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/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 <net/net_namespace.h>
128 #include <net/request_sock.h>
129 #include <net/sock.h>
130 #include <linux/net_tstamp.h>
131 #include <net/xfrm.h>
132 #include <linux/ipsec.h>
133 #include <net/cls_cgroup.h>
134 #include <net/netprio_cgroup.h>
135 #include <linux/sock_diag.h>
137 #include <linux/filter.h>
138 #include <net/sock_reuseport.h>
139 #include <net/bpf_sk_storage.h>
141 #include <trace/events/sock.h>
144 #include <net/busy_poll.h>
145 #include <net/phonet/phonet.h>
147 #include <linux/ethtool.h>
151 static DEFINE_MUTEX(proto_list_mutex);
152 static LIST_HEAD(proto_list);
154 static void sock_def_write_space_wfree(struct sock *sk);
155 static void sock_def_write_space(struct sock *sk);
158 * sk_ns_capable - General socket capability test
159 * @sk: Socket to use a capability on or through
160 * @user_ns: The user namespace of the capability to use
161 * @cap: The capability to use
163 * Test to see if the opener of the socket had when the socket was
164 * created and the current process has the capability @cap in the user
165 * namespace @user_ns.
167 bool sk_ns_capable(const struct sock *sk,
168 struct user_namespace *user_ns, int cap)
170 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
171 ns_capable(user_ns, cap);
173 EXPORT_SYMBOL(sk_ns_capable);
176 * sk_capable - Socket global capability test
177 * @sk: Socket to use a capability on or through
178 * @cap: The global capability to use
180 * Test to see if the opener of the socket had when the socket was
181 * created and the current process has the capability @cap in all user
184 bool sk_capable(const struct sock *sk, int cap)
186 return sk_ns_capable(sk, &init_user_ns, cap);
188 EXPORT_SYMBOL(sk_capable);
191 * sk_net_capable - Network namespace socket capability test
192 * @sk: Socket to use a capability on or through
193 * @cap: The capability to use
195 * Test to see if the opener of the socket had when the socket was created
196 * and the current process has the capability @cap over the network namespace
197 * the socket is a member of.
199 bool sk_net_capable(const struct sock *sk, int cap)
201 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
203 EXPORT_SYMBOL(sk_net_capable);
206 * Each address family might have different locking rules, so we have
207 * one slock key per address family and separate keys for internal and
210 static struct lock_class_key af_family_keys[AF_MAX];
211 static struct lock_class_key af_family_kern_keys[AF_MAX];
212 static struct lock_class_key af_family_slock_keys[AF_MAX];
213 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
216 * Make lock validator output more readable. (we pre-construct these
217 * strings build-time, so that runtime initialization of socket
221 #define _sock_locks(x) \
222 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
223 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
224 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
225 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
226 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
227 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
228 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
229 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
230 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
231 x "27" , x "28" , x "AF_CAN" , \
232 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
233 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
234 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
235 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
236 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
240 static const char *const af_family_key_strings[AF_MAX+1] = {
241 _sock_locks("sk_lock-")
243 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
244 _sock_locks("slock-")
246 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
247 _sock_locks("clock-")
250 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
251 _sock_locks("k-sk_lock-")
253 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
254 _sock_locks("k-slock-")
256 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
257 _sock_locks("k-clock-")
259 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
260 _sock_locks("rlock-")
262 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
263 _sock_locks("wlock-")
265 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
266 _sock_locks("elock-")
270 * sk_callback_lock and sk queues locking rules are per-address-family,
271 * so split the lock classes by using a per-AF key:
273 static struct lock_class_key af_callback_keys[AF_MAX];
274 static struct lock_class_key af_rlock_keys[AF_MAX];
275 static struct lock_class_key af_wlock_keys[AF_MAX];
276 static struct lock_class_key af_elock_keys[AF_MAX];
277 static struct lock_class_key af_kern_callback_keys[AF_MAX];
279 /* Run time adjustable parameters. */
280 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
281 EXPORT_SYMBOL(sysctl_wmem_max);
282 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
283 EXPORT_SYMBOL(sysctl_rmem_max);
284 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
285 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
287 int sysctl_tstamp_allow_data __read_mostly = 1;
289 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
290 EXPORT_SYMBOL_GPL(memalloc_socks_key);
293 * sk_set_memalloc - sets %SOCK_MEMALLOC
294 * @sk: socket to set it on
296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
297 * It's the responsibility of the admin to adjust min_free_kbytes
298 * to meet the requirements
300 void sk_set_memalloc(struct sock *sk)
302 sock_set_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation |= __GFP_MEMALLOC;
304 static_branch_inc(&memalloc_socks_key);
306 EXPORT_SYMBOL_GPL(sk_set_memalloc);
308 void sk_clear_memalloc(struct sock *sk)
310 sock_reset_flag(sk, SOCK_MEMALLOC);
311 sk->sk_allocation &= ~__GFP_MEMALLOC;
312 static_branch_dec(&memalloc_socks_key);
315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
316 * progress of swapping. SOCK_MEMALLOC may be cleared while
317 * it has rmem allocations due to the last swapfile being deactivated
318 * but there is a risk that the socket is unusable due to exceeding
319 * the rmem limits. Reclaim the reserves and obey rmem limits again.
323 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
325 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
328 unsigned int noreclaim_flag;
330 /* these should have been dropped before queueing */
331 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
333 noreclaim_flag = memalloc_noreclaim_save();
334 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
338 memalloc_noreclaim_restore(noreclaim_flag);
342 EXPORT_SYMBOL(__sk_backlog_rcv);
344 void sk_error_report(struct sock *sk)
346 sk->sk_error_report(sk);
348 switch (sk->sk_family) {
352 trace_inet_sk_error_report(sk);
358 EXPORT_SYMBOL(sk_error_report);
360 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
362 struct __kernel_sock_timeval tv;
364 if (timeo == MAX_SCHEDULE_TIMEOUT) {
368 tv.tv_sec = timeo / HZ;
369 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
372 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
373 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
374 *(struct old_timeval32 *)optval = tv32;
379 struct __kernel_old_timeval old_tv;
380 old_tv.tv_sec = tv.tv_sec;
381 old_tv.tv_usec = tv.tv_usec;
382 *(struct __kernel_old_timeval *)optval = old_tv;
383 return sizeof(old_tv);
386 *(struct __kernel_sock_timeval *)optval = tv;
389 EXPORT_SYMBOL(sock_get_timeout);
391 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
392 sockptr_t optval, int optlen, bool old_timeval)
394 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
395 struct old_timeval32 tv32;
397 if (optlen < sizeof(tv32))
400 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
402 tv->tv_sec = tv32.tv_sec;
403 tv->tv_usec = tv32.tv_usec;
404 } else if (old_timeval) {
405 struct __kernel_old_timeval old_tv;
407 if (optlen < sizeof(old_tv))
409 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
411 tv->tv_sec = old_tv.tv_sec;
412 tv->tv_usec = old_tv.tv_usec;
414 if (optlen < sizeof(*tv))
416 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
422 EXPORT_SYMBOL(sock_copy_user_timeval);
424 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
427 struct __kernel_sock_timeval tv;
428 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
434 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
438 static int warned __read_mostly;
440 WRITE_ONCE(*timeo_p, 0);
441 if (warned < 10 && net_ratelimit()) {
443 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
444 __func__, current->comm, task_pid_nr(current));
448 val = MAX_SCHEDULE_TIMEOUT;
449 if ((tv.tv_sec || tv.tv_usec) &&
450 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
451 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
453 WRITE_ONCE(*timeo_p, val);
457 static bool sock_needs_netstamp(const struct sock *sk)
459 switch (sk->sk_family) {
468 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
470 if (sk->sk_flags & flags) {
471 sk->sk_flags &= ~flags;
472 if (sock_needs_netstamp(sk) &&
473 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
474 net_disable_timestamp();
479 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
482 struct sk_buff_head *list = &sk->sk_receive_queue;
484 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
485 atomic_inc(&sk->sk_drops);
486 trace_sock_rcvqueue_full(sk, skb);
490 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
491 atomic_inc(&sk->sk_drops);
496 skb_set_owner_r(skb, sk);
498 /* we escape from rcu protected region, make sure we dont leak
503 spin_lock_irqsave(&list->lock, flags);
504 sock_skb_set_dropcount(sk, skb);
505 __skb_queue_tail(list, skb);
506 spin_unlock_irqrestore(&list->lock, flags);
508 if (!sock_flag(sk, SOCK_DEAD))
509 sk->sk_data_ready(sk);
512 EXPORT_SYMBOL(__sock_queue_rcv_skb);
514 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
515 enum skb_drop_reason *reason)
517 enum skb_drop_reason drop_reason;
520 err = sk_filter(sk, skb);
522 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
525 err = __sock_queue_rcv_skb(sk, skb);
528 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
531 drop_reason = SKB_DROP_REASON_PROTO_MEM;
534 drop_reason = SKB_NOT_DROPPED_YET;
539 *reason = drop_reason;
542 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
544 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
545 const int nested, unsigned int trim_cap, bool refcounted)
547 int rc = NET_RX_SUCCESS;
549 if (sk_filter_trim_cap(sk, skb, trim_cap))
550 goto discard_and_relse;
554 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
555 atomic_inc(&sk->sk_drops);
556 goto discard_and_relse;
559 bh_lock_sock_nested(sk);
562 if (!sock_owned_by_user(sk)) {
564 * trylock + unlock semantics:
566 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
568 rc = sk_backlog_rcv(sk, skb);
570 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
571 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
573 atomic_inc(&sk->sk_drops);
574 goto discard_and_relse;
586 EXPORT_SYMBOL(__sk_receive_skb);
588 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
590 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
592 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
594 struct dst_entry *dst = __sk_dst_get(sk);
596 if (dst && dst->obsolete &&
597 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
598 dst, cookie) == NULL) {
599 sk_tx_queue_clear(sk);
600 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
601 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
608 EXPORT_SYMBOL(__sk_dst_check);
610 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
612 struct dst_entry *dst = sk_dst_get(sk);
614 if (dst && dst->obsolete &&
615 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
616 dst, cookie) == NULL) {
624 EXPORT_SYMBOL(sk_dst_check);
626 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
628 int ret = -ENOPROTOOPT;
629 #ifdef CONFIG_NETDEVICES
630 struct net *net = sock_net(sk);
634 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
641 /* Paired with all READ_ONCE() done locklessly. */
642 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
644 if (sk->sk_prot->rehash)
645 sk->sk_prot->rehash(sk);
656 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
662 ret = sock_bindtoindex_locked(sk, ifindex);
668 EXPORT_SYMBOL(sock_bindtoindex);
670 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
672 int ret = -ENOPROTOOPT;
673 #ifdef CONFIG_NETDEVICES
674 struct net *net = sock_net(sk);
675 char devname[IFNAMSIZ];
682 /* Bind this socket to a particular device like "eth0",
683 * as specified in the passed interface name. If the
684 * name is "" or the option length is zero the socket
687 if (optlen > IFNAMSIZ - 1)
688 optlen = IFNAMSIZ - 1;
689 memset(devname, 0, sizeof(devname));
692 if (copy_from_sockptr(devname, optval, optlen))
696 if (devname[0] != '\0') {
697 struct net_device *dev;
700 dev = dev_get_by_name_rcu(net, devname);
702 index = dev->ifindex;
709 sockopt_lock_sock(sk);
710 ret = sock_bindtoindex_locked(sk, index);
711 sockopt_release_sock(sk);
718 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
719 sockptr_t optlen, int len)
721 int ret = -ENOPROTOOPT;
722 #ifdef CONFIG_NETDEVICES
723 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
724 struct net *net = sock_net(sk);
725 char devname[IFNAMSIZ];
727 if (bound_dev_if == 0) {
736 ret = netdev_get_name(net, devname, bound_dev_if);
740 len = strlen(devname) + 1;
743 if (copy_to_sockptr(optval, devname, len))
748 if (copy_to_sockptr(optlen, &len, sizeof(int)))
759 bool sk_mc_loop(const struct sock *sk)
761 if (dev_recursion_level())
765 /* IPV6_ADDRFORM can change sk->sk_family under us. */
766 switch (READ_ONCE(sk->sk_family)) {
768 return inet_test_bit(MC_LOOP, sk);
769 #if IS_ENABLED(CONFIG_IPV6)
771 return inet6_test_bit(MC6_LOOP, sk);
777 EXPORT_SYMBOL(sk_mc_loop);
779 void sock_set_reuseaddr(struct sock *sk)
782 sk->sk_reuse = SK_CAN_REUSE;
785 EXPORT_SYMBOL(sock_set_reuseaddr);
787 void sock_set_reuseport(struct sock *sk)
790 sk->sk_reuseport = true;
793 EXPORT_SYMBOL(sock_set_reuseport);
795 void sock_no_linger(struct sock *sk)
798 WRITE_ONCE(sk->sk_lingertime, 0);
799 sock_set_flag(sk, SOCK_LINGER);
802 EXPORT_SYMBOL(sock_no_linger);
804 void sock_set_priority(struct sock *sk, u32 priority)
806 WRITE_ONCE(sk->sk_priority, priority);
808 EXPORT_SYMBOL(sock_set_priority);
810 void sock_set_sndtimeo(struct sock *sk, s64 secs)
813 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
814 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
816 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
819 EXPORT_SYMBOL(sock_set_sndtimeo);
821 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
824 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
825 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
826 sock_set_flag(sk, SOCK_RCVTSTAMP);
827 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
829 sock_reset_flag(sk, SOCK_RCVTSTAMP);
830 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
834 void sock_enable_timestamps(struct sock *sk)
837 __sock_set_timestamps(sk, true, false, true);
840 EXPORT_SYMBOL(sock_enable_timestamps);
842 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
845 case SO_TIMESTAMP_OLD:
846 __sock_set_timestamps(sk, valbool, false, false);
848 case SO_TIMESTAMP_NEW:
849 __sock_set_timestamps(sk, valbool, true, false);
851 case SO_TIMESTAMPNS_OLD:
852 __sock_set_timestamps(sk, valbool, false, true);
854 case SO_TIMESTAMPNS_NEW:
855 __sock_set_timestamps(sk, valbool, true, true);
860 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
862 struct net *net = sock_net(sk);
863 struct net_device *dev = NULL;
868 if (sk->sk_bound_dev_if)
869 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
872 pr_err("%s: sock not bind to device\n", __func__);
876 num = ethtool_get_phc_vclocks(dev, &vclock_index);
879 for (i = 0; i < num; i++) {
880 if (*(vclock_index + i) == phc_index) {
892 WRITE_ONCE(sk->sk_bind_phc, phc_index);
897 int sock_set_timestamping(struct sock *sk, int optname,
898 struct so_timestamping timestamping)
900 int val = timestamping.flags;
903 if (val & ~SOF_TIMESTAMPING_MASK)
906 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
907 !(val & SOF_TIMESTAMPING_OPT_ID))
910 if (val & SOF_TIMESTAMPING_OPT_ID &&
911 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
913 if ((1 << sk->sk_state) &
914 (TCPF_CLOSE | TCPF_LISTEN))
916 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
917 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
919 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
921 atomic_set(&sk->sk_tskey, 0);
925 if (val & SOF_TIMESTAMPING_OPT_STATS &&
926 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
929 if (val & SOF_TIMESTAMPING_BIND_PHC) {
930 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
935 WRITE_ONCE(sk->sk_tsflags, val);
936 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
938 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
939 sock_enable_timestamp(sk,
940 SOCK_TIMESTAMPING_RX_SOFTWARE);
942 sock_disable_timestamp(sk,
943 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
947 void sock_set_keepalive(struct sock *sk)
950 if (sk->sk_prot->keepalive)
951 sk->sk_prot->keepalive(sk, true);
952 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
955 EXPORT_SYMBOL(sock_set_keepalive);
957 static void __sock_set_rcvbuf(struct sock *sk, int val)
959 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
960 * as a negative value.
962 val = min_t(int, val, INT_MAX / 2);
963 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
965 /* We double it on the way in to account for "struct sk_buff" etc.
966 * overhead. Applications assume that the SO_RCVBUF setting they make
967 * will allow that much actual data to be received on that socket.
969 * Applications are unaware that "struct sk_buff" and other overheads
970 * allocate from the receive buffer during socket buffer allocation.
972 * And after considering the possible alternatives, returning the value
973 * we actually used in getsockopt is the most desirable behavior.
975 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
978 void sock_set_rcvbuf(struct sock *sk, int val)
981 __sock_set_rcvbuf(sk, val);
984 EXPORT_SYMBOL(sock_set_rcvbuf);
986 static void __sock_set_mark(struct sock *sk, u32 val)
988 if (val != sk->sk_mark) {
989 WRITE_ONCE(sk->sk_mark, val);
994 void sock_set_mark(struct sock *sk, u32 val)
997 __sock_set_mark(sk, val);
1000 EXPORT_SYMBOL(sock_set_mark);
1002 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1004 /* Round down bytes to multiple of pages */
1005 bytes = round_down(bytes, PAGE_SIZE);
1007 WARN_ON(bytes > sk->sk_reserved_mem);
1008 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1012 static int sock_reserve_memory(struct sock *sk, int bytes)
1018 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1024 pages = sk_mem_pages(bytes);
1026 /* pre-charge to memcg */
1027 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1028 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1032 /* pre-charge to forward_alloc */
1033 sk_memory_allocated_add(sk, pages);
1034 allocated = sk_memory_allocated(sk);
1035 /* If the system goes into memory pressure with this
1036 * precharge, give up and return error.
1038 if (allocated > sk_prot_mem_limits(sk, 1)) {
1039 sk_memory_allocated_sub(sk, pages);
1040 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1043 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1045 WRITE_ONCE(sk->sk_reserved_mem,
1046 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1051 void sockopt_lock_sock(struct sock *sk)
1053 /* When current->bpf_ctx is set, the setsockopt is called from
1054 * a bpf prog. bpf has ensured the sk lock has been
1055 * acquired before calling setsockopt().
1057 if (has_current_bpf_ctx())
1062 EXPORT_SYMBOL(sockopt_lock_sock);
1064 void sockopt_release_sock(struct sock *sk)
1066 if (has_current_bpf_ctx())
1071 EXPORT_SYMBOL(sockopt_release_sock);
1073 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1075 return has_current_bpf_ctx() || ns_capable(ns, cap);
1077 EXPORT_SYMBOL(sockopt_ns_capable);
1079 bool sockopt_capable(int cap)
1081 return has_current_bpf_ctx() || capable(cap);
1083 EXPORT_SYMBOL(sockopt_capable);
1086 * This is meant for all protocols to use and covers goings on
1087 * at the socket level. Everything here is generic.
1090 int sk_setsockopt(struct sock *sk, int level, int optname,
1091 sockptr_t optval, unsigned int optlen)
1093 struct so_timestamping timestamping;
1094 struct socket *sock = sk->sk_socket;
1095 struct sock_txtime sk_txtime;
1102 * Options without arguments
1105 if (optname == SO_BINDTODEVICE)
1106 return sock_setbindtodevice(sk, optval, optlen);
1108 if (optlen < sizeof(int))
1111 if (copy_from_sockptr(&val, optval, sizeof(val)))
1114 valbool = val ? 1 : 0;
1116 /* handle options which do not require locking the socket. */
1119 if ((val >= 0 && val <= 6) ||
1120 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1121 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1122 sock_set_priority(sk, val);
1127 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1130 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1133 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1139 return -ENOPROTOOPT;
1140 #ifdef CONFIG_NET_RX_BUSY_POLL
1144 WRITE_ONCE(sk->sk_ll_usec, val);
1146 case SO_PREFER_BUSY_POLL:
1147 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1149 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1151 case SO_BUSY_POLL_BUDGET:
1152 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1153 !sockopt_capable(CAP_NET_ADMIN))
1155 if (val < 0 || val > U16_MAX)
1157 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1160 case SO_MAX_PACING_RATE:
1162 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1163 unsigned long pacing_rate;
1165 if (sizeof(ulval) != sizeof(val) &&
1166 optlen >= sizeof(ulval) &&
1167 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1171 cmpxchg(&sk->sk_pacing_status,
1174 /* Pairs with READ_ONCE() from sk_getsockopt() */
1175 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1176 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1177 if (ulval < pacing_rate)
1178 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1182 if (val < -1 || val > 1)
1184 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1185 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1186 /* Paired with READ_ONCE() in tcp_rtx_synack()
1187 * and sk_getsockopt().
1189 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1193 int (*set_peek_off)(struct sock *sk, int val);
1195 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1197 ret = set_peek_off(sk, val);
1204 sockopt_lock_sock(sk);
1208 if (val && !sockopt_capable(CAP_NET_ADMIN))
1211 sock_valbool_flag(sk, SOCK_DBG, valbool);
1214 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1217 sk->sk_reuseport = valbool;
1220 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1224 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1227 /* Don't error on this BSD doesn't and if you think
1228 * about it this is right. Otherwise apps have to
1229 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1230 * are treated in BSD as hints
1232 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1234 /* Ensure val * 2 fits into an int, to prevent max_t()
1235 * from treating it as a negative value.
1237 val = min_t(int, val, INT_MAX / 2);
1238 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1239 WRITE_ONCE(sk->sk_sndbuf,
1240 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1241 /* Wake up sending tasks if we upped the value. */
1242 sk->sk_write_space(sk);
1245 case SO_SNDBUFFORCE:
1246 if (!sockopt_capable(CAP_NET_ADMIN)) {
1251 /* No negative values (to prevent underflow, as val will be
1259 /* Don't error on this BSD doesn't and if you think
1260 * about it this is right. Otherwise apps have to
1261 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1262 * are treated in BSD as hints
1264 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1267 case SO_RCVBUFFORCE:
1268 if (!sockopt_capable(CAP_NET_ADMIN)) {
1273 /* No negative values (to prevent underflow, as val will be
1276 __sock_set_rcvbuf(sk, max(val, 0));
1280 if (sk->sk_prot->keepalive)
1281 sk->sk_prot->keepalive(sk, valbool);
1282 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1286 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1290 sk->sk_no_check_tx = valbool;
1294 if (optlen < sizeof(ling)) {
1295 ret = -EINVAL; /* 1003.1g */
1298 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1302 if (!ling.l_onoff) {
1303 sock_reset_flag(sk, SOCK_LINGER);
1305 unsigned long t_sec = ling.l_linger;
1307 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1308 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1310 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1311 sock_set_flag(sk, SOCK_LINGER);
1318 case SO_TIMESTAMP_OLD:
1319 case SO_TIMESTAMP_NEW:
1320 case SO_TIMESTAMPNS_OLD:
1321 case SO_TIMESTAMPNS_NEW:
1322 sock_set_timestamp(sk, optname, valbool);
1325 case SO_TIMESTAMPING_NEW:
1326 case SO_TIMESTAMPING_OLD:
1327 if (optlen == sizeof(timestamping)) {
1328 if (copy_from_sockptr(×tamping, optval,
1329 sizeof(timestamping))) {
1334 memset(×tamping, 0, sizeof(timestamping));
1335 timestamping.flags = val;
1337 ret = sock_set_timestamping(sk, optname, timestamping);
1342 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1347 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1349 ret = set_rcvlowat(sk, val);
1351 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1354 case SO_RCVTIMEO_OLD:
1355 case SO_RCVTIMEO_NEW:
1356 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1357 optlen, optname == SO_RCVTIMEO_OLD);
1360 case SO_SNDTIMEO_OLD:
1361 case SO_SNDTIMEO_NEW:
1362 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1363 optlen, optname == SO_SNDTIMEO_OLD);
1366 case SO_ATTACH_FILTER: {
1367 struct sock_fprog fprog;
1369 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1371 ret = sk_attach_filter(&fprog, sk);
1376 if (optlen == sizeof(u32)) {
1380 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1383 ret = sk_attach_bpf(ufd, sk);
1387 case SO_ATTACH_REUSEPORT_CBPF: {
1388 struct sock_fprog fprog;
1390 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1392 ret = sk_reuseport_attach_filter(&fprog, sk);
1395 case SO_ATTACH_REUSEPORT_EBPF:
1397 if (optlen == sizeof(u32)) {
1401 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1404 ret = sk_reuseport_attach_bpf(ufd, sk);
1408 case SO_DETACH_REUSEPORT_BPF:
1409 ret = reuseport_detach_prog(sk);
1412 case SO_DETACH_FILTER:
1413 ret = sk_detach_filter(sk);
1416 case SO_LOCK_FILTER:
1417 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1420 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1424 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1425 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1430 __sock_set_mark(sk, val);
1433 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1437 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1440 case SO_WIFI_STATUS:
1441 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1445 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1448 case SO_SELECT_ERR_QUEUE:
1449 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1453 case SO_INCOMING_CPU:
1454 reuseport_update_incoming_cpu(sk, val);
1459 dst_negative_advice(sk);
1463 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1464 if (!(sk_is_tcp(sk) ||
1465 (sk->sk_type == SOCK_DGRAM &&
1466 sk->sk_protocol == IPPROTO_UDP)))
1468 } else if (sk->sk_family != PF_RDS) {
1472 if (val < 0 || val > 1)
1475 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1480 if (optlen != sizeof(struct sock_txtime)) {
1483 } else if (copy_from_sockptr(&sk_txtime, optval,
1484 sizeof(struct sock_txtime))) {
1487 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1491 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1492 * scheduler has enough safe guards.
1494 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1495 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1499 sock_valbool_flag(sk, SOCK_TXTIME, true);
1500 sk->sk_clockid = sk_txtime.clockid;
1501 sk->sk_txtime_deadline_mode =
1502 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1503 sk->sk_txtime_report_errors =
1504 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1507 case SO_BINDTOIFINDEX:
1508 ret = sock_bindtoindex_locked(sk, val);
1512 if (val & ~SOCK_BUF_LOCK_MASK) {
1516 sk->sk_userlocks = val | (sk->sk_userlocks &
1517 ~SOCK_BUF_LOCK_MASK);
1520 case SO_RESERVE_MEM:
1529 delta = val - sk->sk_reserved_mem;
1531 sock_release_reserved_memory(sk, -delta);
1533 ret = sock_reserve_memory(sk, delta);
1541 sockopt_release_sock(sk);
1545 int sock_setsockopt(struct socket *sock, int level, int optname,
1546 sockptr_t optval, unsigned int optlen)
1548 return sk_setsockopt(sock->sk, level, optname,
1551 EXPORT_SYMBOL(sock_setsockopt);
1553 static const struct cred *sk_get_peer_cred(struct sock *sk)
1555 const struct cred *cred;
1557 spin_lock(&sk->sk_peer_lock);
1558 cred = get_cred(sk->sk_peer_cred);
1559 spin_unlock(&sk->sk_peer_lock);
1564 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1565 struct ucred *ucred)
1567 ucred->pid = pid_vnr(pid);
1568 ucred->uid = ucred->gid = -1;
1570 struct user_namespace *current_ns = current_user_ns();
1572 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1573 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1577 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1579 struct user_namespace *user_ns = current_user_ns();
1582 for (i = 0; i < src->ngroups; i++) {
1583 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1585 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1592 int sk_getsockopt(struct sock *sk, int level, int optname,
1593 sockptr_t optval, sockptr_t optlen)
1595 struct socket *sock = sk->sk_socket;
1600 unsigned long ulval;
1602 struct old_timeval32 tm32;
1603 struct __kernel_old_timeval tm;
1604 struct __kernel_sock_timeval stm;
1605 struct sock_txtime txtime;
1606 struct so_timestamping timestamping;
1609 int lv = sizeof(int);
1612 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1617 memset(&v, 0, sizeof(v));
1621 v.val = sock_flag(sk, SOCK_DBG);
1625 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1629 v.val = sock_flag(sk, SOCK_BROADCAST);
1633 v.val = READ_ONCE(sk->sk_sndbuf);
1637 v.val = READ_ONCE(sk->sk_rcvbuf);
1641 v.val = sk->sk_reuse;
1645 v.val = sk->sk_reuseport;
1649 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1653 v.val = sk->sk_type;
1657 v.val = sk->sk_protocol;
1661 v.val = sk->sk_family;
1665 v.val = -sock_error(sk);
1667 v.val = xchg(&sk->sk_err_soft, 0);
1671 v.val = sock_flag(sk, SOCK_URGINLINE);
1675 v.val = sk->sk_no_check_tx;
1679 v.val = READ_ONCE(sk->sk_priority);
1683 lv = sizeof(v.ling);
1684 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1685 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1691 case SO_TIMESTAMP_OLD:
1692 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1693 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1694 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1697 case SO_TIMESTAMPNS_OLD:
1698 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1701 case SO_TIMESTAMP_NEW:
1702 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1705 case SO_TIMESTAMPNS_NEW:
1706 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1709 case SO_TIMESTAMPING_OLD:
1710 case SO_TIMESTAMPING_NEW:
1711 lv = sizeof(v.timestamping);
1712 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1713 * returning the flags when they were set through the same option.
1714 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1716 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1717 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1718 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1722 case SO_RCVTIMEO_OLD:
1723 case SO_RCVTIMEO_NEW:
1724 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1725 SO_RCVTIMEO_OLD == optname);
1728 case SO_SNDTIMEO_OLD:
1729 case SO_SNDTIMEO_NEW:
1730 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1731 SO_SNDTIMEO_OLD == optname);
1735 v.val = READ_ONCE(sk->sk_rcvlowat);
1743 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1747 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1752 struct ucred peercred;
1753 if (len > sizeof(peercred))
1754 len = sizeof(peercred);
1756 spin_lock(&sk->sk_peer_lock);
1757 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1758 spin_unlock(&sk->sk_peer_lock);
1760 if (copy_to_sockptr(optval, &peercred, len))
1767 struct pid *peer_pid;
1768 struct file *pidfd_file = NULL;
1771 if (len > sizeof(pidfd))
1772 len = sizeof(pidfd);
1774 spin_lock(&sk->sk_peer_lock);
1775 peer_pid = get_pid(sk->sk_peer_pid);
1776 spin_unlock(&sk->sk_peer_lock);
1781 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1786 if (copy_to_sockptr(optval, &pidfd, len) ||
1787 copy_to_sockptr(optlen, &len, sizeof(int))) {
1788 put_unused_fd(pidfd);
1794 fd_install(pidfd, pidfd_file);
1800 const struct cred *cred;
1803 cred = sk_get_peer_cred(sk);
1807 n = cred->group_info->ngroups;
1808 if (len < n * sizeof(gid_t)) {
1809 len = n * sizeof(gid_t);
1811 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1813 len = n * sizeof(gid_t);
1815 ret = groups_to_user(optval, cred->group_info);
1824 struct sockaddr_storage address;
1826 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1831 if (copy_to_sockptr(optval, &address, len))
1836 /* Dubious BSD thing... Probably nobody even uses it, but
1837 * the UNIX standard wants it for whatever reason... -DaveM
1840 v.val = sk->sk_state == TCP_LISTEN;
1844 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1848 return security_socket_getpeersec_stream(sock,
1849 optval, optlen, len);
1852 v.val = READ_ONCE(sk->sk_mark);
1856 v.val = sock_flag(sk, SOCK_RCVMARK);
1860 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1863 case SO_WIFI_STATUS:
1864 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1868 if (!READ_ONCE(sock->ops)->set_peek_off)
1871 v.val = READ_ONCE(sk->sk_peek_off);
1874 v.val = sock_flag(sk, SOCK_NOFCS);
1877 case SO_BINDTODEVICE:
1878 return sock_getbindtodevice(sk, optval, optlen, len);
1881 len = sk_get_filter(sk, optval, len);
1887 case SO_LOCK_FILTER:
1888 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1891 case SO_BPF_EXTENSIONS:
1892 v.val = bpf_tell_extensions();
1895 case SO_SELECT_ERR_QUEUE:
1896 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1899 #ifdef CONFIG_NET_RX_BUSY_POLL
1901 v.val = READ_ONCE(sk->sk_ll_usec);
1903 case SO_PREFER_BUSY_POLL:
1904 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1908 case SO_MAX_PACING_RATE:
1909 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1910 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1911 lv = sizeof(v.ulval);
1912 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1915 v.val = min_t(unsigned long, ~0U,
1916 READ_ONCE(sk->sk_max_pacing_rate));
1920 case SO_INCOMING_CPU:
1921 v.val = READ_ONCE(sk->sk_incoming_cpu);
1926 u32 meminfo[SK_MEMINFO_VARS];
1928 sk_get_meminfo(sk, meminfo);
1930 len = min_t(unsigned int, len, sizeof(meminfo));
1931 if (copy_to_sockptr(optval, &meminfo, len))
1937 #ifdef CONFIG_NET_RX_BUSY_POLL
1938 case SO_INCOMING_NAPI_ID:
1939 v.val = READ_ONCE(sk->sk_napi_id);
1941 /* aggregate non-NAPI IDs down to 0 */
1942 if (v.val < MIN_NAPI_ID)
1952 v.val64 = sock_gen_cookie(sk);
1956 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1960 lv = sizeof(v.txtime);
1961 v.txtime.clockid = sk->sk_clockid;
1962 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1963 SOF_TXTIME_DEADLINE_MODE : 0;
1964 v.txtime.flags |= sk->sk_txtime_report_errors ?
1965 SOF_TXTIME_REPORT_ERRORS : 0;
1968 case SO_BINDTOIFINDEX:
1969 v.val = READ_ONCE(sk->sk_bound_dev_if);
1972 case SO_NETNS_COOKIE:
1976 v.val64 = sock_net(sk)->net_cookie;
1980 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1983 case SO_RESERVE_MEM:
1984 v.val = READ_ONCE(sk->sk_reserved_mem);
1988 /* Paired with WRITE_ONCE() in sk_setsockopt() */
1989 v.val = READ_ONCE(sk->sk_txrehash);
1993 /* We implement the SO_SNDLOWAT etc to not be settable
1996 return -ENOPROTOOPT;
2001 if (copy_to_sockptr(optval, &v, len))
2004 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2010 * Initialize an sk_lock.
2012 * (We also register the sk_lock with the lock validator.)
2014 static inline void sock_lock_init(struct sock *sk)
2016 if (sk->sk_kern_sock)
2017 sock_lock_init_class_and_name(
2019 af_family_kern_slock_key_strings[sk->sk_family],
2020 af_family_kern_slock_keys + sk->sk_family,
2021 af_family_kern_key_strings[sk->sk_family],
2022 af_family_kern_keys + sk->sk_family);
2024 sock_lock_init_class_and_name(
2026 af_family_slock_key_strings[sk->sk_family],
2027 af_family_slock_keys + sk->sk_family,
2028 af_family_key_strings[sk->sk_family],
2029 af_family_keys + sk->sk_family);
2033 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2034 * even temporarly, because of RCU lookups. sk_node should also be left as is.
2035 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2037 static void sock_copy(struct sock *nsk, const struct sock *osk)
2039 const struct proto *prot = READ_ONCE(osk->sk_prot);
2040 #ifdef CONFIG_SECURITY_NETWORK
2041 void *sptr = nsk->sk_security;
2044 /* If we move sk_tx_queue_mapping out of the private section,
2045 * we must check if sk_tx_queue_clear() is called after
2046 * sock_copy() in sk_clone_lock().
2048 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2049 offsetof(struct sock, sk_dontcopy_begin) ||
2050 offsetof(struct sock, sk_tx_queue_mapping) >=
2051 offsetof(struct sock, sk_dontcopy_end));
2053 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2055 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2056 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2058 #ifdef CONFIG_SECURITY_NETWORK
2059 nsk->sk_security = sptr;
2060 security_sk_clone(osk, nsk);
2064 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2068 struct kmem_cache *slab;
2072 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2075 if (want_init_on_alloc(priority))
2076 sk_prot_clear_nulls(sk, prot->obj_size);
2078 sk = kmalloc(prot->obj_size, priority);
2081 if (security_sk_alloc(sk, family, priority))
2084 if (!try_module_get(prot->owner))
2091 security_sk_free(sk);
2094 kmem_cache_free(slab, sk);
2100 static void sk_prot_free(struct proto *prot, struct sock *sk)
2102 struct kmem_cache *slab;
2103 struct module *owner;
2105 owner = prot->owner;
2108 cgroup_sk_free(&sk->sk_cgrp_data);
2109 mem_cgroup_sk_free(sk);
2110 security_sk_free(sk);
2112 kmem_cache_free(slab, sk);
2119 * sk_alloc - All socket objects are allocated here
2120 * @net: the applicable net namespace
2121 * @family: protocol family
2122 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2123 * @prot: struct proto associated with this new sock instance
2124 * @kern: is this to be a kernel socket?
2126 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2127 struct proto *prot, int kern)
2131 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2133 sk->sk_family = family;
2135 * See comment in struct sock definition to understand
2136 * why we need sk_prot_creator -acme
2138 sk->sk_prot = sk->sk_prot_creator = prot;
2139 sk->sk_kern_sock = kern;
2141 sk->sk_net_refcnt = kern ? 0 : 1;
2142 if (likely(sk->sk_net_refcnt)) {
2143 get_net_track(net, &sk->ns_tracker, priority);
2144 sock_inuse_add(net, 1);
2146 __netns_tracker_alloc(net, &sk->ns_tracker,
2150 sock_net_set(sk, net);
2151 refcount_set(&sk->sk_wmem_alloc, 1);
2153 mem_cgroup_sk_alloc(sk);
2154 cgroup_sk_alloc(&sk->sk_cgrp_data);
2155 sock_update_classid(&sk->sk_cgrp_data);
2156 sock_update_netprioidx(&sk->sk_cgrp_data);
2157 sk_tx_queue_clear(sk);
2162 EXPORT_SYMBOL(sk_alloc);
2164 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2165 * grace period. This is the case for UDP sockets and TCP listeners.
2167 static void __sk_destruct(struct rcu_head *head)
2169 struct sock *sk = container_of(head, struct sock, sk_rcu);
2170 struct sk_filter *filter;
2172 if (sk->sk_destruct)
2173 sk->sk_destruct(sk);
2175 filter = rcu_dereference_check(sk->sk_filter,
2176 refcount_read(&sk->sk_wmem_alloc) == 0);
2178 sk_filter_uncharge(sk, filter);
2179 RCU_INIT_POINTER(sk->sk_filter, NULL);
2182 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2184 #ifdef CONFIG_BPF_SYSCALL
2185 bpf_sk_storage_free(sk);
2188 if (atomic_read(&sk->sk_omem_alloc))
2189 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2190 __func__, atomic_read(&sk->sk_omem_alloc));
2192 if (sk->sk_frag.page) {
2193 put_page(sk->sk_frag.page);
2194 sk->sk_frag.page = NULL;
2197 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2198 put_cred(sk->sk_peer_cred);
2199 put_pid(sk->sk_peer_pid);
2201 if (likely(sk->sk_net_refcnt))
2202 put_net_track(sock_net(sk), &sk->ns_tracker);
2204 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2206 sk_prot_free(sk->sk_prot_creator, sk);
2209 void sk_destruct(struct sock *sk)
2211 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2213 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2214 reuseport_detach_sock(sk);
2215 use_call_rcu = true;
2219 call_rcu(&sk->sk_rcu, __sk_destruct);
2221 __sk_destruct(&sk->sk_rcu);
2224 static void __sk_free(struct sock *sk)
2226 if (likely(sk->sk_net_refcnt))
2227 sock_inuse_add(sock_net(sk), -1);
2229 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2230 sock_diag_broadcast_destroy(sk);
2235 void sk_free(struct sock *sk)
2238 * We subtract one from sk_wmem_alloc and can know if
2239 * some packets are still in some tx queue.
2240 * If not null, sock_wfree() will call __sk_free(sk) later
2242 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2245 EXPORT_SYMBOL(sk_free);
2247 static void sk_init_common(struct sock *sk)
2249 skb_queue_head_init(&sk->sk_receive_queue);
2250 skb_queue_head_init(&sk->sk_write_queue);
2251 skb_queue_head_init(&sk->sk_error_queue);
2253 rwlock_init(&sk->sk_callback_lock);
2254 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2255 af_rlock_keys + sk->sk_family,
2256 af_family_rlock_key_strings[sk->sk_family]);
2257 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2258 af_wlock_keys + sk->sk_family,
2259 af_family_wlock_key_strings[sk->sk_family]);
2260 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2261 af_elock_keys + sk->sk_family,
2262 af_family_elock_key_strings[sk->sk_family]);
2263 lockdep_set_class_and_name(&sk->sk_callback_lock,
2264 af_callback_keys + sk->sk_family,
2265 af_family_clock_key_strings[sk->sk_family]);
2269 * sk_clone_lock - clone a socket, and lock its clone
2270 * @sk: the socket to clone
2271 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2273 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2275 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2277 struct proto *prot = READ_ONCE(sk->sk_prot);
2278 struct sk_filter *filter;
2279 bool is_charged = true;
2282 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2286 sock_copy(newsk, sk);
2288 newsk->sk_prot_creator = prot;
2291 if (likely(newsk->sk_net_refcnt)) {
2292 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2293 sock_inuse_add(sock_net(newsk), 1);
2295 /* Kernel sockets are not elevating the struct net refcount.
2296 * Instead, use a tracker to more easily detect if a layer
2297 * is not properly dismantling its kernel sockets at netns
2300 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2303 sk_node_init(&newsk->sk_node);
2304 sock_lock_init(newsk);
2305 bh_lock_sock(newsk);
2306 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2307 newsk->sk_backlog.len = 0;
2309 atomic_set(&newsk->sk_rmem_alloc, 0);
2311 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2312 refcount_set(&newsk->sk_wmem_alloc, 1);
2314 atomic_set(&newsk->sk_omem_alloc, 0);
2315 sk_init_common(newsk);
2317 newsk->sk_dst_cache = NULL;
2318 newsk->sk_dst_pending_confirm = 0;
2319 newsk->sk_wmem_queued = 0;
2320 newsk->sk_forward_alloc = 0;
2321 newsk->sk_reserved_mem = 0;
2322 atomic_set(&newsk->sk_drops, 0);
2323 newsk->sk_send_head = NULL;
2324 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2325 atomic_set(&newsk->sk_zckey, 0);
2327 sock_reset_flag(newsk, SOCK_DONE);
2329 /* sk->sk_memcg will be populated at accept() time */
2330 newsk->sk_memcg = NULL;
2332 cgroup_sk_clone(&newsk->sk_cgrp_data);
2335 filter = rcu_dereference(sk->sk_filter);
2337 /* though it's an empty new sock, the charging may fail
2338 * if sysctl_optmem_max was changed between creation of
2339 * original socket and cloning
2341 is_charged = sk_filter_charge(newsk, filter);
2342 RCU_INIT_POINTER(newsk->sk_filter, filter);
2345 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2346 /* We need to make sure that we don't uncharge the new
2347 * socket if we couldn't charge it in the first place
2348 * as otherwise we uncharge the parent's filter.
2351 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2352 sk_free_unlock_clone(newsk);
2356 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2358 if (bpf_sk_storage_clone(sk, newsk)) {
2359 sk_free_unlock_clone(newsk);
2364 /* Clear sk_user_data if parent had the pointer tagged
2365 * as not suitable for copying when cloning.
2367 if (sk_user_data_is_nocopy(newsk))
2368 newsk->sk_user_data = NULL;
2371 newsk->sk_err_soft = 0;
2372 newsk->sk_priority = 0;
2373 newsk->sk_incoming_cpu = raw_smp_processor_id();
2375 /* Before updating sk_refcnt, we must commit prior changes to memory
2376 * (Documentation/RCU/rculist_nulls.rst for details)
2379 refcount_set(&newsk->sk_refcnt, 2);
2381 sk_set_socket(newsk, NULL);
2382 sk_tx_queue_clear(newsk);
2383 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2385 if (newsk->sk_prot->sockets_allocated)
2386 sk_sockets_allocated_inc(newsk);
2388 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2389 net_enable_timestamp();
2393 EXPORT_SYMBOL_GPL(sk_clone_lock);
2395 void sk_free_unlock_clone(struct sock *sk)
2397 /* It is still raw copy of parent, so invalidate
2398 * destructor and make plain sk_free() */
2399 sk->sk_destruct = NULL;
2403 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2405 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2407 bool is_ipv6 = false;
2410 #if IS_ENABLED(CONFIG_IPV6)
2411 is_ipv6 = (sk->sk_family == AF_INET6 &&
2412 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2414 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2415 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2416 READ_ONCE(dst->dev->gso_ipv4_max_size);
2417 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2418 max_size = GSO_LEGACY_MAX_SIZE;
2420 return max_size - (MAX_TCP_HEADER + 1);
2423 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2427 sk->sk_route_caps = dst->dev->features;
2429 sk->sk_route_caps |= NETIF_F_GSO;
2430 if (sk->sk_route_caps & NETIF_F_GSO)
2431 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2432 if (unlikely(sk->sk_gso_disabled))
2433 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2434 if (sk_can_gso(sk)) {
2435 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2436 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2438 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2439 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2440 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2441 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2444 sk->sk_gso_max_segs = max_segs;
2445 sk_dst_set(sk, dst);
2447 EXPORT_SYMBOL_GPL(sk_setup_caps);
2450 * Simple resource managers for sockets.
2455 * Write buffer destructor automatically called from kfree_skb.
2457 void sock_wfree(struct sk_buff *skb)
2459 struct sock *sk = skb->sk;
2460 unsigned int len = skb->truesize;
2463 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2464 if (sock_flag(sk, SOCK_RCU_FREE) &&
2465 sk->sk_write_space == sock_def_write_space) {
2467 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2468 sock_def_write_space_wfree(sk);
2476 * Keep a reference on sk_wmem_alloc, this will be released
2477 * after sk_write_space() call
2479 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2480 sk->sk_write_space(sk);
2484 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2485 * could not do because of in-flight packets
2487 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2490 EXPORT_SYMBOL(sock_wfree);
2492 /* This variant of sock_wfree() is used by TCP,
2493 * since it sets SOCK_USE_WRITE_QUEUE.
2495 void __sock_wfree(struct sk_buff *skb)
2497 struct sock *sk = skb->sk;
2499 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2503 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2508 if (unlikely(!sk_fullsock(sk))) {
2509 skb->destructor = sock_edemux;
2514 skb->destructor = sock_wfree;
2515 skb_set_hash_from_sk(skb, sk);
2517 * We used to take a refcount on sk, but following operation
2518 * is enough to guarantee sk_free() wont free this sock until
2519 * all in-flight packets are completed
2521 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2523 EXPORT_SYMBOL(skb_set_owner_w);
2525 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2527 #ifdef CONFIG_TLS_DEVICE
2528 /* Drivers depend on in-order delivery for crypto offload,
2529 * partial orphan breaks out-of-order-OK logic.
2534 return (skb->destructor == sock_wfree ||
2535 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2538 /* This helper is used by netem, as it can hold packets in its
2539 * delay queue. We want to allow the owner socket to send more
2540 * packets, as if they were already TX completed by a typical driver.
2541 * But we also want to keep skb->sk set because some packet schedulers
2542 * rely on it (sch_fq for example).
2544 void skb_orphan_partial(struct sk_buff *skb)
2546 if (skb_is_tcp_pure_ack(skb))
2549 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2554 EXPORT_SYMBOL(skb_orphan_partial);
2557 * Read buffer destructor automatically called from kfree_skb.
2559 void sock_rfree(struct sk_buff *skb)
2561 struct sock *sk = skb->sk;
2562 unsigned int len = skb->truesize;
2564 atomic_sub(len, &sk->sk_rmem_alloc);
2565 sk_mem_uncharge(sk, len);
2567 EXPORT_SYMBOL(sock_rfree);
2570 * Buffer destructor for skbs that are not used directly in read or write
2571 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2573 void sock_efree(struct sk_buff *skb)
2577 EXPORT_SYMBOL(sock_efree);
2579 /* Buffer destructor for prefetch/receive path where reference count may
2580 * not be held, e.g. for listen sockets.
2583 void sock_pfree(struct sk_buff *skb)
2585 struct sock *sk = skb->sk;
2587 if (!sk_is_refcounted(sk))
2590 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2591 inet_reqsk(sk)->rsk_listener = NULL;
2592 reqsk_free(inet_reqsk(sk));
2598 EXPORT_SYMBOL(sock_pfree);
2599 #endif /* CONFIG_INET */
2601 kuid_t sock_i_uid(struct sock *sk)
2605 read_lock_bh(&sk->sk_callback_lock);
2606 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2607 read_unlock_bh(&sk->sk_callback_lock);
2610 EXPORT_SYMBOL(sock_i_uid);
2612 unsigned long __sock_i_ino(struct sock *sk)
2616 read_lock(&sk->sk_callback_lock);
2617 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2618 read_unlock(&sk->sk_callback_lock);
2621 EXPORT_SYMBOL(__sock_i_ino);
2623 unsigned long sock_i_ino(struct sock *sk)
2628 ino = __sock_i_ino(sk);
2632 EXPORT_SYMBOL(sock_i_ino);
2635 * Allocate a skb from the socket's send buffer.
2637 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2641 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2642 struct sk_buff *skb = alloc_skb(size, priority);
2645 skb_set_owner_w(skb, sk);
2651 EXPORT_SYMBOL(sock_wmalloc);
2653 static void sock_ofree(struct sk_buff *skb)
2655 struct sock *sk = skb->sk;
2657 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2660 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2663 struct sk_buff *skb;
2665 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2666 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2667 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2670 skb = alloc_skb(size, priority);
2674 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2676 skb->destructor = sock_ofree;
2681 * Allocate a memory block from the socket's option memory buffer.
2683 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2685 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2687 if ((unsigned int)size <= optmem_max &&
2688 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2690 /* First do the add, to avoid the race if kmalloc
2693 atomic_add(size, &sk->sk_omem_alloc);
2694 mem = kmalloc(size, priority);
2697 atomic_sub(size, &sk->sk_omem_alloc);
2701 EXPORT_SYMBOL(sock_kmalloc);
2703 /* Free an option memory block. Note, we actually want the inline
2704 * here as this allows gcc to detect the nullify and fold away the
2705 * condition entirely.
2707 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2710 if (WARN_ON_ONCE(!mem))
2713 kfree_sensitive(mem);
2716 atomic_sub(size, &sk->sk_omem_alloc);
2719 void sock_kfree_s(struct sock *sk, void *mem, int size)
2721 __sock_kfree_s(sk, mem, size, false);
2723 EXPORT_SYMBOL(sock_kfree_s);
2725 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2727 __sock_kfree_s(sk, mem, size, true);
2729 EXPORT_SYMBOL(sock_kzfree_s);
2731 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2732 I think, these locks should be removed for datagram sockets.
2734 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2738 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2742 if (signal_pending(current))
2744 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2745 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2746 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2748 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2750 if (READ_ONCE(sk->sk_err))
2752 timeo = schedule_timeout(timeo);
2754 finish_wait(sk_sleep(sk), &wait);
2760 * Generic send/receive buffer handlers
2763 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2764 unsigned long data_len, int noblock,
2765 int *errcode, int max_page_order)
2767 struct sk_buff *skb;
2771 timeo = sock_sndtimeo(sk, noblock);
2773 err = sock_error(sk);
2778 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2781 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2784 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2785 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2789 if (signal_pending(current))
2791 timeo = sock_wait_for_wmem(sk, timeo);
2793 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2794 errcode, sk->sk_allocation);
2796 skb_set_owner_w(skb, sk);
2800 err = sock_intr_errno(timeo);
2805 EXPORT_SYMBOL(sock_alloc_send_pskb);
2807 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2808 struct sockcm_cookie *sockc)
2812 switch (cmsg->cmsg_type) {
2814 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2815 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2817 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2819 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2821 case SO_TIMESTAMPING_OLD:
2822 case SO_TIMESTAMPING_NEW:
2823 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2826 tsflags = *(u32 *)CMSG_DATA(cmsg);
2827 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2830 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2831 sockc->tsflags |= tsflags;
2834 if (!sock_flag(sk, SOCK_TXTIME))
2836 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2838 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2840 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2842 case SCM_CREDENTIALS:
2849 EXPORT_SYMBOL(__sock_cmsg_send);
2851 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2852 struct sockcm_cookie *sockc)
2854 struct cmsghdr *cmsg;
2857 for_each_cmsghdr(cmsg, msg) {
2858 if (!CMSG_OK(msg, cmsg))
2860 if (cmsg->cmsg_level != SOL_SOCKET)
2862 ret = __sock_cmsg_send(sk, cmsg, sockc);
2868 EXPORT_SYMBOL(sock_cmsg_send);
2870 static void sk_enter_memory_pressure(struct sock *sk)
2872 if (!sk->sk_prot->enter_memory_pressure)
2875 sk->sk_prot->enter_memory_pressure(sk);
2878 static void sk_leave_memory_pressure(struct sock *sk)
2880 if (sk->sk_prot->leave_memory_pressure) {
2881 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2882 tcp_leave_memory_pressure, sk);
2884 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2886 if (memory_pressure && READ_ONCE(*memory_pressure))
2887 WRITE_ONCE(*memory_pressure, 0);
2891 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2894 * skb_page_frag_refill - check that a page_frag contains enough room
2895 * @sz: minimum size of the fragment we want to get
2896 * @pfrag: pointer to page_frag
2897 * @gfp: priority for memory allocation
2899 * Note: While this allocator tries to use high order pages, there is
2900 * no guarantee that allocations succeed. Therefore, @sz MUST be
2901 * less or equal than PAGE_SIZE.
2903 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2906 if (page_ref_count(pfrag->page) == 1) {
2910 if (pfrag->offset + sz <= pfrag->size)
2912 put_page(pfrag->page);
2916 if (SKB_FRAG_PAGE_ORDER &&
2917 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2918 /* Avoid direct reclaim but allow kswapd to wake */
2919 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2920 __GFP_COMP | __GFP_NOWARN |
2922 SKB_FRAG_PAGE_ORDER);
2923 if (likely(pfrag->page)) {
2924 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2928 pfrag->page = alloc_page(gfp);
2929 if (likely(pfrag->page)) {
2930 pfrag->size = PAGE_SIZE;
2935 EXPORT_SYMBOL(skb_page_frag_refill);
2937 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2939 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2942 sk_enter_memory_pressure(sk);
2943 sk_stream_moderate_sndbuf(sk);
2946 EXPORT_SYMBOL(sk_page_frag_refill);
2948 void __lock_sock(struct sock *sk)
2949 __releases(&sk->sk_lock.slock)
2950 __acquires(&sk->sk_lock.slock)
2955 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2956 TASK_UNINTERRUPTIBLE);
2957 spin_unlock_bh(&sk->sk_lock.slock);
2959 spin_lock_bh(&sk->sk_lock.slock);
2960 if (!sock_owned_by_user(sk))
2963 finish_wait(&sk->sk_lock.wq, &wait);
2966 void __release_sock(struct sock *sk)
2967 __releases(&sk->sk_lock.slock)
2968 __acquires(&sk->sk_lock.slock)
2970 struct sk_buff *skb, *next;
2972 while ((skb = sk->sk_backlog.head) != NULL) {
2973 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2975 spin_unlock_bh(&sk->sk_lock.slock);
2980 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2981 skb_mark_not_on_list(skb);
2982 sk_backlog_rcv(sk, skb);
2987 } while (skb != NULL);
2989 spin_lock_bh(&sk->sk_lock.slock);
2993 * Doing the zeroing here guarantee we can not loop forever
2994 * while a wild producer attempts to flood us.
2996 sk->sk_backlog.len = 0;
2999 void __sk_flush_backlog(struct sock *sk)
3001 spin_lock_bh(&sk->sk_lock.slock);
3004 if (sk->sk_prot->release_cb)
3005 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3006 tcp_release_cb, sk);
3008 spin_unlock_bh(&sk->sk_lock.slock);
3010 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3013 * sk_wait_data - wait for data to arrive at sk_receive_queue
3014 * @sk: sock to wait on
3015 * @timeo: for how long
3016 * @skb: last skb seen on sk_receive_queue
3018 * Now socket state including sk->sk_err is changed only under lock,
3019 * hence we may omit checks after joining wait queue.
3020 * We check receive queue before schedule() only as optimization;
3021 * it is very likely that release_sock() added new data.
3023 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3025 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3028 add_wait_queue(sk_sleep(sk), &wait);
3029 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3030 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3031 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3032 remove_wait_queue(sk_sleep(sk), &wait);
3035 EXPORT_SYMBOL(sk_wait_data);
3038 * __sk_mem_raise_allocated - increase memory_allocated
3040 * @size: memory size to allocate
3041 * @amt: pages to allocate
3042 * @kind: allocation type
3044 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3046 * Unlike the globally shared limits among the sockets under same protocol,
3047 * consuming the budget of a memcg won't have direct effect on other ones.
3048 * So be optimistic about memcg's tolerance, and leave the callers to decide
3049 * whether or not to raise allocated through sk_under_memory_pressure() or
3052 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3054 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3055 struct proto *prot = sk->sk_prot;
3056 bool charged = false;
3059 sk_memory_allocated_add(sk, amt);
3060 allocated = sk_memory_allocated(sk);
3063 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3064 goto suppress_allocation;
3069 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3070 sk_leave_memory_pressure(sk);
3074 /* Under pressure. */
3075 if (allocated > sk_prot_mem_limits(sk, 1))
3076 sk_enter_memory_pressure(sk);
3078 /* Over hard limit. */
3079 if (allocated > sk_prot_mem_limits(sk, 2))
3080 goto suppress_allocation;
3082 /* Guarantee minimum buffer size under pressure (either global
3083 * or memcg) to make sure features described in RFC 7323 (TCP
3084 * Extensions for High Performance) work properly.
3086 * This rule does NOT stand when exceeds global or memcg's hard
3087 * limit, or else a DoS attack can be taken place by spawning
3088 * lots of sockets whose usage are under minimum buffer size.
3090 if (kind == SK_MEM_RECV) {
3091 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3094 } else { /* SK_MEM_SEND */
3095 int wmem0 = sk_get_wmem0(sk, prot);
3097 if (sk->sk_type == SOCK_STREAM) {
3098 if (sk->sk_wmem_queued < wmem0)
3100 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3105 if (sk_has_memory_pressure(sk)) {
3108 /* The following 'average' heuristic is within the
3109 * scope of global accounting, so it only makes
3110 * sense for global memory pressure.
3112 if (!sk_under_global_memory_pressure(sk))
3115 /* Try to be fair among all the sockets under global
3116 * pressure by allowing the ones that below average
3119 alloc = sk_sockets_allocated_read_positive(sk);
3120 if (sk_prot_mem_limits(sk, 2) > alloc *
3121 sk_mem_pages(sk->sk_wmem_queued +
3122 atomic_read(&sk->sk_rmem_alloc) +
3123 sk->sk_forward_alloc))
3127 suppress_allocation:
3129 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3130 sk_stream_moderate_sndbuf(sk);
3132 /* Fail only if socket is _under_ its sndbuf.
3133 * In this case we cannot block, so that we have to fail.
3135 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3136 /* Force charge with __GFP_NOFAIL */
3137 if (memcg && !charged) {
3138 mem_cgroup_charge_skmem(memcg, amt,
3139 gfp_memcg_charge() | __GFP_NOFAIL);
3145 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3146 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3148 sk_memory_allocated_sub(sk, amt);
3151 mem_cgroup_uncharge_skmem(memcg, amt);
3157 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3159 * @size: memory size to allocate
3160 * @kind: allocation type
3162 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3163 * rmem allocation. This function assumes that protocols which have
3164 * memory_pressure use sk_wmem_queued as write buffer accounting.
3166 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3168 int ret, amt = sk_mem_pages(size);
3170 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3171 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3173 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3176 EXPORT_SYMBOL(__sk_mem_schedule);
3179 * __sk_mem_reduce_allocated - reclaim memory_allocated
3181 * @amount: number of quanta
3183 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3185 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3187 sk_memory_allocated_sub(sk, amount);
3189 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3190 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3192 if (sk_under_global_memory_pressure(sk) &&
3193 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3194 sk_leave_memory_pressure(sk);
3198 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3200 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3202 void __sk_mem_reclaim(struct sock *sk, int amount)
3204 amount >>= PAGE_SHIFT;
3205 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3206 __sk_mem_reduce_allocated(sk, amount);
3208 EXPORT_SYMBOL(__sk_mem_reclaim);
3210 int sk_set_peek_off(struct sock *sk, int val)
3212 WRITE_ONCE(sk->sk_peek_off, val);
3215 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3218 * Set of default routines for initialising struct proto_ops when
3219 * the protocol does not support a particular function. In certain
3220 * cases where it makes no sense for a protocol to have a "do nothing"
3221 * function, some default processing is provided.
3224 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3228 EXPORT_SYMBOL(sock_no_bind);
3230 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3235 EXPORT_SYMBOL(sock_no_connect);
3237 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3241 EXPORT_SYMBOL(sock_no_socketpair);
3243 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3248 EXPORT_SYMBOL(sock_no_accept);
3250 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3255 EXPORT_SYMBOL(sock_no_getname);
3257 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3261 EXPORT_SYMBOL(sock_no_ioctl);
3263 int sock_no_listen(struct socket *sock, int backlog)
3267 EXPORT_SYMBOL(sock_no_listen);
3269 int sock_no_shutdown(struct socket *sock, int how)
3273 EXPORT_SYMBOL(sock_no_shutdown);
3275 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3279 EXPORT_SYMBOL(sock_no_sendmsg);
3281 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3285 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3287 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3292 EXPORT_SYMBOL(sock_no_recvmsg);
3294 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3296 /* Mirror missing mmap method error code */
3299 EXPORT_SYMBOL(sock_no_mmap);
3302 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3303 * various sock-based usage counts.
3305 void __receive_sock(struct file *file)
3307 struct socket *sock;
3309 sock = sock_from_file(file);
3311 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3312 sock_update_classid(&sock->sk->sk_cgrp_data);
3317 * Default Socket Callbacks
3320 static void sock_def_wakeup(struct sock *sk)
3322 struct socket_wq *wq;
3325 wq = rcu_dereference(sk->sk_wq);
3326 if (skwq_has_sleeper(wq))
3327 wake_up_interruptible_all(&wq->wait);
3331 static void sock_def_error_report(struct sock *sk)
3333 struct socket_wq *wq;
3336 wq = rcu_dereference(sk->sk_wq);
3337 if (skwq_has_sleeper(wq))
3338 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3339 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3343 void sock_def_readable(struct sock *sk)
3345 struct socket_wq *wq;
3347 trace_sk_data_ready(sk);
3350 wq = rcu_dereference(sk->sk_wq);
3351 if (skwq_has_sleeper(wq))
3352 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3353 EPOLLRDNORM | EPOLLRDBAND);
3354 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3358 static void sock_def_write_space(struct sock *sk)
3360 struct socket_wq *wq;
3364 /* Do not wake up a writer until he can make "significant"
3367 if (sock_writeable(sk)) {
3368 wq = rcu_dereference(sk->sk_wq);
3369 if (skwq_has_sleeper(wq))
3370 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3371 EPOLLWRNORM | EPOLLWRBAND);
3373 /* Should agree with poll, otherwise some programs break */
3374 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3380 /* An optimised version of sock_def_write_space(), should only be called
3381 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3384 static void sock_def_write_space_wfree(struct sock *sk)
3386 /* Do not wake up a writer until he can make "significant"
3389 if (sock_writeable(sk)) {
3390 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3392 /* rely on refcount_sub from sock_wfree() */
3393 smp_mb__after_atomic();
3394 if (wq && waitqueue_active(&wq->wait))
3395 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3396 EPOLLWRNORM | EPOLLWRBAND);
3398 /* Should agree with poll, otherwise some programs break */
3399 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3403 static void sock_def_destruct(struct sock *sk)
3407 void sk_send_sigurg(struct sock *sk)
3409 if (sk->sk_socket && sk->sk_socket->file)
3410 if (send_sigurg(&sk->sk_socket->file->f_owner))
3411 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3413 EXPORT_SYMBOL(sk_send_sigurg);
3415 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3416 unsigned long expires)
3418 if (!mod_timer(timer, expires))
3421 EXPORT_SYMBOL(sk_reset_timer);
3423 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3425 if (del_timer(timer))
3428 EXPORT_SYMBOL(sk_stop_timer);
3430 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3432 if (del_timer_sync(timer))
3435 EXPORT_SYMBOL(sk_stop_timer_sync);
3437 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3440 sk->sk_send_head = NULL;
3442 timer_setup(&sk->sk_timer, NULL, 0);
3444 sk->sk_allocation = GFP_KERNEL;
3445 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3446 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3447 sk->sk_state = TCP_CLOSE;
3448 sk->sk_use_task_frag = true;
3449 sk_set_socket(sk, sock);
3451 sock_set_flag(sk, SOCK_ZAPPED);
3454 sk->sk_type = sock->type;
3455 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3458 RCU_INIT_POINTER(sk->sk_wq, NULL);
3462 rwlock_init(&sk->sk_callback_lock);
3463 if (sk->sk_kern_sock)
3464 lockdep_set_class_and_name(
3465 &sk->sk_callback_lock,
3466 af_kern_callback_keys + sk->sk_family,
3467 af_family_kern_clock_key_strings[sk->sk_family]);
3469 lockdep_set_class_and_name(
3470 &sk->sk_callback_lock,
3471 af_callback_keys + sk->sk_family,
3472 af_family_clock_key_strings[sk->sk_family]);
3474 sk->sk_state_change = sock_def_wakeup;
3475 sk->sk_data_ready = sock_def_readable;
3476 sk->sk_write_space = sock_def_write_space;
3477 sk->sk_error_report = sock_def_error_report;
3478 sk->sk_destruct = sock_def_destruct;
3480 sk->sk_frag.page = NULL;
3481 sk->sk_frag.offset = 0;
3482 sk->sk_peek_off = -1;
3484 sk->sk_peer_pid = NULL;
3485 sk->sk_peer_cred = NULL;
3486 spin_lock_init(&sk->sk_peer_lock);
3488 sk->sk_write_pending = 0;
3489 sk->sk_rcvlowat = 1;
3490 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3491 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3493 sk->sk_stamp = SK_DEFAULT_STAMP;
3494 #if BITS_PER_LONG==32
3495 seqlock_init(&sk->sk_stamp_seq);
3497 atomic_set(&sk->sk_zckey, 0);
3499 #ifdef CONFIG_NET_RX_BUSY_POLL
3501 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3504 sk->sk_max_pacing_rate = ~0UL;
3505 sk->sk_pacing_rate = ~0UL;
3506 WRITE_ONCE(sk->sk_pacing_shift, 10);
3507 sk->sk_incoming_cpu = -1;
3509 sk_rx_queue_clear(sk);
3511 * Before updating sk_refcnt, we must commit prior changes to memory
3512 * (Documentation/RCU/rculist_nulls.rst for details)
3515 refcount_set(&sk->sk_refcnt, 1);
3516 atomic_set(&sk->sk_drops, 0);
3518 EXPORT_SYMBOL(sock_init_data_uid);
3520 void sock_init_data(struct socket *sock, struct sock *sk)
3523 SOCK_INODE(sock)->i_uid :
3524 make_kuid(sock_net(sk)->user_ns, 0);
3526 sock_init_data_uid(sock, sk, uid);
3528 EXPORT_SYMBOL(sock_init_data);
3530 void lock_sock_nested(struct sock *sk, int subclass)
3532 /* The sk_lock has mutex_lock() semantics here. */
3533 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3536 spin_lock_bh(&sk->sk_lock.slock);
3537 if (sock_owned_by_user_nocheck(sk))
3539 sk->sk_lock.owned = 1;
3540 spin_unlock_bh(&sk->sk_lock.slock);
3542 EXPORT_SYMBOL(lock_sock_nested);
3544 void release_sock(struct sock *sk)
3546 spin_lock_bh(&sk->sk_lock.slock);
3547 if (sk->sk_backlog.tail)
3550 if (sk->sk_prot->release_cb)
3551 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3552 tcp_release_cb, sk);
3554 sock_release_ownership(sk);
3555 if (waitqueue_active(&sk->sk_lock.wq))
3556 wake_up(&sk->sk_lock.wq);
3557 spin_unlock_bh(&sk->sk_lock.slock);
3559 EXPORT_SYMBOL(release_sock);
3561 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3564 spin_lock_bh(&sk->sk_lock.slock);
3566 if (!sock_owned_by_user_nocheck(sk)) {
3568 * Fast path return with bottom halves disabled and
3569 * sock::sk_lock.slock held.
3571 * The 'mutex' is not contended and holding
3572 * sock::sk_lock.slock prevents all other lockers to
3573 * proceed so the corresponding unlock_sock_fast() can
3574 * avoid the slow path of release_sock() completely and
3575 * just release slock.
3577 * From a semantical POV this is equivalent to 'acquiring'
3578 * the 'mutex', hence the corresponding lockdep
3579 * mutex_release() has to happen in the fast path of
3580 * unlock_sock_fast().
3586 sk->sk_lock.owned = 1;
3587 __acquire(&sk->sk_lock.slock);
3588 spin_unlock_bh(&sk->sk_lock.slock);
3591 EXPORT_SYMBOL(__lock_sock_fast);
3593 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3594 bool timeval, bool time32)
3596 struct sock *sk = sock->sk;
3597 struct timespec64 ts;
3599 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3600 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3601 if (ts.tv_sec == -1)
3603 if (ts.tv_sec == 0) {
3604 ktime_t kt = ktime_get_real();
3605 sock_write_timestamp(sk, kt);
3606 ts = ktime_to_timespec64(kt);
3612 #ifdef CONFIG_COMPAT_32BIT_TIME
3614 return put_old_timespec32(&ts, userstamp);
3616 #ifdef CONFIG_SPARC64
3617 /* beware of padding in sparc64 timeval */
3618 if (timeval && !in_compat_syscall()) {
3619 struct __kernel_old_timeval __user tv = {
3620 .tv_sec = ts.tv_sec,
3621 .tv_usec = ts.tv_nsec,
3623 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3628 return put_timespec64(&ts, userstamp);
3630 EXPORT_SYMBOL(sock_gettstamp);
3632 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3634 if (!sock_flag(sk, flag)) {
3635 unsigned long previous_flags = sk->sk_flags;
3637 sock_set_flag(sk, flag);
3639 * we just set one of the two flags which require net
3640 * time stamping, but time stamping might have been on
3641 * already because of the other one
3643 if (sock_needs_netstamp(sk) &&
3644 !(previous_flags & SK_FLAGS_TIMESTAMP))
3645 net_enable_timestamp();
3649 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3650 int level, int type)
3652 struct sock_exterr_skb *serr;
3653 struct sk_buff *skb;
3657 skb = sock_dequeue_err_skb(sk);
3663 msg->msg_flags |= MSG_TRUNC;
3666 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3670 sock_recv_timestamp(msg, sk, skb);
3672 serr = SKB_EXT_ERR(skb);
3673 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3675 msg->msg_flags |= MSG_ERRQUEUE;
3683 EXPORT_SYMBOL(sock_recv_errqueue);
3686 * Get a socket option on an socket.
3688 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3689 * asynchronous errors should be reported by getsockopt. We assume
3690 * this means if you specify SO_ERROR (otherwise whats the point of it).
3692 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3693 char __user *optval, int __user *optlen)
3695 struct sock *sk = sock->sk;
3697 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3698 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3700 EXPORT_SYMBOL(sock_common_getsockopt);
3702 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3705 struct sock *sk = sock->sk;
3709 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3711 msg->msg_namelen = addr_len;
3714 EXPORT_SYMBOL(sock_common_recvmsg);
3717 * Set socket options on an inet socket.
3719 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3720 sockptr_t optval, unsigned int optlen)
3722 struct sock *sk = sock->sk;
3724 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3725 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3727 EXPORT_SYMBOL(sock_common_setsockopt);
3729 void sk_common_release(struct sock *sk)
3731 if (sk->sk_prot->destroy)
3732 sk->sk_prot->destroy(sk);
3735 * Observation: when sk_common_release is called, processes have
3736 * no access to socket. But net still has.
3737 * Step one, detach it from networking:
3739 * A. Remove from hash tables.
3742 sk->sk_prot->unhash(sk);
3745 * In this point socket cannot receive new packets, but it is possible
3746 * that some packets are in flight because some CPU runs receiver and
3747 * did hash table lookup before we unhashed socket. They will achieve
3748 * receive queue and will be purged by socket destructor.
3750 * Also we still have packets pending on receive queue and probably,
3751 * our own packets waiting in device queues. sock_destroy will drain
3752 * receive queue, but transmitted packets will delay socket destruction
3753 * until the last reference will be released.
3758 xfrm_sk_free_policy(sk);
3762 EXPORT_SYMBOL(sk_common_release);
3764 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3766 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3768 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3769 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3770 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3771 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3772 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3773 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3774 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3775 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3776 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3779 #ifdef CONFIG_PROC_FS
3780 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3782 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3784 int cpu, idx = prot->inuse_idx;
3787 for_each_possible_cpu(cpu)
3788 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3790 return res >= 0 ? res : 0;
3792 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3794 int sock_inuse_get(struct net *net)
3798 for_each_possible_cpu(cpu)
3799 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3804 EXPORT_SYMBOL_GPL(sock_inuse_get);
3806 static int __net_init sock_inuse_init_net(struct net *net)
3808 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3809 if (net->core.prot_inuse == NULL)
3814 static void __net_exit sock_inuse_exit_net(struct net *net)
3816 free_percpu(net->core.prot_inuse);
3819 static struct pernet_operations net_inuse_ops = {
3820 .init = sock_inuse_init_net,
3821 .exit = sock_inuse_exit_net,
3824 static __init int net_inuse_init(void)
3826 if (register_pernet_subsys(&net_inuse_ops))
3827 panic("Cannot initialize net inuse counters");
3832 core_initcall(net_inuse_init);
3834 static int assign_proto_idx(struct proto *prot)
3836 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3838 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3839 pr_err("PROTO_INUSE_NR exhausted\n");
3843 set_bit(prot->inuse_idx, proto_inuse_idx);
3847 static void release_proto_idx(struct proto *prot)
3849 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3850 clear_bit(prot->inuse_idx, proto_inuse_idx);
3853 static inline int assign_proto_idx(struct proto *prot)
3858 static inline void release_proto_idx(struct proto *prot)
3864 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3868 kfree(twsk_prot->twsk_slab_name);
3869 twsk_prot->twsk_slab_name = NULL;
3870 kmem_cache_destroy(twsk_prot->twsk_slab);
3871 twsk_prot->twsk_slab = NULL;
3874 static int tw_prot_init(const struct proto *prot)
3876 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3881 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3883 if (!twsk_prot->twsk_slab_name)
3886 twsk_prot->twsk_slab =
3887 kmem_cache_create(twsk_prot->twsk_slab_name,
3888 twsk_prot->twsk_obj_size, 0,
3889 SLAB_ACCOUNT | prot->slab_flags,
3891 if (!twsk_prot->twsk_slab) {
3892 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3900 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3904 kfree(rsk_prot->slab_name);
3905 rsk_prot->slab_name = NULL;
3906 kmem_cache_destroy(rsk_prot->slab);
3907 rsk_prot->slab = NULL;
3910 static int req_prot_init(const struct proto *prot)
3912 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3917 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3919 if (!rsk_prot->slab_name)
3922 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3923 rsk_prot->obj_size, 0,
3924 SLAB_ACCOUNT | prot->slab_flags,
3927 if (!rsk_prot->slab) {
3928 pr_crit("%s: Can't create request sock SLAB cache!\n",
3935 int proto_register(struct proto *prot, int alloc_slab)
3939 if (prot->memory_allocated && !prot->sysctl_mem) {
3940 pr_err("%s: missing sysctl_mem\n", prot->name);
3943 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3944 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3948 prot->slab = kmem_cache_create_usercopy(prot->name,
3950 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3952 prot->useroffset, prot->usersize,
3955 if (prot->slab == NULL) {
3956 pr_crit("%s: Can't create sock SLAB cache!\n",
3961 if (req_prot_init(prot))
3962 goto out_free_request_sock_slab;
3964 if (tw_prot_init(prot))
3965 goto out_free_timewait_sock_slab;
3968 mutex_lock(&proto_list_mutex);
3969 ret = assign_proto_idx(prot);
3971 mutex_unlock(&proto_list_mutex);
3972 goto out_free_timewait_sock_slab;
3974 list_add(&prot->node, &proto_list);
3975 mutex_unlock(&proto_list_mutex);
3978 out_free_timewait_sock_slab:
3980 tw_prot_cleanup(prot->twsk_prot);
3981 out_free_request_sock_slab:
3983 req_prot_cleanup(prot->rsk_prot);
3985 kmem_cache_destroy(prot->slab);
3991 EXPORT_SYMBOL(proto_register);
3993 void proto_unregister(struct proto *prot)
3995 mutex_lock(&proto_list_mutex);
3996 release_proto_idx(prot);
3997 list_del(&prot->node);
3998 mutex_unlock(&proto_list_mutex);
4000 kmem_cache_destroy(prot->slab);
4003 req_prot_cleanup(prot->rsk_prot);
4004 tw_prot_cleanup(prot->twsk_prot);
4006 EXPORT_SYMBOL(proto_unregister);
4008 int sock_load_diag_module(int family, int protocol)
4011 if (!sock_is_registered(family))
4014 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4015 NETLINK_SOCK_DIAG, family);
4019 if (family == AF_INET &&
4020 protocol != IPPROTO_RAW &&
4021 protocol < MAX_INET_PROTOS &&
4022 !rcu_access_pointer(inet_protos[protocol]))
4026 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4027 NETLINK_SOCK_DIAG, family, protocol);
4029 EXPORT_SYMBOL(sock_load_diag_module);
4031 #ifdef CONFIG_PROC_FS
4032 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4033 __acquires(proto_list_mutex)
4035 mutex_lock(&proto_list_mutex);
4036 return seq_list_start_head(&proto_list, *pos);
4039 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4041 return seq_list_next(v, &proto_list, pos);
4044 static void proto_seq_stop(struct seq_file *seq, void *v)
4045 __releases(proto_list_mutex)
4047 mutex_unlock(&proto_list_mutex);
4050 static char proto_method_implemented(const void *method)
4052 return method == NULL ? 'n' : 'y';
4054 static long sock_prot_memory_allocated(struct proto *proto)
4056 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4059 static const char *sock_prot_memory_pressure(struct proto *proto)
4061 return proto->memory_pressure != NULL ?
4062 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4065 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4068 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4069 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4072 sock_prot_inuse_get(seq_file_net(seq), proto),
4073 sock_prot_memory_allocated(proto),
4074 sock_prot_memory_pressure(proto),
4076 proto->slab == NULL ? "no" : "yes",
4077 module_name(proto->owner),
4078 proto_method_implemented(proto->close),
4079 proto_method_implemented(proto->connect),
4080 proto_method_implemented(proto->disconnect),
4081 proto_method_implemented(proto->accept),
4082 proto_method_implemented(proto->ioctl),
4083 proto_method_implemented(proto->init),
4084 proto_method_implemented(proto->destroy),
4085 proto_method_implemented(proto->shutdown),
4086 proto_method_implemented(proto->setsockopt),
4087 proto_method_implemented(proto->getsockopt),
4088 proto_method_implemented(proto->sendmsg),
4089 proto_method_implemented(proto->recvmsg),
4090 proto_method_implemented(proto->bind),
4091 proto_method_implemented(proto->backlog_rcv),
4092 proto_method_implemented(proto->hash),
4093 proto_method_implemented(proto->unhash),
4094 proto_method_implemented(proto->get_port),
4095 proto_method_implemented(proto->enter_memory_pressure));
4098 static int proto_seq_show(struct seq_file *seq, void *v)
4100 if (v == &proto_list)
4101 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4110 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4112 proto_seq_printf(seq, list_entry(v, struct proto, node));
4116 static const struct seq_operations proto_seq_ops = {
4117 .start = proto_seq_start,
4118 .next = proto_seq_next,
4119 .stop = proto_seq_stop,
4120 .show = proto_seq_show,
4123 static __net_init int proto_init_net(struct net *net)
4125 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4126 sizeof(struct seq_net_private)))
4132 static __net_exit void proto_exit_net(struct net *net)
4134 remove_proc_entry("protocols", net->proc_net);
4138 static __net_initdata struct pernet_operations proto_net_ops = {
4139 .init = proto_init_net,
4140 .exit = proto_exit_net,
4143 static int __init proto_init(void)
4145 return register_pernet_subsys(&proto_net_ops);
4148 subsys_initcall(proto_init);
4150 #endif /* PROC_FS */
4152 #ifdef CONFIG_NET_RX_BUSY_POLL
4153 bool sk_busy_loop_end(void *p, unsigned long start_time)
4155 struct sock *sk = p;
4157 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4160 if (sk_is_udp(sk) &&
4161 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4164 return sk_busy_loop_timeout(sk, start_time);
4166 EXPORT_SYMBOL(sk_busy_loop_end);
4167 #endif /* CONFIG_NET_RX_BUSY_POLL */
4169 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4171 if (!sk->sk_prot->bind_add)
4173 return sk->sk_prot->bind_add(sk, addr, addr_len);
4175 EXPORT_SYMBOL(sock_bind_add);
4177 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4178 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4179 void __user *arg, void *karg, size_t size)
4183 if (copy_from_user(karg, arg, size))
4186 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4190 if (copy_to_user(arg, karg, size))
4195 EXPORT_SYMBOL(sock_ioctl_inout);
4197 /* This is the most common ioctl prep function, where the result (4 bytes) is
4198 * copied back to userspace if the ioctl() returns successfully. No input is
4199 * copied from userspace as input argument.
4201 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4205 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4209 return put_user(karg, (int __user *)arg);
4212 /* A wrapper around sock ioctls, which copies the data from userspace
4213 * (depending on the protocol/ioctl), and copies back the result to userspace.
4214 * The main motivation for this function is to pass kernel memory to the
4215 * protocol ioctl callbacks, instead of userspace memory.
4217 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4221 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4222 rc = ipmr_sk_ioctl(sk, cmd, arg);
4223 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4224 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4225 else if (sk_is_phonet(sk))
4226 rc = phonet_sk_ioctl(sk, cmd, arg);
4228 /* If ioctl was processed, returns its value */
4232 /* Otherwise call the default handler */
4233 return sock_ioctl_out(sk, cmd, arg);
4235 EXPORT_SYMBOL(sk_ioctl);
4237 static int __init sock_struct_check(void)
4239 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4240 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4241 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4242 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4243 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4245 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4246 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4247 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4248 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4249 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4250 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4251 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4252 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4253 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4255 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4256 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4257 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4259 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4260 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4261 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4262 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4264 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4265 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4266 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf);
4267 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4268 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4269 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4270 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4271 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4272 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4273 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm);
4274 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status);
4275 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4276 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4277 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4278 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4279 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4281 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4282 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4283 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4284 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4285 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4286 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4287 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4288 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4289 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4290 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4291 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4292 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4293 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4297 core_initcall(sock_struct_check);