1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <linux/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/udp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
114 #include <linux/static_key.h>
115 #include <linux/memcontrol.h>
116 #include <linux/prefetch.h>
117 #include <linux/compat.h>
118 #include <linux/mroute.h>
119 #include <linux/mroute6.h>
120 #include <linux/icmpv6.h>
122 #include <linux/uaccess.h>
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <linux/skbuff_ref.h>
128 #include <net/net_namespace.h>
129 #include <net/request_sock.h>
130 #include <net/sock.h>
131 #include <net/proto_memory.h>
132 #include <linux/net_tstamp.h>
133 #include <net/xfrm.h>
134 #include <linux/ipsec.h>
135 #include <net/cls_cgroup.h>
136 #include <net/netprio_cgroup.h>
137 #include <linux/sock_diag.h>
139 #include <linux/filter.h>
140 #include <net/sock_reuseport.h>
141 #include <net/bpf_sk_storage.h>
143 #include <trace/events/sock.h>
146 #include <net/busy_poll.h>
147 #include <net/phonet/phonet.h>
149 #include <linux/ethtool.h>
153 static DEFINE_MUTEX(proto_list_mutex);
154 static LIST_HEAD(proto_list);
156 static void sock_def_write_space_wfree(struct sock *sk);
157 static void sock_def_write_space(struct sock *sk);
160 * sk_ns_capable - General socket capability test
161 * @sk: Socket to use a capability on or through
162 * @user_ns: The user namespace of the capability to use
163 * @cap: The capability to use
165 * Test to see if the opener of the socket had when the socket was
166 * created and the current process has the capability @cap in the user
167 * namespace @user_ns.
169 bool sk_ns_capable(const struct sock *sk,
170 struct user_namespace *user_ns, int cap)
172 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
173 ns_capable(user_ns, cap);
175 EXPORT_SYMBOL(sk_ns_capable);
178 * sk_capable - Socket global capability test
179 * @sk: Socket to use a capability on or through
180 * @cap: The global capability to use
182 * Test to see if the opener of the socket had when the socket was
183 * created and the current process has the capability @cap in all user
186 bool sk_capable(const struct sock *sk, int cap)
188 return sk_ns_capable(sk, &init_user_ns, cap);
190 EXPORT_SYMBOL(sk_capable);
193 * sk_net_capable - Network namespace socket capability test
194 * @sk: Socket to use a capability on or through
195 * @cap: The capability to use
197 * Test to see if the opener of the socket had when the socket was created
198 * and the current process has the capability @cap over the network namespace
199 * the socket is a member of.
201 bool sk_net_capable(const struct sock *sk, int cap)
203 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
205 EXPORT_SYMBOL(sk_net_capable);
208 * Each address family might have different locking rules, so we have
209 * one slock key per address family and separate keys for internal and
212 static struct lock_class_key af_family_keys[AF_MAX];
213 static struct lock_class_key af_family_kern_keys[AF_MAX];
214 static struct lock_class_key af_family_slock_keys[AF_MAX];
215 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
218 * Make lock validator output more readable. (we pre-construct these
219 * strings build-time, so that runtime initialization of socket
223 #define _sock_locks(x) \
224 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
225 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
226 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
227 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
228 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
229 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
230 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
231 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
232 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
233 x "27" , x "28" , x "AF_CAN" , \
234 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
235 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
236 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
237 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
238 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
242 static const char *const af_family_key_strings[AF_MAX+1] = {
243 _sock_locks("sk_lock-")
245 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
246 _sock_locks("slock-")
248 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
249 _sock_locks("clock-")
252 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
253 _sock_locks("k-sk_lock-")
255 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
256 _sock_locks("k-slock-")
258 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
259 _sock_locks("k-clock-")
261 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
262 _sock_locks("rlock-")
264 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
265 _sock_locks("wlock-")
267 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
268 _sock_locks("elock-")
272 * sk_callback_lock and sk queues locking rules are per-address-family,
273 * so split the lock classes by using a per-AF key:
275 static struct lock_class_key af_callback_keys[AF_MAX];
276 static struct lock_class_key af_rlock_keys[AF_MAX];
277 static struct lock_class_key af_wlock_keys[AF_MAX];
278 static struct lock_class_key af_elock_keys[AF_MAX];
279 static struct lock_class_key af_kern_callback_keys[AF_MAX];
281 /* Run time adjustable parameters. */
282 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
283 EXPORT_SYMBOL(sysctl_wmem_max);
284 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
285 EXPORT_SYMBOL(sysctl_rmem_max);
286 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
287 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
289 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
290 EXPORT_SYMBOL_GPL(memalloc_socks_key);
293 * sk_set_memalloc - sets %SOCK_MEMALLOC
294 * @sk: socket to set it on
296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
297 * It's the responsibility of the admin to adjust min_free_kbytes
298 * to meet the requirements
300 void sk_set_memalloc(struct sock *sk)
302 sock_set_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation |= __GFP_MEMALLOC;
304 static_branch_inc(&memalloc_socks_key);
306 EXPORT_SYMBOL_GPL(sk_set_memalloc);
308 void sk_clear_memalloc(struct sock *sk)
310 sock_reset_flag(sk, SOCK_MEMALLOC);
311 sk->sk_allocation &= ~__GFP_MEMALLOC;
312 static_branch_dec(&memalloc_socks_key);
315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
316 * progress of swapping. SOCK_MEMALLOC may be cleared while
317 * it has rmem allocations due to the last swapfile being deactivated
318 * but there is a risk that the socket is unusable due to exceeding
319 * the rmem limits. Reclaim the reserves and obey rmem limits again.
323 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
325 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
328 unsigned int noreclaim_flag;
330 /* these should have been dropped before queueing */
331 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
333 noreclaim_flag = memalloc_noreclaim_save();
334 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
338 memalloc_noreclaim_restore(noreclaim_flag);
342 EXPORT_SYMBOL(__sk_backlog_rcv);
344 void sk_error_report(struct sock *sk)
346 sk->sk_error_report(sk);
348 switch (sk->sk_family) {
352 trace_inet_sk_error_report(sk);
358 EXPORT_SYMBOL(sk_error_report);
360 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
362 struct __kernel_sock_timeval tv;
364 if (timeo == MAX_SCHEDULE_TIMEOUT) {
368 tv.tv_sec = timeo / HZ;
369 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
372 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
373 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
374 *(struct old_timeval32 *)optval = tv32;
379 struct __kernel_old_timeval old_tv;
380 old_tv.tv_sec = tv.tv_sec;
381 old_tv.tv_usec = tv.tv_usec;
382 *(struct __kernel_old_timeval *)optval = old_tv;
383 return sizeof(old_tv);
386 *(struct __kernel_sock_timeval *)optval = tv;
389 EXPORT_SYMBOL(sock_get_timeout);
391 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
392 sockptr_t optval, int optlen, bool old_timeval)
394 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
395 struct old_timeval32 tv32;
397 if (optlen < sizeof(tv32))
400 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
402 tv->tv_sec = tv32.tv_sec;
403 tv->tv_usec = tv32.tv_usec;
404 } else if (old_timeval) {
405 struct __kernel_old_timeval old_tv;
407 if (optlen < sizeof(old_tv))
409 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
411 tv->tv_sec = old_tv.tv_sec;
412 tv->tv_usec = old_tv.tv_usec;
414 if (optlen < sizeof(*tv))
416 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
422 EXPORT_SYMBOL(sock_copy_user_timeval);
424 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
427 struct __kernel_sock_timeval tv;
428 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
434 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
438 static int warned __read_mostly;
440 WRITE_ONCE(*timeo_p, 0);
441 if (warned < 10 && net_ratelimit()) {
443 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
444 __func__, current->comm, task_pid_nr(current));
448 val = MAX_SCHEDULE_TIMEOUT;
449 if ((tv.tv_sec || tv.tv_usec) &&
450 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
451 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
453 WRITE_ONCE(*timeo_p, val);
457 static bool sock_needs_netstamp(const struct sock *sk)
459 switch (sk->sk_family) {
468 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
470 if (sk->sk_flags & flags) {
471 sk->sk_flags &= ~flags;
472 if (sock_needs_netstamp(sk) &&
473 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
474 net_disable_timestamp();
479 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
482 struct sk_buff_head *list = &sk->sk_receive_queue;
484 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) {
485 atomic_inc(&sk->sk_drops);
486 trace_sock_rcvqueue_full(sk, skb);
490 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
491 atomic_inc(&sk->sk_drops);
496 skb_set_owner_r(skb, sk);
498 /* we escape from rcu protected region, make sure we dont leak
503 spin_lock_irqsave(&list->lock, flags);
504 sock_skb_set_dropcount(sk, skb);
505 __skb_queue_tail(list, skb);
506 spin_unlock_irqrestore(&list->lock, flags);
508 if (!sock_flag(sk, SOCK_DEAD))
509 sk->sk_data_ready(sk);
512 EXPORT_SYMBOL(__sock_queue_rcv_skb);
514 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
515 enum skb_drop_reason *reason)
517 enum skb_drop_reason drop_reason;
520 err = sk_filter(sk, skb);
522 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
525 err = __sock_queue_rcv_skb(sk, skb);
528 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
531 drop_reason = SKB_DROP_REASON_PROTO_MEM;
534 drop_reason = SKB_NOT_DROPPED_YET;
539 *reason = drop_reason;
542 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
544 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
545 const int nested, unsigned int trim_cap, bool refcounted)
547 int rc = NET_RX_SUCCESS;
549 if (sk_filter_trim_cap(sk, skb, trim_cap))
550 goto discard_and_relse;
554 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) {
555 atomic_inc(&sk->sk_drops);
556 goto discard_and_relse;
559 bh_lock_sock_nested(sk);
562 if (!sock_owned_by_user(sk)) {
564 * trylock + unlock semantics:
566 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
568 rc = sk_backlog_rcv(sk, skb);
570 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
571 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
573 atomic_inc(&sk->sk_drops);
574 goto discard_and_relse;
586 EXPORT_SYMBOL(__sk_receive_skb);
588 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
590 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
592 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
594 struct dst_entry *dst = __sk_dst_get(sk);
596 if (dst && dst->obsolete &&
597 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
598 dst, cookie) == NULL) {
599 sk_tx_queue_clear(sk);
600 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
601 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
608 EXPORT_SYMBOL(__sk_dst_check);
610 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
612 struct dst_entry *dst = sk_dst_get(sk);
614 if (dst && dst->obsolete &&
615 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
616 dst, cookie) == NULL) {
624 EXPORT_SYMBOL(sk_dst_check);
626 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
628 int ret = -ENOPROTOOPT;
629 #ifdef CONFIG_NETDEVICES
630 struct net *net = sock_net(sk);
634 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
641 /* Paired with all READ_ONCE() done locklessly. */
642 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
644 if (sk->sk_prot->rehash)
645 sk->sk_prot->rehash(sk);
656 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
662 ret = sock_bindtoindex_locked(sk, ifindex);
668 EXPORT_SYMBOL(sock_bindtoindex);
670 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
672 int ret = -ENOPROTOOPT;
673 #ifdef CONFIG_NETDEVICES
674 struct net *net = sock_net(sk);
675 char devname[IFNAMSIZ];
682 /* Bind this socket to a particular device like "eth0",
683 * as specified in the passed interface name. If the
684 * name is "" or the option length is zero the socket
687 if (optlen > IFNAMSIZ - 1)
688 optlen = IFNAMSIZ - 1;
689 memset(devname, 0, sizeof(devname));
692 if (copy_from_sockptr(devname, optval, optlen))
696 if (devname[0] != '\0') {
697 struct net_device *dev;
700 dev = dev_get_by_name_rcu(net, devname);
702 index = dev->ifindex;
709 sockopt_lock_sock(sk);
710 ret = sock_bindtoindex_locked(sk, index);
711 sockopt_release_sock(sk);
718 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
719 sockptr_t optlen, int len)
721 int ret = -ENOPROTOOPT;
722 #ifdef CONFIG_NETDEVICES
723 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
724 struct net *net = sock_net(sk);
725 char devname[IFNAMSIZ];
727 if (bound_dev_if == 0) {
736 ret = netdev_get_name(net, devname, bound_dev_if);
740 len = strlen(devname) + 1;
743 if (copy_to_sockptr(optval, devname, len))
748 if (copy_to_sockptr(optlen, &len, sizeof(int)))
759 bool sk_mc_loop(const struct sock *sk)
761 if (dev_recursion_level())
765 /* IPV6_ADDRFORM can change sk->sk_family under us. */
766 switch (READ_ONCE(sk->sk_family)) {
768 return inet_test_bit(MC_LOOP, sk);
769 #if IS_ENABLED(CONFIG_IPV6)
771 return inet6_test_bit(MC6_LOOP, sk);
777 EXPORT_SYMBOL(sk_mc_loop);
779 void sock_set_reuseaddr(struct sock *sk)
782 sk->sk_reuse = SK_CAN_REUSE;
785 EXPORT_SYMBOL(sock_set_reuseaddr);
787 void sock_set_reuseport(struct sock *sk)
790 sk->sk_reuseport = true;
793 EXPORT_SYMBOL(sock_set_reuseport);
795 void sock_no_linger(struct sock *sk)
798 WRITE_ONCE(sk->sk_lingertime, 0);
799 sock_set_flag(sk, SOCK_LINGER);
802 EXPORT_SYMBOL(sock_no_linger);
804 void sock_set_priority(struct sock *sk, u32 priority)
806 WRITE_ONCE(sk->sk_priority, priority);
808 EXPORT_SYMBOL(sock_set_priority);
810 void sock_set_sndtimeo(struct sock *sk, s64 secs)
813 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
814 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
816 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
819 EXPORT_SYMBOL(sock_set_sndtimeo);
821 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
823 sock_valbool_flag(sk, SOCK_RCVTSTAMP, val);
824 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns);
826 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
827 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
831 void sock_enable_timestamps(struct sock *sk)
834 __sock_set_timestamps(sk, true, false, true);
837 EXPORT_SYMBOL(sock_enable_timestamps);
839 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
842 case SO_TIMESTAMP_OLD:
843 __sock_set_timestamps(sk, valbool, false, false);
845 case SO_TIMESTAMP_NEW:
846 __sock_set_timestamps(sk, valbool, true, false);
848 case SO_TIMESTAMPNS_OLD:
849 __sock_set_timestamps(sk, valbool, false, true);
851 case SO_TIMESTAMPNS_NEW:
852 __sock_set_timestamps(sk, valbool, true, true);
857 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
859 struct net *net = sock_net(sk);
860 struct net_device *dev = NULL;
865 if (sk->sk_bound_dev_if)
866 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
869 pr_err("%s: sock not bind to device\n", __func__);
873 num = ethtool_get_phc_vclocks(dev, &vclock_index);
876 for (i = 0; i < num; i++) {
877 if (*(vclock_index + i) == phc_index) {
889 WRITE_ONCE(sk->sk_bind_phc, phc_index);
894 int sock_set_timestamping(struct sock *sk, int optname,
895 struct so_timestamping timestamping)
897 int val = timestamping.flags;
900 if (val & ~SOF_TIMESTAMPING_MASK)
903 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
904 !(val & SOF_TIMESTAMPING_OPT_ID))
907 if (val & SOF_TIMESTAMPING_OPT_ID &&
908 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
910 if ((1 << sk->sk_state) &
911 (TCPF_CLOSE | TCPF_LISTEN))
913 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
914 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
916 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
918 atomic_set(&sk->sk_tskey, 0);
922 if (val & SOF_TIMESTAMPING_OPT_STATS &&
923 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
926 if (val & SOF_TIMESTAMPING_BIND_PHC) {
927 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
932 WRITE_ONCE(sk->sk_tsflags, val);
933 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
935 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
936 sock_enable_timestamp(sk,
937 SOCK_TIMESTAMPING_RX_SOFTWARE);
939 sock_disable_timestamp(sk,
940 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
944 void sock_set_keepalive(struct sock *sk)
947 if (sk->sk_prot->keepalive)
948 sk->sk_prot->keepalive(sk, true);
949 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
952 EXPORT_SYMBOL(sock_set_keepalive);
954 static void __sock_set_rcvbuf(struct sock *sk, int val)
956 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
957 * as a negative value.
959 val = min_t(int, val, INT_MAX / 2);
960 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
962 /* We double it on the way in to account for "struct sk_buff" etc.
963 * overhead. Applications assume that the SO_RCVBUF setting they make
964 * will allow that much actual data to be received on that socket.
966 * Applications are unaware that "struct sk_buff" and other overheads
967 * allocate from the receive buffer during socket buffer allocation.
969 * And after considering the possible alternatives, returning the value
970 * we actually used in getsockopt is the most desirable behavior.
972 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
975 void sock_set_rcvbuf(struct sock *sk, int val)
978 __sock_set_rcvbuf(sk, val);
981 EXPORT_SYMBOL(sock_set_rcvbuf);
983 static void __sock_set_mark(struct sock *sk, u32 val)
985 if (val != sk->sk_mark) {
986 WRITE_ONCE(sk->sk_mark, val);
991 void sock_set_mark(struct sock *sk, u32 val)
994 __sock_set_mark(sk, val);
997 EXPORT_SYMBOL(sock_set_mark);
999 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1001 /* Round down bytes to multiple of pages */
1002 bytes = round_down(bytes, PAGE_SIZE);
1004 WARN_ON(bytes > sk->sk_reserved_mem);
1005 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1009 static int sock_reserve_memory(struct sock *sk, int bytes)
1015 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1021 pages = sk_mem_pages(bytes);
1023 /* pre-charge to memcg */
1024 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1025 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1029 /* pre-charge to forward_alloc */
1030 sk_memory_allocated_add(sk, pages);
1031 allocated = sk_memory_allocated(sk);
1032 /* If the system goes into memory pressure with this
1033 * precharge, give up and return error.
1035 if (allocated > sk_prot_mem_limits(sk, 1)) {
1036 sk_memory_allocated_sub(sk, pages);
1037 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1040 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1042 WRITE_ONCE(sk->sk_reserved_mem,
1043 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1048 #ifdef CONFIG_PAGE_POOL
1050 /* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED
1051 * in 1 syscall. The limit exists to limit the amount of memory the kernel
1052 * allocates to copy these tokens, and to prevent looping over the frags for
1055 #define MAX_DONTNEED_TOKENS 128
1056 #define MAX_DONTNEED_FRAGS 1024
1058 static noinline_for_stack int
1059 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen)
1061 unsigned int num_tokens, i, j, k, netmem_num = 0;
1062 struct dmabuf_token *tokens;
1063 int ret = 0, num_frags = 0;
1064 netmem_ref netmems[16];
1069 if (optlen % sizeof(*tokens) ||
1070 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS)
1073 num_tokens = optlen / sizeof(*tokens);
1074 tokens = kvmalloc_array(num_tokens, sizeof(*tokens), GFP_KERNEL);
1078 if (copy_from_sockptr(tokens, optval, optlen)) {
1083 xa_lock_bh(&sk->sk_user_frags);
1084 for (i = 0; i < num_tokens; i++) {
1085 for (j = 0; j < tokens[i].token_count; j++) {
1086 if (++num_frags > MAX_DONTNEED_FRAGS)
1087 goto frag_limit_reached;
1089 netmem_ref netmem = (__force netmem_ref)__xa_erase(
1090 &sk->sk_user_frags, tokens[i].token_start + j);
1092 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem)))
1095 netmems[netmem_num++] = netmem;
1096 if (netmem_num == ARRAY_SIZE(netmems)) {
1097 xa_unlock_bh(&sk->sk_user_frags);
1098 for (k = 0; k < netmem_num; k++)
1099 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1101 xa_lock_bh(&sk->sk_user_frags);
1108 xa_unlock_bh(&sk->sk_user_frags);
1109 for (k = 0; k < netmem_num; k++)
1110 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1117 void sockopt_lock_sock(struct sock *sk)
1119 /* When current->bpf_ctx is set, the setsockopt is called from
1120 * a bpf prog. bpf has ensured the sk lock has been
1121 * acquired before calling setsockopt().
1123 if (has_current_bpf_ctx())
1128 EXPORT_SYMBOL(sockopt_lock_sock);
1130 void sockopt_release_sock(struct sock *sk)
1132 if (has_current_bpf_ctx())
1137 EXPORT_SYMBOL(sockopt_release_sock);
1139 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1141 return has_current_bpf_ctx() || ns_capable(ns, cap);
1143 EXPORT_SYMBOL(sockopt_ns_capable);
1145 bool sockopt_capable(int cap)
1147 return has_current_bpf_ctx() || capable(cap);
1149 EXPORT_SYMBOL(sockopt_capable);
1151 static int sockopt_validate_clockid(__kernel_clockid_t value)
1154 case CLOCK_REALTIME:
1155 case CLOCK_MONOTONIC:
1163 * This is meant for all protocols to use and covers goings on
1164 * at the socket level. Everything here is generic.
1167 int sk_setsockopt(struct sock *sk, int level, int optname,
1168 sockptr_t optval, unsigned int optlen)
1170 struct so_timestamping timestamping;
1171 struct socket *sock = sk->sk_socket;
1172 struct sock_txtime sk_txtime;
1179 * Options without arguments
1182 if (optname == SO_BINDTODEVICE)
1183 return sock_setbindtodevice(sk, optval, optlen);
1185 if (optlen < sizeof(int))
1188 if (copy_from_sockptr(&val, optval, sizeof(val)))
1191 valbool = val ? 1 : 0;
1193 /* handle options which do not require locking the socket. */
1196 if ((val >= 0 && val <= 6) ||
1197 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1198 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1199 sock_set_priority(sk, val);
1204 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1207 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1210 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1216 return -ENOPROTOOPT;
1217 #ifdef CONFIG_NET_RX_BUSY_POLL
1221 WRITE_ONCE(sk->sk_ll_usec, val);
1223 case SO_PREFER_BUSY_POLL:
1224 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1226 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1228 case SO_BUSY_POLL_BUDGET:
1229 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1230 !sockopt_capable(CAP_NET_ADMIN))
1232 if (val < 0 || val > U16_MAX)
1234 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1237 case SO_MAX_PACING_RATE:
1239 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1240 unsigned long pacing_rate;
1242 if (sizeof(ulval) != sizeof(val) &&
1243 optlen >= sizeof(ulval) &&
1244 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1248 cmpxchg(&sk->sk_pacing_status,
1251 /* Pairs with READ_ONCE() from sk_getsockopt() */
1252 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1253 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1254 if (ulval < pacing_rate)
1255 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1259 if (val < -1 || val > 1)
1261 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1262 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1263 /* Paired with READ_ONCE() in tcp_rtx_synack()
1264 * and sk_getsockopt().
1266 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1270 int (*set_peek_off)(struct sock *sk, int val);
1272 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1274 ret = set_peek_off(sk, val);
1279 #ifdef CONFIG_PAGE_POOL
1280 case SO_DEVMEM_DONTNEED:
1281 return sock_devmem_dontneed(sk, optval, optlen);
1285 sockopt_lock_sock(sk);
1289 if (val && !sockopt_capable(CAP_NET_ADMIN))
1292 sock_valbool_flag(sk, SOCK_DBG, valbool);
1295 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1298 if (valbool && !sk_is_inet(sk))
1301 sk->sk_reuseport = valbool;
1304 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1308 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1311 /* Don't error on this BSD doesn't and if you think
1312 * about it this is right. Otherwise apps have to
1313 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1314 * are treated in BSD as hints
1316 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1318 /* Ensure val * 2 fits into an int, to prevent max_t()
1319 * from treating it as a negative value.
1321 val = min_t(int, val, INT_MAX / 2);
1322 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1323 WRITE_ONCE(sk->sk_sndbuf,
1324 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1325 /* Wake up sending tasks if we upped the value. */
1326 sk->sk_write_space(sk);
1329 case SO_SNDBUFFORCE:
1330 if (!sockopt_capable(CAP_NET_ADMIN)) {
1335 /* No negative values (to prevent underflow, as val will be
1343 /* Don't error on this BSD doesn't and if you think
1344 * about it this is right. Otherwise apps have to
1345 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1346 * are treated in BSD as hints
1348 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1351 case SO_RCVBUFFORCE:
1352 if (!sockopt_capable(CAP_NET_ADMIN)) {
1357 /* No negative values (to prevent underflow, as val will be
1360 __sock_set_rcvbuf(sk, max(val, 0));
1364 if (sk->sk_prot->keepalive)
1365 sk->sk_prot->keepalive(sk, valbool);
1366 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1370 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1374 sk->sk_no_check_tx = valbool;
1378 if (optlen < sizeof(ling)) {
1379 ret = -EINVAL; /* 1003.1g */
1382 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1386 if (!ling.l_onoff) {
1387 sock_reset_flag(sk, SOCK_LINGER);
1389 unsigned long t_sec = ling.l_linger;
1391 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1392 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1394 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1395 sock_set_flag(sk, SOCK_LINGER);
1402 case SO_TIMESTAMP_OLD:
1403 case SO_TIMESTAMP_NEW:
1404 case SO_TIMESTAMPNS_OLD:
1405 case SO_TIMESTAMPNS_NEW:
1406 sock_set_timestamp(sk, optname, valbool);
1409 case SO_TIMESTAMPING_NEW:
1410 case SO_TIMESTAMPING_OLD:
1411 if (optlen == sizeof(timestamping)) {
1412 if (copy_from_sockptr(×tamping, optval,
1413 sizeof(timestamping))) {
1418 memset(×tamping, 0, sizeof(timestamping));
1419 timestamping.flags = val;
1421 ret = sock_set_timestamping(sk, optname, timestamping);
1426 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1431 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1433 ret = set_rcvlowat(sk, val);
1435 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1438 case SO_RCVTIMEO_OLD:
1439 case SO_RCVTIMEO_NEW:
1440 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1441 optlen, optname == SO_RCVTIMEO_OLD);
1444 case SO_SNDTIMEO_OLD:
1445 case SO_SNDTIMEO_NEW:
1446 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1447 optlen, optname == SO_SNDTIMEO_OLD);
1450 case SO_ATTACH_FILTER: {
1451 struct sock_fprog fprog;
1453 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1455 ret = sk_attach_filter(&fprog, sk);
1460 if (optlen == sizeof(u32)) {
1464 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1467 ret = sk_attach_bpf(ufd, sk);
1471 case SO_ATTACH_REUSEPORT_CBPF: {
1472 struct sock_fprog fprog;
1474 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1476 ret = sk_reuseport_attach_filter(&fprog, sk);
1479 case SO_ATTACH_REUSEPORT_EBPF:
1481 if (optlen == sizeof(u32)) {
1485 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1488 ret = sk_reuseport_attach_bpf(ufd, sk);
1492 case SO_DETACH_REUSEPORT_BPF:
1493 ret = reuseport_detach_prog(sk);
1496 case SO_DETACH_FILTER:
1497 ret = sk_detach_filter(sk);
1500 case SO_LOCK_FILTER:
1501 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1504 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1508 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1509 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1514 __sock_set_mark(sk, val);
1517 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1521 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1524 case SO_WIFI_STATUS:
1525 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1529 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1532 case SO_SELECT_ERR_QUEUE:
1533 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1537 case SO_INCOMING_CPU:
1538 reuseport_update_incoming_cpu(sk, val);
1543 dst_negative_advice(sk);
1547 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1548 if (!(sk_is_tcp(sk) ||
1549 (sk->sk_type == SOCK_DGRAM &&
1550 sk->sk_protocol == IPPROTO_UDP)))
1552 } else if (sk->sk_family != PF_RDS) {
1556 if (val < 0 || val > 1)
1559 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1564 if (optlen != sizeof(struct sock_txtime)) {
1567 } else if (copy_from_sockptr(&sk_txtime, optval,
1568 sizeof(struct sock_txtime))) {
1571 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1575 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1576 * scheduler has enough safe guards.
1578 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1579 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1584 ret = sockopt_validate_clockid(sk_txtime.clockid);
1588 sock_valbool_flag(sk, SOCK_TXTIME, true);
1589 sk->sk_clockid = sk_txtime.clockid;
1590 sk->sk_txtime_deadline_mode =
1591 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1592 sk->sk_txtime_report_errors =
1593 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1596 case SO_BINDTOIFINDEX:
1597 ret = sock_bindtoindex_locked(sk, val);
1601 if (val & ~SOCK_BUF_LOCK_MASK) {
1605 sk->sk_userlocks = val | (sk->sk_userlocks &
1606 ~SOCK_BUF_LOCK_MASK);
1609 case SO_RESERVE_MEM:
1618 delta = val - sk->sk_reserved_mem;
1620 sock_release_reserved_memory(sk, -delta);
1622 ret = sock_reserve_memory(sk, delta);
1630 sockopt_release_sock(sk);
1634 int sock_setsockopt(struct socket *sock, int level, int optname,
1635 sockptr_t optval, unsigned int optlen)
1637 return sk_setsockopt(sock->sk, level, optname,
1640 EXPORT_SYMBOL(sock_setsockopt);
1642 static const struct cred *sk_get_peer_cred(struct sock *sk)
1644 const struct cred *cred;
1646 spin_lock(&sk->sk_peer_lock);
1647 cred = get_cred(sk->sk_peer_cred);
1648 spin_unlock(&sk->sk_peer_lock);
1653 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1654 struct ucred *ucred)
1656 ucred->pid = pid_vnr(pid);
1657 ucred->uid = ucred->gid = -1;
1659 struct user_namespace *current_ns = current_user_ns();
1661 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1662 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1666 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1668 struct user_namespace *user_ns = current_user_ns();
1671 for (i = 0; i < src->ngroups; i++) {
1672 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1674 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1681 int sk_getsockopt(struct sock *sk, int level, int optname,
1682 sockptr_t optval, sockptr_t optlen)
1684 struct socket *sock = sk->sk_socket;
1689 unsigned long ulval;
1691 struct old_timeval32 tm32;
1692 struct __kernel_old_timeval tm;
1693 struct __kernel_sock_timeval stm;
1694 struct sock_txtime txtime;
1695 struct so_timestamping timestamping;
1698 int lv = sizeof(int);
1701 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1706 memset(&v, 0, sizeof(v));
1710 v.val = sock_flag(sk, SOCK_DBG);
1714 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1718 v.val = sock_flag(sk, SOCK_BROADCAST);
1722 v.val = READ_ONCE(sk->sk_sndbuf);
1726 v.val = READ_ONCE(sk->sk_rcvbuf);
1730 v.val = sk->sk_reuse;
1734 v.val = sk->sk_reuseport;
1738 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1742 v.val = sk->sk_type;
1746 v.val = sk->sk_protocol;
1750 v.val = sk->sk_family;
1754 v.val = -sock_error(sk);
1756 v.val = xchg(&sk->sk_err_soft, 0);
1760 v.val = sock_flag(sk, SOCK_URGINLINE);
1764 v.val = sk->sk_no_check_tx;
1768 v.val = READ_ONCE(sk->sk_priority);
1772 lv = sizeof(v.ling);
1773 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1774 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1780 case SO_TIMESTAMP_OLD:
1781 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1782 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1783 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1786 case SO_TIMESTAMPNS_OLD:
1787 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1790 case SO_TIMESTAMP_NEW:
1791 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1794 case SO_TIMESTAMPNS_NEW:
1795 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1798 case SO_TIMESTAMPING_OLD:
1799 case SO_TIMESTAMPING_NEW:
1800 lv = sizeof(v.timestamping);
1801 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1802 * returning the flags when they were set through the same option.
1803 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1805 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1806 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1807 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1811 case SO_RCVTIMEO_OLD:
1812 case SO_RCVTIMEO_NEW:
1813 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1814 SO_RCVTIMEO_OLD == optname);
1817 case SO_SNDTIMEO_OLD:
1818 case SO_SNDTIMEO_NEW:
1819 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1820 SO_SNDTIMEO_OLD == optname);
1824 v.val = READ_ONCE(sk->sk_rcvlowat);
1832 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1836 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1841 struct ucred peercred;
1842 if (len > sizeof(peercred))
1843 len = sizeof(peercred);
1845 spin_lock(&sk->sk_peer_lock);
1846 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1847 spin_unlock(&sk->sk_peer_lock);
1849 if (copy_to_sockptr(optval, &peercred, len))
1856 struct pid *peer_pid;
1857 struct file *pidfd_file = NULL;
1860 if (len > sizeof(pidfd))
1861 len = sizeof(pidfd);
1863 spin_lock(&sk->sk_peer_lock);
1864 peer_pid = get_pid(sk->sk_peer_pid);
1865 spin_unlock(&sk->sk_peer_lock);
1870 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1875 if (copy_to_sockptr(optval, &pidfd, len) ||
1876 copy_to_sockptr(optlen, &len, sizeof(int))) {
1877 put_unused_fd(pidfd);
1883 fd_install(pidfd, pidfd_file);
1889 const struct cred *cred;
1892 cred = sk_get_peer_cred(sk);
1896 n = cred->group_info->ngroups;
1897 if (len < n * sizeof(gid_t)) {
1898 len = n * sizeof(gid_t);
1900 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1902 len = n * sizeof(gid_t);
1904 ret = groups_to_user(optval, cred->group_info);
1913 struct sockaddr_storage address;
1915 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1920 if (copy_to_sockptr(optval, &address, len))
1925 /* Dubious BSD thing... Probably nobody even uses it, but
1926 * the UNIX standard wants it for whatever reason... -DaveM
1929 v.val = sk->sk_state == TCP_LISTEN;
1933 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1937 return security_socket_getpeersec_stream(sock,
1938 optval, optlen, len);
1941 v.val = READ_ONCE(sk->sk_mark);
1945 v.val = sock_flag(sk, SOCK_RCVMARK);
1949 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1952 case SO_WIFI_STATUS:
1953 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1957 if (!READ_ONCE(sock->ops)->set_peek_off)
1960 v.val = READ_ONCE(sk->sk_peek_off);
1963 v.val = sock_flag(sk, SOCK_NOFCS);
1966 case SO_BINDTODEVICE:
1967 return sock_getbindtodevice(sk, optval, optlen, len);
1970 len = sk_get_filter(sk, optval, len);
1976 case SO_LOCK_FILTER:
1977 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1980 case SO_BPF_EXTENSIONS:
1981 v.val = bpf_tell_extensions();
1984 case SO_SELECT_ERR_QUEUE:
1985 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1988 #ifdef CONFIG_NET_RX_BUSY_POLL
1990 v.val = READ_ONCE(sk->sk_ll_usec);
1992 case SO_PREFER_BUSY_POLL:
1993 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1997 case SO_MAX_PACING_RATE:
1998 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1999 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
2000 lv = sizeof(v.ulval);
2001 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
2004 v.val = min_t(unsigned long, ~0U,
2005 READ_ONCE(sk->sk_max_pacing_rate));
2009 case SO_INCOMING_CPU:
2010 v.val = READ_ONCE(sk->sk_incoming_cpu);
2015 u32 meminfo[SK_MEMINFO_VARS];
2017 sk_get_meminfo(sk, meminfo);
2019 len = min_t(unsigned int, len, sizeof(meminfo));
2020 if (copy_to_sockptr(optval, &meminfo, len))
2026 #ifdef CONFIG_NET_RX_BUSY_POLL
2027 case SO_INCOMING_NAPI_ID:
2028 v.val = READ_ONCE(sk->sk_napi_id);
2030 /* aggregate non-NAPI IDs down to 0 */
2031 if (v.val < MIN_NAPI_ID)
2041 v.val64 = sock_gen_cookie(sk);
2045 v.val = sock_flag(sk, SOCK_ZEROCOPY);
2049 lv = sizeof(v.txtime);
2050 v.txtime.clockid = sk->sk_clockid;
2051 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
2052 SOF_TXTIME_DEADLINE_MODE : 0;
2053 v.txtime.flags |= sk->sk_txtime_report_errors ?
2054 SOF_TXTIME_REPORT_ERRORS : 0;
2057 case SO_BINDTOIFINDEX:
2058 v.val = READ_ONCE(sk->sk_bound_dev_if);
2061 case SO_NETNS_COOKIE:
2065 v.val64 = sock_net(sk)->net_cookie;
2069 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
2072 case SO_RESERVE_MEM:
2073 v.val = READ_ONCE(sk->sk_reserved_mem);
2077 /* Paired with WRITE_ONCE() in sk_setsockopt() */
2078 v.val = READ_ONCE(sk->sk_txrehash);
2082 /* We implement the SO_SNDLOWAT etc to not be settable
2085 return -ENOPROTOOPT;
2090 if (copy_to_sockptr(optval, &v, len))
2093 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2099 * Initialize an sk_lock.
2101 * (We also register the sk_lock with the lock validator.)
2103 static inline void sock_lock_init(struct sock *sk)
2105 if (sk->sk_kern_sock)
2106 sock_lock_init_class_and_name(
2108 af_family_kern_slock_key_strings[sk->sk_family],
2109 af_family_kern_slock_keys + sk->sk_family,
2110 af_family_kern_key_strings[sk->sk_family],
2111 af_family_kern_keys + sk->sk_family);
2113 sock_lock_init_class_and_name(
2115 af_family_slock_key_strings[sk->sk_family],
2116 af_family_slock_keys + sk->sk_family,
2117 af_family_key_strings[sk->sk_family],
2118 af_family_keys + sk->sk_family);
2122 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2123 * even temporarily, because of RCU lookups. sk_node should also be left as is.
2124 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2126 static void sock_copy(struct sock *nsk, const struct sock *osk)
2128 const struct proto *prot = READ_ONCE(osk->sk_prot);
2129 #ifdef CONFIG_SECURITY_NETWORK
2130 void *sptr = nsk->sk_security;
2133 /* If we move sk_tx_queue_mapping out of the private section,
2134 * we must check if sk_tx_queue_clear() is called after
2135 * sock_copy() in sk_clone_lock().
2137 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2138 offsetof(struct sock, sk_dontcopy_begin) ||
2139 offsetof(struct sock, sk_tx_queue_mapping) >=
2140 offsetof(struct sock, sk_dontcopy_end));
2142 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2144 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2145 prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2146 /* alloc is larger than struct, see sk_prot_alloc() */);
2148 #ifdef CONFIG_SECURITY_NETWORK
2149 nsk->sk_security = sptr;
2150 security_sk_clone(osk, nsk);
2154 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2158 struct kmem_cache *slab;
2162 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2165 if (want_init_on_alloc(priority))
2166 sk_prot_clear_nulls(sk, prot->obj_size);
2168 sk = kmalloc(prot->obj_size, priority);
2171 if (security_sk_alloc(sk, family, priority))
2174 if (!try_module_get(prot->owner))
2181 security_sk_free(sk);
2184 kmem_cache_free(slab, sk);
2190 static void sk_prot_free(struct proto *prot, struct sock *sk)
2192 struct kmem_cache *slab;
2193 struct module *owner;
2195 owner = prot->owner;
2198 cgroup_sk_free(&sk->sk_cgrp_data);
2199 mem_cgroup_sk_free(sk);
2200 security_sk_free(sk);
2202 kmem_cache_free(slab, sk);
2209 * sk_alloc - All socket objects are allocated here
2210 * @net: the applicable net namespace
2211 * @family: protocol family
2212 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2213 * @prot: struct proto associated with this new sock instance
2214 * @kern: is this to be a kernel socket?
2216 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2217 struct proto *prot, int kern)
2221 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2223 sk->sk_family = family;
2225 * See comment in struct sock definition to understand
2226 * why we need sk_prot_creator -acme
2228 sk->sk_prot = sk->sk_prot_creator = prot;
2229 sk->sk_kern_sock = kern;
2231 sk->sk_net_refcnt = kern ? 0 : 1;
2232 if (likely(sk->sk_net_refcnt)) {
2233 get_net_track(net, &sk->ns_tracker, priority);
2234 sock_inuse_add(net, 1);
2236 __netns_tracker_alloc(net, &sk->ns_tracker,
2240 sock_net_set(sk, net);
2241 refcount_set(&sk->sk_wmem_alloc, 1);
2243 mem_cgroup_sk_alloc(sk);
2244 cgroup_sk_alloc(&sk->sk_cgrp_data);
2245 sock_update_classid(&sk->sk_cgrp_data);
2246 sock_update_netprioidx(&sk->sk_cgrp_data);
2247 sk_tx_queue_clear(sk);
2252 EXPORT_SYMBOL(sk_alloc);
2254 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2255 * grace period. This is the case for UDP sockets and TCP listeners.
2257 static void __sk_destruct(struct rcu_head *head)
2259 struct sock *sk = container_of(head, struct sock, sk_rcu);
2260 struct sk_filter *filter;
2262 if (sk->sk_destruct)
2263 sk->sk_destruct(sk);
2265 filter = rcu_dereference_check(sk->sk_filter,
2266 refcount_read(&sk->sk_wmem_alloc) == 0);
2268 sk_filter_uncharge(sk, filter);
2269 RCU_INIT_POINTER(sk->sk_filter, NULL);
2272 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2274 #ifdef CONFIG_BPF_SYSCALL
2275 bpf_sk_storage_free(sk);
2278 if (atomic_read(&sk->sk_omem_alloc))
2279 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2280 __func__, atomic_read(&sk->sk_omem_alloc));
2282 if (sk->sk_frag.page) {
2283 put_page(sk->sk_frag.page);
2284 sk->sk_frag.page = NULL;
2287 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2288 put_cred(sk->sk_peer_cred);
2289 put_pid(sk->sk_peer_pid);
2291 if (likely(sk->sk_net_refcnt))
2292 put_net_track(sock_net(sk), &sk->ns_tracker);
2294 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2296 sk_prot_free(sk->sk_prot_creator, sk);
2299 void sk_destruct(struct sock *sk)
2301 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2303 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2304 reuseport_detach_sock(sk);
2305 use_call_rcu = true;
2309 call_rcu(&sk->sk_rcu, __sk_destruct);
2311 __sk_destruct(&sk->sk_rcu);
2314 static void __sk_free(struct sock *sk)
2316 if (likely(sk->sk_net_refcnt))
2317 sock_inuse_add(sock_net(sk), -1);
2319 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2320 sock_diag_broadcast_destroy(sk);
2325 void sk_free(struct sock *sk)
2328 * We subtract one from sk_wmem_alloc and can know if
2329 * some packets are still in some tx queue.
2330 * If not null, sock_wfree() will call __sk_free(sk) later
2332 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2335 EXPORT_SYMBOL(sk_free);
2337 static void sk_init_common(struct sock *sk)
2339 skb_queue_head_init(&sk->sk_receive_queue);
2340 skb_queue_head_init(&sk->sk_write_queue);
2341 skb_queue_head_init(&sk->sk_error_queue);
2343 rwlock_init(&sk->sk_callback_lock);
2344 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2345 af_rlock_keys + sk->sk_family,
2346 af_family_rlock_key_strings[sk->sk_family]);
2347 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2348 af_wlock_keys + sk->sk_family,
2349 af_family_wlock_key_strings[sk->sk_family]);
2350 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2351 af_elock_keys + sk->sk_family,
2352 af_family_elock_key_strings[sk->sk_family]);
2353 if (sk->sk_kern_sock)
2354 lockdep_set_class_and_name(&sk->sk_callback_lock,
2355 af_kern_callback_keys + sk->sk_family,
2356 af_family_kern_clock_key_strings[sk->sk_family]);
2358 lockdep_set_class_and_name(&sk->sk_callback_lock,
2359 af_callback_keys + sk->sk_family,
2360 af_family_clock_key_strings[sk->sk_family]);
2364 * sk_clone_lock - clone a socket, and lock its clone
2365 * @sk: the socket to clone
2366 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2368 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2370 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2372 struct proto *prot = READ_ONCE(sk->sk_prot);
2373 struct sk_filter *filter;
2374 bool is_charged = true;
2377 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2381 sock_copy(newsk, sk);
2383 newsk->sk_prot_creator = prot;
2386 if (likely(newsk->sk_net_refcnt)) {
2387 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2388 sock_inuse_add(sock_net(newsk), 1);
2390 /* Kernel sockets are not elevating the struct net refcount.
2391 * Instead, use a tracker to more easily detect if a layer
2392 * is not properly dismantling its kernel sockets at netns
2395 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2398 sk_node_init(&newsk->sk_node);
2399 sock_lock_init(newsk);
2400 bh_lock_sock(newsk);
2401 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2402 newsk->sk_backlog.len = 0;
2404 atomic_set(&newsk->sk_rmem_alloc, 0);
2406 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2407 refcount_set(&newsk->sk_wmem_alloc, 1);
2409 atomic_set(&newsk->sk_omem_alloc, 0);
2410 sk_init_common(newsk);
2412 newsk->sk_dst_cache = NULL;
2413 newsk->sk_dst_pending_confirm = 0;
2414 newsk->sk_wmem_queued = 0;
2415 newsk->sk_forward_alloc = 0;
2416 newsk->sk_reserved_mem = 0;
2417 atomic_set(&newsk->sk_drops, 0);
2418 newsk->sk_send_head = NULL;
2419 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2420 atomic_set(&newsk->sk_zckey, 0);
2422 sock_reset_flag(newsk, SOCK_DONE);
2424 /* sk->sk_memcg will be populated at accept() time */
2425 newsk->sk_memcg = NULL;
2427 cgroup_sk_clone(&newsk->sk_cgrp_data);
2430 filter = rcu_dereference(sk->sk_filter);
2432 /* though it's an empty new sock, the charging may fail
2433 * if sysctl_optmem_max was changed between creation of
2434 * original socket and cloning
2436 is_charged = sk_filter_charge(newsk, filter);
2437 RCU_INIT_POINTER(newsk->sk_filter, filter);
2440 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2441 /* We need to make sure that we don't uncharge the new
2442 * socket if we couldn't charge it in the first place
2443 * as otherwise we uncharge the parent's filter.
2446 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2447 sk_free_unlock_clone(newsk);
2451 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2453 if (bpf_sk_storage_clone(sk, newsk)) {
2454 sk_free_unlock_clone(newsk);
2459 /* Clear sk_user_data if parent had the pointer tagged
2460 * as not suitable for copying when cloning.
2462 if (sk_user_data_is_nocopy(newsk))
2463 newsk->sk_user_data = NULL;
2466 newsk->sk_err_soft = 0;
2467 newsk->sk_priority = 0;
2468 newsk->sk_incoming_cpu = raw_smp_processor_id();
2470 /* Before updating sk_refcnt, we must commit prior changes to memory
2471 * (Documentation/RCU/rculist_nulls.rst for details)
2474 refcount_set(&newsk->sk_refcnt, 2);
2476 sk_set_socket(newsk, NULL);
2477 sk_tx_queue_clear(newsk);
2478 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2480 if (newsk->sk_prot->sockets_allocated)
2481 sk_sockets_allocated_inc(newsk);
2483 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2484 net_enable_timestamp();
2488 EXPORT_SYMBOL_GPL(sk_clone_lock);
2490 void sk_free_unlock_clone(struct sock *sk)
2492 /* It is still raw copy of parent, so invalidate
2493 * destructor and make plain sk_free() */
2494 sk->sk_destruct = NULL;
2498 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2500 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2502 bool is_ipv6 = false;
2505 #if IS_ENABLED(CONFIG_IPV6)
2506 is_ipv6 = (sk->sk_family == AF_INET6 &&
2507 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2509 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2510 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2511 READ_ONCE(dst->dev->gso_ipv4_max_size);
2512 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2513 max_size = GSO_LEGACY_MAX_SIZE;
2515 return max_size - (MAX_TCP_HEADER + 1);
2518 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2522 sk->sk_route_caps = dst->dev->features;
2524 sk->sk_route_caps |= NETIF_F_GSO;
2525 if (sk->sk_route_caps & NETIF_F_GSO)
2526 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2527 if (unlikely(sk->sk_gso_disabled))
2528 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2529 if (sk_can_gso(sk)) {
2530 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2531 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2533 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2534 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2535 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2536 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2539 sk->sk_gso_max_segs = max_segs;
2540 sk_dst_set(sk, dst);
2542 EXPORT_SYMBOL_GPL(sk_setup_caps);
2545 * Simple resource managers for sockets.
2550 * Write buffer destructor automatically called from kfree_skb.
2552 void sock_wfree(struct sk_buff *skb)
2554 struct sock *sk = skb->sk;
2555 unsigned int len = skb->truesize;
2558 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2559 if (sock_flag(sk, SOCK_RCU_FREE) &&
2560 sk->sk_write_space == sock_def_write_space) {
2562 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2563 sock_def_write_space_wfree(sk);
2571 * Keep a reference on sk_wmem_alloc, this will be released
2572 * after sk_write_space() call
2574 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2575 sk->sk_write_space(sk);
2579 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2580 * could not do because of in-flight packets
2582 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2585 EXPORT_SYMBOL(sock_wfree);
2587 /* This variant of sock_wfree() is used by TCP,
2588 * since it sets SOCK_USE_WRITE_QUEUE.
2590 void __sock_wfree(struct sk_buff *skb)
2592 struct sock *sk = skb->sk;
2594 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2598 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2602 if (unlikely(!sk_fullsock(sk)))
2603 return skb_set_owner_edemux(skb, sk);
2606 skb->destructor = sock_wfree;
2607 skb_set_hash_from_sk(skb, sk);
2609 * We used to take a refcount on sk, but following operation
2610 * is enough to guarantee sk_free() won't free this sock until
2611 * all in-flight packets are completed
2613 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2615 EXPORT_SYMBOL(skb_set_owner_w);
2617 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2619 /* Drivers depend on in-order delivery for crypto offload,
2620 * partial orphan breaks out-of-order-OK logic.
2622 if (skb_is_decrypted(skb))
2625 return (skb->destructor == sock_wfree ||
2626 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2629 /* This helper is used by netem, as it can hold packets in its
2630 * delay queue. We want to allow the owner socket to send more
2631 * packets, as if they were already TX completed by a typical driver.
2632 * But we also want to keep skb->sk set because some packet schedulers
2633 * rely on it (sch_fq for example).
2635 void skb_orphan_partial(struct sk_buff *skb)
2637 if (skb_is_tcp_pure_ack(skb))
2640 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2645 EXPORT_SYMBOL(skb_orphan_partial);
2648 * Read buffer destructor automatically called from kfree_skb.
2650 void sock_rfree(struct sk_buff *skb)
2652 struct sock *sk = skb->sk;
2653 unsigned int len = skb->truesize;
2655 atomic_sub(len, &sk->sk_rmem_alloc);
2656 sk_mem_uncharge(sk, len);
2658 EXPORT_SYMBOL(sock_rfree);
2661 * Buffer destructor for skbs that are not used directly in read or write
2662 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2664 void sock_efree(struct sk_buff *skb)
2668 EXPORT_SYMBOL(sock_efree);
2670 /* Buffer destructor for prefetch/receive path where reference count may
2671 * not be held, e.g. for listen sockets.
2674 void sock_pfree(struct sk_buff *skb)
2676 struct sock *sk = skb->sk;
2678 if (!sk_is_refcounted(sk))
2681 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2682 inet_reqsk(sk)->rsk_listener = NULL;
2683 reqsk_free(inet_reqsk(sk));
2689 EXPORT_SYMBOL(sock_pfree);
2690 #endif /* CONFIG_INET */
2692 kuid_t sock_i_uid(struct sock *sk)
2696 read_lock_bh(&sk->sk_callback_lock);
2697 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2698 read_unlock_bh(&sk->sk_callback_lock);
2701 EXPORT_SYMBOL(sock_i_uid);
2703 unsigned long __sock_i_ino(struct sock *sk)
2707 read_lock(&sk->sk_callback_lock);
2708 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2709 read_unlock(&sk->sk_callback_lock);
2712 EXPORT_SYMBOL(__sock_i_ino);
2714 unsigned long sock_i_ino(struct sock *sk)
2719 ino = __sock_i_ino(sk);
2723 EXPORT_SYMBOL(sock_i_ino);
2726 * Allocate a skb from the socket's send buffer.
2728 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2732 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2733 struct sk_buff *skb = alloc_skb(size, priority);
2736 skb_set_owner_w(skb, sk);
2742 EXPORT_SYMBOL(sock_wmalloc);
2744 static void sock_ofree(struct sk_buff *skb)
2746 struct sock *sk = skb->sk;
2748 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2751 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2754 struct sk_buff *skb;
2756 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2757 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2758 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2761 skb = alloc_skb(size, priority);
2765 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2767 skb->destructor = sock_ofree;
2772 * Allocate a memory block from the socket's option memory buffer.
2774 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2776 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2778 if ((unsigned int)size <= optmem_max &&
2779 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2781 /* First do the add, to avoid the race if kmalloc
2784 atomic_add(size, &sk->sk_omem_alloc);
2785 mem = kmalloc(size, priority);
2788 atomic_sub(size, &sk->sk_omem_alloc);
2792 EXPORT_SYMBOL(sock_kmalloc);
2794 /* Free an option memory block. Note, we actually want the inline
2795 * here as this allows gcc to detect the nullify and fold away the
2796 * condition entirely.
2798 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2801 if (WARN_ON_ONCE(!mem))
2804 kfree_sensitive(mem);
2807 atomic_sub(size, &sk->sk_omem_alloc);
2810 void sock_kfree_s(struct sock *sk, void *mem, int size)
2812 __sock_kfree_s(sk, mem, size, false);
2814 EXPORT_SYMBOL(sock_kfree_s);
2816 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2818 __sock_kfree_s(sk, mem, size, true);
2820 EXPORT_SYMBOL(sock_kzfree_s);
2822 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2823 I think, these locks should be removed for datagram sockets.
2825 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2829 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2833 if (signal_pending(current))
2835 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2836 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2837 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2839 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2841 if (READ_ONCE(sk->sk_err))
2843 timeo = schedule_timeout(timeo);
2845 finish_wait(sk_sleep(sk), &wait);
2851 * Generic send/receive buffer handlers
2854 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2855 unsigned long data_len, int noblock,
2856 int *errcode, int max_page_order)
2858 struct sk_buff *skb;
2862 timeo = sock_sndtimeo(sk, noblock);
2864 err = sock_error(sk);
2869 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2872 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2875 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2876 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2880 if (signal_pending(current))
2882 timeo = sock_wait_for_wmem(sk, timeo);
2884 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2885 errcode, sk->sk_allocation);
2887 skb_set_owner_w(skb, sk);
2891 err = sock_intr_errno(timeo);
2896 EXPORT_SYMBOL(sock_alloc_send_pskb);
2898 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2899 struct sockcm_cookie *sockc)
2903 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31));
2905 switch (cmsg->cmsg_type) {
2907 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2908 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2910 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2912 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2914 case SO_TIMESTAMPING_OLD:
2915 case SO_TIMESTAMPING_NEW:
2916 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2919 tsflags = *(u32 *)CMSG_DATA(cmsg);
2920 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2923 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2924 sockc->tsflags |= tsflags;
2927 if (!sock_flag(sk, SOCK_TXTIME))
2929 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2931 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2936 tsflags = READ_ONCE(sk->sk_tsflags);
2937 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID))
2939 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2941 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg);
2942 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID;
2944 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2946 case SCM_CREDENTIALS:
2953 EXPORT_SYMBOL(__sock_cmsg_send);
2955 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2956 struct sockcm_cookie *sockc)
2958 struct cmsghdr *cmsg;
2961 for_each_cmsghdr(cmsg, msg) {
2962 if (!CMSG_OK(msg, cmsg))
2964 if (cmsg->cmsg_level != SOL_SOCKET)
2966 ret = __sock_cmsg_send(sk, cmsg, sockc);
2972 EXPORT_SYMBOL(sock_cmsg_send);
2974 static void sk_enter_memory_pressure(struct sock *sk)
2976 if (!sk->sk_prot->enter_memory_pressure)
2979 sk->sk_prot->enter_memory_pressure(sk);
2982 static void sk_leave_memory_pressure(struct sock *sk)
2984 if (sk->sk_prot->leave_memory_pressure) {
2985 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2986 tcp_leave_memory_pressure, sk);
2988 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2990 if (memory_pressure && READ_ONCE(*memory_pressure))
2991 WRITE_ONCE(*memory_pressure, 0);
2995 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2998 * skb_page_frag_refill - check that a page_frag contains enough room
2999 * @sz: minimum size of the fragment we want to get
3000 * @pfrag: pointer to page_frag
3001 * @gfp: priority for memory allocation
3003 * Note: While this allocator tries to use high order pages, there is
3004 * no guarantee that allocations succeed. Therefore, @sz MUST be
3005 * less or equal than PAGE_SIZE.
3007 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
3010 if (page_ref_count(pfrag->page) == 1) {
3014 if (pfrag->offset + sz <= pfrag->size)
3016 put_page(pfrag->page);
3020 if (SKB_FRAG_PAGE_ORDER &&
3021 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
3022 /* Avoid direct reclaim but allow kswapd to wake */
3023 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
3024 __GFP_COMP | __GFP_NOWARN |
3026 SKB_FRAG_PAGE_ORDER);
3027 if (likely(pfrag->page)) {
3028 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
3032 pfrag->page = alloc_page(gfp);
3033 if (likely(pfrag->page)) {
3034 pfrag->size = PAGE_SIZE;
3039 EXPORT_SYMBOL(skb_page_frag_refill);
3041 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
3043 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
3046 sk_enter_memory_pressure(sk);
3047 sk_stream_moderate_sndbuf(sk);
3050 EXPORT_SYMBOL(sk_page_frag_refill);
3052 void __lock_sock(struct sock *sk)
3053 __releases(&sk->sk_lock.slock)
3054 __acquires(&sk->sk_lock.slock)
3059 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
3060 TASK_UNINTERRUPTIBLE);
3061 spin_unlock_bh(&sk->sk_lock.slock);
3063 spin_lock_bh(&sk->sk_lock.slock);
3064 if (!sock_owned_by_user(sk))
3067 finish_wait(&sk->sk_lock.wq, &wait);
3070 void __release_sock(struct sock *sk)
3071 __releases(&sk->sk_lock.slock)
3072 __acquires(&sk->sk_lock.slock)
3074 struct sk_buff *skb, *next;
3076 while ((skb = sk->sk_backlog.head) != NULL) {
3077 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
3079 spin_unlock_bh(&sk->sk_lock.slock);
3084 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
3085 skb_mark_not_on_list(skb);
3086 sk_backlog_rcv(sk, skb);
3091 } while (skb != NULL);
3093 spin_lock_bh(&sk->sk_lock.slock);
3097 * Doing the zeroing here guarantee we can not loop forever
3098 * while a wild producer attempts to flood us.
3100 sk->sk_backlog.len = 0;
3103 void __sk_flush_backlog(struct sock *sk)
3105 spin_lock_bh(&sk->sk_lock.slock);
3108 if (sk->sk_prot->release_cb)
3109 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3110 tcp_release_cb, sk);
3112 spin_unlock_bh(&sk->sk_lock.slock);
3114 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3117 * sk_wait_data - wait for data to arrive at sk_receive_queue
3118 * @sk: sock to wait on
3119 * @timeo: for how long
3120 * @skb: last skb seen on sk_receive_queue
3122 * Now socket state including sk->sk_err is changed only under lock,
3123 * hence we may omit checks after joining wait queue.
3124 * We check receive queue before schedule() only as optimization;
3125 * it is very likely that release_sock() added new data.
3127 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3129 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3132 add_wait_queue(sk_sleep(sk), &wait);
3133 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3134 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3135 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3136 remove_wait_queue(sk_sleep(sk), &wait);
3139 EXPORT_SYMBOL(sk_wait_data);
3142 * __sk_mem_raise_allocated - increase memory_allocated
3144 * @size: memory size to allocate
3145 * @amt: pages to allocate
3146 * @kind: allocation type
3148 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3150 * Unlike the globally shared limits among the sockets under same protocol,
3151 * consuming the budget of a memcg won't have direct effect on other ones.
3152 * So be optimistic about memcg's tolerance, and leave the callers to decide
3153 * whether or not to raise allocated through sk_under_memory_pressure() or
3156 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3158 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3159 struct proto *prot = sk->sk_prot;
3160 bool charged = false;
3163 sk_memory_allocated_add(sk, amt);
3164 allocated = sk_memory_allocated(sk);
3167 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3168 goto suppress_allocation;
3173 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3174 sk_leave_memory_pressure(sk);
3178 /* Under pressure. */
3179 if (allocated > sk_prot_mem_limits(sk, 1))
3180 sk_enter_memory_pressure(sk);
3182 /* Over hard limit. */
3183 if (allocated > sk_prot_mem_limits(sk, 2))
3184 goto suppress_allocation;
3186 /* Guarantee minimum buffer size under pressure (either global
3187 * or memcg) to make sure features described in RFC 7323 (TCP
3188 * Extensions for High Performance) work properly.
3190 * This rule does NOT stand when exceeds global or memcg's hard
3191 * limit, or else a DoS attack can be taken place by spawning
3192 * lots of sockets whose usage are under minimum buffer size.
3194 if (kind == SK_MEM_RECV) {
3195 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3198 } else { /* SK_MEM_SEND */
3199 int wmem0 = sk_get_wmem0(sk, prot);
3201 if (sk->sk_type == SOCK_STREAM) {
3202 if (sk->sk_wmem_queued < wmem0)
3204 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3209 if (sk_has_memory_pressure(sk)) {
3212 /* The following 'average' heuristic is within the
3213 * scope of global accounting, so it only makes
3214 * sense for global memory pressure.
3216 if (!sk_under_global_memory_pressure(sk))
3219 /* Try to be fair among all the sockets under global
3220 * pressure by allowing the ones that below average
3223 alloc = sk_sockets_allocated_read_positive(sk);
3224 if (sk_prot_mem_limits(sk, 2) > alloc *
3225 sk_mem_pages(sk->sk_wmem_queued +
3226 atomic_read(&sk->sk_rmem_alloc) +
3227 sk->sk_forward_alloc))
3231 suppress_allocation:
3233 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3234 sk_stream_moderate_sndbuf(sk);
3236 /* Fail only if socket is _under_ its sndbuf.
3237 * In this case we cannot block, so that we have to fail.
3239 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3240 /* Force charge with __GFP_NOFAIL */
3241 if (memcg && !charged) {
3242 mem_cgroup_charge_skmem(memcg, amt,
3243 gfp_memcg_charge() | __GFP_NOFAIL);
3249 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3250 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3252 sk_memory_allocated_sub(sk, amt);
3255 mem_cgroup_uncharge_skmem(memcg, amt);
3261 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3263 * @size: memory size to allocate
3264 * @kind: allocation type
3266 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3267 * rmem allocation. This function assumes that protocols which have
3268 * memory_pressure use sk_wmem_queued as write buffer accounting.
3270 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3272 int ret, amt = sk_mem_pages(size);
3274 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3275 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3277 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3280 EXPORT_SYMBOL(__sk_mem_schedule);
3283 * __sk_mem_reduce_allocated - reclaim memory_allocated
3285 * @amount: number of quanta
3287 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3289 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3291 sk_memory_allocated_sub(sk, amount);
3293 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3294 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3296 if (sk_under_global_memory_pressure(sk) &&
3297 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3298 sk_leave_memory_pressure(sk);
3302 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3304 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3306 void __sk_mem_reclaim(struct sock *sk, int amount)
3308 amount >>= PAGE_SHIFT;
3309 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3310 __sk_mem_reduce_allocated(sk, amount);
3312 EXPORT_SYMBOL(__sk_mem_reclaim);
3314 int sk_set_peek_off(struct sock *sk, int val)
3316 WRITE_ONCE(sk->sk_peek_off, val);
3319 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3322 * Set of default routines for initialising struct proto_ops when
3323 * the protocol does not support a particular function. In certain
3324 * cases where it makes no sense for a protocol to have a "do nothing"
3325 * function, some default processing is provided.
3328 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3332 EXPORT_SYMBOL(sock_no_bind);
3334 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3339 EXPORT_SYMBOL(sock_no_connect);
3341 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3345 EXPORT_SYMBOL(sock_no_socketpair);
3347 int sock_no_accept(struct socket *sock, struct socket *newsock,
3348 struct proto_accept_arg *arg)
3352 EXPORT_SYMBOL(sock_no_accept);
3354 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3359 EXPORT_SYMBOL(sock_no_getname);
3361 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3365 EXPORT_SYMBOL(sock_no_ioctl);
3367 int sock_no_listen(struct socket *sock, int backlog)
3371 EXPORT_SYMBOL(sock_no_listen);
3373 int sock_no_shutdown(struct socket *sock, int how)
3377 EXPORT_SYMBOL(sock_no_shutdown);
3379 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3383 EXPORT_SYMBOL(sock_no_sendmsg);
3385 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3389 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3391 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3396 EXPORT_SYMBOL(sock_no_recvmsg);
3398 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3400 /* Mirror missing mmap method error code */
3403 EXPORT_SYMBOL(sock_no_mmap);
3406 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3407 * various sock-based usage counts.
3409 void __receive_sock(struct file *file)
3411 struct socket *sock;
3413 sock = sock_from_file(file);
3415 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3416 sock_update_classid(&sock->sk->sk_cgrp_data);
3421 * Default Socket Callbacks
3424 static void sock_def_wakeup(struct sock *sk)
3426 struct socket_wq *wq;
3429 wq = rcu_dereference(sk->sk_wq);
3430 if (skwq_has_sleeper(wq))
3431 wake_up_interruptible_all(&wq->wait);
3435 static void sock_def_error_report(struct sock *sk)
3437 struct socket_wq *wq;
3440 wq = rcu_dereference(sk->sk_wq);
3441 if (skwq_has_sleeper(wq))
3442 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3443 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR);
3447 void sock_def_readable(struct sock *sk)
3449 struct socket_wq *wq;
3451 trace_sk_data_ready(sk);
3454 wq = rcu_dereference(sk->sk_wq);
3455 if (skwq_has_sleeper(wq))
3456 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3457 EPOLLRDNORM | EPOLLRDBAND);
3458 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
3462 static void sock_def_write_space(struct sock *sk)
3464 struct socket_wq *wq;
3468 /* Do not wake up a writer until he can make "significant"
3471 if (sock_writeable(sk)) {
3472 wq = rcu_dereference(sk->sk_wq);
3473 if (skwq_has_sleeper(wq))
3474 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3475 EPOLLWRNORM | EPOLLWRBAND);
3477 /* Should agree with poll, otherwise some programs break */
3478 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3484 /* An optimised version of sock_def_write_space(), should only be called
3485 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3488 static void sock_def_write_space_wfree(struct sock *sk)
3490 /* Do not wake up a writer until he can make "significant"
3493 if (sock_writeable(sk)) {
3494 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3496 /* rely on refcount_sub from sock_wfree() */
3497 smp_mb__after_atomic();
3498 if (wq && waitqueue_active(&wq->wait))
3499 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3500 EPOLLWRNORM | EPOLLWRBAND);
3502 /* Should agree with poll, otherwise some programs break */
3503 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3507 static void sock_def_destruct(struct sock *sk)
3511 void sk_send_sigurg(struct sock *sk)
3513 if (sk->sk_socket && sk->sk_socket->file)
3514 if (send_sigurg(sk->sk_socket->file))
3515 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3517 EXPORT_SYMBOL(sk_send_sigurg);
3519 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3520 unsigned long expires)
3522 if (!mod_timer(timer, expires))
3525 EXPORT_SYMBOL(sk_reset_timer);
3527 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3529 if (del_timer(timer))
3532 EXPORT_SYMBOL(sk_stop_timer);
3534 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3536 if (del_timer_sync(timer))
3539 EXPORT_SYMBOL(sk_stop_timer_sync);
3541 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3544 sk->sk_send_head = NULL;
3546 timer_setup(&sk->sk_timer, NULL, 0);
3548 sk->sk_allocation = GFP_KERNEL;
3549 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3550 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3551 sk->sk_state = TCP_CLOSE;
3552 sk->sk_use_task_frag = true;
3553 sk_set_socket(sk, sock);
3555 sock_set_flag(sk, SOCK_ZAPPED);
3558 sk->sk_type = sock->type;
3559 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3562 RCU_INIT_POINTER(sk->sk_wq, NULL);
3566 sk->sk_state_change = sock_def_wakeup;
3567 sk->sk_data_ready = sock_def_readable;
3568 sk->sk_write_space = sock_def_write_space;
3569 sk->sk_error_report = sock_def_error_report;
3570 sk->sk_destruct = sock_def_destruct;
3572 sk->sk_frag.page = NULL;
3573 sk->sk_frag.offset = 0;
3574 sk->sk_peek_off = -1;
3576 sk->sk_peer_pid = NULL;
3577 sk->sk_peer_cred = NULL;
3578 spin_lock_init(&sk->sk_peer_lock);
3580 sk->sk_write_pending = 0;
3581 sk->sk_rcvlowat = 1;
3582 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3583 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3585 sk->sk_stamp = SK_DEFAULT_STAMP;
3586 #if BITS_PER_LONG==32
3587 seqlock_init(&sk->sk_stamp_seq);
3589 atomic_set(&sk->sk_zckey, 0);
3591 #ifdef CONFIG_NET_RX_BUSY_POLL
3593 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3596 sk->sk_max_pacing_rate = ~0UL;
3597 sk->sk_pacing_rate = ~0UL;
3598 WRITE_ONCE(sk->sk_pacing_shift, 10);
3599 sk->sk_incoming_cpu = -1;
3601 sk_rx_queue_clear(sk);
3603 * Before updating sk_refcnt, we must commit prior changes to memory
3604 * (Documentation/RCU/rculist_nulls.rst for details)
3607 refcount_set(&sk->sk_refcnt, 1);
3608 atomic_set(&sk->sk_drops, 0);
3610 EXPORT_SYMBOL(sock_init_data_uid);
3612 void sock_init_data(struct socket *sock, struct sock *sk)
3615 SOCK_INODE(sock)->i_uid :
3616 make_kuid(sock_net(sk)->user_ns, 0);
3618 sock_init_data_uid(sock, sk, uid);
3620 EXPORT_SYMBOL(sock_init_data);
3622 void lock_sock_nested(struct sock *sk, int subclass)
3624 /* The sk_lock has mutex_lock() semantics here. */
3625 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3628 spin_lock_bh(&sk->sk_lock.slock);
3629 if (sock_owned_by_user_nocheck(sk))
3631 sk->sk_lock.owned = 1;
3632 spin_unlock_bh(&sk->sk_lock.slock);
3634 EXPORT_SYMBOL(lock_sock_nested);
3636 void release_sock(struct sock *sk)
3638 spin_lock_bh(&sk->sk_lock.slock);
3639 if (sk->sk_backlog.tail)
3642 if (sk->sk_prot->release_cb)
3643 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3644 tcp_release_cb, sk);
3646 sock_release_ownership(sk);
3647 if (waitqueue_active(&sk->sk_lock.wq))
3648 wake_up(&sk->sk_lock.wq);
3649 spin_unlock_bh(&sk->sk_lock.slock);
3651 EXPORT_SYMBOL(release_sock);
3653 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3656 spin_lock_bh(&sk->sk_lock.slock);
3658 if (!sock_owned_by_user_nocheck(sk)) {
3660 * Fast path return with bottom halves disabled and
3661 * sock::sk_lock.slock held.
3663 * The 'mutex' is not contended and holding
3664 * sock::sk_lock.slock prevents all other lockers to
3665 * proceed so the corresponding unlock_sock_fast() can
3666 * avoid the slow path of release_sock() completely and
3667 * just release slock.
3669 * From a semantical POV this is equivalent to 'acquiring'
3670 * the 'mutex', hence the corresponding lockdep
3671 * mutex_release() has to happen in the fast path of
3672 * unlock_sock_fast().
3678 sk->sk_lock.owned = 1;
3679 __acquire(&sk->sk_lock.slock);
3680 spin_unlock_bh(&sk->sk_lock.slock);
3683 EXPORT_SYMBOL(__lock_sock_fast);
3685 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3686 bool timeval, bool time32)
3688 struct sock *sk = sock->sk;
3689 struct timespec64 ts;
3691 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3692 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3693 if (ts.tv_sec == -1)
3695 if (ts.tv_sec == 0) {
3696 ktime_t kt = ktime_get_real();
3697 sock_write_timestamp(sk, kt);
3698 ts = ktime_to_timespec64(kt);
3704 #ifdef CONFIG_COMPAT_32BIT_TIME
3706 return put_old_timespec32(&ts, userstamp);
3708 #ifdef CONFIG_SPARC64
3709 /* beware of padding in sparc64 timeval */
3710 if (timeval && !in_compat_syscall()) {
3711 struct __kernel_old_timeval __user tv = {
3712 .tv_sec = ts.tv_sec,
3713 .tv_usec = ts.tv_nsec,
3715 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3720 return put_timespec64(&ts, userstamp);
3722 EXPORT_SYMBOL(sock_gettstamp);
3724 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3726 if (!sock_flag(sk, flag)) {
3727 unsigned long previous_flags = sk->sk_flags;
3729 sock_set_flag(sk, flag);
3731 * we just set one of the two flags which require net
3732 * time stamping, but time stamping might have been on
3733 * already because of the other one
3735 if (sock_needs_netstamp(sk) &&
3736 !(previous_flags & SK_FLAGS_TIMESTAMP))
3737 net_enable_timestamp();
3741 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3742 int level, int type)
3744 struct sock_exterr_skb *serr;
3745 struct sk_buff *skb;
3749 skb = sock_dequeue_err_skb(sk);
3755 msg->msg_flags |= MSG_TRUNC;
3758 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3762 sock_recv_timestamp(msg, sk, skb);
3764 serr = SKB_EXT_ERR(skb);
3765 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3767 msg->msg_flags |= MSG_ERRQUEUE;
3775 EXPORT_SYMBOL(sock_recv_errqueue);
3778 * Get a socket option on an socket.
3780 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3781 * asynchronous errors should be reported by getsockopt. We assume
3782 * this means if you specify SO_ERROR (otherwise what is the point of it).
3784 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3785 char __user *optval, int __user *optlen)
3787 struct sock *sk = sock->sk;
3789 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3790 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3792 EXPORT_SYMBOL(sock_common_getsockopt);
3794 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3797 struct sock *sk = sock->sk;
3801 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3803 msg->msg_namelen = addr_len;
3806 EXPORT_SYMBOL(sock_common_recvmsg);
3809 * Set socket options on an inet socket.
3811 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3812 sockptr_t optval, unsigned int optlen)
3814 struct sock *sk = sock->sk;
3816 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3817 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3819 EXPORT_SYMBOL(sock_common_setsockopt);
3821 void sk_common_release(struct sock *sk)
3823 if (sk->sk_prot->destroy)
3824 sk->sk_prot->destroy(sk);
3827 * Observation: when sk_common_release is called, processes have
3828 * no access to socket. But net still has.
3829 * Step one, detach it from networking:
3831 * A. Remove from hash tables.
3834 sk->sk_prot->unhash(sk);
3837 * In this point socket cannot receive new packets, but it is possible
3838 * that some packets are in flight because some CPU runs receiver and
3839 * did hash table lookup before we unhashed socket. They will achieve
3840 * receive queue and will be purged by socket destructor.
3842 * Also we still have packets pending on receive queue and probably,
3843 * our own packets waiting in device queues. sock_destroy will drain
3844 * receive queue, but transmitted packets will delay socket destruction
3845 * until the last reference will be released.
3850 xfrm_sk_free_policy(sk);
3854 EXPORT_SYMBOL(sk_common_release);
3856 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3858 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3860 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3861 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3862 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3863 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3864 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3865 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3866 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3867 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3868 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3871 #ifdef CONFIG_PROC_FS
3872 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3874 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3876 int cpu, idx = prot->inuse_idx;
3879 for_each_possible_cpu(cpu)
3880 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3882 return res >= 0 ? res : 0;
3884 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3886 int sock_inuse_get(struct net *net)
3890 for_each_possible_cpu(cpu)
3891 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3896 EXPORT_SYMBOL_GPL(sock_inuse_get);
3898 static int __net_init sock_inuse_init_net(struct net *net)
3900 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3901 if (net->core.prot_inuse == NULL)
3906 static void __net_exit sock_inuse_exit_net(struct net *net)
3908 free_percpu(net->core.prot_inuse);
3911 static struct pernet_operations net_inuse_ops = {
3912 .init = sock_inuse_init_net,
3913 .exit = sock_inuse_exit_net,
3916 static __init int net_inuse_init(void)
3918 if (register_pernet_subsys(&net_inuse_ops))
3919 panic("Cannot initialize net inuse counters");
3924 core_initcall(net_inuse_init);
3926 static int assign_proto_idx(struct proto *prot)
3928 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3930 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3931 pr_err("PROTO_INUSE_NR exhausted\n");
3935 set_bit(prot->inuse_idx, proto_inuse_idx);
3939 static void release_proto_idx(struct proto *prot)
3941 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3942 clear_bit(prot->inuse_idx, proto_inuse_idx);
3945 static inline int assign_proto_idx(struct proto *prot)
3950 static inline void release_proto_idx(struct proto *prot)
3956 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3960 kfree(twsk_prot->twsk_slab_name);
3961 twsk_prot->twsk_slab_name = NULL;
3962 kmem_cache_destroy(twsk_prot->twsk_slab);
3963 twsk_prot->twsk_slab = NULL;
3966 static int tw_prot_init(const struct proto *prot)
3968 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3973 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3975 if (!twsk_prot->twsk_slab_name)
3978 twsk_prot->twsk_slab =
3979 kmem_cache_create(twsk_prot->twsk_slab_name,
3980 twsk_prot->twsk_obj_size, 0,
3981 SLAB_ACCOUNT | prot->slab_flags,
3983 if (!twsk_prot->twsk_slab) {
3984 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3992 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3996 kfree(rsk_prot->slab_name);
3997 rsk_prot->slab_name = NULL;
3998 kmem_cache_destroy(rsk_prot->slab);
3999 rsk_prot->slab = NULL;
4002 static int req_prot_init(const struct proto *prot)
4004 struct request_sock_ops *rsk_prot = prot->rsk_prot;
4009 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
4011 if (!rsk_prot->slab_name)
4014 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
4015 rsk_prot->obj_size, 0,
4016 SLAB_ACCOUNT | prot->slab_flags,
4019 if (!rsk_prot->slab) {
4020 pr_crit("%s: Can't create request sock SLAB cache!\n",
4027 int proto_register(struct proto *prot, int alloc_slab)
4031 if (prot->memory_allocated && !prot->sysctl_mem) {
4032 pr_err("%s: missing sysctl_mem\n", prot->name);
4035 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
4036 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
4040 prot->slab = kmem_cache_create_usercopy(prot->name,
4042 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
4044 prot->useroffset, prot->usersize,
4047 if (prot->slab == NULL) {
4048 pr_crit("%s: Can't create sock SLAB cache!\n",
4053 if (req_prot_init(prot))
4054 goto out_free_request_sock_slab;
4056 if (tw_prot_init(prot))
4057 goto out_free_timewait_sock_slab;
4060 mutex_lock(&proto_list_mutex);
4061 ret = assign_proto_idx(prot);
4063 mutex_unlock(&proto_list_mutex);
4064 goto out_free_timewait_sock_slab;
4066 list_add(&prot->node, &proto_list);
4067 mutex_unlock(&proto_list_mutex);
4070 out_free_timewait_sock_slab:
4072 tw_prot_cleanup(prot->twsk_prot);
4073 out_free_request_sock_slab:
4075 req_prot_cleanup(prot->rsk_prot);
4077 kmem_cache_destroy(prot->slab);
4083 EXPORT_SYMBOL(proto_register);
4085 void proto_unregister(struct proto *prot)
4087 mutex_lock(&proto_list_mutex);
4088 release_proto_idx(prot);
4089 list_del(&prot->node);
4090 mutex_unlock(&proto_list_mutex);
4092 kmem_cache_destroy(prot->slab);
4095 req_prot_cleanup(prot->rsk_prot);
4096 tw_prot_cleanup(prot->twsk_prot);
4098 EXPORT_SYMBOL(proto_unregister);
4100 int sock_load_diag_module(int family, int protocol)
4103 if (!sock_is_registered(family))
4106 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4107 NETLINK_SOCK_DIAG, family);
4111 if (family == AF_INET &&
4112 protocol != IPPROTO_RAW &&
4113 protocol < MAX_INET_PROTOS &&
4114 !rcu_access_pointer(inet_protos[protocol]))
4118 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4119 NETLINK_SOCK_DIAG, family, protocol);
4121 EXPORT_SYMBOL(sock_load_diag_module);
4123 #ifdef CONFIG_PROC_FS
4124 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4125 __acquires(proto_list_mutex)
4127 mutex_lock(&proto_list_mutex);
4128 return seq_list_start_head(&proto_list, *pos);
4131 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4133 return seq_list_next(v, &proto_list, pos);
4136 static void proto_seq_stop(struct seq_file *seq, void *v)
4137 __releases(proto_list_mutex)
4139 mutex_unlock(&proto_list_mutex);
4142 static char proto_method_implemented(const void *method)
4144 return method == NULL ? 'n' : 'y';
4146 static long sock_prot_memory_allocated(struct proto *proto)
4148 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4151 static const char *sock_prot_memory_pressure(struct proto *proto)
4153 return proto->memory_pressure != NULL ?
4154 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4157 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4160 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4161 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4164 sock_prot_inuse_get(seq_file_net(seq), proto),
4165 sock_prot_memory_allocated(proto),
4166 sock_prot_memory_pressure(proto),
4168 proto->slab == NULL ? "no" : "yes",
4169 module_name(proto->owner),
4170 proto_method_implemented(proto->close),
4171 proto_method_implemented(proto->connect),
4172 proto_method_implemented(proto->disconnect),
4173 proto_method_implemented(proto->accept),
4174 proto_method_implemented(proto->ioctl),
4175 proto_method_implemented(proto->init),
4176 proto_method_implemented(proto->destroy),
4177 proto_method_implemented(proto->shutdown),
4178 proto_method_implemented(proto->setsockopt),
4179 proto_method_implemented(proto->getsockopt),
4180 proto_method_implemented(proto->sendmsg),
4181 proto_method_implemented(proto->recvmsg),
4182 proto_method_implemented(proto->bind),
4183 proto_method_implemented(proto->backlog_rcv),
4184 proto_method_implemented(proto->hash),
4185 proto_method_implemented(proto->unhash),
4186 proto_method_implemented(proto->get_port),
4187 proto_method_implemented(proto->enter_memory_pressure));
4190 static int proto_seq_show(struct seq_file *seq, void *v)
4192 if (v == &proto_list)
4193 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4202 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4204 proto_seq_printf(seq, list_entry(v, struct proto, node));
4208 static const struct seq_operations proto_seq_ops = {
4209 .start = proto_seq_start,
4210 .next = proto_seq_next,
4211 .stop = proto_seq_stop,
4212 .show = proto_seq_show,
4215 static __net_init int proto_init_net(struct net *net)
4217 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4218 sizeof(struct seq_net_private)))
4224 static __net_exit void proto_exit_net(struct net *net)
4226 remove_proc_entry("protocols", net->proc_net);
4230 static __net_initdata struct pernet_operations proto_net_ops = {
4231 .init = proto_init_net,
4232 .exit = proto_exit_net,
4235 static int __init proto_init(void)
4237 return register_pernet_subsys(&proto_net_ops);
4240 subsys_initcall(proto_init);
4242 #endif /* PROC_FS */
4244 #ifdef CONFIG_NET_RX_BUSY_POLL
4245 bool sk_busy_loop_end(void *p, unsigned long start_time)
4247 struct sock *sk = p;
4249 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4252 if (sk_is_udp(sk) &&
4253 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4256 return sk_busy_loop_timeout(sk, start_time);
4258 EXPORT_SYMBOL(sk_busy_loop_end);
4259 #endif /* CONFIG_NET_RX_BUSY_POLL */
4261 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4263 if (!sk->sk_prot->bind_add)
4265 return sk->sk_prot->bind_add(sk, addr, addr_len);
4267 EXPORT_SYMBOL(sock_bind_add);
4269 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4270 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4271 void __user *arg, void *karg, size_t size)
4275 if (copy_from_user(karg, arg, size))
4278 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4282 if (copy_to_user(arg, karg, size))
4287 EXPORT_SYMBOL(sock_ioctl_inout);
4289 /* This is the most common ioctl prep function, where the result (4 bytes) is
4290 * copied back to userspace if the ioctl() returns successfully. No input is
4291 * copied from userspace as input argument.
4293 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4297 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4301 return put_user(karg, (int __user *)arg);
4304 /* A wrapper around sock ioctls, which copies the data from userspace
4305 * (depending on the protocol/ioctl), and copies back the result to userspace.
4306 * The main motivation for this function is to pass kernel memory to the
4307 * protocol ioctl callbacks, instead of userspace memory.
4309 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4313 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4314 rc = ipmr_sk_ioctl(sk, cmd, arg);
4315 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4316 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4317 else if (sk_is_phonet(sk))
4318 rc = phonet_sk_ioctl(sk, cmd, arg);
4320 /* If ioctl was processed, returns its value */
4324 /* Otherwise call the default handler */
4325 return sock_ioctl_out(sk, cmd, arg);
4327 EXPORT_SYMBOL(sk_ioctl);
4329 static int __init sock_struct_check(void)
4331 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4332 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4333 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4334 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4335 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4337 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4338 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4339 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4340 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4341 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4342 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4343 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4344 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4345 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4347 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4348 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4349 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4351 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4352 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4353 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4354 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4356 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4357 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4358 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf);
4359 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4360 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4361 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4362 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4363 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4364 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4365 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm);
4366 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status);
4367 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4368 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4369 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4370 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4371 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4373 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4374 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4375 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4376 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4377 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4378 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4379 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4380 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4381 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4382 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4383 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4384 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4385 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4389 core_initcall(sock_struct_check);