_sock_rcvqueue_full(sk, skb);
482 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
483 atomic_inc(&sk->sk_drops);
488 skb_set_owner_r(skb, sk);
490 /* we escape from rcu protected region, make sure we dont leak
495 spin_lock_irqsave(&list->lock, flags);
496 sock_skb_set_dropcount(sk, skb);
497 __skb_queue_tail(list, skb);
498 spin_unlock_irqrestore(&list->lock, flags);
500 if (!sock_flag(sk, SOCK_DEAD))
501 sk->sk_data_ready(sk);
504 EXPORT_SYMBOL(__sock_queue_rcv_skb);
506 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
510 err = sk_filter(sk, skb);
514 return __sock_queue_rcv_skb(sk, skb);
516 EXPORT_SYMBOL(sock_queue_rcv_skb);
518 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
519 const int nested, unsigned int trim_cap, bool refcounted)
521 int rc = NET_RX_SUCCESS;
523 if (sk_filter_trim_cap(sk, skb, trim_cap))
524 goto discard_and_relse;
528 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
529 atomic_inc(&sk->sk_drops);
530 goto discard_and_relse;
533 bh_lock_sock_nested(sk);
536 if (!sock_owned_by_user(sk)) {
538 * trylock + unlock semantics:
540 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
542 rc = sk_backlog_rcv(sk, skb);
544 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
545 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
547 atomic_inc(&sk->sk_drops);
548 goto discard_and_relse;
560 EXPORT_SYMBOL(__sk_receive_skb);
562 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
564 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
566 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
568 struct dst_entry *dst = __sk_dst_get(sk);
570 if (dst && dst->obsolete &&
571 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
572 dst, cookie) == NULL) {
573 sk_tx_queue_clear(sk);
574 sk->sk_dst_pending_confirm = 0;
575 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
582 EXPORT_SYMBOL(__sk_dst_check);
584 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
586 struct dst_entry *dst = sk_dst_get(sk);
588 if (dst && dst->obsolete &&
589 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
590 dst, cookie) == NULL) {
598 EXPORT_SYMBOL(sk_dst_check);
600 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
602 int ret = -ENOPROTOOPT;
603 #ifdef CONFIG_NETDEVICES
604 struct net *net = sock_net(sk);
608 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
615 sk->sk_bound_dev_if = ifindex;
616 if (sk->sk_prot->rehash)
617 sk->sk_prot->rehash(sk);
628 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
634 ret = sock_bindtoindex_locked(sk, ifindex);
640 EXPORT_SYMBOL(sock_bindtoindex);
642 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
644 int ret = -ENOPROTOOPT;
645 #ifdef CONFIG_NETDEVICES
646 struct net *net = sock_net(sk);
647 char devname[IFNAMSIZ];
654 /* Bind this socket to a particular device like "eth0",
655 * as specified in the passed interface name. If the
656 * name is "" or the option length is zero the socket
659 if (optlen > IFNAMSIZ - 1)
660 optlen = IFNAMSIZ - 1;
661 memset(devname, 0, sizeof(devname));
664 if (copy_from_sockptr(devname, optval, optlen))
668 if (devname[0] != '\0') {
669 struct net_device *dev;
672 dev = dev_get_by_name_rcu(net, devname);
674 index = dev->ifindex;
681 return sock_bindtoindex(sk, index, true);
688 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
689 int __user *optlen, int len)
691 int ret = -ENOPROTOOPT;
692 #ifdef CONFIG_NETDEVICES
693 struct net *net = sock_net(sk);
694 char devname[IFNAMSIZ];
696 if (sk->sk_bound_dev_if == 0) {
705 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
709 len = strlen(devname) + 1;
712 if (copy_to_user(optval, devname, len))
717 if (put_user(len, optlen))
728 bool sk_mc_loop(struct sock *sk)
730 if (dev_recursion_level())
734 switch (sk->sk_family) {
736 return inet_sk(sk)->mc_loop;
737 #if IS_ENABLED(CONFIG_IPV6)
739 return inet6_sk(sk)->mc_loop;
745 EXPORT_SYMBOL(sk_mc_loop);
747 void sock_set_reuseaddr(struct sock *sk)
750 sk->sk_reuse = SK_CAN_REUSE;
753 EXPORT_SYMBOL(sock_set_reuseaddr);
755 void sock_set_reuseport(struct sock *sk)
758 sk->sk_reuseport = true;
761 EXPORT_SYMBOL(sock_set_reuseport);
763 void sock_no_linger(struct sock *sk)
766 sk->sk_lingertime = 0;
767 sock_set_flag(sk, SOCK_LINGER);
770 EXPORT_SYMBOL(sock_no_linger);
772 void sock_set_priority(struct sock *sk, u32 priority)
775 sk->sk_priority = priority;
778 EXPORT_SYMBOL(sock_set_priority);
780 void sock_set_sndtimeo(struct sock *sk, s64 secs)
783 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
784 sk->sk_sndtimeo = secs * HZ;
786 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
789 EXPORT_SYMBOL(sock_set_sndtimeo);
791 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
794 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
795 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
796 sock_set_flag(sk, SOCK_RCVTSTAMP);
797 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
799 sock_reset_flag(sk, SOCK_RCVTSTAMP);
800 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
804 void sock_enable_timestamps(struct sock *sk)
807 __sock_set_timestamps(sk, true, false, true);
810 EXPORT_SYMBOL(sock_enable_timestamps);
812 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
815 case SO_TIMESTAMP_OLD:
816 __sock_set_timestamps(sk, valbool, false, false);
818 case SO_TIMESTAMP_NEW:
819 __sock_set_timestamps(sk, valbool, true, false);
821 case SO_TIMESTAMPNS_OLD:
822 __sock_set_timestamps(sk, valbool, false, true);
824 case SO_TIMESTAMPNS_NEW:
825 __sock_set_timestamps(sk, valbool, true, true);
830 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
832 struct net *net = sock_net(sk);
833 struct net_device *dev = NULL;
838 if (sk->sk_bound_dev_if)
839 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
842 pr_err("%s: sock not bind to device\n", __func__);
846 num = ethtool_get_phc_vclocks(dev, &vclock_index);
849 for (i = 0; i < num; i++) {
850 if (*(vclock_index + i) == phc_index) {
862 sk->sk_bind_phc = phc_index;
867 int sock_set_timestamping(struct sock *sk, int optname,
868 struct so_timestamping timestamping)
870 int val = timestamping.flags;
873 if (val & ~SOF_TIMESTAMPING_MASK)
876 if (val & SOF_TIMESTAMPING_OPT_ID &&
877 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
879 if ((1 << sk->sk_state) &
880 (TCPF_CLOSE | TCPF_LISTEN))
882 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
884 atomic_set(&sk->sk_tskey, 0);
888 if (val & SOF_TIMESTAMPING_OPT_STATS &&
889 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
892 if (val & SOF_TIMESTAMPING_BIND_PHC) {
893 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
898 sk->sk_tsflags = val;
899 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
901 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
902 sock_enable_timestamp(sk,
903 SOCK_TIMESTAMPING_RX_SOFTWARE);
905 sock_disable_timestamp(sk,
906 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
910 void sock_set_keepalive(struct sock *sk)
913 if (sk->sk_prot->keepalive)
914 sk->sk_prot->keepalive(sk, true);
915 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
918 EXPORT_SYMBOL(sock_set_keepalive);
920 static void __sock_set_rcvbuf(struct sock *sk, int val)
922 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
923 * as a negative value.
925 val = min_t(int, val, INT_MAX / 2);
926 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
928 /* We double it on the way in to account for "struct sk_buff" etc.
929 * overhead. Applications assume that the SO_RCVBUF setting they make
930 * will allow that much actual data to be received on that socket.
932 * Applications are unaware that "struct sk_buff" and other overheads
933 * allocate from the receive buffer during socket buffer allocation.
935 * And after considering the possible alternatives, returning the value
936 * we actually used in getsockopt is the most desirable behavior.
938 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
941 void sock_set_rcvbuf(struct sock *sk, int val)
944 __sock_set_rcvbuf(sk, val);
947 EXPORT_SYMBOL(sock_set_rcvbuf);
949 static void __sock_set_mark(struct sock *sk, u32 val)
951 if (val != sk->sk_mark) {
957 void sock_set_mark(struct sock *sk, u32 val)
960 __sock_set_mark(sk, val);
963 EXPORT_SYMBOL(sock_set_mark);
965 static void sock_release_reserved_memory(struct sock *sk, int bytes)
967 /* Round down bytes to multiple of pages */
968 bytes &= ~(SK_MEM_QUANTUM - 1);
970 WARN_ON(bytes > sk->sk_reserved_mem);
971 sk->sk_reserved_mem -= bytes;
975 static int sock_reserve_memory(struct sock *sk, int bytes)
981 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
987 pages = sk_mem_pages(bytes);
989 /* pre-charge to memcg */
990 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
991 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
995 /* pre-charge to forward_alloc */
996 allocated = sk_memory_allocated_add(sk, pages);
997 /* If the system goes into memory pressure with this
998 * precharge, give up and return error.
1000 if (allocated > sk_prot_mem_limits(sk, 1)) {
1001 sk_memory_allocated_sub(sk, pages);
1002 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1005 sk->sk_forward_alloc += pages << SK_MEM_QUANTUM_SHIFT;
1007 sk->sk_reserved_mem += pages << SK_MEM_QUANTUM_SHIFT;
1013 * This is meant for all protocols to use and covers goings on
1014 * at the socket level. Everything here is generic.
1017 int sock_setsockopt(struct socket *sock, int level, int optname,
1018 sockptr_t optval, unsigned int optlen)
1020 struct so_timestamping timestamping;
1021 struct sock_txtime sk_txtime;
1022 struct sock *sk = sock->sk;
1029 * Options without arguments
1032 if (optname == SO_BINDTODEVICE)
1033 return sock_setbindtodevice(sk, optval, optlen);
1035 if (optlen < sizeof(int))
1038 if (copy_from_sockptr(&val, optval, sizeof(val)))
1041 valbool = val ? 1 : 0;
1047 if (val && !capable(CAP_NET_ADMIN))
1050 sock_valbool_flag(sk, SOCK_DBG, valbool);
1053 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1056 sk->sk_reuseport = valbool;
1065 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1069 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1072 /* Don't error on this BSD doesn't and if you think
1073 * about it this is right. Otherwise apps have to
1074 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1075 * are treated in BSD as hints
1077 val = min_t(u32, val, sysctl_wmem_max);
1079 /* Ensure val * 2 fits into an int, to prevent max_t()
1080 * from treating it as a negative value.
1082 val = min_t(int, val, INT_MAX / 2);
1083 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1084 WRITE_ONCE(sk->sk_sndbuf,
1085 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1086 /* Wake up sending tasks if we upped the value. */
1087 sk->sk_write_space(sk);
1090 case SO_SNDBUFFORCE:
1091 if (!capable(CAP_NET_ADMIN)) {
1096 /* No negative values (to prevent underflow, as val will be
1104 /* Don't error on this BSD doesn't and if you think
1105 * about it this is right. Otherwise apps have to
1106 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1107 * are treated in BSD as hints
1109 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
1112 case SO_RCVBUFFORCE:
1113 if (!capable(CAP_NET_ADMIN)) {
1118 /* No negative values (to prevent underflow, as val will be
1121 __sock_set_rcvbuf(sk, max(val, 0));
1125 if (sk->sk_prot->keepalive)
1126 sk->sk_prot->keepalive(sk, valbool);
1127 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1131 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1135 sk->sk_no_check_tx = valbool;
1139 if ((val >= 0 && val <= 6) ||
1140 ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1141 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1142 sk->sk_priority = val;
1148 if (optlen < sizeof(ling)) {
1149 ret = -EINVAL; /* 1003.1g */
1152 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1157 sock_reset_flag(sk, SOCK_LINGER);
1159 #if (BITS_PER_LONG == 32)
1160 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1161 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1164 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1165 sock_set_flag(sk, SOCK_LINGER);
1174 set_bit(SOCK_PASSCRED, &sock->flags);
1176 clear_bit(SOCK_PASSCRED, &sock->flags);
1179 case SO_TIMESTAMP_OLD:
1180 case SO_TIMESTAMP_NEW:
1181 case SO_TIMESTAMPNS_OLD:
1182 case SO_TIMESTAMPNS_NEW:
1183 sock_set_timestamp(sk, optname, valbool);
1186 case SO_TIMESTAMPING_NEW:
1187 case SO_TIMESTAMPING_OLD:
1188 if (optlen == sizeof(timestamping)) {
1189 if (copy_from_sockptr(×tamping, optval,
1190 sizeof(timestamping))) {
1195 memset(×tamping, 0, sizeof(timestamping));
1196 timestamping.flags = val;
1198 ret = sock_set_timestamping(sk, optname, timestamping);
1204 if (sock->ops->set_rcvlowat)
1205 ret = sock->ops->set_rcvlowat(sk, val);
1207 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1210 case SO_RCVTIMEO_OLD:
1211 case SO_RCVTIMEO_NEW:
1212 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1213 optlen, optname == SO_RCVTIMEO_OLD);
1216 case SO_SNDTIMEO_OLD:
1217 case SO_SNDTIMEO_NEW:
1218 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1219 optlen, optname == SO_SNDTIMEO_OLD);
1222 case SO_ATTACH_FILTER: {
1223 struct sock_fprog fprog;
1225 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1227 ret = sk_attach_filter(&fprog, sk);
1232 if (optlen == sizeof(u32)) {
1236 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1239 ret = sk_attach_bpf(ufd, sk);
1243 case SO_ATTACH_REUSEPORT_CBPF: {
1244 struct sock_fprog fprog;
1246 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1248 ret = sk_reuseport_attach_filter(&fprog, sk);
1251 case SO_ATTACH_REUSEPORT_EBPF:
1253 if (optlen == sizeof(u32)) {
1257 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1260 ret = sk_reuseport_attach_bpf(ufd, sk);
1264 case SO_DETACH_REUSEPORT_BPF:
1265 ret = reuseport_detach_prog(sk);
1268 case SO_DETACH_FILTER:
1269 ret = sk_detach_filter(sk);
1272 case SO_LOCK_FILTER:
1273 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1276 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1281 set_bit(SOCK_PASSSEC, &sock->flags);
1283 clear_bit(SOCK_PASSSEC, &sock->flags);
1286 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1287 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1292 __sock_set_mark(sk, val);
1296 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1299 case SO_WIFI_STATUS:
1300 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1304 if (sock->ops->set_peek_off)
1305 ret = sock->ops->set_peek_off(sk, val);
1311 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1314 case SO_SELECT_ERR_QUEUE:
1315 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1318 #ifdef CONFIG_NET_RX_BUSY_POLL
1320 /* allow unprivileged users to decrease the value */
1321 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1327 WRITE_ONCE(sk->sk_ll_usec, val);
1330 case SO_PREFER_BUSY_POLL:
1331 if (valbool && !capable(CAP_NET_ADMIN))
1334 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1336 case SO_BUSY_POLL_BUDGET:
1337 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1340 if (val < 0 || val > U16_MAX)
1343 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1348 case SO_MAX_PACING_RATE:
1350 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1352 if (sizeof(ulval) != sizeof(val) &&
1353 optlen >= sizeof(ulval) &&
1354 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1359 cmpxchg(&sk->sk_pacing_status,
1362 sk->sk_max_pacing_rate = ulval;
1363 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1366 case SO_INCOMING_CPU:
1367 WRITE_ONCE(sk->sk_incoming_cpu, val);
1372 dst_negative_advice(sk);
1376 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1377 if (!(sk_is_tcp(sk) ||
1378 (sk->sk_type == SOCK_DGRAM &&
1379 sk->sk_protocol == IPPROTO_UDP)))
1381 } else if (sk->sk_family != PF_RDS) {
1385 if (val < 0 || val > 1)
1388 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1393 if (optlen != sizeof(struct sock_txtime)) {
1396 } else if (copy_from_sockptr(&sk_txtime, optval,
1397 sizeof(struct sock_txtime))) {
1400 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1404 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1405 * scheduler has enough safe guards.
1407 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1408 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1412 sock_valbool_flag(sk, SOCK_TXTIME, true);
1413 sk->sk_clockid = sk_txtime.clockid;
1414 sk->sk_txtime_deadline_mode =
1415 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1416 sk->sk_txtime_report_errors =
1417 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1420 case SO_BINDTOIFINDEX:
1421 ret = sock_bindtoindex_locked(sk, val);
1425 if (val & ~SOCK_BUF_LOCK_MASK) {
1429 sk->sk_userlocks = val | (sk->sk_userlocks &
1430 ~SOCK_BUF_LOCK_MASK);
1433 case SO_RESERVE_MEM:
1442 delta = val - sk->sk_reserved_mem;
1444 sock_release_reserved_memory(sk, -delta);
1446 ret = sock_reserve_memory(sk, delta);
1451 if (val < -1 || val > 1) {
1455 /* Paired with READ_ONCE() in tcp_rtx_synack() */
1456 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1466 EXPORT_SYMBOL(sock_setsockopt);
1468 static const struct cred *sk_get_peer_cred(struct sock *sk)
1470 const struct cred *cred;
1472 spin_lock(&sk->sk_peer_lock);
1473 cred = get_cred(sk->sk_peer_cred);
1474 spin_unlock(&sk->sk_peer_lock);
1479 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1480 struct ucred *ucred)
1482 ucred->pid = pid_vnr(pid);
1483 ucred->uid = ucred->gid = -1;
1485 struct user_namespace *current_ns = current_user_ns();
1487 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1488 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1492 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1494 struct user_namespace *user_ns = current_user_ns();
1497 for (i = 0; i < src->ngroups; i++)
1498 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1504 int sock_getsockopt(struct socket *sock, int level, int optname,
1505 char __user *optval, int __user *optlen)
1507 struct sock *sk = sock->sk;
1512 unsigned long ulval;
1514 struct old_timeval32 tm32;
1515 struct __kernel_old_timeval tm;
1516 struct __kernel_sock_timeval stm;
1517 struct sock_txtime txtime;
1518 struct so_timestamping timestamping;
1521 int lv = sizeof(int);
1524 if (get_user(len, optlen))
1529 memset(&v, 0, sizeof(v));
1533 v.val = sock_flag(sk, SOCK_DBG);
1537 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1541 v.val = sock_flag(sk, SOCK_BROADCAST);
1545 v.val = sk->sk_sndbuf;
1549 v.val = sk->sk_rcvbuf;
1553 v.val = sk->sk_reuse;
1557 v.val = sk->sk_reuseport;
1561 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1565 v.val = sk->sk_type;
1569 v.val = sk->sk_protocol;
1573 v.val = sk->sk_family;
1577 v.val = -sock_error(sk);
1579 v.val = xchg(&sk->sk_err_soft, 0);
1583 v.val = sock_flag(sk, SOCK_URGINLINE);
1587 v.val = sk->sk_no_check_tx;
1591 v.val = sk->sk_priority;
1595 lv = sizeof(v.ling);
1596 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1597 v.ling.l_linger = sk->sk_lingertime / HZ;
1603 case SO_TIMESTAMP_OLD:
1604 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1605 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1606 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1609 case SO_TIMESTAMPNS_OLD:
1610 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1613 case SO_TIMESTAMP_NEW:
1614 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1617 case SO_TIMESTAMPNS_NEW:
1618 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1621 case SO_TIMESTAMPING_OLD:
1622 lv = sizeof(v.timestamping);
1623 v.timestamping.flags = sk->sk_tsflags;
1624 v.timestamping.bind_phc = sk->sk_bind_phc;
1627 case SO_RCVTIMEO_OLD:
1628 case SO_RCVTIMEO_NEW:
1629 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1632 case SO_SNDTIMEO_OLD:
1633 case SO_SNDTIMEO_NEW:
1634 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1638 v.val = sk->sk_rcvlowat;
1646 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1651 struct ucred peercred;
1652 if (len > sizeof(peercred))
1653 len = sizeof(peercred);
1655 spin_lock(&sk->sk_peer_lock);
1656 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1657 spin_unlock(&sk->sk_peer_lock);
1659 if (copy_to_user(optval, &peercred, len))
1666 const struct cred *cred;
1669 cred = sk_get_peer_cred(sk);
1673 n = cred->group_info->ngroups;
1674 if (len < n * sizeof(gid_t)) {
1675 len = n * sizeof(gid_t);
1677 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1679 len = n * sizeof(gid_t);
1681 ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1692 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1697 if (copy_to_user(optval, address, len))
1702 /* Dubious BSD thing... Probably nobody even uses it, but
1703 * the UNIX standard wants it for whatever reason... -DaveM
1706 v.val = sk->sk_state == TCP_LISTEN;
1710 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1714 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1717 v.val = sk->sk_mark;
1721 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1724 case SO_WIFI_STATUS:
1725 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1729 if (!sock->ops->set_peek_off)
1732 v.val = sk->sk_peek_off;
1735 v.val = sock_flag(sk, SOCK_NOFCS);
1738 case SO_BINDTODEVICE:
1739 return sock_getbindtodevice(sk, optval, optlen, len);
1742 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1748 case SO_LOCK_FILTER:
1749 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1752 case SO_BPF_EXTENSIONS:
1753 v.val = bpf_tell_extensions();
1756 case SO_SELECT_ERR_QUEUE:
1757 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1760 #ifdef CONFIG_NET_RX_BUSY_POLL
1762 v.val = sk->sk_ll_usec;
1764 case SO_PREFER_BUSY_POLL:
1765 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1769 case SO_MAX_PACING_RATE:
1770 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1771 lv = sizeof(v.ulval);
1772 v.ulval = sk->sk_max_pacing_rate;
1775 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1779 case SO_INCOMING_CPU:
1780 v.val = READ_ONCE(sk->sk_incoming_cpu);
1785 u32 meminfo[SK_MEMINFO_VARS];
1787 sk_get_meminfo(sk, meminfo);
1789 len = min_t(unsigned int, len, sizeof(meminfo));
1790 if (copy_to_user(optval, &meminfo, len))
1796 #ifdef CONFIG_NET_RX_BUSY_POLL
1797 case SO_INCOMING_NAPI_ID:
1798 v.val = READ_ONCE(sk->sk_napi_id);
1800 /* aggregate non-NAPI IDs down to 0 */
1801 if (v.val < MIN_NAPI_ID)
1811 v.val64 = sock_gen_cookie(sk);
1815 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1819 lv = sizeof(v.txtime);
1820 v.txtime.clockid = sk->sk_clockid;
1821 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1822 SOF_TXTIME_DEADLINE_MODE : 0;
1823 v.txtime.flags |= sk->sk_txtime_report_errors ?
1824 SOF_TXTIME_REPORT_ERRORS : 0;
1827 case SO_BINDTOIFINDEX:
1828 v.val = sk->sk_bound_dev_if;
1831 case SO_NETNS_COOKIE:
1835 v.val64 = sock_net(sk)->net_cookie;
1839 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1842 case SO_RESERVE_MEM:
1843 v.val = sk->sk_reserved_mem;
1847 v.val = sk->sk_txrehash;
1851 /* We implement the SO_SNDLOWAT etc to not be settable
1854 return -ENOPROTOOPT;
1859 if (copy_to_user(optval, &v, len))
1862 if (put_user(len, optlen))
1868 * Initialize an sk_lock.
1870 * (We also register the sk_lock with the lock validator.)
1872 static inline void sock_lock_init(struct sock *sk)
1874 if (sk->sk_kern_sock)
1875 sock_lock_init_class_and_name(
1877 af_family_kern_slock_key_strings[sk->sk_family],
1878 af_family_kern_slock_keys + sk->sk_family,
1879 af_family_kern_key_strings[sk->sk_family],
1880 af_family_kern_keys + sk->sk_family);
1882 sock_lock_init_class_and_name(
1884 af_family_slock_key_strings[sk->sk_family],
1885 af_family_slock_keys + sk->sk_family,
1886 af_family_key_strings[sk->sk_family],
1887 af_family_keys + sk->sk_family);
1891 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1892 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1893 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1895 static void sock_copy(struct sock *nsk, const struct sock *osk)
1897 const struct proto *prot = READ_ONCE(osk->sk_prot);
1898 #ifdef CONFIG_SECURITY_NETWORK
1899 void *sptr = nsk->sk_security;
1902 /* If we move sk_tx_queue_mapping out of the private section,
1903 * we must check if sk_tx_queue_clear() is called after
1904 * sock_copy() in sk_clone_lock().
1906 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1907 offsetof(struct sock, sk_dontcopy_begin) ||
1908 offsetof(struct sock, sk_tx_queue_mapping) >=
1909 offsetof(struct sock, sk_dontcopy_end));
1911 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1913 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1914 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1916 #ifdef CONFIG_SECURITY_NETWORK
1917 nsk->sk_security = sptr;
1918 security_sk_clone(osk, nsk);
1922 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1926 struct kmem_cache *slab;
1930 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1933 if (want_init_on_alloc(priority))
1934 sk_prot_clear_nulls(sk, prot->obj_size);
1936 sk = kmalloc(prot->obj_size, priority);
1939 if (security_sk_alloc(sk, family, priority))
1942 if (!try_module_get(prot->owner))
1949 security_sk_free(sk);
1952 kmem_cache_free(slab, sk);
1958 static void sk_prot_free(struct proto *prot, struct sock *sk)
1960 struct kmem_cache *slab;
1961 struct module *owner;
1963 owner = prot->owner;
1966 cgroup_sk_free(&sk->sk_cgrp_data);
1967 mem_cgroup_sk_free(sk);
1968 security_sk_free(sk);
1970 kmem_cache_free(slab, sk);
1977 * sk_alloc - All socket objects are allocated here
1978 * @net: the applicable net namespace
1979 * @family: protocol family
1980 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1981 * @prot: struct proto associated with this new sock instance
1982 * @kern: is this to be a kernel socket?
1984 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1985 struct proto *prot, int kern)
1989 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1991 sk->sk_family = family;
1993 * See comment in struct sock definition to understand
1994 * why we need sk_prot_creator -acme
1996 sk->sk_prot = sk->sk_prot_creator = prot;
1997 sk->sk_kern_sock = kern;
1999 sk->sk_net_refcnt = kern ? 0 : 1;
2000 if (likely(sk->sk_net_refcnt)) {
2001 get_net_track(net, &sk->ns_tracker, priority);
2002 sock_inuse_add(net, 1);
2005 sock_net_set(sk, net);
2006 refcount_set(&sk->sk_wmem_alloc, 1);
2008 mem_cgroup_sk_alloc(sk);
2009 cgroup_sk_alloc(&sk->sk_cgrp_data);
2010 sock_update_classid(&sk->sk_cgrp_data);
2011 sock_update_netprioidx(&sk->sk_cgrp_data);
2012 sk_tx_queue_clear(sk);
2017 EXPORT_SYMBOL(sk_alloc);
2019 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2020 * grace period. This is the case for UDP sockets and TCP listeners.
2022 static void __sk_destruct(struct rcu_head *head)
2024 struct sock *sk = container_of(head, struct sock, sk_rcu);
2025 struct sk_filter *filter;
2027 if (sk->sk_destruct)
2028 sk->sk_destruct(sk);
2030 filter = rcu_dereference_check(sk->sk_filter,
2031 refcount_read(&sk->sk_wmem_alloc) == 0);
2033 sk_filter_uncharge(sk, filter);
2034 RCU_INIT_POINTER(sk->sk_filter, NULL);
2037 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2039 #ifdef CONFIG_BPF_SYSCALL
2040 bpf_sk_storage_free(sk);
2043 if (atomic_read(&sk->sk_omem_alloc))
2044 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2045 __func__, atomic_read(&sk->sk_omem_alloc));
2047 if (sk->sk_frag.page) {
2048 put_page(sk->sk_frag.page);
2049 sk->sk_frag.page = NULL;
2052 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2053 put_cred(sk->sk_peer_cred);
2054 put_pid(sk->sk_peer_pid);
2056 if (likely(sk->sk_net_refcnt))
2057 put_net_track(sock_net(sk), &sk->ns_tracker);
2058 sk_prot_free(sk->sk_prot_creator, sk);
2061 void sk_destruct(struct sock *sk)
2063 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2065 WARN_ON_ONCE(!llist_empty(&sk->defer_list));
2066 sk_defer_free_flush(sk);
2068 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2069 reuseport_detach_sock(sk);
2070 use_call_rcu = true;
2074 call_rcu(&sk->sk_rcu, __sk_destruct);
2076 __sk_destruct(&sk->sk_rcu);
2079 static void __sk_free(struct sock *sk)
2081 if (likely(sk->sk_net_refcnt))
2082 sock_inuse_add(sock_net(sk), -1);
2084 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2085 sock_diag_broadcast_destroy(sk);
2090 void sk_free(struct sock *sk)
2093 * We subtract one from sk_wmem_alloc and can know if
2094 * some packets are still in some tx queue.
2095 * If not null, sock_wfree() will call __sk_free(sk) later
2097 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2100 EXPORT_SYMBOL(sk_free);
2102 static void sk_init_common(struct sock *sk)
2104 skb_queue_head_init(&sk->sk_receive_queue);
2105 skb_queue_head_init(&sk->sk_write_queue);
2106 skb_queue_head_init(&sk->sk_error_queue);
2108 rwlock_init(&sk->sk_callback_lock);
2109 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2110 af_rlock_keys + sk->sk_family,
2111 af_family_rlock_key_strings[sk->sk_family]);
2112 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2113 af_wlock_keys + sk->sk_family,
2114 af_family_wlock_key_strings[sk->sk_family]);
2115 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2116 af_elock_keys + sk->sk_family,
2117 af_family_elock_key_strings[sk->sk_family]);
2118 lockdep_set_class_and_name(&sk->sk_callback_lock,
2119 af_callback_keys + sk->sk_family,
2120 af_family_clock_key_strings[sk->sk_family]);
2124 * sk_clone_lock - clone a socket, and lock its clone
2125 * @sk: the socket to clone
2126 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2128 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2130 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2132 struct proto *prot = READ_ONCE(sk->sk_prot);
2133 struct sk_filter *filter;
2134 bool is_charged = true;
2137 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2141 sock_copy(newsk, sk);
2143 newsk->sk_prot_creator = prot;
2146 if (likely(newsk->sk_net_refcnt)) {
2147 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2148 sock_inuse_add(sock_net(newsk), 1);
2150 sk_node_init(&newsk->sk_node);
2151 sock_lock_init(newsk);
2152 bh_lock_sock(newsk);
2153 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2154 newsk->sk_backlog.len = 0;
2156 atomic_set(&newsk->sk_rmem_alloc, 0);
2158 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2159 refcount_set(&newsk->sk_wmem_alloc, 1);
2161 atomic_set(&newsk->sk_omem_alloc, 0);
2162 sk_init_common(newsk);
2164 newsk->sk_dst_cache = NULL;
2165 newsk->sk_dst_pending_confirm = 0;
2166 newsk->sk_wmem_queued = 0;
2167 newsk->sk_forward_alloc = 0;
2168 newsk->sk_reserved_mem = 0;
2169 atomic_set(&newsk->sk_drops, 0);
2170 newsk->sk_send_head = NULL;
2171 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2172 atomic_set(&newsk->sk_zckey, 0);
2174 sock_reset_flag(newsk, SOCK_DONE);
2176 /* sk->sk_memcg will be populated at accept() time */
2177 newsk->sk_memcg = NULL;
2179 cgroup_sk_clone(&newsk->sk_cgrp_data);
2182 filter = rcu_dereference(sk->sk_filter);
2184 /* though it's an empty new sock, the charging may fail
2185 * if sysctl_optmem_max was changed between creation of
2186 * original socket and cloning
2188 is_charged = sk_filter_charge(newsk, filter);
2189 RCU_INIT_POINTER(newsk->sk_filter, filter);
2192 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2193 /* We need to make sure that we don't uncharge the new
2194 * socket if we couldn't charge it in the first place
2195 * as otherwise we uncharge the parent's filter.
2198 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2199 sk_free_unlock_clone(newsk);
2203 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2205 if (bpf_sk_storage_clone(sk, newsk)) {
2206 sk_free_unlock_clone(newsk);
2211 /* Clear sk_user_data if parent had the pointer tagged
2212 * as not suitable for copying when cloning.
2214 if (sk_user_data_is_nocopy(newsk))
2215 newsk->sk_user_data = NULL;
2218 newsk->sk_err_soft = 0;
2219 newsk->sk_priority = 0;
2220 newsk->sk_incoming_cpu = raw_smp_processor_id();
2222 /* Before updating sk_refcnt, we must commit prior changes to memory
2223 * (Documentation/RCU/rculist_nulls.rst for details)
2226 refcount_set(&newsk->sk_refcnt, 2);
2228 /* Increment the counter in the same struct proto as the master
2229 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2230 * is the same as sk->sk_prot->socks, as this field was copied
2233 * This _changes_ the previous behaviour, where
2234 * tcp_create_openreq_child always was incrementing the
2235 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2236 * to be taken into account in all callers. -acme
2238 sk_refcnt_debug_inc(newsk);
2239 sk_set_socket(newsk, NULL);
2240 sk_tx_queue_clear(newsk);
2241 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2243 if (newsk->sk_prot->sockets_allocated)
2244 sk_sockets_allocated_inc(newsk);
2246 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2247 net_enable_timestamp();
2251 EXPORT_SYMBOL_GPL(sk_clone_lock);
2253 void sk_free_unlock_clone(struct sock *sk)
2255 /* It is still raw copy of parent, so invalidate
2256 * destructor and make plain sk_free() */
2257 sk->sk_destruct = NULL;
2261 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2263 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2267 sk_dst_set(sk, dst);
2268 sk->sk_route_caps = dst->dev->features;
2270 sk->sk_route_caps |= NETIF_F_GSO;
2271 if (sk->sk_route_caps & NETIF_F_GSO)
2272 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2273 if (unlikely(sk->sk_gso_disabled))
2274 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2275 if (sk_can_gso(sk)) {
2276 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2277 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2279 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2280 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */
2281 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size);
2282 sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1);
2283 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2284 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2287 sk->sk_gso_max_segs = max_segs;
2289 EXPORT_SYMBOL_GPL(sk_setup_caps);
2292 * Simple resource managers for sockets.
2297 * Write buffer destructor automatically called from kfree_skb.
2299 void sock_wfree(struct sk_buff *skb)
2301 struct sock *sk = skb->sk;
2302 unsigned int len = skb->truesize;
2304 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2306 * Keep a reference on sk_wmem_alloc, this will be released
2307 * after sk_write_space() call
2309 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2310 sk->sk_write_space(sk);
2314 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2315 * could not do because of in-flight packets
2317 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2320 EXPORT_SYMBOL(sock_wfree);
2322 /* This variant of sock_wfree() is used by TCP,
2323 * since it sets SOCK_USE_WRITE_QUEUE.
2325 void __sock_wfree(struct sk_buff *skb)
2327 struct sock *sk = skb->sk;
2329 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2333 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2338 if (unlikely(!sk_fullsock(sk))) {
2339 skb->destructor = sock_edemux;
2344 skb->destructor = sock_wfree;
2345 skb_set_hash_from_sk(skb, sk);
2347 * We used to take a refcount on sk, but following operation
2348 * is enough to guarantee sk_free() wont free this sock until
2349 * all in-flight packets are completed
2351 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2353 EXPORT_SYMBOL(skb_set_owner_w);
2355 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2357 #ifdef CONFIG_TLS_DEVICE
2358 /* Drivers depend on in-order delivery for crypto offload,
2359 * partial orphan breaks out-of-order-OK logic.
2364 return (skb->destructor == sock_wfree ||
2365 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2368 /* This helper is used by netem, as it can hold packets in its
2369 * delay queue. We want to allow the owner socket to send more
2370 * packets, as if they were already TX completed by a typical driver.
2371 * But we also want to keep skb->sk set because some packet schedulers
2372 * rely on it (sch_fq for example).
2374 void skb_orphan_partial(struct sk_buff *skb)
2376 if (skb_is_tcp_pure_ack(skb))
2379 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2384 EXPORT_SYMBOL(skb_orphan_partial);
2387 * Read buffer destructor automatically called from kfree_skb.
2389 void sock_rfree(struct sk_buff *skb)
2391 struct sock *sk = skb->sk;
2392 unsigned int len = skb->truesize;
2394 atomic_sub(len, &sk->sk_rmem_alloc);
2395 sk_mem_uncharge(sk, len);
2397 EXPORT_SYMBOL(sock_rfree);
2400 * Buffer destructor for skbs that are not used directly in read or write
2401 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2403 void sock_efree(struct sk_buff *skb)
2407 EXPORT_SYMBOL(sock_efree);
2409 /* Buffer destructor for prefetch/receive path where reference count may
2410 * not be held, e.g. for listen sockets.
2413 void sock_pfree(struct sk_buff *skb)
2415 if (sk_is_refcounted(skb->sk))
2416 sock_gen_put(skb->sk);
2418 EXPORT_SYMBOL(sock_pfree);
2419 #endif /* CONFIG_INET */
2421 kuid_t sock_i_uid(struct sock *sk)
2425 read_lock_bh(&sk->sk_callback_lock);
2426 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2427 read_unlock_bh(&sk->sk_callback_lock);
2430 EXPORT_SYMBOL(sock_i_uid);
2432 unsigned long sock_i_ino(struct sock *sk)
2436 read_lock_bh(&sk->sk_callback_lock);
2437 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2438 read_unlock_bh(&sk->sk_callback_lock);
2441 EXPORT_SYMBOL(sock_i_ino);
2444 * Allocate a skb from the socket's send buffer.
2446 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2450 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2451 struct sk_buff *skb = alloc_skb(size, priority);
2454 skb_set_owner_w(skb, sk);
2460 EXPORT_SYMBOL(sock_wmalloc);
2462 static void sock_ofree(struct sk_buff *skb)
2464 struct sock *sk = skb->sk;
2466 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2469 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2472 struct sk_buff *skb;
2474 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2475 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2479 skb = alloc_skb(size, priority);
2483 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2485 skb->destructor = sock_ofree;
2490 * Allocate a memory block from the socket's option memory buffer.
2492 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2494 if ((unsigned int)size <= sysctl_optmem_max &&
2495 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2497 /* First do the add, to avoid the race if kmalloc
2500 atomic_add(size, &sk->sk_omem_alloc);
2501 mem = kmalloc(size, priority);
2504 atomic_sub(size, &sk->sk_omem_alloc);
2508 EXPORT_SYMBOL(sock_kmalloc);
2510 /* Free an option memory block. Note, we actually want the inline
2511 * here as this allows gcc to detect the nullify and fold away the
2512 * condition entirely.
2514 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2517 if (WARN_ON_ONCE(!mem))
2520 kfree_sensitive(mem);
2523 atomic_sub(size, &sk->sk_omem_alloc);
2526 void sock_kfree_s(struct sock *sk, void *mem, int size)
2528 __sock_kfree_s(sk, mem, size, false);
2530 EXPORT_SYMBOL(sock_kfree_s);
2532 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2534 __sock_kfree_s(sk, mem, size, true);
2536 EXPORT_SYMBOL(sock_kzfree_s);
2538 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2539 I think, these locks should be removed for datagram sockets.
2541 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2545 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2549 if (signal_pending(current))
2551 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2552 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2553 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2555 if (sk->sk_shutdown & SEND_SHUTDOWN)
2559 timeo = schedule_timeout(timeo);
2561 finish_wait(sk_sleep(sk), &wait);
2567 * Generic send/receive buffer handlers
2570 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2571 unsigned long data_len, int noblock,
2572 int *errcode, int max_page_order)
2574 struct sk_buff *skb;
2578 timeo = sock_sndtimeo(sk, noblock);
2580 err = sock_error(sk);
2585 if (sk->sk_shutdown & SEND_SHUTDOWN)
2588 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2591 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2592 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2596 if (signal_pending(current))
2598 timeo = sock_wait_for_wmem(sk, timeo);
2600 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2601 errcode, sk->sk_allocation);
2603 skb_set_owner_w(skb, sk);
2607 err = sock_intr_errno(timeo);
2612 EXPORT_SYMBOL(sock_alloc_send_pskb);
2614 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2615 int noblock, int *errcode)
2617 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2619 EXPORT_SYMBOL(sock_alloc_send_skb);
2621 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2622 struct sockcm_cookie *sockc)
2626 switch (cmsg->cmsg_type) {
2628 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2629 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2631 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2633 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2635 case SO_TIMESTAMPING_OLD:
2636 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2639 tsflags = *(u32 *)CMSG_DATA(cmsg);
2640 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2643 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2644 sockc->tsflags |= tsflags;
2647 if (!sock_flag(sk, SOCK_TXTIME))
2649 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2651 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2653 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2655 case SCM_CREDENTIALS:
2662 EXPORT_SYMBOL(__sock_cmsg_send);
2664 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2665 struct sockcm_cookie *sockc)
2667 struct cmsghdr *cmsg;
2670 for_each_cmsghdr(cmsg, msg) {
2671 if (!CMSG_OK(msg, cmsg))
2673 if (cmsg->cmsg_level != SOL_SOCKET)
2675 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2681 EXPORT_SYMBOL(sock_cmsg_send);
2683 static void sk_enter_memory_pressure(struct sock *sk)
2685 if (!sk->sk_prot->enter_memory_pressure)
2688 sk->sk_prot->enter_memory_pressure(sk);
2691 static void sk_leave_memory_pressure(struct sock *sk)
2693 if (sk->sk_prot->leave_memory_pressure) {
2694 sk->sk_prot->leave_memory_pressure(sk);
2696 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2698 if (memory_pressure && READ_ONCE(*memory_pressure))
2699 WRITE_ONCE(*memory_pressure, 0);
2703 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2706 * skb_page_frag_refill - check that a page_frag contains enough room
2707 * @sz: minimum size of the fragment we want to get
2708 * @pfrag: pointer to page_frag
2709 * @gfp: priority for memory allocation
2711 * Note: While this allocator tries to use high order pages, there is
2712 * no guarantee that allocations succeed. Therefore, @sz MUST be
2713 * less or equal than PAGE_SIZE.
2715 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2718 if (page_ref_count(pfrag->page) == 1) {
2722 if (pfrag->offset + sz <= pfrag->size)
2724 put_page(pfrag->page);
2728 if (SKB_FRAG_PAGE_ORDER &&
2729 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2730 /* Avoid direct reclaim but allow kswapd to wake */
2731 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2732 __GFP_COMP | __GFP_NOWARN |
2734 SKB_FRAG_PAGE_ORDER);
2735 if (likely(pfrag->page)) {
2736 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2740 pfrag->page = alloc_page(gfp);
2741 if (likely(pfrag->page)) {
2742 pfrag->size = PAGE_SIZE;
2747 EXPORT_SYMBOL(skb_page_frag_refill);
2749 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2751 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2754 sk_enter_memory_pressure(sk);
2755 sk_stream_moderate_sndbuf(sk);
2758 EXPORT_SYMBOL(sk_page_frag_refill);
2760 void __lock_sock(struct sock *sk)
2761 __releases(&sk->sk_lock.slock)
2762 __acquires(&sk->sk_lock.slock)
2767 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2768 TASK_UNINTERRUPTIBLE);
2769 spin_unlock_bh(&sk->sk_lock.slock);
2771 spin_lock_bh(&sk->sk_lock.slock);
2772 if (!sock_owned_by_user(sk))
2775 finish_wait(&sk->sk_lock.wq, &wait);
2778 void __release_sock(struct sock *sk)
2779 __releases(&sk->sk_lock.slock)
2780 __acquires(&sk->sk_lock.slock)
2782 struct sk_buff *skb, *next;
2784 while ((skb = sk->sk_backlog.head) != NULL) {
2785 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2787 spin_unlock_bh(&sk->sk_lock.slock);
2792 WARN_ON_ONCE(skb_dst_is_noref(skb));
2793 skb_mark_not_on_list(skb);
2794 sk_backlog_rcv(sk, skb);
2799 } while (skb != NULL);
2801 spin_lock_bh(&sk->sk_lock.slock);
2805 * Doing the zeroing here guarantee we can not loop forever
2806 * while a wild producer attempts to flood us.
2808 sk->sk_backlog.len = 0;
2811 void __sk_flush_backlog(struct sock *sk)
2813 spin_lock_bh(&sk->sk_lock.slock);
2815 spin_unlock_bh(&sk->sk_lock.slock);
2819 * sk_wait_data - wait for data to arrive at sk_receive_queue
2820 * @sk: sock to wait on
2821 * @timeo: for how long
2822 * @skb: last skb seen on sk_receive_queue
2824 * Now socket state including sk->sk_err is changed only under lock,
2825 * hence we may omit checks after joining wait queue.
2826 * We check receive queue before schedule() only as optimization;
2827 * it is very likely that release_sock() added new data.
2829 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2831 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2834 add_wait_queue(sk_sleep(sk), &wait);
2835 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2836 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2837 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2838 remove_wait_queue(sk_sleep(sk), &wait);
2841 EXPORT_SYMBOL(sk_wait_data);
2844 * __sk_mem_raise_allocated - increase memory_allocated
2846 * @size: memory size to allocate
2847 * @amt: pages to allocate
2848 * @kind: allocation type
2850 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2852 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2854 struct proto *prot = sk->sk_prot;
2855 long allocated = sk_memory_allocated_add(sk, amt);
2856 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2857 bool charged = true;
2860 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2861 gfp_memcg_charge())))
2862 goto suppress_allocation;
2865 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2866 sk_leave_memory_pressure(sk);
2870 /* Under pressure. */
2871 if (allocated > sk_prot_mem_limits(sk, 1))
2872 sk_enter_memory_pressure(sk);
2874 /* Over hard limit. */
2875 if (allocated > sk_prot_mem_limits(sk, 2))
2876 goto suppress_allocation;
2878 /* guarantee minimum buffer size under pressure */
2879 if (kind == SK_MEM_RECV) {
2880 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2883 } else { /* SK_MEM_SEND */
2884 int wmem0 = sk_get_wmem0(sk, prot);
2886 if (sk->sk_type == SOCK_STREAM) {
2887 if (sk->sk_wmem_queued < wmem0)
2889 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2894 if (sk_has_memory_pressure(sk)) {
2897 if (!sk_under_memory_pressure(sk))
2899 alloc = sk_sockets_allocated_read_positive(sk);
2900 if (sk_prot_mem_limits(sk, 2) > alloc *
2901 sk_mem_pages(sk->sk_wmem_queued +
2902 atomic_read(&sk->sk_rmem_alloc) +
2903 sk->sk_forward_alloc))
2907 suppress_allocation:
2909 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2910 sk_stream_moderate_sndbuf(sk);
2912 /* Fail only if socket is _under_ its sndbuf.
2913 * In this case we cannot block, so that we have to fail.
2915 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
2916 /* Force charge with __GFP_NOFAIL */
2917 if (memcg_charge && !charged) {
2918 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2919 gfp_memcg_charge() | __GFP_NOFAIL);
2925 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2926 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2928 sk_memory_allocated_sub(sk, amt);
2930 if (memcg_charge && charged)
2931 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2935 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2938 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2940 * @size: memory size to allocate
2941 * @kind: allocation type
2943 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2944 * rmem allocation. This function assumes that protocols which have
2945 * memory_pressure use sk_wmem_queued as write buffer accounting.
2947 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2949 int ret, amt = sk_mem_pages(size);
2951 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2952 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2954 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2957 EXPORT_SYMBOL(__sk_mem_schedule);
2960 * __sk_mem_reduce_allocated - reclaim memory_allocated
2962 * @amount: number of quanta
2964 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2966 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2968 sk_memory_allocated_sub(sk, amount);
2970 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2971 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2973 if (sk_under_memory_pressure(sk) &&
2974 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2975 sk_leave_memory_pressure(sk);
2977 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2980 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2982 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2984 void __sk_mem_reclaim(struct sock *sk, int amount)
2986 amount >>= SK_MEM_QUANTUM_SHIFT;
2987 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2988 __sk_mem_reduce_allocated(sk, amount);
2990 EXPORT_SYMBOL(__sk_mem_reclaim);
2992 int sk_set_peek_off(struct sock *sk, int val)
2994 sk->sk_peek_off = val;
2997 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3000 * Set of default routines for initialising struct proto_ops when
3001 * the protocol does not support a particular function. In certain
3002 * cases where it makes no sense for a protocol to have a "do nothing"
3003 * function, some default processing is provided.
3006 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3010 EXPORT_SYMBOL(sock_no_bind);
3012 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3017 EXPORT_SYMBOL(sock_no_connect);
3019 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3023 EXPORT_SYMBOL(sock_no_socketpair);
3025 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3030 EXPORT_SYMBOL(sock_no_accept);
3032 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3037 EXPORT_SYMBOL(sock_no_getname);
3039 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3043 EXPORT_SYMBOL(sock_no_ioctl);
3045 int sock_no_listen(struct socket *sock, int backlog)
3049 EXPORT_SYMBOL(sock_no_listen);
3051 int sock_no_shutdown(struct socket *sock, int how)
3055 EXPORT_SYMBOL(sock_no_shutdown);
3057 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3061 EXPORT_SYMBOL(sock_no_sendmsg);
3063 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3067 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3069 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3074 EXPORT_SYMBOL(sock_no_recvmsg);
3076 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3078 /* Mirror missing mmap method error code */
3081 EXPORT_SYMBOL(sock_no_mmap);
3084 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3085 * various sock-based usage counts.
3087 void __receive_sock(struct file *file)
3089 struct socket *sock;
3091 sock = sock_from_file(file);
3093 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3094 sock_update_classid(&sock->sk->sk_cgrp_data);
3098 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3101 struct msghdr msg = {.msg_flags = flags};
3103 char *kaddr = kmap(page);
3104 iov.iov_base = kaddr + offset;
3106 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3110 EXPORT_SYMBOL(sock_no_sendpage);
3112 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3113 int offset, size_t size, int flags)
3116 struct msghdr msg = {.msg_flags = flags};
3118 char *kaddr = kmap(page);
3120 iov.iov_base = kaddr + offset;
3122 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3126 EXPORT_SYMBOL(sock_no_sendpage_locked);
3129 * Default Socket Callbacks
3132 static void sock_def_wakeup(struct sock *sk)
3134 struct socket_wq *wq;
3137 wq = rcu_dereference(sk->sk_wq);
3138 if (skwq_has_sleeper(wq))
3139 wake_up_interruptible_all(&wq->wait);
3143 static void sock_def_error_report(struct sock *sk)
3145 struct socket_wq *wq;
3148 wq = rcu_dereference(sk->sk_wq);
3149 if (skwq_has_sleeper(wq))
3150 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3151 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3155 void sock_def_readable(struct sock *sk)
3157 struct socket_wq *wq;
3160 wq = rcu_dereference(sk->sk_wq);
3161 if (skwq_has_sleeper(wq))
3162 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3163 EPOLLRDNORM | EPOLLRDBAND);
3164 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3168 static void sock_def_write_space(struct sock *sk)
3170 struct socket_wq *wq;
3174 /* Do not wake up a writer until he can make "significant"
3177 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
3178 wq = rcu_dereference(sk->sk_wq);
3179 if (skwq_has_sleeper(wq))
3180 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3181 EPOLLWRNORM | EPOLLWRBAND);
3183 /* Should agree with poll, otherwise some programs break */
3184 if (sock_writeable(sk))
3185 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3191 static void sock_def_destruct(struct sock *sk)
3195 void sk_send_sigurg(struct sock *sk)
3197 if (sk->sk_socket && sk->sk_socket->file)
3198 if (send_sigurg(&sk->sk_socket->file->f_owner))
3199 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3201 EXPORT_SYMBOL(sk_send_sigurg);
3203 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3204 unsigned long expires)
3206 if (!mod_timer(timer, expires))
3209 EXPORT_SYMBOL(sk_reset_timer);
3211 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3213 if (del_timer(timer))
3216 EXPORT_SYMBOL(sk_stop_timer);
3218 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3220 if (del_timer_sync(timer))
3223 EXPORT_SYMBOL(sk_stop_timer_sync);
3225 void sock_init_data(struct socket *sock, struct sock *sk)
3228 sk->sk_send_head = NULL;
3230 timer_setup(&sk->sk_timer, NULL, 0);
3232 sk->sk_allocation = GFP_KERNEL;
3233 sk->sk_rcvbuf = sysctl_rmem_default;
3234 sk->sk_sndbuf = sysctl_wmem_default;
3235 sk->sk_state = TCP_CLOSE;
3236 sk_set_socket(sk, sock);
3238 sock_set_flag(sk, SOCK_ZAPPED);
3241 sk->sk_type = sock->type;
3242 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3244 sk->sk_uid = SOCK_INODE(sock)->i_uid;
3246 RCU_INIT_POINTER(sk->sk_wq, NULL);
3247 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3250 rwlock_init(&sk->sk_callback_lock);
3251 if (sk->sk_kern_sock)
3252 lockdep_set_class_and_name(
3253 &sk->sk_callback_lock,
3254 af_kern_callback_keys + sk->sk_family,
3255 af_family_kern_clock_key_strings[sk->sk_family]);
3257 lockdep_set_class_and_name(
3258 &sk->sk_callback_lock,
3259 af_callback_keys + sk->sk_family,
3260 af_family_clock_key_strings[sk->sk_family]);
3262 sk->sk_state_change = sock_def_wakeup;
3263 sk->sk_data_ready = sock_def_readable;
3264 sk->sk_write_space = sock_def_write_space;
3265 sk->sk_error_report = sock_def_error_report;
3266 sk->sk_destruct = sock_def_destruct;
3268 sk->sk_frag.page = NULL;
3269 sk->sk_frag.offset = 0;
3270 sk->sk_peek_off = -1;
3272 sk->sk_peer_pid = NULL;
3273 sk->sk_peer_cred = NULL;
3274 spin_lock_init(&sk->sk_peer_lock);
3276 sk->sk_write_pending = 0;
3277 sk->sk_rcvlowat = 1;
3278 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3279 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3281 sk->sk_stamp = SK_DEFAULT_STAMP;
3282 #if BITS_PER_LONG==32
3283 seqlock_init(&sk->sk_stamp_seq);
3285 atomic_set(&sk->sk_zckey, 0);
3287 #ifdef CONFIG_NET_RX_BUSY_POLL
3289 sk->sk_ll_usec = sysctl_net_busy_read;
3292 sk->sk_max_pacing_rate = ~0UL;
3293 sk->sk_pacing_rate = ~0UL;
3294 WRITE_ONCE(sk->sk_pacing_shift, 10);
3295 sk->sk_incoming_cpu = -1;
3296 sk->sk_txrehash = SOCK_TXREHASH_DEFAULT;
3298 sk_rx_queue_clear(sk);
3300 * Before updating sk_refcnt, we must commit prior changes to memory
3301 * (Documentation/RCU/rculist_nulls.rst for details)
3304 refcount_set(&sk->sk_refcnt, 1);
3305 atomic_set(&sk->sk_drops, 0);
3307 EXPORT_SYMBOL(sock_init_data);
3309 void lock_sock_nested(struct sock *sk, int subclass)
3311 /* The sk_lock has mutex_lock() semantics here. */
3312 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3315 spin_lock_bh(&sk->sk_lock.slock);
3316 if (sock_owned_by_user_nocheck(sk))
3318 sk->sk_lock.owned = 1;
3319 spin_unlock_bh(&sk->sk_lock.slock);
3321 EXPORT_SYMBOL(lock_sock_nested);
3323 void release_sock(struct sock *sk)
3325 spin_lock_bh(&sk->sk_lock.slock);
3326 if (sk->sk_backlog.tail)
3329 /* Warning : release_cb() might need to release sk ownership,
3330 * ie call sock_release_ownership(sk) before us.
3332 if (sk->sk_prot->release_cb)
3333 sk->sk_prot->release_cb(sk);
3335 sock_release_ownership(sk);
3336 if (waitqueue_active(&sk->sk_lock.wq))
3337 wake_up(&sk->sk_lock.wq);
3338 spin_unlock_bh(&sk->sk_lock.slock);
3340 EXPORT_SYMBOL(release_sock);
3342 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3345 spin_lock_bh(&sk->sk_lock.slock);
3347 if (!sock_owned_by_user_nocheck(sk)) {
3349 * Fast path return with bottom halves disabled and
3350 * sock::sk_lock.slock held.
3352 * The 'mutex' is not contended and holding
3353 * sock::sk_lock.slock prevents all other lockers to
3354 * proceed so the corresponding unlock_sock_fast() can
3355 * avoid the slow path of release_sock() completely and
3356 * just release slock.
3358 * From a semantical POV this is equivalent to 'acquiring'
3359 * the 'mutex', hence the corresponding lockdep
3360 * mutex_release() has to happen in the fast path of
3361 * unlock_sock_fast().
3367 sk->sk_lock.owned = 1;
3368 __acquire(&sk->sk_lock.slock);
3369 spin_unlock_bh(&sk->sk_lock.slock);
3372 EXPORT_SYMBOL(__lock_sock_fast);
3374 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3375 bool timeval, bool time32)
3377 struct sock *sk = sock->sk;
3378 struct timespec64 ts;
3380 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3381 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3382 if (ts.tv_sec == -1)
3384 if (ts.tv_sec == 0) {
3385 ktime_t kt = ktime_get_real();
3386 sock_write_timestamp(sk, kt);
3387 ts = ktime_to_timespec64(kt);
3393 #ifdef CONFIG_COMPAT_32BIT_TIME
3395 return put_old_timespec32(&ts, userstamp);
3397 #ifdef CONFIG_SPARC64
3398 /* beware of padding in sparc64 timeval */
3399 if (timeval && !in_compat_syscall()) {
3400 struct __kernel_old_timeval __user tv = {
3401 .tv_sec = ts.tv_sec,
3402 .tv_usec = ts.tv_nsec,
3404 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3409 return put_timespec64(&ts, userstamp);
3411 EXPORT_SYMBOL(sock_gettstamp);
3413 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3415 if (!sock_flag(sk, flag)) {
3416 unsigned long previous_flags = sk->sk_flags;
3418 sock_set_flag(sk, flag);
3420 * we just set one of the two flags which require net
3421 * time stamping, but time stamping might have been on
3422 * already because of the other one
3424 if (sock_needs_netstamp(sk) &&
3425 !(previous_flags & SK_FLAGS_TIMESTAMP))
3426 net_enable_timestamp();
3430 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3431 int level, int type)
3433 struct sock_exterr_skb *serr;
3434 struct sk_buff *skb;
3438 skb = sock_dequeue_err_skb(sk);
3444 msg->msg_flags |= MSG_TRUNC;
3447 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3451 sock_recv_timestamp(msg, sk, skb);
3453 serr = SKB_EXT_ERR(skb);
3454 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3456 msg->msg_flags |= MSG_ERRQUEUE;
3464 EXPORT_SYMBOL(sock_recv_errqueue);
3467 * Get a socket option on an socket.
3469 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3470 * asynchronous errors should be reported by getsockopt. We assume
3471 * this means if you specify SO_ERROR (otherwise whats the point of it).
3473 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3474 char __user *optval, int __user *optlen)
3476 struct sock *sk = sock->sk;
3478 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3480 EXPORT_SYMBOL(sock_common_getsockopt);
3482 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3485 struct sock *sk = sock->sk;
3489 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3490 flags & ~MSG_DONTWAIT, &addr_len);
3492 msg->msg_namelen = addr_len;
3495 EXPORT_SYMBOL(sock_common_recvmsg);
3498 * Set socket options on an inet socket.
3500 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3501 sockptr_t optval, unsigned int optlen)
3503 struct sock *sk = sock->sk;
3505 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3507 EXPORT_SYMBOL(sock_common_setsockopt);
3509 void sk_common_release(struct sock *sk)
3511 if (sk->sk_prot->destroy)
3512 sk->sk_prot->destroy(sk);
3515 * Observation: when sk_common_release is called, processes have
3516 * no access to socket. But net still has.
3517 * Step one, detach it from networking:
3519 * A. Remove from hash tables.
3522 sk->sk_prot->unhash(sk);
3525 * In this point socket cannot receive new packets, but it is possible
3526 * that some packets are in flight because some CPU runs receiver and
3527 * did hash table lookup before we unhashed socket. They will achieve
3528 * receive queue and will be purged by socket destructor.
3530 * Also we still have packets pending on receive queue and probably,
3531 * our own packets waiting in device queues. sock_destroy will drain
3532 * receive queue, but transmitted packets will delay socket destruction
3533 * until the last reference will be released.
3538 xfrm_sk_free_policy(sk);
3540 sk_refcnt_debug_release(sk);
3544 EXPORT_SYMBOL(sk_common_release);
3546 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3548 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3550 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3551 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3552 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3553 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3554 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3555 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3556 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3557 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3558 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3561 #ifdef CONFIG_PROC_FS
3562 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3564 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3566 int cpu, idx = prot->inuse_idx;
3569 for_each_possible_cpu(cpu)
3570 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3572 return res >= 0 ? res : 0;
3574 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3576 int sock_inuse_get(struct net *net)
3580 for_each_possible_cpu(cpu)
3581 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3586 EXPORT_SYMBOL_GPL(sock_inuse_get);
3588 static int __net_init sock_inuse_init_net(struct net *net)
3590 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3591 if (net->core.prot_inuse == NULL)
3596 static void __net_exit sock_inuse_exit_net(struct net *net)
3598 free_percpu(net->core.prot_inuse);
3601 static struct pernet_operations net_inuse_ops = {
3602 .init = sock_inuse_init_net,
3603 .exit = sock_inuse_exit_net,
3606 static __init int net_inuse_init(void)
3608 if (register_pernet_subsys(&net_inuse_ops))
3609 panic("Cannot initialize net inuse counters");
3614 core_initcall(net_inuse_init);
3616 static int assign_proto_idx(struct proto *prot)
3618 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3620 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3621 pr_err("PROTO_INUSE_NR exhausted\n");
3625 set_bit(prot->inuse_idx, proto_inuse_idx);
3629 static void release_proto_idx(struct proto *prot)
3631 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3632 clear_bit(prot->inuse_idx, proto_inuse_idx);
3635 static inline int assign_proto_idx(struct proto *prot)
3640 static inline void release_proto_idx(struct proto *prot)
3646 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3650 kfree(twsk_prot->twsk_slab_name);
3651 twsk_prot->twsk_slab_name = NULL;
3652 kmem_cache_destroy(twsk_prot->twsk_slab);
3653 twsk_prot->twsk_slab = NULL;
3656 static int tw_prot_init(const struct proto *prot)
3658 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3663 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3665 if (!twsk_prot->twsk_slab_name)
3668 twsk_prot->twsk_slab =
3669 kmem_cache_create(twsk_prot->twsk_slab_name,
3670 twsk_prot->twsk_obj_size, 0,
3671 SLAB_ACCOUNT | prot->slab_flags,
3673 if (!twsk_prot->twsk_slab) {
3674 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3682 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3686 kfree(rsk_prot->slab_name);
3687 rsk_prot->slab_name = NULL;
3688 kmem_cache_destroy(rsk_prot->slab);
3689 rsk_prot->slab = NULL;
3692 static int req_prot_init(const struct proto *prot)
3694 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3699 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3701 if (!rsk_prot->slab_name)
3704 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3705 rsk_prot->obj_size, 0,
3706 SLAB_ACCOUNT | prot->slab_flags,
3709 if (!rsk_prot->slab) {
3710 pr_crit("%s: Can't create request sock SLAB cache!\n",
3717 int proto_register(struct proto *prot, int alloc_slab)
3721 if (prot->memory_allocated && !prot->sysctl_mem) {
3722 pr_err("%s: missing sysctl_mem\n", prot->name);
3726 prot->slab = kmem_cache_create_usercopy(prot->name,
3728 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3730 prot->useroffset, prot->usersize,
3733 if (prot->slab == NULL) {
3734 pr_crit("%s: Can't create sock SLAB cache!\n",
3739 if (req_prot_init(prot))
3740 goto out_free_request_sock_slab;
3742 if (tw_prot_init(prot))
3743 goto out_free_timewait_sock_slab;
3746 mutex_lock(&proto_list_mutex);
3747 ret = assign_proto_idx(prot);
3749 mutex_unlock(&proto_list_mutex);
3750 goto out_free_timewait_sock_slab;
3752 list_add(&prot->node, &proto_list);
3753 mutex_unlock(&proto_list_mutex);
3756 out_free_timewait_sock_slab:
3758 tw_prot_cleanup(prot->twsk_prot);
3759 out_free_request_sock_slab:
3761 req_prot_cleanup(prot->rsk_prot);
3763 kmem_cache_destroy(prot->slab);
3769 EXPORT_SYMBOL(proto_register);
3771 void proto_unregister(struct proto *prot)
3773 mutex_lock(&proto_list_mutex);
3774 release_proto_idx(prot);
3775 list_del(&prot->node);
3776 mutex_unlock(&proto_list_mutex);
3778 kmem_cache_destroy(prot->slab);
3781 req_prot_cleanup(prot->rsk_prot);
3782 tw_prot_cleanup(prot->twsk_prot);
3784 EXPORT_SYMBOL(proto_unregister);
3786 int sock_load_diag_module(int family, int protocol)
3789 if (!sock_is_registered(family))
3792 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3793 NETLINK_SOCK_DIAG, family);
3797 if (family == AF_INET &&
3798 protocol != IPPROTO_RAW &&
3799 protocol < MAX_INET_PROTOS &&
3800 !rcu_access_pointer(inet_protos[protocol]))
3804 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3805 NETLINK_SOCK_DIAG, family, protocol);
3807 EXPORT_SYMBOL(sock_load_diag_module);
3809 #ifdef CONFIG_PROC_FS
3810 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3811 __acquires(proto_list_mutex)
3813 mutex_lock(&proto_list_mutex);
3814 return seq_list_start_head(&proto_list, *pos);
3817 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3819 return seq_list_next(v, &proto_list, pos);
3822 static void proto_seq_stop(struct seq_file *seq, void *v)
3823 __releases(proto_list_mutex)
3825 mutex_unlock(&proto_list_mutex);
3828 static char proto_method_implemented(const void *method)
3830 return method == NULL ? 'n' : 'y';
3832 static long sock_prot_memory_allocated(struct proto *proto)
3834 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3837 static const char *sock_prot_memory_pressure(struct proto *proto)
3839 return proto->memory_pressure != NULL ?
3840 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3843 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3846 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3847 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3850 sock_prot_inuse_get(seq_file_net(seq), proto),
3851 sock_prot_memory_allocated(proto),
3852 sock_prot_memory_pressure(proto),
3854 proto->slab == NULL ? "no" : "yes",
3855 module_name(proto->owner),
3856 proto_method_implemented(proto->close),
3857 proto_method_implemented(proto->connect),
3858 proto_method_implemented(proto->disconnect),
3859 proto_method_implemented(proto->accept),
3860 proto_method_implemented(proto->ioctl),
3861 proto_method_implemented(proto->init),
3862 proto_method_implemented(proto->destroy),
3863 proto_method_implemented(proto->shutdown),
3864 proto_method_implemented(proto->setsockopt),
3865 proto_method_implemented(proto->getsockopt),
3866 proto_method_implemented(proto->sendmsg),
3867 proto_method_implemented(proto->recvmsg),
3868 proto_method_implemented(proto->sendpage),
3869 proto_method_implemented(proto->bind),
3870 proto_method_implemented(proto->backlog_rcv),
3871 proto_method_implemented(proto->hash),
3872 proto_method_implemented(proto->unhash),
3873 proto_method_implemented(proto->get_port),
3874 proto_method_implemented(proto->enter_memory_pressure));
3877 static int proto_seq_show(struct seq_file *seq, void *v)
3879 if (v == &proto_list)
3880 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3889 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3891 proto_seq_printf(seq, list_entry(v, struct proto, node));
3895 static const struct seq_operations proto_seq_ops = {
3896 .start = proto_seq_start,
3897 .next = proto_seq_next,
3898 .stop = proto_seq_stop,
3899 .show = proto_seq_show,
3902 static __net_init int proto_init_net(struct net *net)
3904 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3905 sizeof(struct seq_net_private)))
3911 static __net_exit void proto_exit_net(struct net *net)
3913 remove_proc_entry("protocols", net->proc_net);
3917 static __net_initdata struct pernet_operations proto_net_ops = {
3918 .init = proto_init_net,
3919 .exit = proto_exit_net,
3922 static int __init proto_init(void)
3924 return register_pernet_subsys(&proto_net_ops);
3927 subsys_initcall(proto_init);
3929 #endif /* PROC_FS */
3931 #ifdef CONFIG_NET_RX_BUSY_POLL
3932 bool sk_busy_loop_end(void *p, unsigned long start_time)
3934 struct sock *sk = p;
3936 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3937 sk_busy_loop_timeout(sk, start_time);
3939 EXPORT_SYMBOL(sk_busy_loop_end);
3940 #endif /* CONFIG_NET_RX_BUSY_POLL */
3942 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3944 if (!sk->sk_prot->bind_add)
3946 return sk->sk_prot->bind_add(sk, addr, addr_len);
3948 EXPORT_SYMBOL(sock_bind_add);