1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
117 #include <linux/mroute.h>
118 #include <linux/mroute6.h>
119 #include <linux/icmpv6.h>
121 #include <linux/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
138 #include <net/bpf_sk_storage.h>
140 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
144 #include <net/phonet/phonet.h>
146 #include <linux/ethtool.h>
150 static DEFINE_MUTEX(proto_list_mutex);
151 static LIST_HEAD(proto_list);
153 static void sock_def_write_space_wfree(struct sock *sk);
154 static void sock_def_write_space(struct sock *sk);
157 * sk_ns_capable - General socket capability test
158 * @sk: Socket to use a capability on or through
159 * @user_ns: The user namespace of the capability to use
160 * @cap: The capability to use
162 * Test to see if the opener of the socket had when the socket was
163 * created and the current process has the capability @cap in the user
164 * namespace @user_ns.
166 bool sk_ns_capable(const struct sock *sk,
167 struct user_namespace *user_ns, int cap)
169 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
170 ns_capable(user_ns, cap);
172 EXPORT_SYMBOL(sk_ns_capable);
175 * sk_capable - Socket global capability test
176 * @sk: Socket to use a capability on or through
177 * @cap: The global capability to use
179 * Test to see if the opener of the socket had when the socket was
180 * created and the current process has the capability @cap in all user
183 bool sk_capable(const struct sock *sk, int cap)
185 return sk_ns_capable(sk, &init_user_ns, cap);
187 EXPORT_SYMBOL(sk_capable);
190 * sk_net_capable - Network namespace socket capability test
191 * @sk: Socket to use a capability on or through
192 * @cap: The capability to use
194 * Test to see if the opener of the socket had when the socket was created
195 * and the current process has the capability @cap over the network namespace
196 * the socket is a member of.
198 bool sk_net_capable(const struct sock *sk, int cap)
200 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
202 EXPORT_SYMBOL(sk_net_capable);
205 * Each address family might have different locking rules, so we have
206 * one slock key per address family and separate keys for internal and
209 static struct lock_class_key af_family_keys[AF_MAX];
210 static struct lock_class_key af_family_kern_keys[AF_MAX];
211 static struct lock_class_key af_family_slock_keys[AF_MAX];
212 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
215 * Make lock validator output more readable. (we pre-construct these
216 * strings build-time, so that runtime initialization of socket
220 #define _sock_locks(x) \
221 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
222 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
223 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
224 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
225 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
226 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
227 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
228 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
229 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
230 x "27" , x "28" , x "AF_CAN" , \
231 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
232 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
233 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
234 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
235 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
239 static const char *const af_family_key_strings[AF_MAX+1] = {
240 _sock_locks("sk_lock-")
242 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("slock-")
245 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("clock-")
249 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
250 _sock_locks("k-sk_lock-")
252 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
253 _sock_locks("k-slock-")
255 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
256 _sock_locks("k-clock-")
258 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
259 _sock_locks("rlock-")
261 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
262 _sock_locks("wlock-")
264 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
265 _sock_locks("elock-")
269 * sk_callback_lock and sk queues locking rules are per-address-family,
270 * so split the lock classes by using a per-AF key:
272 static struct lock_class_key af_callback_keys[AF_MAX];
273 static struct lock_class_key af_rlock_keys[AF_MAX];
274 static struct lock_class_key af_wlock_keys[AF_MAX];
275 static struct lock_class_key af_elock_keys[AF_MAX];
276 static struct lock_class_key af_kern_callback_keys[AF_MAX];
278 /* Run time adjustable parameters. */
279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
280 EXPORT_SYMBOL(sysctl_wmem_max);
281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
282 EXPORT_SYMBOL(sysctl_rmem_max);
283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
286 /* Maximal space eaten by iovec or ancillary data plus some space */
287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
288 EXPORT_SYMBOL(sysctl_optmem_max);
290 int sysctl_tstamp_allow_data __read_mostly = 1;
292 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
293 EXPORT_SYMBOL_GPL(memalloc_socks_key);
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
303 void sk_set_memalloc(struct sock *sk)
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
307 static_branch_inc(&memalloc_socks_key);
309 EXPORT_SYMBOL_GPL(sk_set_memalloc);
311 void sk_clear_memalloc(struct sock *sk)
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
315 static_branch_dec(&memalloc_socks_key);
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
326 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
331 unsigned int noreclaim_flag;
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
336 noreclaim_flag = memalloc_noreclaim_save();
337 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
341 memalloc_noreclaim_restore(noreclaim_flag);
345 EXPORT_SYMBOL(__sk_backlog_rcv);
347 void sk_error_report(struct sock *sk)
349 sk->sk_error_report(sk);
351 switch (sk->sk_family) {
355 trace_inet_sk_error_report(sk);
361 EXPORT_SYMBOL(sk_error_report);
363 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
365 struct __kernel_sock_timeval tv;
367 if (timeo == MAX_SCHEDULE_TIMEOUT) {
371 tv.tv_sec = timeo / HZ;
372 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
375 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
376 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
377 *(struct old_timeval32 *)optval = tv32;
382 struct __kernel_old_timeval old_tv;
383 old_tv.tv_sec = tv.tv_sec;
384 old_tv.tv_usec = tv.tv_usec;
385 *(struct __kernel_old_timeval *)optval = old_tv;
386 return sizeof(old_tv);
389 *(struct __kernel_sock_timeval *)optval = tv;
392 EXPORT_SYMBOL(sock_get_timeout);
394 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
395 sockptr_t optval, int optlen, bool old_timeval)
397 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
398 struct old_timeval32 tv32;
400 if (optlen < sizeof(tv32))
403 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
405 tv->tv_sec = tv32.tv_sec;
406 tv->tv_usec = tv32.tv_usec;
407 } else if (old_timeval) {
408 struct __kernel_old_timeval old_tv;
410 if (optlen < sizeof(old_tv))
412 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
414 tv->tv_sec = old_tv.tv_sec;
415 tv->tv_usec = old_tv.tv_usec;
417 if (optlen < sizeof(*tv))
419 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
425 EXPORT_SYMBOL(sock_copy_user_timeval);
427 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
430 struct __kernel_sock_timeval tv;
431 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
437 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
441 static int warned __read_mostly;
443 WRITE_ONCE(*timeo_p, 0);
444 if (warned < 10 && net_ratelimit()) {
446 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
447 __func__, current->comm, task_pid_nr(current));
451 val = MAX_SCHEDULE_TIMEOUT;
452 if ((tv.tv_sec || tv.tv_usec) &&
453 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
454 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
456 WRITE_ONCE(*timeo_p, val);
460 static bool sock_needs_netstamp(const struct sock *sk)
462 switch (sk->sk_family) {
471 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
473 if (sk->sk_flags & flags) {
474 sk->sk_flags &= ~flags;
475 if (sock_needs_netstamp(sk) &&
476 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
477 net_disable_timestamp();
482 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
485 struct sk_buff_head *list = &sk->sk_receive_queue;
487 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
488 atomic_inc(&sk->sk_drops);
489 trace_sock_rcvqueue_full(sk, skb);
493 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
494 atomic_inc(&sk->sk_drops);
499 skb_set_owner_r(skb, sk);
501 /* we escape from rcu protected region, make sure we dont leak
506 spin_lock_irqsave(&list->lock, flags);
507 sock_skb_set_dropcount(sk, skb);
508 __skb_queue_tail(list, skb);
509 spin_unlock_irqrestore(&list->lock, flags);
511 if (!sock_flag(sk, SOCK_DEAD))
512 sk->sk_data_ready(sk);
515 EXPORT_SYMBOL(__sock_queue_rcv_skb);
517 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
518 enum skb_drop_reason *reason)
520 enum skb_drop_reason drop_reason;
523 err = sk_filter(sk, skb);
525 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
528 err = __sock_queue_rcv_skb(sk, skb);
531 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
534 drop_reason = SKB_DROP_REASON_PROTO_MEM;
537 drop_reason = SKB_NOT_DROPPED_YET;
542 *reason = drop_reason;
545 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
547 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
548 const int nested, unsigned int trim_cap, bool refcounted)
550 int rc = NET_RX_SUCCESS;
552 if (sk_filter_trim_cap(sk, skb, trim_cap))
553 goto discard_and_relse;
557 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
558 atomic_inc(&sk->sk_drops);
559 goto discard_and_relse;
562 bh_lock_sock_nested(sk);
565 if (!sock_owned_by_user(sk)) {
567 * trylock + unlock semantics:
569 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
571 rc = sk_backlog_rcv(sk, skb);
573 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
574 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
576 atomic_inc(&sk->sk_drops);
577 goto discard_and_relse;
589 EXPORT_SYMBOL(__sk_receive_skb);
591 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
593 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
595 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
597 struct dst_entry *dst = __sk_dst_get(sk);
599 if (dst && dst->obsolete &&
600 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
601 dst, cookie) == NULL) {
602 sk_tx_queue_clear(sk);
603 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
604 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
611 EXPORT_SYMBOL(__sk_dst_check);
613 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
615 struct dst_entry *dst = sk_dst_get(sk);
617 if (dst && dst->obsolete &&
618 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
619 dst, cookie) == NULL) {
627 EXPORT_SYMBOL(sk_dst_check);
629 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
631 int ret = -ENOPROTOOPT;
632 #ifdef CONFIG_NETDEVICES
633 struct net *net = sock_net(sk);
637 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
644 /* Paired with all READ_ONCE() done locklessly. */
645 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
647 if (sk->sk_prot->rehash)
648 sk->sk_prot->rehash(sk);
659 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
665 ret = sock_bindtoindex_locked(sk, ifindex);
671 EXPORT_SYMBOL(sock_bindtoindex);
673 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
675 int ret = -ENOPROTOOPT;
676 #ifdef CONFIG_NETDEVICES
677 struct net *net = sock_net(sk);
678 char devname[IFNAMSIZ];
685 /* Bind this socket to a particular device like "eth0",
686 * as specified in the passed interface name. If the
687 * name is "" or the option length is zero the socket
690 if (optlen > IFNAMSIZ - 1)
691 optlen = IFNAMSIZ - 1;
692 memset(devname, 0, sizeof(devname));
695 if (copy_from_sockptr(devname, optval, optlen))
699 if (devname[0] != '\0') {
700 struct net_device *dev;
703 dev = dev_get_by_name_rcu(net, devname);
705 index = dev->ifindex;
712 sockopt_lock_sock(sk);
713 ret = sock_bindtoindex_locked(sk, index);
714 sockopt_release_sock(sk);
721 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
722 sockptr_t optlen, int len)
724 int ret = -ENOPROTOOPT;
725 #ifdef CONFIG_NETDEVICES
726 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
727 struct net *net = sock_net(sk);
728 char devname[IFNAMSIZ];
730 if (bound_dev_if == 0) {
739 ret = netdev_get_name(net, devname, bound_dev_if);
743 len = strlen(devname) + 1;
746 if (copy_to_sockptr(optval, devname, len))
751 if (copy_to_sockptr(optlen, &len, sizeof(int)))
762 bool sk_mc_loop(struct sock *sk)
764 if (dev_recursion_level())
768 /* IPV6_ADDRFORM can change sk->sk_family under us. */
769 switch (READ_ONCE(sk->sk_family)) {
771 return inet_test_bit(MC_LOOP, sk);
772 #if IS_ENABLED(CONFIG_IPV6)
774 return inet6_sk(sk)->mc_loop;
780 EXPORT_SYMBOL(sk_mc_loop);
782 void sock_set_reuseaddr(struct sock *sk)
785 sk->sk_reuse = SK_CAN_REUSE;
788 EXPORT_SYMBOL(sock_set_reuseaddr);
790 void sock_set_reuseport(struct sock *sk)
793 sk->sk_reuseport = true;
796 EXPORT_SYMBOL(sock_set_reuseport);
798 void sock_no_linger(struct sock *sk)
801 WRITE_ONCE(sk->sk_lingertime, 0);
802 sock_set_flag(sk, SOCK_LINGER);
805 EXPORT_SYMBOL(sock_no_linger);
807 void sock_set_priority(struct sock *sk, u32 priority)
810 WRITE_ONCE(sk->sk_priority, priority);
813 EXPORT_SYMBOL(sock_set_priority);
815 void sock_set_sndtimeo(struct sock *sk, s64 secs)
818 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
819 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
821 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
824 EXPORT_SYMBOL(sock_set_sndtimeo);
826 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
829 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
830 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
831 sock_set_flag(sk, SOCK_RCVTSTAMP);
832 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
834 sock_reset_flag(sk, SOCK_RCVTSTAMP);
835 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
839 void sock_enable_timestamps(struct sock *sk)
842 __sock_set_timestamps(sk, true, false, true);
845 EXPORT_SYMBOL(sock_enable_timestamps);
847 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
850 case SO_TIMESTAMP_OLD:
851 __sock_set_timestamps(sk, valbool, false, false);
853 case SO_TIMESTAMP_NEW:
854 __sock_set_timestamps(sk, valbool, true, false);
856 case SO_TIMESTAMPNS_OLD:
857 __sock_set_timestamps(sk, valbool, false, true);
859 case SO_TIMESTAMPNS_NEW:
860 __sock_set_timestamps(sk, valbool, true, true);
865 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
867 struct net *net = sock_net(sk);
868 struct net_device *dev = NULL;
873 if (sk->sk_bound_dev_if)
874 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
877 pr_err("%s: sock not bind to device\n", __func__);
881 num = ethtool_get_phc_vclocks(dev, &vclock_index);
884 for (i = 0; i < num; i++) {
885 if (*(vclock_index + i) == phc_index) {
897 WRITE_ONCE(sk->sk_bind_phc, phc_index);
902 int sock_set_timestamping(struct sock *sk, int optname,
903 struct so_timestamping timestamping)
905 int val = timestamping.flags;
908 if (val & ~SOF_TIMESTAMPING_MASK)
911 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
912 !(val & SOF_TIMESTAMPING_OPT_ID))
915 if (val & SOF_TIMESTAMPING_OPT_ID &&
916 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
918 if ((1 << sk->sk_state) &
919 (TCPF_CLOSE | TCPF_LISTEN))
921 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
922 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
924 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
926 atomic_set(&sk->sk_tskey, 0);
930 if (val & SOF_TIMESTAMPING_OPT_STATS &&
931 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
934 if (val & SOF_TIMESTAMPING_BIND_PHC) {
935 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
940 WRITE_ONCE(sk->sk_tsflags, val);
941 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
943 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
944 sock_enable_timestamp(sk,
945 SOCK_TIMESTAMPING_RX_SOFTWARE);
947 sock_disable_timestamp(sk,
948 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
952 void sock_set_keepalive(struct sock *sk)
955 if (sk->sk_prot->keepalive)
956 sk->sk_prot->keepalive(sk, true);
957 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
960 EXPORT_SYMBOL(sock_set_keepalive);
962 static void __sock_set_rcvbuf(struct sock *sk, int val)
964 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
965 * as a negative value.
967 val = min_t(int, val, INT_MAX / 2);
968 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
970 /* We double it on the way in to account for "struct sk_buff" etc.
971 * overhead. Applications assume that the SO_RCVBUF setting they make
972 * will allow that much actual data to be received on that socket.
974 * Applications are unaware that "struct sk_buff" and other overheads
975 * allocate from the receive buffer during socket buffer allocation.
977 * And after considering the possible alternatives, returning the value
978 * we actually used in getsockopt is the most desirable behavior.
980 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
983 void sock_set_rcvbuf(struct sock *sk, int val)
986 __sock_set_rcvbuf(sk, val);
989 EXPORT_SYMBOL(sock_set_rcvbuf);
991 static void __sock_set_mark(struct sock *sk, u32 val)
993 if (val != sk->sk_mark) {
994 WRITE_ONCE(sk->sk_mark, val);
999 void sock_set_mark(struct sock *sk, u32 val)
1002 __sock_set_mark(sk, val);
1005 EXPORT_SYMBOL(sock_set_mark);
1007 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1009 /* Round down bytes to multiple of pages */
1010 bytes = round_down(bytes, PAGE_SIZE);
1012 WARN_ON(bytes > sk->sk_reserved_mem);
1013 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1017 static int sock_reserve_memory(struct sock *sk, int bytes)
1023 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1029 pages = sk_mem_pages(bytes);
1031 /* pre-charge to memcg */
1032 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1033 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1037 /* pre-charge to forward_alloc */
1038 sk_memory_allocated_add(sk, pages);
1039 allocated = sk_memory_allocated(sk);
1040 /* If the system goes into memory pressure with this
1041 * precharge, give up and return error.
1043 if (allocated > sk_prot_mem_limits(sk, 1)) {
1044 sk_memory_allocated_sub(sk, pages);
1045 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1048 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1050 WRITE_ONCE(sk->sk_reserved_mem,
1051 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1056 void sockopt_lock_sock(struct sock *sk)
1058 /* When current->bpf_ctx is set, the setsockopt is called from
1059 * a bpf prog. bpf has ensured the sk lock has been
1060 * acquired before calling setsockopt().
1062 if (has_current_bpf_ctx())
1067 EXPORT_SYMBOL(sockopt_lock_sock);
1069 void sockopt_release_sock(struct sock *sk)
1071 if (has_current_bpf_ctx())
1076 EXPORT_SYMBOL(sockopt_release_sock);
1078 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1080 return has_current_bpf_ctx() || ns_capable(ns, cap);
1082 EXPORT_SYMBOL(sockopt_ns_capable);
1084 bool sockopt_capable(int cap)
1086 return has_current_bpf_ctx() || capable(cap);
1088 EXPORT_SYMBOL(sockopt_capable);
1091 * This is meant for all protocols to use and covers goings on
1092 * at the socket level. Everything here is generic.
1095 int sk_setsockopt(struct sock *sk, int level, int optname,
1096 sockptr_t optval, unsigned int optlen)
1098 struct so_timestamping timestamping;
1099 struct socket *sock = sk->sk_socket;
1100 struct sock_txtime sk_txtime;
1107 * Options without arguments
1110 if (optname == SO_BINDTODEVICE)
1111 return sock_setbindtodevice(sk, optval, optlen);
1113 if (optlen < sizeof(int))
1116 if (copy_from_sockptr(&val, optval, sizeof(val)))
1119 valbool = val ? 1 : 0;
1121 sockopt_lock_sock(sk);
1125 if (val && !sockopt_capable(CAP_NET_ADMIN))
1128 sock_valbool_flag(sk, SOCK_DBG, valbool);
1131 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1134 sk->sk_reuseport = valbool;
1143 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1147 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1150 /* Don't error on this BSD doesn't and if you think
1151 * about it this is right. Otherwise apps have to
1152 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1153 * are treated in BSD as hints
1155 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1157 /* Ensure val * 2 fits into an int, to prevent max_t()
1158 * from treating it as a negative value.
1160 val = min_t(int, val, INT_MAX / 2);
1161 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1162 WRITE_ONCE(sk->sk_sndbuf,
1163 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1164 /* Wake up sending tasks if we upped the value. */
1165 sk->sk_write_space(sk);
1168 case SO_SNDBUFFORCE:
1169 if (!sockopt_capable(CAP_NET_ADMIN)) {
1174 /* No negative values (to prevent underflow, as val will be
1182 /* Don't error on this BSD doesn't and if you think
1183 * about it this is right. Otherwise apps have to
1184 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1185 * are treated in BSD as hints
1187 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1190 case SO_RCVBUFFORCE:
1191 if (!sockopt_capable(CAP_NET_ADMIN)) {
1196 /* No negative values (to prevent underflow, as val will be
1199 __sock_set_rcvbuf(sk, max(val, 0));
1203 if (sk->sk_prot->keepalive)
1204 sk->sk_prot->keepalive(sk, valbool);
1205 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1209 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1213 sk->sk_no_check_tx = valbool;
1217 if ((val >= 0 && val <= 6) ||
1218 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1219 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1220 WRITE_ONCE(sk->sk_priority, val);
1226 if (optlen < sizeof(ling)) {
1227 ret = -EINVAL; /* 1003.1g */
1230 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1234 if (!ling.l_onoff) {
1235 sock_reset_flag(sk, SOCK_LINGER);
1237 unsigned long t_sec = ling.l_linger;
1239 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1240 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1242 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1243 sock_set_flag(sk, SOCK_LINGER);
1251 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1255 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1258 case SO_TIMESTAMP_OLD:
1259 case SO_TIMESTAMP_NEW:
1260 case SO_TIMESTAMPNS_OLD:
1261 case SO_TIMESTAMPNS_NEW:
1262 sock_set_timestamp(sk, optname, valbool);
1265 case SO_TIMESTAMPING_NEW:
1266 case SO_TIMESTAMPING_OLD:
1267 if (optlen == sizeof(timestamping)) {
1268 if (copy_from_sockptr(×tamping, optval,
1269 sizeof(timestamping))) {
1274 memset(×tamping, 0, sizeof(timestamping));
1275 timestamping.flags = val;
1277 ret = sock_set_timestamping(sk, optname, timestamping);
1282 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1287 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1289 ret = set_rcvlowat(sk, val);
1291 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1294 case SO_RCVTIMEO_OLD:
1295 case SO_RCVTIMEO_NEW:
1296 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1297 optlen, optname == SO_RCVTIMEO_OLD);
1300 case SO_SNDTIMEO_OLD:
1301 case SO_SNDTIMEO_NEW:
1302 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1303 optlen, optname == SO_SNDTIMEO_OLD);
1306 case SO_ATTACH_FILTER: {
1307 struct sock_fprog fprog;
1309 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1311 ret = sk_attach_filter(&fprog, sk);
1316 if (optlen == sizeof(u32)) {
1320 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1323 ret = sk_attach_bpf(ufd, sk);
1327 case SO_ATTACH_REUSEPORT_CBPF: {
1328 struct sock_fprog fprog;
1330 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1332 ret = sk_reuseport_attach_filter(&fprog, sk);
1335 case SO_ATTACH_REUSEPORT_EBPF:
1337 if (optlen == sizeof(u32)) {
1341 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1344 ret = sk_reuseport_attach_bpf(ufd, sk);
1348 case SO_DETACH_REUSEPORT_BPF:
1349 ret = reuseport_detach_prog(sk);
1352 case SO_DETACH_FILTER:
1353 ret = sk_detach_filter(sk);
1356 case SO_LOCK_FILTER:
1357 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1360 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1364 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1367 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1368 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1373 __sock_set_mark(sk, val);
1376 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1380 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1383 case SO_WIFI_STATUS:
1384 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1389 int (*set_peek_off)(struct sock *sk, int val);
1391 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1393 ret = set_peek_off(sk, val);
1400 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1403 case SO_SELECT_ERR_QUEUE:
1404 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1407 #ifdef CONFIG_NET_RX_BUSY_POLL
1412 WRITE_ONCE(sk->sk_ll_usec, val);
1414 case SO_PREFER_BUSY_POLL:
1415 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1418 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1420 case SO_BUSY_POLL_BUDGET:
1421 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) {
1424 if (val < 0 || val > U16_MAX)
1427 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1432 case SO_MAX_PACING_RATE:
1434 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1436 if (sizeof(ulval) != sizeof(val) &&
1437 optlen >= sizeof(ulval) &&
1438 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1443 cmpxchg(&sk->sk_pacing_status,
1446 /* Pairs with READ_ONCE() from sk_getsockopt() */
1447 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1448 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1451 case SO_INCOMING_CPU:
1452 reuseport_update_incoming_cpu(sk, val);
1457 dst_negative_advice(sk);
1461 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1462 if (!(sk_is_tcp(sk) ||
1463 (sk->sk_type == SOCK_DGRAM &&
1464 sk->sk_protocol == IPPROTO_UDP)))
1466 } else if (sk->sk_family != PF_RDS) {
1470 if (val < 0 || val > 1)
1473 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1478 if (optlen != sizeof(struct sock_txtime)) {
1481 } else if (copy_from_sockptr(&sk_txtime, optval,
1482 sizeof(struct sock_txtime))) {
1485 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1489 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1490 * scheduler has enough safe guards.
1492 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1493 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1497 sock_valbool_flag(sk, SOCK_TXTIME, true);
1498 sk->sk_clockid = sk_txtime.clockid;
1499 sk->sk_txtime_deadline_mode =
1500 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1501 sk->sk_txtime_report_errors =
1502 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1505 case SO_BINDTOIFINDEX:
1506 ret = sock_bindtoindex_locked(sk, val);
1510 if (val & ~SOCK_BUF_LOCK_MASK) {
1514 sk->sk_userlocks = val | (sk->sk_userlocks &
1515 ~SOCK_BUF_LOCK_MASK);
1518 case SO_RESERVE_MEM:
1527 delta = val - sk->sk_reserved_mem;
1529 sock_release_reserved_memory(sk, -delta);
1531 ret = sock_reserve_memory(sk, delta);
1536 if (val < -1 || val > 1) {
1540 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1541 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1542 /* Paired with READ_ONCE() in tcp_rtx_synack()
1543 * and sk_getsockopt().
1545 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1552 sockopt_release_sock(sk);
1556 int sock_setsockopt(struct socket *sock, int level, int optname,
1557 sockptr_t optval, unsigned int optlen)
1559 return sk_setsockopt(sock->sk, level, optname,
1562 EXPORT_SYMBOL(sock_setsockopt);
1564 static const struct cred *sk_get_peer_cred(struct sock *sk)
1566 const struct cred *cred;
1568 spin_lock(&sk->sk_peer_lock);
1569 cred = get_cred(sk->sk_peer_cred);
1570 spin_unlock(&sk->sk_peer_lock);
1575 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1576 struct ucred *ucred)
1578 ucred->pid = pid_vnr(pid);
1579 ucred->uid = ucred->gid = -1;
1581 struct user_namespace *current_ns = current_user_ns();
1583 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1584 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1588 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1590 struct user_namespace *user_ns = current_user_ns();
1593 for (i = 0; i < src->ngroups; i++) {
1594 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1596 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1603 int sk_getsockopt(struct sock *sk, int level, int optname,
1604 sockptr_t optval, sockptr_t optlen)
1606 struct socket *sock = sk->sk_socket;
1611 unsigned long ulval;
1613 struct old_timeval32 tm32;
1614 struct __kernel_old_timeval tm;
1615 struct __kernel_sock_timeval stm;
1616 struct sock_txtime txtime;
1617 struct so_timestamping timestamping;
1620 int lv = sizeof(int);
1623 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1628 memset(&v, 0, sizeof(v));
1632 v.val = sock_flag(sk, SOCK_DBG);
1636 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1640 v.val = sock_flag(sk, SOCK_BROADCAST);
1644 v.val = READ_ONCE(sk->sk_sndbuf);
1648 v.val = READ_ONCE(sk->sk_rcvbuf);
1652 v.val = sk->sk_reuse;
1656 v.val = sk->sk_reuseport;
1660 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1664 v.val = sk->sk_type;
1668 v.val = sk->sk_protocol;
1672 v.val = sk->sk_family;
1676 v.val = -sock_error(sk);
1678 v.val = xchg(&sk->sk_err_soft, 0);
1682 v.val = sock_flag(sk, SOCK_URGINLINE);
1686 v.val = sk->sk_no_check_tx;
1690 v.val = READ_ONCE(sk->sk_priority);
1694 lv = sizeof(v.ling);
1695 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1696 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1702 case SO_TIMESTAMP_OLD:
1703 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1704 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1705 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1708 case SO_TIMESTAMPNS_OLD:
1709 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1712 case SO_TIMESTAMP_NEW:
1713 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1716 case SO_TIMESTAMPNS_NEW:
1717 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1720 case SO_TIMESTAMPING_OLD:
1721 case SO_TIMESTAMPING_NEW:
1722 lv = sizeof(v.timestamping);
1723 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1724 * returning the flags when they were set through the same option.
1725 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1727 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1728 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1729 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1733 case SO_RCVTIMEO_OLD:
1734 case SO_RCVTIMEO_NEW:
1735 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1736 SO_RCVTIMEO_OLD == optname);
1739 case SO_SNDTIMEO_OLD:
1740 case SO_SNDTIMEO_NEW:
1741 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1742 SO_SNDTIMEO_OLD == optname);
1746 v.val = READ_ONCE(sk->sk_rcvlowat);
1754 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1758 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1763 struct ucred peercred;
1764 if (len > sizeof(peercred))
1765 len = sizeof(peercred);
1767 spin_lock(&sk->sk_peer_lock);
1768 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1769 spin_unlock(&sk->sk_peer_lock);
1771 if (copy_to_sockptr(optval, &peercred, len))
1778 struct pid *peer_pid;
1779 struct file *pidfd_file = NULL;
1782 if (len > sizeof(pidfd))
1783 len = sizeof(pidfd);
1785 spin_lock(&sk->sk_peer_lock);
1786 peer_pid = get_pid(sk->sk_peer_pid);
1787 spin_unlock(&sk->sk_peer_lock);
1792 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1797 if (copy_to_sockptr(optval, &pidfd, len) ||
1798 copy_to_sockptr(optlen, &len, sizeof(int))) {
1799 put_unused_fd(pidfd);
1805 fd_install(pidfd, pidfd_file);
1811 const struct cred *cred;
1814 cred = sk_get_peer_cred(sk);
1818 n = cred->group_info->ngroups;
1819 if (len < n * sizeof(gid_t)) {
1820 len = n * sizeof(gid_t);
1822 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1824 len = n * sizeof(gid_t);
1826 ret = groups_to_user(optval, cred->group_info);
1835 struct sockaddr_storage address;
1837 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1842 if (copy_to_sockptr(optval, &address, len))
1847 /* Dubious BSD thing... Probably nobody even uses it, but
1848 * the UNIX standard wants it for whatever reason... -DaveM
1851 v.val = sk->sk_state == TCP_LISTEN;
1855 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1859 return security_socket_getpeersec_stream(sock,
1860 optval, optlen, len);
1863 v.val = READ_ONCE(sk->sk_mark);
1867 v.val = sock_flag(sk, SOCK_RCVMARK);
1871 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1874 case SO_WIFI_STATUS:
1875 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1879 if (!READ_ONCE(sock->ops)->set_peek_off)
1882 v.val = READ_ONCE(sk->sk_peek_off);
1885 v.val = sock_flag(sk, SOCK_NOFCS);
1888 case SO_BINDTODEVICE:
1889 return sock_getbindtodevice(sk, optval, optlen, len);
1892 len = sk_get_filter(sk, optval, len);
1898 case SO_LOCK_FILTER:
1899 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1902 case SO_BPF_EXTENSIONS:
1903 v.val = bpf_tell_extensions();
1906 case SO_SELECT_ERR_QUEUE:
1907 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1910 #ifdef CONFIG_NET_RX_BUSY_POLL
1912 v.val = READ_ONCE(sk->sk_ll_usec);
1914 case SO_PREFER_BUSY_POLL:
1915 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1919 case SO_MAX_PACING_RATE:
1920 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1921 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1922 lv = sizeof(v.ulval);
1923 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1926 v.val = min_t(unsigned long, ~0U,
1927 READ_ONCE(sk->sk_max_pacing_rate));
1931 case SO_INCOMING_CPU:
1932 v.val = READ_ONCE(sk->sk_incoming_cpu);
1937 u32 meminfo[SK_MEMINFO_VARS];
1939 sk_get_meminfo(sk, meminfo);
1941 len = min_t(unsigned int, len, sizeof(meminfo));
1942 if (copy_to_sockptr(optval, &meminfo, len))
1948 #ifdef CONFIG_NET_RX_BUSY_POLL
1949 case SO_INCOMING_NAPI_ID:
1950 v.val = READ_ONCE(sk->sk_napi_id);
1952 /* aggregate non-NAPI IDs down to 0 */
1953 if (v.val < MIN_NAPI_ID)
1963 v.val64 = sock_gen_cookie(sk);
1967 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1971 lv = sizeof(v.txtime);
1972 v.txtime.clockid = sk->sk_clockid;
1973 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1974 SOF_TXTIME_DEADLINE_MODE : 0;
1975 v.txtime.flags |= sk->sk_txtime_report_errors ?
1976 SOF_TXTIME_REPORT_ERRORS : 0;
1979 case SO_BINDTOIFINDEX:
1980 v.val = READ_ONCE(sk->sk_bound_dev_if);
1983 case SO_NETNS_COOKIE:
1987 v.val64 = sock_net(sk)->net_cookie;
1991 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1994 case SO_RESERVE_MEM:
1995 v.val = READ_ONCE(sk->sk_reserved_mem);
1999 /* Paired with WRITE_ONCE() in sk_setsockopt() */
2000 v.val = READ_ONCE(sk->sk_txrehash);
2004 /* We implement the SO_SNDLOWAT etc to not be settable
2007 return -ENOPROTOOPT;
2012 if (copy_to_sockptr(optval, &v, len))
2015 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2020 int sock_getsockopt(struct socket *sock, int level, int optname,
2021 char __user *optval, int __user *optlen)
2023 return sk_getsockopt(sock->sk, level, optname,
2024 USER_SOCKPTR(optval),
2025 USER_SOCKPTR(optlen));
2029 * Initialize an sk_lock.
2031 * (We also register the sk_lock with the lock validator.)
2033 static inline void sock_lock_init(struct sock *sk)
2035 if (sk->sk_kern_sock)
2036 sock_lock_init_class_and_name(
2038 af_family_kern_slock_key_strings[sk->sk_family],
2039 af_family_kern_slock_keys + sk->sk_family,
2040 af_family_kern_key_strings[sk->sk_family],
2041 af_family_kern_keys + sk->sk_family);
2043 sock_lock_init_class_and_name(
2045 af_family_slock_key_strings[sk->sk_family],
2046 af_family_slock_keys + sk->sk_family,
2047 af_family_key_strings[sk->sk_family],
2048 af_family_keys + sk->sk_family);
2052 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2053 * even temporarly, because of RCU lookups. sk_node should also be left as is.
2054 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2056 static void sock_copy(struct sock *nsk, const struct sock *osk)
2058 const struct proto *prot = READ_ONCE(osk->sk_prot);
2059 #ifdef CONFIG_SECURITY_NETWORK
2060 void *sptr = nsk->sk_security;
2063 /* If we move sk_tx_queue_mapping out of the private section,
2064 * we must check if sk_tx_queue_clear() is called after
2065 * sock_copy() in sk_clone_lock().
2067 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2068 offsetof(struct sock, sk_dontcopy_begin) ||
2069 offsetof(struct sock, sk_tx_queue_mapping) >=
2070 offsetof(struct sock, sk_dontcopy_end));
2072 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2074 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2075 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2077 #ifdef CONFIG_SECURITY_NETWORK
2078 nsk->sk_security = sptr;
2079 security_sk_clone(osk, nsk);
2083 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2087 struct kmem_cache *slab;
2091 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2094 if (want_init_on_alloc(priority))
2095 sk_prot_clear_nulls(sk, prot->obj_size);
2097 sk = kmalloc(prot->obj_size, priority);
2100 if (security_sk_alloc(sk, family, priority))
2103 if (!try_module_get(prot->owner))
2110 security_sk_free(sk);
2113 kmem_cache_free(slab, sk);
2119 static void sk_prot_free(struct proto *prot, struct sock *sk)
2121 struct kmem_cache *slab;
2122 struct module *owner;
2124 owner = prot->owner;
2127 cgroup_sk_free(&sk->sk_cgrp_data);
2128 mem_cgroup_sk_free(sk);
2129 security_sk_free(sk);
2131 kmem_cache_free(slab, sk);
2138 * sk_alloc - All socket objects are allocated here
2139 * @net: the applicable net namespace
2140 * @family: protocol family
2141 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2142 * @prot: struct proto associated with this new sock instance
2143 * @kern: is this to be a kernel socket?
2145 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2146 struct proto *prot, int kern)
2150 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2152 sk->sk_family = family;
2154 * See comment in struct sock definition to understand
2155 * why we need sk_prot_creator -acme
2157 sk->sk_prot = sk->sk_prot_creator = prot;
2158 sk->sk_kern_sock = kern;
2160 sk->sk_net_refcnt = kern ? 0 : 1;
2161 if (likely(sk->sk_net_refcnt)) {
2162 get_net_track(net, &sk->ns_tracker, priority);
2163 sock_inuse_add(net, 1);
2165 __netns_tracker_alloc(net, &sk->ns_tracker,
2169 sock_net_set(sk, net);
2170 refcount_set(&sk->sk_wmem_alloc, 1);
2172 mem_cgroup_sk_alloc(sk);
2173 cgroup_sk_alloc(&sk->sk_cgrp_data);
2174 sock_update_classid(&sk->sk_cgrp_data);
2175 sock_update_netprioidx(&sk->sk_cgrp_data);
2176 sk_tx_queue_clear(sk);
2181 EXPORT_SYMBOL(sk_alloc);
2183 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2184 * grace period. This is the case for UDP sockets and TCP listeners.
2186 static void __sk_destruct(struct rcu_head *head)
2188 struct sock *sk = container_of(head, struct sock, sk_rcu);
2189 struct sk_filter *filter;
2191 if (sk->sk_destruct)
2192 sk->sk_destruct(sk);
2194 filter = rcu_dereference_check(sk->sk_filter,
2195 refcount_read(&sk->sk_wmem_alloc) == 0);
2197 sk_filter_uncharge(sk, filter);
2198 RCU_INIT_POINTER(sk->sk_filter, NULL);
2201 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2203 #ifdef CONFIG_BPF_SYSCALL
2204 bpf_sk_storage_free(sk);
2207 if (atomic_read(&sk->sk_omem_alloc))
2208 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2209 __func__, atomic_read(&sk->sk_omem_alloc));
2211 if (sk->sk_frag.page) {
2212 put_page(sk->sk_frag.page);
2213 sk->sk_frag.page = NULL;
2216 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2217 put_cred(sk->sk_peer_cred);
2218 put_pid(sk->sk_peer_pid);
2220 if (likely(sk->sk_net_refcnt))
2221 put_net_track(sock_net(sk), &sk->ns_tracker);
2223 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2225 sk_prot_free(sk->sk_prot_creator, sk);
2228 void sk_destruct(struct sock *sk)
2230 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2232 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2233 reuseport_detach_sock(sk);
2234 use_call_rcu = true;
2238 call_rcu(&sk->sk_rcu, __sk_destruct);
2240 __sk_destruct(&sk->sk_rcu);
2243 static void __sk_free(struct sock *sk)
2245 if (likely(sk->sk_net_refcnt))
2246 sock_inuse_add(sock_net(sk), -1);
2248 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2249 sock_diag_broadcast_destroy(sk);
2254 void sk_free(struct sock *sk)
2257 * We subtract one from sk_wmem_alloc and can know if
2258 * some packets are still in some tx queue.
2259 * If not null, sock_wfree() will call __sk_free(sk) later
2261 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2264 EXPORT_SYMBOL(sk_free);
2266 static void sk_init_common(struct sock *sk)
2268 skb_queue_head_init(&sk->sk_receive_queue);
2269 skb_queue_head_init(&sk->sk_write_queue);
2270 skb_queue_head_init(&sk->sk_error_queue);
2272 rwlock_init(&sk->sk_callback_lock);
2273 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2274 af_rlock_keys + sk->sk_family,
2275 af_family_rlock_key_strings[sk->sk_family]);
2276 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2277 af_wlock_keys + sk->sk_family,
2278 af_family_wlock_key_strings[sk->sk_family]);
2279 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2280 af_elock_keys + sk->sk_family,
2281 af_family_elock_key_strings[sk->sk_family]);
2282 lockdep_set_class_and_name(&sk->sk_callback_lock,
2283 af_callback_keys + sk->sk_family,
2284 af_family_clock_key_strings[sk->sk_family]);
2288 * sk_clone_lock - clone a socket, and lock its clone
2289 * @sk: the socket to clone
2290 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2292 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2294 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2296 struct proto *prot = READ_ONCE(sk->sk_prot);
2297 struct sk_filter *filter;
2298 bool is_charged = true;
2301 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2305 sock_copy(newsk, sk);
2307 newsk->sk_prot_creator = prot;
2310 if (likely(newsk->sk_net_refcnt)) {
2311 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2312 sock_inuse_add(sock_net(newsk), 1);
2314 /* Kernel sockets are not elevating the struct net refcount.
2315 * Instead, use a tracker to more easily detect if a layer
2316 * is not properly dismantling its kernel sockets at netns
2319 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2322 sk_node_init(&newsk->sk_node);
2323 sock_lock_init(newsk);
2324 bh_lock_sock(newsk);
2325 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2326 newsk->sk_backlog.len = 0;
2328 atomic_set(&newsk->sk_rmem_alloc, 0);
2330 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2331 refcount_set(&newsk->sk_wmem_alloc, 1);
2333 atomic_set(&newsk->sk_omem_alloc, 0);
2334 sk_init_common(newsk);
2336 newsk->sk_dst_cache = NULL;
2337 newsk->sk_dst_pending_confirm = 0;
2338 newsk->sk_wmem_queued = 0;
2339 newsk->sk_forward_alloc = 0;
2340 newsk->sk_reserved_mem = 0;
2341 atomic_set(&newsk->sk_drops, 0);
2342 newsk->sk_send_head = NULL;
2343 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2344 atomic_set(&newsk->sk_zckey, 0);
2346 sock_reset_flag(newsk, SOCK_DONE);
2348 /* sk->sk_memcg will be populated at accept() time */
2349 newsk->sk_memcg = NULL;
2351 cgroup_sk_clone(&newsk->sk_cgrp_data);
2354 filter = rcu_dereference(sk->sk_filter);
2356 /* though it's an empty new sock, the charging may fail
2357 * if sysctl_optmem_max was changed between creation of
2358 * original socket and cloning
2360 is_charged = sk_filter_charge(newsk, filter);
2361 RCU_INIT_POINTER(newsk->sk_filter, filter);
2364 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2365 /* We need to make sure that we don't uncharge the new
2366 * socket if we couldn't charge it in the first place
2367 * as otherwise we uncharge the parent's filter.
2370 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2371 sk_free_unlock_clone(newsk);
2375 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2377 if (bpf_sk_storage_clone(sk, newsk)) {
2378 sk_free_unlock_clone(newsk);
2383 /* Clear sk_user_data if parent had the pointer tagged
2384 * as not suitable for copying when cloning.
2386 if (sk_user_data_is_nocopy(newsk))
2387 newsk->sk_user_data = NULL;
2390 newsk->sk_err_soft = 0;
2391 newsk->sk_priority = 0;
2392 newsk->sk_incoming_cpu = raw_smp_processor_id();
2394 /* Before updating sk_refcnt, we must commit prior changes to memory
2395 * (Documentation/RCU/rculist_nulls.rst for details)
2398 refcount_set(&newsk->sk_refcnt, 2);
2400 sk_set_socket(newsk, NULL);
2401 sk_tx_queue_clear(newsk);
2402 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2404 if (newsk->sk_prot->sockets_allocated)
2405 sk_sockets_allocated_inc(newsk);
2407 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2408 net_enable_timestamp();
2412 EXPORT_SYMBOL_GPL(sk_clone_lock);
2414 void sk_free_unlock_clone(struct sock *sk)
2416 /* It is still raw copy of parent, so invalidate
2417 * destructor and make plain sk_free() */
2418 sk->sk_destruct = NULL;
2422 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2424 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2426 bool is_ipv6 = false;
2429 #if IS_ENABLED(CONFIG_IPV6)
2430 is_ipv6 = (sk->sk_family == AF_INET6 &&
2431 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2433 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2434 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2435 READ_ONCE(dst->dev->gso_ipv4_max_size);
2436 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2437 max_size = GSO_LEGACY_MAX_SIZE;
2439 return max_size - (MAX_TCP_HEADER + 1);
2442 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2446 sk->sk_route_caps = dst->dev->features;
2448 sk->sk_route_caps |= NETIF_F_GSO;
2449 if (sk->sk_route_caps & NETIF_F_GSO)
2450 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2451 if (unlikely(sk->sk_gso_disabled))
2452 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2453 if (sk_can_gso(sk)) {
2454 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2455 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2457 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2458 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2459 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2460 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2463 sk->sk_gso_max_segs = max_segs;
2464 sk_dst_set(sk, dst);
2466 EXPORT_SYMBOL_GPL(sk_setup_caps);
2469 * Simple resource managers for sockets.
2474 * Write buffer destructor automatically called from kfree_skb.
2476 void sock_wfree(struct sk_buff *skb)
2478 struct sock *sk = skb->sk;
2479 unsigned int len = skb->truesize;
2482 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2483 if (sock_flag(sk, SOCK_RCU_FREE) &&
2484 sk->sk_write_space == sock_def_write_space) {
2486 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2487 sock_def_write_space_wfree(sk);
2495 * Keep a reference on sk_wmem_alloc, this will be released
2496 * after sk_write_space() call
2498 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2499 sk->sk_write_space(sk);
2503 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2504 * could not do because of in-flight packets
2506 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2509 EXPORT_SYMBOL(sock_wfree);
2511 /* This variant of sock_wfree() is used by TCP,
2512 * since it sets SOCK_USE_WRITE_QUEUE.
2514 void __sock_wfree(struct sk_buff *skb)
2516 struct sock *sk = skb->sk;
2518 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2522 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2527 if (unlikely(!sk_fullsock(sk))) {
2528 skb->destructor = sock_edemux;
2533 skb->destructor = sock_wfree;
2534 skb_set_hash_from_sk(skb, sk);
2536 * We used to take a refcount on sk, but following operation
2537 * is enough to guarantee sk_free() wont free this sock until
2538 * all in-flight packets are completed
2540 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2542 EXPORT_SYMBOL(skb_set_owner_w);
2544 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2546 #ifdef CONFIG_TLS_DEVICE
2547 /* Drivers depend on in-order delivery for crypto offload,
2548 * partial orphan breaks out-of-order-OK logic.
2553 return (skb->destructor == sock_wfree ||
2554 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2557 /* This helper is used by netem, as it can hold packets in its
2558 * delay queue. We want to allow the owner socket to send more
2559 * packets, as if they were already TX completed by a typical driver.
2560 * But we also want to keep skb->sk set because some packet schedulers
2561 * rely on it (sch_fq for example).
2563 void skb_orphan_partial(struct sk_buff *skb)
2565 if (skb_is_tcp_pure_ack(skb))
2568 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2573 EXPORT_SYMBOL(skb_orphan_partial);
2576 * Read buffer destructor automatically called from kfree_skb.
2578 void sock_rfree(struct sk_buff *skb)
2580 struct sock *sk = skb->sk;
2581 unsigned int len = skb->truesize;
2583 atomic_sub(len, &sk->sk_rmem_alloc);
2584 sk_mem_uncharge(sk, len);
2586 EXPORT_SYMBOL(sock_rfree);
2589 * Buffer destructor for skbs that are not used directly in read or write
2590 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2592 void sock_efree(struct sk_buff *skb)
2596 EXPORT_SYMBOL(sock_efree);
2598 /* Buffer destructor for prefetch/receive path where reference count may
2599 * not be held, e.g. for listen sockets.
2602 void sock_pfree(struct sk_buff *skb)
2604 if (sk_is_refcounted(skb->sk))
2605 sock_gen_put(skb->sk);
2607 EXPORT_SYMBOL(sock_pfree);
2608 #endif /* CONFIG_INET */
2610 kuid_t sock_i_uid(struct sock *sk)
2614 read_lock_bh(&sk->sk_callback_lock);
2615 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2616 read_unlock_bh(&sk->sk_callback_lock);
2619 EXPORT_SYMBOL(sock_i_uid);
2621 unsigned long __sock_i_ino(struct sock *sk)
2625 read_lock(&sk->sk_callback_lock);
2626 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2627 read_unlock(&sk->sk_callback_lock);
2630 EXPORT_SYMBOL(__sock_i_ino);
2632 unsigned long sock_i_ino(struct sock *sk)
2637 ino = __sock_i_ino(sk);
2641 EXPORT_SYMBOL(sock_i_ino);
2644 * Allocate a skb from the socket's send buffer.
2646 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2650 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2651 struct sk_buff *skb = alloc_skb(size, priority);
2654 skb_set_owner_w(skb, sk);
2660 EXPORT_SYMBOL(sock_wmalloc);
2662 static void sock_ofree(struct sk_buff *skb)
2664 struct sock *sk = skb->sk;
2666 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2669 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2672 struct sk_buff *skb;
2674 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2675 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2676 READ_ONCE(sysctl_optmem_max))
2679 skb = alloc_skb(size, priority);
2683 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2685 skb->destructor = sock_ofree;
2690 * Allocate a memory block from the socket's option memory buffer.
2692 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2694 int optmem_max = READ_ONCE(sysctl_optmem_max);
2696 if ((unsigned int)size <= optmem_max &&
2697 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2699 /* First do the add, to avoid the race if kmalloc
2702 atomic_add(size, &sk->sk_omem_alloc);
2703 mem = kmalloc(size, priority);
2706 atomic_sub(size, &sk->sk_omem_alloc);
2710 EXPORT_SYMBOL(sock_kmalloc);
2712 /* Free an option memory block. Note, we actually want the inline
2713 * here as this allows gcc to detect the nullify and fold away the
2714 * condition entirely.
2716 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2719 if (WARN_ON_ONCE(!mem))
2722 kfree_sensitive(mem);
2725 atomic_sub(size, &sk->sk_omem_alloc);
2728 void sock_kfree_s(struct sock *sk, void *mem, int size)
2730 __sock_kfree_s(sk, mem, size, false);
2732 EXPORT_SYMBOL(sock_kfree_s);
2734 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2736 __sock_kfree_s(sk, mem, size, true);
2738 EXPORT_SYMBOL(sock_kzfree_s);
2740 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2741 I think, these locks should be removed for datagram sockets.
2743 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2747 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2751 if (signal_pending(current))
2753 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2754 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2755 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2757 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2759 if (READ_ONCE(sk->sk_err))
2761 timeo = schedule_timeout(timeo);
2763 finish_wait(sk_sleep(sk), &wait);
2769 * Generic send/receive buffer handlers
2772 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2773 unsigned long data_len, int noblock,
2774 int *errcode, int max_page_order)
2776 struct sk_buff *skb;
2780 timeo = sock_sndtimeo(sk, noblock);
2782 err = sock_error(sk);
2787 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2790 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2793 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2794 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2798 if (signal_pending(current))
2800 timeo = sock_wait_for_wmem(sk, timeo);
2802 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2803 errcode, sk->sk_allocation);
2805 skb_set_owner_w(skb, sk);
2809 err = sock_intr_errno(timeo);
2814 EXPORT_SYMBOL(sock_alloc_send_pskb);
2816 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2817 struct sockcm_cookie *sockc)
2821 switch (cmsg->cmsg_type) {
2823 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2824 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2826 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2828 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2830 case SO_TIMESTAMPING_OLD:
2831 case SO_TIMESTAMPING_NEW:
2832 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2835 tsflags = *(u32 *)CMSG_DATA(cmsg);
2836 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2839 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2840 sockc->tsflags |= tsflags;
2843 if (!sock_flag(sk, SOCK_TXTIME))
2845 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2847 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2849 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2851 case SCM_CREDENTIALS:
2858 EXPORT_SYMBOL(__sock_cmsg_send);
2860 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2861 struct sockcm_cookie *sockc)
2863 struct cmsghdr *cmsg;
2866 for_each_cmsghdr(cmsg, msg) {
2867 if (!CMSG_OK(msg, cmsg))
2869 if (cmsg->cmsg_level != SOL_SOCKET)
2871 ret = __sock_cmsg_send(sk, cmsg, sockc);
2877 EXPORT_SYMBOL(sock_cmsg_send);
2879 static void sk_enter_memory_pressure(struct sock *sk)
2881 if (!sk->sk_prot->enter_memory_pressure)
2884 sk->sk_prot->enter_memory_pressure(sk);
2887 static void sk_leave_memory_pressure(struct sock *sk)
2889 if (sk->sk_prot->leave_memory_pressure) {
2890 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2891 tcp_leave_memory_pressure, sk);
2893 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2895 if (memory_pressure && READ_ONCE(*memory_pressure))
2896 WRITE_ONCE(*memory_pressure, 0);
2900 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2903 * skb_page_frag_refill - check that a page_frag contains enough room
2904 * @sz: minimum size of the fragment we want to get
2905 * @pfrag: pointer to page_frag
2906 * @gfp: priority for memory allocation
2908 * Note: While this allocator tries to use high order pages, there is
2909 * no guarantee that allocations succeed. Therefore, @sz MUST be
2910 * less or equal than PAGE_SIZE.
2912 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2915 if (page_ref_count(pfrag->page) == 1) {
2919 if (pfrag->offset + sz <= pfrag->size)
2921 put_page(pfrag->page);
2925 if (SKB_FRAG_PAGE_ORDER &&
2926 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2927 /* Avoid direct reclaim but allow kswapd to wake */
2928 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2929 __GFP_COMP | __GFP_NOWARN |
2931 SKB_FRAG_PAGE_ORDER);
2932 if (likely(pfrag->page)) {
2933 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2937 pfrag->page = alloc_page(gfp);
2938 if (likely(pfrag->page)) {
2939 pfrag->size = PAGE_SIZE;
2944 EXPORT_SYMBOL(skb_page_frag_refill);
2946 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2948 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2951 sk_enter_memory_pressure(sk);
2952 sk_stream_moderate_sndbuf(sk);
2955 EXPORT_SYMBOL(sk_page_frag_refill);
2957 void __lock_sock(struct sock *sk)
2958 __releases(&sk->sk_lock.slock)
2959 __acquires(&sk->sk_lock.slock)
2964 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2965 TASK_UNINTERRUPTIBLE);
2966 spin_unlock_bh(&sk->sk_lock.slock);
2968 spin_lock_bh(&sk->sk_lock.slock);
2969 if (!sock_owned_by_user(sk))
2972 finish_wait(&sk->sk_lock.wq, &wait);
2975 void __release_sock(struct sock *sk)
2976 __releases(&sk->sk_lock.slock)
2977 __acquires(&sk->sk_lock.slock)
2979 struct sk_buff *skb, *next;
2981 while ((skb = sk->sk_backlog.head) != NULL) {
2982 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2984 spin_unlock_bh(&sk->sk_lock.slock);
2989 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2990 skb_mark_not_on_list(skb);
2991 sk_backlog_rcv(sk, skb);
2996 } while (skb != NULL);
2998 spin_lock_bh(&sk->sk_lock.slock);
3002 * Doing the zeroing here guarantee we can not loop forever
3003 * while a wild producer attempts to flood us.
3005 sk->sk_backlog.len = 0;
3008 void __sk_flush_backlog(struct sock *sk)
3010 spin_lock_bh(&sk->sk_lock.slock);
3012 spin_unlock_bh(&sk->sk_lock.slock);
3014 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3017 * sk_wait_data - wait for data to arrive at sk_receive_queue
3018 * @sk: sock to wait on
3019 * @timeo: for how long
3020 * @skb: last skb seen on sk_receive_queue
3022 * Now socket state including sk->sk_err is changed only under lock,
3023 * hence we may omit checks after joining wait queue.
3024 * We check receive queue before schedule() only as optimization;
3025 * it is very likely that release_sock() added new data.
3027 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3029 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3032 add_wait_queue(sk_sleep(sk), &wait);
3033 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3034 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3035 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3036 remove_wait_queue(sk_sleep(sk), &wait);
3039 EXPORT_SYMBOL(sk_wait_data);
3042 * __sk_mem_raise_allocated - increase memory_allocated
3044 * @size: memory size to allocate
3045 * @amt: pages to allocate
3046 * @kind: allocation type
3048 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
3050 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3052 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
3053 struct proto *prot = sk->sk_prot;
3054 bool charged = true;
3057 sk_memory_allocated_add(sk, amt);
3058 allocated = sk_memory_allocated(sk);
3060 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3061 gfp_memcg_charge())))
3062 goto suppress_allocation;
3065 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3066 sk_leave_memory_pressure(sk);
3070 /* Under pressure. */
3071 if (allocated > sk_prot_mem_limits(sk, 1))
3072 sk_enter_memory_pressure(sk);
3074 /* Over hard limit. */
3075 if (allocated > sk_prot_mem_limits(sk, 2))
3076 goto suppress_allocation;
3078 /* guarantee minimum buffer size under pressure */
3079 if (kind == SK_MEM_RECV) {
3080 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3083 } else { /* SK_MEM_SEND */
3084 int wmem0 = sk_get_wmem0(sk, prot);
3086 if (sk->sk_type == SOCK_STREAM) {
3087 if (sk->sk_wmem_queued < wmem0)
3089 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3094 if (sk_has_memory_pressure(sk)) {
3097 if (!sk_under_memory_pressure(sk))
3099 alloc = sk_sockets_allocated_read_positive(sk);
3100 if (sk_prot_mem_limits(sk, 2) > alloc *
3101 sk_mem_pages(sk->sk_wmem_queued +
3102 atomic_read(&sk->sk_rmem_alloc) +
3103 sk->sk_forward_alloc))
3107 suppress_allocation:
3109 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3110 sk_stream_moderate_sndbuf(sk);
3112 /* Fail only if socket is _under_ its sndbuf.
3113 * In this case we cannot block, so that we have to fail.
3115 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3116 /* Force charge with __GFP_NOFAIL */
3117 if (memcg_charge && !charged) {
3118 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3119 gfp_memcg_charge() | __GFP_NOFAIL);
3125 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3126 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3128 sk_memory_allocated_sub(sk, amt);
3130 if (memcg_charge && charged)
3131 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
3137 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3139 * @size: memory size to allocate
3140 * @kind: allocation type
3142 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3143 * rmem allocation. This function assumes that protocols which have
3144 * memory_pressure use sk_wmem_queued as write buffer accounting.
3146 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3148 int ret, amt = sk_mem_pages(size);
3150 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3151 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3153 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3156 EXPORT_SYMBOL(__sk_mem_schedule);
3159 * __sk_mem_reduce_allocated - reclaim memory_allocated
3161 * @amount: number of quanta
3163 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3165 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3167 sk_memory_allocated_sub(sk, amount);
3169 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3170 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3172 if (sk_under_global_memory_pressure(sk) &&
3173 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3174 sk_leave_memory_pressure(sk);
3178 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3180 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3182 void __sk_mem_reclaim(struct sock *sk, int amount)
3184 amount >>= PAGE_SHIFT;
3185 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3186 __sk_mem_reduce_allocated(sk, amount);
3188 EXPORT_SYMBOL(__sk_mem_reclaim);
3190 int sk_set_peek_off(struct sock *sk, int val)
3192 WRITE_ONCE(sk->sk_peek_off, val);
3195 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3198 * Set of default routines for initialising struct proto_ops when
3199 * the protocol does not support a particular function. In certain
3200 * cases where it makes no sense for a protocol to have a "do nothing"
3201 * function, some default processing is provided.
3204 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3208 EXPORT_SYMBOL(sock_no_bind);
3210 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3215 EXPORT_SYMBOL(sock_no_connect);
3217 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3221 EXPORT_SYMBOL(sock_no_socketpair);
3223 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3228 EXPORT_SYMBOL(sock_no_accept);
3230 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3235 EXPORT_SYMBOL(sock_no_getname);
3237 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3241 EXPORT_SYMBOL(sock_no_ioctl);
3243 int sock_no_listen(struct socket *sock, int backlog)
3247 EXPORT_SYMBOL(sock_no_listen);
3249 int sock_no_shutdown(struct socket *sock, int how)
3253 EXPORT_SYMBOL(sock_no_shutdown);
3255 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3259 EXPORT_SYMBOL(sock_no_sendmsg);
3261 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3265 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3267 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3272 EXPORT_SYMBOL(sock_no_recvmsg);
3274 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3276 /* Mirror missing mmap method error code */
3279 EXPORT_SYMBOL(sock_no_mmap);
3282 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3283 * various sock-based usage counts.
3285 void __receive_sock(struct file *file)
3287 struct socket *sock;
3289 sock = sock_from_file(file);
3291 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3292 sock_update_classid(&sock->sk->sk_cgrp_data);
3297 * Default Socket Callbacks
3300 static void sock_def_wakeup(struct sock *sk)
3302 struct socket_wq *wq;
3305 wq = rcu_dereference(sk->sk_wq);
3306 if (skwq_has_sleeper(wq))
3307 wake_up_interruptible_all(&wq->wait);
3311 static void sock_def_error_report(struct sock *sk)
3313 struct socket_wq *wq;
3316 wq = rcu_dereference(sk->sk_wq);
3317 if (skwq_has_sleeper(wq))
3318 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3319 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3323 void sock_def_readable(struct sock *sk)
3325 struct socket_wq *wq;
3327 trace_sk_data_ready(sk);
3330 wq = rcu_dereference(sk->sk_wq);
3331 if (skwq_has_sleeper(wq))
3332 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3333 EPOLLRDNORM | EPOLLRDBAND);
3334 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3338 static void sock_def_write_space(struct sock *sk)
3340 struct socket_wq *wq;
3344 /* Do not wake up a writer until he can make "significant"
3347 if (sock_writeable(sk)) {
3348 wq = rcu_dereference(sk->sk_wq);
3349 if (skwq_has_sleeper(wq))
3350 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3351 EPOLLWRNORM | EPOLLWRBAND);
3353 /* Should agree with poll, otherwise some programs break */
3354 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3360 /* An optimised version of sock_def_write_space(), should only be called
3361 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3364 static void sock_def_write_space_wfree(struct sock *sk)
3366 /* Do not wake up a writer until he can make "significant"
3369 if (sock_writeable(sk)) {
3370 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3372 /* rely on refcount_sub from sock_wfree() */
3373 smp_mb__after_atomic();
3374 if (wq && waitqueue_active(&wq->wait))
3375 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3376 EPOLLWRNORM | EPOLLWRBAND);
3378 /* Should agree with poll, otherwise some programs break */
3379 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3383 static void sock_def_destruct(struct sock *sk)
3387 void sk_send_sigurg(struct sock *sk)
3389 if (sk->sk_socket && sk->sk_socket->file)
3390 if (send_sigurg(&sk->sk_socket->file->f_owner))
3391 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3393 EXPORT_SYMBOL(sk_send_sigurg);
3395 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3396 unsigned long expires)
3398 if (!mod_timer(timer, expires))
3401 EXPORT_SYMBOL(sk_reset_timer);
3403 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3405 if (del_timer(timer))
3408 EXPORT_SYMBOL(sk_stop_timer);
3410 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3412 if (del_timer_sync(timer))
3415 EXPORT_SYMBOL(sk_stop_timer_sync);
3417 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3420 sk->sk_send_head = NULL;
3422 timer_setup(&sk->sk_timer, NULL, 0);
3424 sk->sk_allocation = GFP_KERNEL;
3425 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3426 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3427 sk->sk_state = TCP_CLOSE;
3428 sk->sk_use_task_frag = true;
3429 sk_set_socket(sk, sock);
3431 sock_set_flag(sk, SOCK_ZAPPED);
3434 sk->sk_type = sock->type;
3435 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3438 RCU_INIT_POINTER(sk->sk_wq, NULL);
3442 rwlock_init(&sk->sk_callback_lock);
3443 if (sk->sk_kern_sock)
3444 lockdep_set_class_and_name(
3445 &sk->sk_callback_lock,
3446 af_kern_callback_keys + sk->sk_family,
3447 af_family_kern_clock_key_strings[sk->sk_family]);
3449 lockdep_set_class_and_name(
3450 &sk->sk_callback_lock,
3451 af_callback_keys + sk->sk_family,
3452 af_family_clock_key_strings[sk->sk_family]);
3454 sk->sk_state_change = sock_def_wakeup;
3455 sk->sk_data_ready = sock_def_readable;
3456 sk->sk_write_space = sock_def_write_space;
3457 sk->sk_error_report = sock_def_error_report;
3458 sk->sk_destruct = sock_def_destruct;
3460 sk->sk_frag.page = NULL;
3461 sk->sk_frag.offset = 0;
3462 sk->sk_peek_off = -1;
3464 sk->sk_peer_pid = NULL;
3465 sk->sk_peer_cred = NULL;
3466 spin_lock_init(&sk->sk_peer_lock);
3468 sk->sk_write_pending = 0;
3469 sk->sk_rcvlowat = 1;
3470 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3471 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3473 sk->sk_stamp = SK_DEFAULT_STAMP;
3474 #if BITS_PER_LONG==32
3475 seqlock_init(&sk->sk_stamp_seq);
3477 atomic_set(&sk->sk_zckey, 0);
3479 #ifdef CONFIG_NET_RX_BUSY_POLL
3481 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3484 sk->sk_max_pacing_rate = ~0UL;
3485 sk->sk_pacing_rate = ~0UL;
3486 WRITE_ONCE(sk->sk_pacing_shift, 10);
3487 sk->sk_incoming_cpu = -1;
3489 sk_rx_queue_clear(sk);
3491 * Before updating sk_refcnt, we must commit prior changes to memory
3492 * (Documentation/RCU/rculist_nulls.rst for details)
3495 refcount_set(&sk->sk_refcnt, 1);
3496 atomic_set(&sk->sk_drops, 0);
3498 EXPORT_SYMBOL(sock_init_data_uid);
3500 void sock_init_data(struct socket *sock, struct sock *sk)
3503 SOCK_INODE(sock)->i_uid :
3504 make_kuid(sock_net(sk)->user_ns, 0);
3506 sock_init_data_uid(sock, sk, uid);
3508 EXPORT_SYMBOL(sock_init_data);
3510 void lock_sock_nested(struct sock *sk, int subclass)
3512 /* The sk_lock has mutex_lock() semantics here. */
3513 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3516 spin_lock_bh(&sk->sk_lock.slock);
3517 if (sock_owned_by_user_nocheck(sk))
3519 sk->sk_lock.owned = 1;
3520 spin_unlock_bh(&sk->sk_lock.slock);
3522 EXPORT_SYMBOL(lock_sock_nested);
3524 void release_sock(struct sock *sk)
3526 spin_lock_bh(&sk->sk_lock.slock);
3527 if (sk->sk_backlog.tail)
3530 /* Warning : release_cb() might need to release sk ownership,
3531 * ie call sock_release_ownership(sk) before us.
3533 if (sk->sk_prot->release_cb)
3534 sk->sk_prot->release_cb(sk);
3536 sock_release_ownership(sk);
3537 if (waitqueue_active(&sk->sk_lock.wq))
3538 wake_up(&sk->sk_lock.wq);
3539 spin_unlock_bh(&sk->sk_lock.slock);
3541 EXPORT_SYMBOL(release_sock);
3543 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3546 spin_lock_bh(&sk->sk_lock.slock);
3548 if (!sock_owned_by_user_nocheck(sk)) {
3550 * Fast path return with bottom halves disabled and
3551 * sock::sk_lock.slock held.
3553 * The 'mutex' is not contended and holding
3554 * sock::sk_lock.slock prevents all other lockers to
3555 * proceed so the corresponding unlock_sock_fast() can
3556 * avoid the slow path of release_sock() completely and
3557 * just release slock.
3559 * From a semantical POV this is equivalent to 'acquiring'
3560 * the 'mutex', hence the corresponding lockdep
3561 * mutex_release() has to happen in the fast path of
3562 * unlock_sock_fast().
3568 sk->sk_lock.owned = 1;
3569 __acquire(&sk->sk_lock.slock);
3570 spin_unlock_bh(&sk->sk_lock.slock);
3573 EXPORT_SYMBOL(__lock_sock_fast);
3575 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3576 bool timeval, bool time32)
3578 struct sock *sk = sock->sk;
3579 struct timespec64 ts;
3581 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3582 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3583 if (ts.tv_sec == -1)
3585 if (ts.tv_sec == 0) {
3586 ktime_t kt = ktime_get_real();
3587 sock_write_timestamp(sk, kt);
3588 ts = ktime_to_timespec64(kt);
3594 #ifdef CONFIG_COMPAT_32BIT_TIME
3596 return put_old_timespec32(&ts, userstamp);
3598 #ifdef CONFIG_SPARC64
3599 /* beware of padding in sparc64 timeval */
3600 if (timeval && !in_compat_syscall()) {
3601 struct __kernel_old_timeval __user tv = {
3602 .tv_sec = ts.tv_sec,
3603 .tv_usec = ts.tv_nsec,
3605 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3610 return put_timespec64(&ts, userstamp);
3612 EXPORT_SYMBOL(sock_gettstamp);
3614 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3616 if (!sock_flag(sk, flag)) {
3617 unsigned long previous_flags = sk->sk_flags;
3619 sock_set_flag(sk, flag);
3621 * we just set one of the two flags which require net
3622 * time stamping, but time stamping might have been on
3623 * already because of the other one
3625 if (sock_needs_netstamp(sk) &&
3626 !(previous_flags & SK_FLAGS_TIMESTAMP))
3627 net_enable_timestamp();
3631 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3632 int level, int type)
3634 struct sock_exterr_skb *serr;
3635 struct sk_buff *skb;
3639 skb = sock_dequeue_err_skb(sk);
3645 msg->msg_flags |= MSG_TRUNC;
3648 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3652 sock_recv_timestamp(msg, sk, skb);
3654 serr = SKB_EXT_ERR(skb);
3655 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3657 msg->msg_flags |= MSG_ERRQUEUE;
3665 EXPORT_SYMBOL(sock_recv_errqueue);
3668 * Get a socket option on an socket.
3670 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3671 * asynchronous errors should be reported by getsockopt. We assume
3672 * this means if you specify SO_ERROR (otherwise whats the point of it).
3674 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3675 char __user *optval, int __user *optlen)
3677 struct sock *sk = sock->sk;
3679 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3680 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3682 EXPORT_SYMBOL(sock_common_getsockopt);
3684 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3687 struct sock *sk = sock->sk;
3691 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3693 msg->msg_namelen = addr_len;
3696 EXPORT_SYMBOL(sock_common_recvmsg);
3699 * Set socket options on an inet socket.
3701 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3702 sockptr_t optval, unsigned int optlen)
3704 struct sock *sk = sock->sk;
3706 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3707 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3709 EXPORT_SYMBOL(sock_common_setsockopt);
3711 void sk_common_release(struct sock *sk)
3713 if (sk->sk_prot->destroy)
3714 sk->sk_prot->destroy(sk);
3717 * Observation: when sk_common_release is called, processes have
3718 * no access to socket. But net still has.
3719 * Step one, detach it from networking:
3721 * A. Remove from hash tables.
3724 sk->sk_prot->unhash(sk);
3727 * In this point socket cannot receive new packets, but it is possible
3728 * that some packets are in flight because some CPU runs receiver and
3729 * did hash table lookup before we unhashed socket. They will achieve
3730 * receive queue and will be purged by socket destructor.
3732 * Also we still have packets pending on receive queue and probably,
3733 * our own packets waiting in device queues. sock_destroy will drain
3734 * receive queue, but transmitted packets will delay socket destruction
3735 * until the last reference will be released.
3740 xfrm_sk_free_policy(sk);
3744 EXPORT_SYMBOL(sk_common_release);
3746 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3748 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3750 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3751 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3752 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3753 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3754 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3755 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3756 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3757 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3758 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3761 #ifdef CONFIG_PROC_FS
3762 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3764 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3766 int cpu, idx = prot->inuse_idx;
3769 for_each_possible_cpu(cpu)
3770 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3772 return res >= 0 ? res : 0;
3774 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3776 int sock_inuse_get(struct net *net)
3780 for_each_possible_cpu(cpu)
3781 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3786 EXPORT_SYMBOL_GPL(sock_inuse_get);
3788 static int __net_init sock_inuse_init_net(struct net *net)
3790 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3791 if (net->core.prot_inuse == NULL)
3796 static void __net_exit sock_inuse_exit_net(struct net *net)
3798 free_percpu(net->core.prot_inuse);
3801 static struct pernet_operations net_inuse_ops = {
3802 .init = sock_inuse_init_net,
3803 .exit = sock_inuse_exit_net,
3806 static __init int net_inuse_init(void)
3808 if (register_pernet_subsys(&net_inuse_ops))
3809 panic("Cannot initialize net inuse counters");
3814 core_initcall(net_inuse_init);
3816 static int assign_proto_idx(struct proto *prot)
3818 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3820 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3821 pr_err("PROTO_INUSE_NR exhausted\n");
3825 set_bit(prot->inuse_idx, proto_inuse_idx);
3829 static void release_proto_idx(struct proto *prot)
3831 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3832 clear_bit(prot->inuse_idx, proto_inuse_idx);
3835 static inline int assign_proto_idx(struct proto *prot)
3840 static inline void release_proto_idx(struct proto *prot)
3846 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3850 kfree(twsk_prot->twsk_slab_name);
3851 twsk_prot->twsk_slab_name = NULL;
3852 kmem_cache_destroy(twsk_prot->twsk_slab);
3853 twsk_prot->twsk_slab = NULL;
3856 static int tw_prot_init(const struct proto *prot)
3858 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3863 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3865 if (!twsk_prot->twsk_slab_name)
3868 twsk_prot->twsk_slab =
3869 kmem_cache_create(twsk_prot->twsk_slab_name,
3870 twsk_prot->twsk_obj_size, 0,
3871 SLAB_ACCOUNT | prot->slab_flags,
3873 if (!twsk_prot->twsk_slab) {
3874 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3882 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3886 kfree(rsk_prot->slab_name);
3887 rsk_prot->slab_name = NULL;
3888 kmem_cache_destroy(rsk_prot->slab);
3889 rsk_prot->slab = NULL;
3892 static int req_prot_init(const struct proto *prot)
3894 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3899 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3901 if (!rsk_prot->slab_name)
3904 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3905 rsk_prot->obj_size, 0,
3906 SLAB_ACCOUNT | prot->slab_flags,
3909 if (!rsk_prot->slab) {
3910 pr_crit("%s: Can't create request sock SLAB cache!\n",
3917 int proto_register(struct proto *prot, int alloc_slab)
3921 if (prot->memory_allocated && !prot->sysctl_mem) {
3922 pr_err("%s: missing sysctl_mem\n", prot->name);
3925 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3926 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3930 prot->slab = kmem_cache_create_usercopy(prot->name,
3932 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3934 prot->useroffset, prot->usersize,
3937 if (prot->slab == NULL) {
3938 pr_crit("%s: Can't create sock SLAB cache!\n",
3943 if (req_prot_init(prot))
3944 goto out_free_request_sock_slab;
3946 if (tw_prot_init(prot))
3947 goto out_free_timewait_sock_slab;
3950 mutex_lock(&proto_list_mutex);
3951 ret = assign_proto_idx(prot);
3953 mutex_unlock(&proto_list_mutex);
3954 goto out_free_timewait_sock_slab;
3956 list_add(&prot->node, &proto_list);
3957 mutex_unlock(&proto_list_mutex);
3960 out_free_timewait_sock_slab:
3962 tw_prot_cleanup(prot->twsk_prot);
3963 out_free_request_sock_slab:
3965 req_prot_cleanup(prot->rsk_prot);
3967 kmem_cache_destroy(prot->slab);
3973 EXPORT_SYMBOL(proto_register);
3975 void proto_unregister(struct proto *prot)
3977 mutex_lock(&proto_list_mutex);
3978 release_proto_idx(prot);
3979 list_del(&prot->node);
3980 mutex_unlock(&proto_list_mutex);
3982 kmem_cache_destroy(prot->slab);
3985 req_prot_cleanup(prot->rsk_prot);
3986 tw_prot_cleanup(prot->twsk_prot);
3988 EXPORT_SYMBOL(proto_unregister);
3990 int sock_load_diag_module(int family, int protocol)
3993 if (!sock_is_registered(family))
3996 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3997 NETLINK_SOCK_DIAG, family);
4001 if (family == AF_INET &&
4002 protocol != IPPROTO_RAW &&
4003 protocol < MAX_INET_PROTOS &&
4004 !rcu_access_pointer(inet_protos[protocol]))
4008 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4009 NETLINK_SOCK_DIAG, family, protocol);
4011 EXPORT_SYMBOL(sock_load_diag_module);
4013 #ifdef CONFIG_PROC_FS
4014 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4015 __acquires(proto_list_mutex)
4017 mutex_lock(&proto_list_mutex);
4018 return seq_list_start_head(&proto_list, *pos);
4021 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4023 return seq_list_next(v, &proto_list, pos);
4026 static void proto_seq_stop(struct seq_file *seq, void *v)
4027 __releases(proto_list_mutex)
4029 mutex_unlock(&proto_list_mutex);
4032 static char proto_method_implemented(const void *method)
4034 return method == NULL ? 'n' : 'y';
4036 static long sock_prot_memory_allocated(struct proto *proto)
4038 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4041 static const char *sock_prot_memory_pressure(struct proto *proto)
4043 return proto->memory_pressure != NULL ?
4044 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4047 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4050 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4051 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4054 sock_prot_inuse_get(seq_file_net(seq), proto),
4055 sock_prot_memory_allocated(proto),
4056 sock_prot_memory_pressure(proto),
4058 proto->slab == NULL ? "no" : "yes",
4059 module_name(proto->owner),
4060 proto_method_implemented(proto->close),
4061 proto_method_implemented(proto->connect),
4062 proto_method_implemented(proto->disconnect),
4063 proto_method_implemented(proto->accept),
4064 proto_method_implemented(proto->ioctl),
4065 proto_method_implemented(proto->init),
4066 proto_method_implemented(proto->destroy),
4067 proto_method_implemented(proto->shutdown),
4068 proto_method_implemented(proto->setsockopt),
4069 proto_method_implemented(proto->getsockopt),
4070 proto_method_implemented(proto->sendmsg),
4071 proto_method_implemented(proto->recvmsg),
4072 proto_method_implemented(proto->bind),
4073 proto_method_implemented(proto->backlog_rcv),
4074 proto_method_implemented(proto->hash),
4075 proto_method_implemented(proto->unhash),
4076 proto_method_implemented(proto->get_port),
4077 proto_method_implemented(proto->enter_memory_pressure));
4080 static int proto_seq_show(struct seq_file *seq, void *v)
4082 if (v == &proto_list)
4083 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4092 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4094 proto_seq_printf(seq, list_entry(v, struct proto, node));
4098 static const struct seq_operations proto_seq_ops = {
4099 .start = proto_seq_start,
4100 .next = proto_seq_next,
4101 .stop = proto_seq_stop,
4102 .show = proto_seq_show,
4105 static __net_init int proto_init_net(struct net *net)
4107 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4108 sizeof(struct seq_net_private)))
4114 static __net_exit void proto_exit_net(struct net *net)
4116 remove_proc_entry("protocols", net->proc_net);
4120 static __net_initdata struct pernet_operations proto_net_ops = {
4121 .init = proto_init_net,
4122 .exit = proto_exit_net,
4125 static int __init proto_init(void)
4127 return register_pernet_subsys(&proto_net_ops);
4130 subsys_initcall(proto_init);
4132 #endif /* PROC_FS */
4134 #ifdef CONFIG_NET_RX_BUSY_POLL
4135 bool sk_busy_loop_end(void *p, unsigned long start_time)
4137 struct sock *sk = p;
4139 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
4140 sk_busy_loop_timeout(sk, start_time);
4142 EXPORT_SYMBOL(sk_busy_loop_end);
4143 #endif /* CONFIG_NET_RX_BUSY_POLL */
4145 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4147 if (!sk->sk_prot->bind_add)
4149 return sk->sk_prot->bind_add(sk, addr, addr_len);
4151 EXPORT_SYMBOL(sock_bind_add);
4153 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4154 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4155 void __user *arg, void *karg, size_t size)
4159 if (copy_from_user(karg, arg, size))
4162 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4166 if (copy_to_user(arg, karg, size))
4171 EXPORT_SYMBOL(sock_ioctl_inout);
4173 /* This is the most common ioctl prep function, where the result (4 bytes) is
4174 * copied back to userspace if the ioctl() returns successfully. No input is
4175 * copied from userspace as input argument.
4177 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4181 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4185 return put_user(karg, (int __user *)arg);
4188 /* A wrapper around sock ioctls, which copies the data from userspace
4189 * (depending on the protocol/ioctl), and copies back the result to userspace.
4190 * The main motivation for this function is to pass kernel memory to the
4191 * protocol ioctl callbacks, instead of userspace memory.
4193 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4197 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4198 rc = ipmr_sk_ioctl(sk, cmd, arg);
4199 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4200 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4201 else if (sk_is_phonet(sk))
4202 rc = phonet_sk_ioctl(sk, cmd, arg);
4204 /* If ioctl was processed, returns its value */
4208 /* Otherwise call the default handler */
4209 return sock_ioctl_out(sk, cmd, arg);
4211 EXPORT_SYMBOL(sk_ioctl);