1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
118 #include <linux/uaccess.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
137 #include <trace/events/sock.h>
140 #include <net/busy_poll.h>
142 #include <linux/ethtool.h>
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
149 static void sock_def_write_space_wfree(struct sock *sk);
150 static void sock_def_write_space(struct sock *sk);
153 * sk_ns_capable - General socket capability test
154 * @sk: Socket to use a capability on or through
155 * @user_ns: The user namespace of the capability to use
156 * @cap: The capability to use
158 * Test to see if the opener of the socket had when the socket was
159 * created and the current process has the capability @cap in the user
160 * namespace @user_ns.
162 bool sk_ns_capable(const struct sock *sk,
163 struct user_namespace *user_ns, int cap)
165 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
166 ns_capable(user_ns, cap);
168 EXPORT_SYMBOL(sk_ns_capable);
171 * sk_capable - Socket global capability test
172 * @sk: Socket to use a capability on or through
173 * @cap: The global capability to use
175 * Test to see if the opener of the socket had when the socket was
176 * created and the current process has the capability @cap in all user
179 bool sk_capable(const struct sock *sk, int cap)
181 return sk_ns_capable(sk, &init_user_ns, cap);
183 EXPORT_SYMBOL(sk_capable);
186 * sk_net_capable - Network namespace socket capability test
187 * @sk: Socket to use a capability on or through
188 * @cap: The capability to use
190 * Test to see if the opener of the socket had when the socket was created
191 * and the current process has the capability @cap over the network namespace
192 * the socket is a member of.
194 bool sk_net_capable(const struct sock *sk, int cap)
196 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
198 EXPORT_SYMBOL(sk_net_capable);
201 * Each address family might have different locking rules, so we have
202 * one slock key per address family and separate keys for internal and
205 static struct lock_class_key af_family_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_keys[AF_MAX];
207 static struct lock_class_key af_family_slock_keys[AF_MAX];
208 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
211 * Make lock validator output more readable. (we pre-construct these
212 * strings build-time, so that runtime initialization of socket
216 #define _sock_locks(x) \
217 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
218 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
219 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
220 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
221 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
222 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
223 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
224 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
225 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
226 x "27" , x "28" , x "AF_CAN" , \
227 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
228 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
229 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
230 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
231 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
235 static const char *const af_family_key_strings[AF_MAX+1] = {
236 _sock_locks("sk_lock-")
238 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
239 _sock_locks("slock-")
241 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
242 _sock_locks("clock-")
245 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
246 _sock_locks("k-sk_lock-")
248 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
249 _sock_locks("k-slock-")
251 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
252 _sock_locks("k-clock-")
254 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
255 _sock_locks("rlock-")
257 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
258 _sock_locks("wlock-")
260 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
261 _sock_locks("elock-")
265 * sk_callback_lock and sk queues locking rules are per-address-family,
266 * so split the lock classes by using a per-AF key:
268 static struct lock_class_key af_callback_keys[AF_MAX];
269 static struct lock_class_key af_rlock_keys[AF_MAX];
270 static struct lock_class_key af_wlock_keys[AF_MAX];
271 static struct lock_class_key af_elock_keys[AF_MAX];
272 static struct lock_class_key af_kern_callback_keys[AF_MAX];
274 /* Run time adjustable parameters. */
275 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
276 EXPORT_SYMBOL(sysctl_wmem_max);
277 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
278 EXPORT_SYMBOL(sysctl_rmem_max);
279 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
280 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
282 /* Maximal space eaten by iovec or ancillary data plus some space */
283 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
284 EXPORT_SYMBOL(sysctl_optmem_max);
286 int sysctl_tstamp_allow_data __read_mostly = 1;
288 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
289 EXPORT_SYMBOL_GPL(memalloc_socks_key);
292 * sk_set_memalloc - sets %SOCK_MEMALLOC
293 * @sk: socket to set it on
295 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
296 * It's the responsibility of the admin to adjust min_free_kbytes
297 * to meet the requirements
299 void sk_set_memalloc(struct sock *sk)
301 sock_set_flag(sk, SOCK_MEMALLOC);
302 sk->sk_allocation |= __GFP_MEMALLOC;
303 static_branch_inc(&memalloc_socks_key);
305 EXPORT_SYMBOL_GPL(sk_set_memalloc);
307 void sk_clear_memalloc(struct sock *sk)
309 sock_reset_flag(sk, SOCK_MEMALLOC);
310 sk->sk_allocation &= ~__GFP_MEMALLOC;
311 static_branch_dec(&memalloc_socks_key);
314 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
315 * progress of swapping. SOCK_MEMALLOC may be cleared while
316 * it has rmem allocations due to the last swapfile being deactivated
317 * but there is a risk that the socket is unusable due to exceeding
318 * the rmem limits. Reclaim the reserves and obey rmem limits again.
322 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
324 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
327 unsigned int noreclaim_flag;
329 /* these should have been dropped before queueing */
330 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
332 noreclaim_flag = memalloc_noreclaim_save();
333 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
337 memalloc_noreclaim_restore(noreclaim_flag);
341 EXPORT_SYMBOL(__sk_backlog_rcv);
343 void sk_error_report(struct sock *sk)
345 sk->sk_error_report(sk);
347 switch (sk->sk_family) {
351 trace_inet_sk_error_report(sk);
357 EXPORT_SYMBOL(sk_error_report);
359 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
361 struct __kernel_sock_timeval tv;
363 if (timeo == MAX_SCHEDULE_TIMEOUT) {
367 tv.tv_sec = timeo / HZ;
368 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
371 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
372 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
373 *(struct old_timeval32 *)optval = tv32;
378 struct __kernel_old_timeval old_tv;
379 old_tv.tv_sec = tv.tv_sec;
380 old_tv.tv_usec = tv.tv_usec;
381 *(struct __kernel_old_timeval *)optval = old_tv;
382 return sizeof(old_tv);
385 *(struct __kernel_sock_timeval *)optval = tv;
388 EXPORT_SYMBOL(sock_get_timeout);
390 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
391 sockptr_t optval, int optlen, bool old_timeval)
393 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
394 struct old_timeval32 tv32;
396 if (optlen < sizeof(tv32))
399 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
401 tv->tv_sec = tv32.tv_sec;
402 tv->tv_usec = tv32.tv_usec;
403 } else if (old_timeval) {
404 struct __kernel_old_timeval old_tv;
406 if (optlen < sizeof(old_tv))
408 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
410 tv->tv_sec = old_tv.tv_sec;
411 tv->tv_usec = old_tv.tv_usec;
413 if (optlen < sizeof(*tv))
415 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
421 EXPORT_SYMBOL(sock_copy_user_timeval);
423 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
426 struct __kernel_sock_timeval tv;
427 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
432 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
436 static int warned __read_mostly;
439 if (warned < 10 && net_ratelimit()) {
441 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
442 __func__, current->comm, task_pid_nr(current));
446 *timeo_p = MAX_SCHEDULE_TIMEOUT;
447 if (tv.tv_sec == 0 && tv.tv_usec == 0)
449 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
450 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
454 static bool sock_needs_netstamp(const struct sock *sk)
456 switch (sk->sk_family) {
465 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
467 if (sk->sk_flags & flags) {
468 sk->sk_flags &= ~flags;
469 if (sock_needs_netstamp(sk) &&
470 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
471 net_disable_timestamp();
476 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
479 struct sk_buff_head *list = &sk->sk_receive_queue;
481 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
482 atomic_inc(&sk->sk_drops);
483 trace_sock_rcvqueue_full(sk, skb);
487 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
488 atomic_inc(&sk->sk_drops);
493 skb_set_owner_r(skb, sk);
495 /* we escape from rcu protected region, make sure we dont leak
500 spin_lock_irqsave(&list->lock, flags);
501 sock_skb_set_dropcount(sk, skb);
502 __skb_queue_tail(list, skb);
503 spin_unlock_irqrestore(&list->lock, flags);
505 if (!sock_flag(sk, SOCK_DEAD))
506 sk->sk_data_ready(sk);
509 EXPORT_SYMBOL(__sock_queue_rcv_skb);
511 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
512 enum skb_drop_reason *reason)
514 enum skb_drop_reason drop_reason;
517 err = sk_filter(sk, skb);
519 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
522 err = __sock_queue_rcv_skb(sk, skb);
525 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
528 drop_reason = SKB_DROP_REASON_PROTO_MEM;
531 drop_reason = SKB_NOT_DROPPED_YET;
536 *reason = drop_reason;
539 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
541 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
542 const int nested, unsigned int trim_cap, bool refcounted)
544 int rc = NET_RX_SUCCESS;
546 if (sk_filter_trim_cap(sk, skb, trim_cap))
547 goto discard_and_relse;
551 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
552 atomic_inc(&sk->sk_drops);
553 goto discard_and_relse;
556 bh_lock_sock_nested(sk);
559 if (!sock_owned_by_user(sk)) {
561 * trylock + unlock semantics:
563 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
565 rc = sk_backlog_rcv(sk, skb);
567 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
568 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
570 atomic_inc(&sk->sk_drops);
571 goto discard_and_relse;
583 EXPORT_SYMBOL(__sk_receive_skb);
585 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
587 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
589 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
591 struct dst_entry *dst = __sk_dst_get(sk);
593 if (dst && dst->obsolete &&
594 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
595 dst, cookie) == NULL) {
596 sk_tx_queue_clear(sk);
597 sk->sk_dst_pending_confirm = 0;
598 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
605 EXPORT_SYMBOL(__sk_dst_check);
607 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
609 struct dst_entry *dst = sk_dst_get(sk);
611 if (dst && dst->obsolete &&
612 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
613 dst, cookie) == NULL) {
621 EXPORT_SYMBOL(sk_dst_check);
623 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
625 int ret = -ENOPROTOOPT;
626 #ifdef CONFIG_NETDEVICES
627 struct net *net = sock_net(sk);
631 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
638 /* Paired with all READ_ONCE() done locklessly. */
639 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
641 if (sk->sk_prot->rehash)
642 sk->sk_prot->rehash(sk);
653 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
659 ret = sock_bindtoindex_locked(sk, ifindex);
665 EXPORT_SYMBOL(sock_bindtoindex);
667 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
669 int ret = -ENOPROTOOPT;
670 #ifdef CONFIG_NETDEVICES
671 struct net *net = sock_net(sk);
672 char devname[IFNAMSIZ];
679 /* Bind this socket to a particular device like "eth0",
680 * as specified in the passed interface name. If the
681 * name is "" or the option length is zero the socket
684 if (optlen > IFNAMSIZ - 1)
685 optlen = IFNAMSIZ - 1;
686 memset(devname, 0, sizeof(devname));
689 if (copy_from_sockptr(devname, optval, optlen))
693 if (devname[0] != '\0') {
694 struct net_device *dev;
697 dev = dev_get_by_name_rcu(net, devname);
699 index = dev->ifindex;
706 return sock_bindtoindex(sk, index, true);
713 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
714 int __user *optlen, int len)
716 int ret = -ENOPROTOOPT;
717 #ifdef CONFIG_NETDEVICES
718 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
719 struct net *net = sock_net(sk);
720 char devname[IFNAMSIZ];
722 if (bound_dev_if == 0) {
731 ret = netdev_get_name(net, devname, bound_dev_if);
735 len = strlen(devname) + 1;
738 if (copy_to_user(optval, devname, len))
743 if (put_user(len, optlen))
754 bool sk_mc_loop(struct sock *sk)
756 if (dev_recursion_level())
760 switch (sk->sk_family) {
762 return inet_sk(sk)->mc_loop;
763 #if IS_ENABLED(CONFIG_IPV6)
765 return inet6_sk(sk)->mc_loop;
771 EXPORT_SYMBOL(sk_mc_loop);
773 void sock_set_reuseaddr(struct sock *sk)
776 sk->sk_reuse = SK_CAN_REUSE;
779 EXPORT_SYMBOL(sock_set_reuseaddr);
781 void sock_set_reuseport(struct sock *sk)
784 sk->sk_reuseport = true;
787 EXPORT_SYMBOL(sock_set_reuseport);
789 void sock_no_linger(struct sock *sk)
792 sk->sk_lingertime = 0;
793 sock_set_flag(sk, SOCK_LINGER);
796 EXPORT_SYMBOL(sock_no_linger);
798 void sock_set_priority(struct sock *sk, u32 priority)
801 sk->sk_priority = priority;
804 EXPORT_SYMBOL(sock_set_priority);
806 void sock_set_sndtimeo(struct sock *sk, s64 secs)
809 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
810 sk->sk_sndtimeo = secs * HZ;
812 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
815 EXPORT_SYMBOL(sock_set_sndtimeo);
817 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
820 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
821 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
822 sock_set_flag(sk, SOCK_RCVTSTAMP);
823 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
825 sock_reset_flag(sk, SOCK_RCVTSTAMP);
826 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
830 void sock_enable_timestamps(struct sock *sk)
833 __sock_set_timestamps(sk, true, false, true);
836 EXPORT_SYMBOL(sock_enable_timestamps);
838 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
841 case SO_TIMESTAMP_OLD:
842 __sock_set_timestamps(sk, valbool, false, false);
844 case SO_TIMESTAMP_NEW:
845 __sock_set_timestamps(sk, valbool, true, false);
847 case SO_TIMESTAMPNS_OLD:
848 __sock_set_timestamps(sk, valbool, false, true);
850 case SO_TIMESTAMPNS_NEW:
851 __sock_set_timestamps(sk, valbool, true, true);
856 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
858 struct net *net = sock_net(sk);
859 struct net_device *dev = NULL;
864 if (sk->sk_bound_dev_if)
865 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
868 pr_err("%s: sock not bind to device\n", __func__);
872 num = ethtool_get_phc_vclocks(dev, &vclock_index);
875 for (i = 0; i < num; i++) {
876 if (*(vclock_index + i) == phc_index) {
888 sk->sk_bind_phc = phc_index;
893 int sock_set_timestamping(struct sock *sk, int optname,
894 struct so_timestamping timestamping)
896 int val = timestamping.flags;
899 if (val & ~SOF_TIMESTAMPING_MASK)
902 if (val & SOF_TIMESTAMPING_OPT_ID &&
903 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
905 if ((1 << sk->sk_state) &
906 (TCPF_CLOSE | TCPF_LISTEN))
908 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
910 atomic_set(&sk->sk_tskey, 0);
914 if (val & SOF_TIMESTAMPING_OPT_STATS &&
915 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
918 if (val & SOF_TIMESTAMPING_BIND_PHC) {
919 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
924 sk->sk_tsflags = val;
925 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
927 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
928 sock_enable_timestamp(sk,
929 SOCK_TIMESTAMPING_RX_SOFTWARE);
931 sock_disable_timestamp(sk,
932 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
936 void sock_set_keepalive(struct sock *sk)
939 if (sk->sk_prot->keepalive)
940 sk->sk_prot->keepalive(sk, true);
941 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
944 EXPORT_SYMBOL(sock_set_keepalive);
946 static void __sock_set_rcvbuf(struct sock *sk, int val)
948 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
949 * as a negative value.
951 val = min_t(int, val, INT_MAX / 2);
952 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
954 /* We double it on the way in to account for "struct sk_buff" etc.
955 * overhead. Applications assume that the SO_RCVBUF setting they make
956 * will allow that much actual data to be received on that socket.
958 * Applications are unaware that "struct sk_buff" and other overheads
959 * allocate from the receive buffer during socket buffer allocation.
961 * And after considering the possible alternatives, returning the value
962 * we actually used in getsockopt is the most desirable behavior.
964 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
967 void sock_set_rcvbuf(struct sock *sk, int val)
970 __sock_set_rcvbuf(sk, val);
973 EXPORT_SYMBOL(sock_set_rcvbuf);
975 static void __sock_set_mark(struct sock *sk, u32 val)
977 if (val != sk->sk_mark) {
983 void sock_set_mark(struct sock *sk, u32 val)
986 __sock_set_mark(sk, val);
989 EXPORT_SYMBOL(sock_set_mark);
991 static void sock_release_reserved_memory(struct sock *sk, int bytes)
993 /* Round down bytes to multiple of pages */
994 bytes &= ~(SK_MEM_QUANTUM - 1);
996 WARN_ON(bytes > sk->sk_reserved_mem);
997 sk->sk_reserved_mem -= bytes;
1001 static int sock_reserve_memory(struct sock *sk, int bytes)
1007 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1013 pages = sk_mem_pages(bytes);
1015 /* pre-charge to memcg */
1016 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1017 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1021 /* pre-charge to forward_alloc */
1022 allocated = sk_memory_allocated_add(sk, pages);
1023 /* If the system goes into memory pressure with this
1024 * precharge, give up and return error.
1026 if (allocated > sk_prot_mem_limits(sk, 1)) {
1027 sk_memory_allocated_sub(sk, pages);
1028 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1031 sk->sk_forward_alloc += pages << SK_MEM_QUANTUM_SHIFT;
1033 sk->sk_reserved_mem += pages << SK_MEM_QUANTUM_SHIFT;
1039 * This is meant for all protocols to use and covers goings on
1040 * at the socket level. Everything here is generic.
1043 int sock_setsockopt(struct socket *sock, int level, int optname,
1044 sockptr_t optval, unsigned int optlen)
1046 struct so_timestamping timestamping;
1047 struct sock_txtime sk_txtime;
1048 struct sock *sk = sock->sk;
1055 * Options without arguments
1058 if (optname == SO_BINDTODEVICE)
1059 return sock_setbindtodevice(sk, optval, optlen);
1061 if (optlen < sizeof(int))
1064 if (copy_from_sockptr(&val, optval, sizeof(val)))
1067 valbool = val ? 1 : 0;
1073 if (val && !capable(CAP_NET_ADMIN))
1076 sock_valbool_flag(sk, SOCK_DBG, valbool);
1079 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1082 sk->sk_reuseport = valbool;
1091 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1095 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1098 /* Don't error on this BSD doesn't and if you think
1099 * about it this is right. Otherwise apps have to
1100 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1101 * are treated in BSD as hints
1103 val = min_t(u32, val, sysctl_wmem_max);
1105 /* Ensure val * 2 fits into an int, to prevent max_t()
1106 * from treating it as a negative value.
1108 val = min_t(int, val, INT_MAX / 2);
1109 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1110 WRITE_ONCE(sk->sk_sndbuf,
1111 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1112 /* Wake up sending tasks if we upped the value. */
1113 sk->sk_write_space(sk);
1116 case SO_SNDBUFFORCE:
1117 if (!capable(CAP_NET_ADMIN)) {
1122 /* No negative values (to prevent underflow, as val will be
1130 /* Don't error on this BSD doesn't and if you think
1131 * about it this is right. Otherwise apps have to
1132 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1133 * are treated in BSD as hints
1135 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
1138 case SO_RCVBUFFORCE:
1139 if (!capable(CAP_NET_ADMIN)) {
1144 /* No negative values (to prevent underflow, as val will be
1147 __sock_set_rcvbuf(sk, max(val, 0));
1151 if (sk->sk_prot->keepalive)
1152 sk->sk_prot->keepalive(sk, valbool);
1153 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1157 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1161 sk->sk_no_check_tx = valbool;
1165 if ((val >= 0 && val <= 6) ||
1166 ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1167 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1168 sk->sk_priority = val;
1174 if (optlen < sizeof(ling)) {
1175 ret = -EINVAL; /* 1003.1g */
1178 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1183 sock_reset_flag(sk, SOCK_LINGER);
1185 #if (BITS_PER_LONG == 32)
1186 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1187 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1190 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1191 sock_set_flag(sk, SOCK_LINGER);
1200 set_bit(SOCK_PASSCRED, &sock->flags);
1202 clear_bit(SOCK_PASSCRED, &sock->flags);
1205 case SO_TIMESTAMP_OLD:
1206 case SO_TIMESTAMP_NEW:
1207 case SO_TIMESTAMPNS_OLD:
1208 case SO_TIMESTAMPNS_NEW:
1209 sock_set_timestamp(sk, optname, valbool);
1212 case SO_TIMESTAMPING_NEW:
1213 case SO_TIMESTAMPING_OLD:
1214 if (optlen == sizeof(timestamping)) {
1215 if (copy_from_sockptr(×tamping, optval,
1216 sizeof(timestamping))) {
1221 memset(×tamping, 0, sizeof(timestamping));
1222 timestamping.flags = val;
1224 ret = sock_set_timestamping(sk, optname, timestamping);
1230 if (sock->ops->set_rcvlowat)
1231 ret = sock->ops->set_rcvlowat(sk, val);
1233 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1236 case SO_RCVTIMEO_OLD:
1237 case SO_RCVTIMEO_NEW:
1238 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1239 optlen, optname == SO_RCVTIMEO_OLD);
1242 case SO_SNDTIMEO_OLD:
1243 case SO_SNDTIMEO_NEW:
1244 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1245 optlen, optname == SO_SNDTIMEO_OLD);
1248 case SO_ATTACH_FILTER: {
1249 struct sock_fprog fprog;
1251 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1253 ret = sk_attach_filter(&fprog, sk);
1258 if (optlen == sizeof(u32)) {
1262 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1265 ret = sk_attach_bpf(ufd, sk);
1269 case SO_ATTACH_REUSEPORT_CBPF: {
1270 struct sock_fprog fprog;
1272 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1274 ret = sk_reuseport_attach_filter(&fprog, sk);
1277 case SO_ATTACH_REUSEPORT_EBPF:
1279 if (optlen == sizeof(u32)) {
1283 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1286 ret = sk_reuseport_attach_bpf(ufd, sk);
1290 case SO_DETACH_REUSEPORT_BPF:
1291 ret = reuseport_detach_prog(sk);
1294 case SO_DETACH_FILTER:
1295 ret = sk_detach_filter(sk);
1298 case SO_LOCK_FILTER:
1299 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1302 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1307 set_bit(SOCK_PASSSEC, &sock->flags);
1309 clear_bit(SOCK_PASSSEC, &sock->flags);
1312 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1313 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1318 __sock_set_mark(sk, val);
1321 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1322 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1327 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1331 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1334 case SO_WIFI_STATUS:
1335 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1339 if (sock->ops->set_peek_off)
1340 ret = sock->ops->set_peek_off(sk, val);
1346 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1349 case SO_SELECT_ERR_QUEUE:
1350 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1353 #ifdef CONFIG_NET_RX_BUSY_POLL
1355 /* allow unprivileged users to decrease the value */
1356 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1362 WRITE_ONCE(sk->sk_ll_usec, val);
1365 case SO_PREFER_BUSY_POLL:
1366 if (valbool && !capable(CAP_NET_ADMIN))
1369 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1371 case SO_BUSY_POLL_BUDGET:
1372 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1375 if (val < 0 || val > U16_MAX)
1378 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1383 case SO_MAX_PACING_RATE:
1385 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1387 if (sizeof(ulval) != sizeof(val) &&
1388 optlen >= sizeof(ulval) &&
1389 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1394 cmpxchg(&sk->sk_pacing_status,
1397 sk->sk_max_pacing_rate = ulval;
1398 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1401 case SO_INCOMING_CPU:
1402 WRITE_ONCE(sk->sk_incoming_cpu, val);
1407 dst_negative_advice(sk);
1411 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1412 if (!(sk_is_tcp(sk) ||
1413 (sk->sk_type == SOCK_DGRAM &&
1414 sk->sk_protocol == IPPROTO_UDP)))
1416 } else if (sk->sk_family != PF_RDS) {
1420 if (val < 0 || val > 1)
1423 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1428 if (optlen != sizeof(struct sock_txtime)) {
1431 } else if (copy_from_sockptr(&sk_txtime, optval,
1432 sizeof(struct sock_txtime))) {
1435 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1439 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1440 * scheduler has enough safe guards.
1442 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1443 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1447 sock_valbool_flag(sk, SOCK_TXTIME, true);
1448 sk->sk_clockid = sk_txtime.clockid;
1449 sk->sk_txtime_deadline_mode =
1450 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1451 sk->sk_txtime_report_errors =
1452 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1455 case SO_BINDTOIFINDEX:
1456 ret = sock_bindtoindex_locked(sk, val);
1460 if (val & ~SOCK_BUF_LOCK_MASK) {
1464 sk->sk_userlocks = val | (sk->sk_userlocks &
1465 ~SOCK_BUF_LOCK_MASK);
1468 case SO_RESERVE_MEM:
1477 delta = val - sk->sk_reserved_mem;
1479 sock_release_reserved_memory(sk, -delta);
1481 ret = sock_reserve_memory(sk, delta);
1486 if (val < -1 || val > 1) {
1490 /* Paired with READ_ONCE() in tcp_rtx_synack() */
1491 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1501 EXPORT_SYMBOL(sock_setsockopt);
1503 static const struct cred *sk_get_peer_cred(struct sock *sk)
1505 const struct cred *cred;
1507 spin_lock(&sk->sk_peer_lock);
1508 cred = get_cred(sk->sk_peer_cred);
1509 spin_unlock(&sk->sk_peer_lock);
1514 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1515 struct ucred *ucred)
1517 ucred->pid = pid_vnr(pid);
1518 ucred->uid = ucred->gid = -1;
1520 struct user_namespace *current_ns = current_user_ns();
1522 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1523 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1527 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1529 struct user_namespace *user_ns = current_user_ns();
1532 for (i = 0; i < src->ngroups; i++)
1533 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1539 int sock_getsockopt(struct socket *sock, int level, int optname,
1540 char __user *optval, int __user *optlen)
1542 struct sock *sk = sock->sk;
1547 unsigned long ulval;
1549 struct old_timeval32 tm32;
1550 struct __kernel_old_timeval tm;
1551 struct __kernel_sock_timeval stm;
1552 struct sock_txtime txtime;
1553 struct so_timestamping timestamping;
1556 int lv = sizeof(int);
1559 if (get_user(len, optlen))
1564 memset(&v, 0, sizeof(v));
1568 v.val = sock_flag(sk, SOCK_DBG);
1572 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1576 v.val = sock_flag(sk, SOCK_BROADCAST);
1580 v.val = sk->sk_sndbuf;
1584 v.val = sk->sk_rcvbuf;
1588 v.val = sk->sk_reuse;
1592 v.val = sk->sk_reuseport;
1596 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1600 v.val = sk->sk_type;
1604 v.val = sk->sk_protocol;
1608 v.val = sk->sk_family;
1612 v.val = -sock_error(sk);
1614 v.val = xchg(&sk->sk_err_soft, 0);
1618 v.val = sock_flag(sk, SOCK_URGINLINE);
1622 v.val = sk->sk_no_check_tx;
1626 v.val = sk->sk_priority;
1630 lv = sizeof(v.ling);
1631 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1632 v.ling.l_linger = sk->sk_lingertime / HZ;
1638 case SO_TIMESTAMP_OLD:
1639 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1640 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1641 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1644 case SO_TIMESTAMPNS_OLD:
1645 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1648 case SO_TIMESTAMP_NEW:
1649 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1652 case SO_TIMESTAMPNS_NEW:
1653 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1656 case SO_TIMESTAMPING_OLD:
1657 lv = sizeof(v.timestamping);
1658 v.timestamping.flags = sk->sk_tsflags;
1659 v.timestamping.bind_phc = sk->sk_bind_phc;
1662 case SO_RCVTIMEO_OLD:
1663 case SO_RCVTIMEO_NEW:
1664 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1667 case SO_SNDTIMEO_OLD:
1668 case SO_SNDTIMEO_NEW:
1669 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1673 v.val = sk->sk_rcvlowat;
1681 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1686 struct ucred peercred;
1687 if (len > sizeof(peercred))
1688 len = sizeof(peercred);
1690 spin_lock(&sk->sk_peer_lock);
1691 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1692 spin_unlock(&sk->sk_peer_lock);
1694 if (copy_to_user(optval, &peercred, len))
1701 const struct cred *cred;
1704 cred = sk_get_peer_cred(sk);
1708 n = cred->group_info->ngroups;
1709 if (len < n * sizeof(gid_t)) {
1710 len = n * sizeof(gid_t);
1712 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1714 len = n * sizeof(gid_t);
1716 ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1727 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1732 if (copy_to_user(optval, address, len))
1737 /* Dubious BSD thing... Probably nobody even uses it, but
1738 * the UNIX standard wants it for whatever reason... -DaveM
1741 v.val = sk->sk_state == TCP_LISTEN;
1745 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1749 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1752 v.val = sk->sk_mark;
1756 v.val = sock_flag(sk, SOCK_RCVMARK);
1760 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1763 case SO_WIFI_STATUS:
1764 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1768 if (!sock->ops->set_peek_off)
1771 v.val = sk->sk_peek_off;
1774 v.val = sock_flag(sk, SOCK_NOFCS);
1777 case SO_BINDTODEVICE:
1778 return sock_getbindtodevice(sk, optval, optlen, len);
1781 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1787 case SO_LOCK_FILTER:
1788 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1791 case SO_BPF_EXTENSIONS:
1792 v.val = bpf_tell_extensions();
1795 case SO_SELECT_ERR_QUEUE:
1796 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1799 #ifdef CONFIG_NET_RX_BUSY_POLL
1801 v.val = sk->sk_ll_usec;
1803 case SO_PREFER_BUSY_POLL:
1804 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1808 case SO_MAX_PACING_RATE:
1809 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1810 lv = sizeof(v.ulval);
1811 v.ulval = sk->sk_max_pacing_rate;
1814 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1818 case SO_INCOMING_CPU:
1819 v.val = READ_ONCE(sk->sk_incoming_cpu);
1824 u32 meminfo[SK_MEMINFO_VARS];
1826 sk_get_meminfo(sk, meminfo);
1828 len = min_t(unsigned int, len, sizeof(meminfo));
1829 if (copy_to_user(optval, &meminfo, len))
1835 #ifdef CONFIG_NET_RX_BUSY_POLL
1836 case SO_INCOMING_NAPI_ID:
1837 v.val = READ_ONCE(sk->sk_napi_id);
1839 /* aggregate non-NAPI IDs down to 0 */
1840 if (v.val < MIN_NAPI_ID)
1850 v.val64 = sock_gen_cookie(sk);
1854 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1858 lv = sizeof(v.txtime);
1859 v.txtime.clockid = sk->sk_clockid;
1860 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1861 SOF_TXTIME_DEADLINE_MODE : 0;
1862 v.txtime.flags |= sk->sk_txtime_report_errors ?
1863 SOF_TXTIME_REPORT_ERRORS : 0;
1866 case SO_BINDTOIFINDEX:
1867 v.val = READ_ONCE(sk->sk_bound_dev_if);
1870 case SO_NETNS_COOKIE:
1874 v.val64 = sock_net(sk)->net_cookie;
1878 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1881 case SO_RESERVE_MEM:
1882 v.val = sk->sk_reserved_mem;
1886 v.val = sk->sk_txrehash;
1890 /* We implement the SO_SNDLOWAT etc to not be settable
1893 return -ENOPROTOOPT;
1898 if (copy_to_user(optval, &v, len))
1901 if (put_user(len, optlen))
1907 * Initialize an sk_lock.
1909 * (We also register the sk_lock with the lock validator.)
1911 static inline void sock_lock_init(struct sock *sk)
1913 if (sk->sk_kern_sock)
1914 sock_lock_init_class_and_name(
1916 af_family_kern_slock_key_strings[sk->sk_family],
1917 af_family_kern_slock_keys + sk->sk_family,
1918 af_family_kern_key_strings[sk->sk_family],
1919 af_family_kern_keys + sk->sk_family);
1921 sock_lock_init_class_and_name(
1923 af_family_slock_key_strings[sk->sk_family],
1924 af_family_slock_keys + sk->sk_family,
1925 af_family_key_strings[sk->sk_family],
1926 af_family_keys + sk->sk_family);
1930 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1931 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1932 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1934 static void sock_copy(struct sock *nsk, const struct sock *osk)
1936 const struct proto *prot = READ_ONCE(osk->sk_prot);
1937 #ifdef CONFIG_SECURITY_NETWORK
1938 void *sptr = nsk->sk_security;
1941 /* If we move sk_tx_queue_mapping out of the private section,
1942 * we must check if sk_tx_queue_clear() is called after
1943 * sock_copy() in sk_clone_lock().
1945 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1946 offsetof(struct sock, sk_dontcopy_begin) ||
1947 offsetof(struct sock, sk_tx_queue_mapping) >=
1948 offsetof(struct sock, sk_dontcopy_end));
1950 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1952 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1953 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1955 #ifdef CONFIG_SECURITY_NETWORK
1956 nsk->sk_security = sptr;
1957 security_sk_clone(osk, nsk);
1961 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1965 struct kmem_cache *slab;
1969 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1972 if (want_init_on_alloc(priority))
1973 sk_prot_clear_nulls(sk, prot->obj_size);
1975 sk = kmalloc(prot->obj_size, priority);
1978 if (security_sk_alloc(sk, family, priority))
1981 if (!try_module_get(prot->owner))
1988 security_sk_free(sk);
1991 kmem_cache_free(slab, sk);
1997 static void sk_prot_free(struct proto *prot, struct sock *sk)
1999 struct kmem_cache *slab;
2000 struct module *owner;
2002 owner = prot->owner;
2005 cgroup_sk_free(&sk->sk_cgrp_data);
2006 mem_cgroup_sk_free(sk);
2007 security_sk_free(sk);
2009 kmem_cache_free(slab, sk);
2016 * sk_alloc - All socket objects are allocated here
2017 * @net: the applicable net namespace
2018 * @family: protocol family
2019 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2020 * @prot: struct proto associated with this new sock instance
2021 * @kern: is this to be a kernel socket?
2023 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2024 struct proto *prot, int kern)
2028 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2030 sk->sk_family = family;
2032 * See comment in struct sock definition to understand
2033 * why we need sk_prot_creator -acme
2035 sk->sk_prot = sk->sk_prot_creator = prot;
2036 sk->sk_kern_sock = kern;
2038 sk->sk_net_refcnt = kern ? 0 : 1;
2039 if (likely(sk->sk_net_refcnt)) {
2040 get_net_track(net, &sk->ns_tracker, priority);
2041 sock_inuse_add(net, 1);
2044 sock_net_set(sk, net);
2045 refcount_set(&sk->sk_wmem_alloc, 1);
2047 mem_cgroup_sk_alloc(sk);
2048 cgroup_sk_alloc(&sk->sk_cgrp_data);
2049 sock_update_classid(&sk->sk_cgrp_data);
2050 sock_update_netprioidx(&sk->sk_cgrp_data);
2051 sk_tx_queue_clear(sk);
2056 EXPORT_SYMBOL(sk_alloc);
2058 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2059 * grace period. This is the case for UDP sockets and TCP listeners.
2061 static void __sk_destruct(struct rcu_head *head)
2063 struct sock *sk = container_of(head, struct sock, sk_rcu);
2064 struct sk_filter *filter;
2066 if (sk->sk_destruct)
2067 sk->sk_destruct(sk);
2069 filter = rcu_dereference_check(sk->sk_filter,
2070 refcount_read(&sk->sk_wmem_alloc) == 0);
2072 sk_filter_uncharge(sk, filter);
2073 RCU_INIT_POINTER(sk->sk_filter, NULL);
2076 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2078 #ifdef CONFIG_BPF_SYSCALL
2079 bpf_sk_storage_free(sk);
2082 if (atomic_read(&sk->sk_omem_alloc))
2083 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2084 __func__, atomic_read(&sk->sk_omem_alloc));
2086 if (sk->sk_frag.page) {
2087 put_page(sk->sk_frag.page);
2088 sk->sk_frag.page = NULL;
2091 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2092 put_cred(sk->sk_peer_cred);
2093 put_pid(sk->sk_peer_pid);
2095 if (likely(sk->sk_net_refcnt))
2096 put_net_track(sock_net(sk), &sk->ns_tracker);
2097 sk_prot_free(sk->sk_prot_creator, sk);
2100 void sk_destruct(struct sock *sk)
2102 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2104 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2105 reuseport_detach_sock(sk);
2106 use_call_rcu = true;
2110 call_rcu(&sk->sk_rcu, __sk_destruct);
2112 __sk_destruct(&sk->sk_rcu);
2115 static void __sk_free(struct sock *sk)
2117 if (likely(sk->sk_net_refcnt))
2118 sock_inuse_add(sock_net(sk), -1);
2120 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2121 sock_diag_broadcast_destroy(sk);
2126 void sk_free(struct sock *sk)
2129 * We subtract one from sk_wmem_alloc and can know if
2130 * some packets are still in some tx queue.
2131 * If not null, sock_wfree() will call __sk_free(sk) later
2133 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2136 EXPORT_SYMBOL(sk_free);
2138 static void sk_init_common(struct sock *sk)
2140 skb_queue_head_init(&sk->sk_receive_queue);
2141 skb_queue_head_init(&sk->sk_write_queue);
2142 skb_queue_head_init(&sk->sk_error_queue);
2144 rwlock_init(&sk->sk_callback_lock);
2145 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2146 af_rlock_keys + sk->sk_family,
2147 af_family_rlock_key_strings[sk->sk_family]);
2148 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2149 af_wlock_keys + sk->sk_family,
2150 af_family_wlock_key_strings[sk->sk_family]);
2151 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2152 af_elock_keys + sk->sk_family,
2153 af_family_elock_key_strings[sk->sk_family]);
2154 lockdep_set_class_and_name(&sk->sk_callback_lock,
2155 af_callback_keys + sk->sk_family,
2156 af_family_clock_key_strings[sk->sk_family]);
2160 * sk_clone_lock - clone a socket, and lock its clone
2161 * @sk: the socket to clone
2162 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2164 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2166 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2168 struct proto *prot = READ_ONCE(sk->sk_prot);
2169 struct sk_filter *filter;
2170 bool is_charged = true;
2173 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2177 sock_copy(newsk, sk);
2179 newsk->sk_prot_creator = prot;
2182 if (likely(newsk->sk_net_refcnt)) {
2183 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2184 sock_inuse_add(sock_net(newsk), 1);
2186 sk_node_init(&newsk->sk_node);
2187 sock_lock_init(newsk);
2188 bh_lock_sock(newsk);
2189 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2190 newsk->sk_backlog.len = 0;
2192 atomic_set(&newsk->sk_rmem_alloc, 0);
2194 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2195 refcount_set(&newsk->sk_wmem_alloc, 1);
2197 atomic_set(&newsk->sk_omem_alloc, 0);
2198 sk_init_common(newsk);
2200 newsk->sk_dst_cache = NULL;
2201 newsk->sk_dst_pending_confirm = 0;
2202 newsk->sk_wmem_queued = 0;
2203 newsk->sk_forward_alloc = 0;
2204 newsk->sk_reserved_mem = 0;
2205 atomic_set(&newsk->sk_drops, 0);
2206 newsk->sk_send_head = NULL;
2207 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2208 atomic_set(&newsk->sk_zckey, 0);
2210 sock_reset_flag(newsk, SOCK_DONE);
2212 /* sk->sk_memcg will be populated at accept() time */
2213 newsk->sk_memcg = NULL;
2215 cgroup_sk_clone(&newsk->sk_cgrp_data);
2218 filter = rcu_dereference(sk->sk_filter);
2220 /* though it's an empty new sock, the charging may fail
2221 * if sysctl_optmem_max was changed between creation of
2222 * original socket and cloning
2224 is_charged = sk_filter_charge(newsk, filter);
2225 RCU_INIT_POINTER(newsk->sk_filter, filter);
2228 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2229 /* We need to make sure that we don't uncharge the new
2230 * socket if we couldn't charge it in the first place
2231 * as otherwise we uncharge the parent's filter.
2234 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2235 sk_free_unlock_clone(newsk);
2239 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2241 if (bpf_sk_storage_clone(sk, newsk)) {
2242 sk_free_unlock_clone(newsk);
2247 /* Clear sk_user_data if parent had the pointer tagged
2248 * as not suitable for copying when cloning.
2250 if (sk_user_data_is_nocopy(newsk))
2251 newsk->sk_user_data = NULL;
2254 newsk->sk_err_soft = 0;
2255 newsk->sk_priority = 0;
2256 newsk->sk_incoming_cpu = raw_smp_processor_id();
2258 /* Before updating sk_refcnt, we must commit prior changes to memory
2259 * (Documentation/RCU/rculist_nulls.rst for details)
2262 refcount_set(&newsk->sk_refcnt, 2);
2264 /* Increment the counter in the same struct proto as the master
2265 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2266 * is the same as sk->sk_prot->socks, as this field was copied
2269 * This _changes_ the previous behaviour, where
2270 * tcp_create_openreq_child always was incrementing the
2271 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2272 * to be taken into account in all callers. -acme
2274 sk_refcnt_debug_inc(newsk);
2275 sk_set_socket(newsk, NULL);
2276 sk_tx_queue_clear(newsk);
2277 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2279 if (newsk->sk_prot->sockets_allocated)
2280 sk_sockets_allocated_inc(newsk);
2282 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2283 net_enable_timestamp();
2287 EXPORT_SYMBOL_GPL(sk_clone_lock);
2289 void sk_free_unlock_clone(struct sock *sk)
2291 /* It is still raw copy of parent, so invalidate
2292 * destructor and make plain sk_free() */
2293 sk->sk_destruct = NULL;
2297 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2299 static void sk_trim_gso_size(struct sock *sk)
2301 if (sk->sk_gso_max_size <= GSO_LEGACY_MAX_SIZE)
2303 #if IS_ENABLED(CONFIG_IPV6)
2304 if (sk->sk_family == AF_INET6 &&
2306 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
2309 sk->sk_gso_max_size = GSO_LEGACY_MAX_SIZE;
2312 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2316 sk_dst_set(sk, dst);
2317 sk->sk_route_caps = dst->dev->features;
2319 sk->sk_route_caps |= NETIF_F_GSO;
2320 if (sk->sk_route_caps & NETIF_F_GSO)
2321 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2322 if (unlikely(sk->sk_gso_disabled))
2323 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2324 if (sk_can_gso(sk)) {
2325 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2326 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2328 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2329 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */
2330 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size);
2331 sk_trim_gso_size(sk);
2332 sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1);
2333 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2334 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2337 sk->sk_gso_max_segs = max_segs;
2339 EXPORT_SYMBOL_GPL(sk_setup_caps);
2342 * Simple resource managers for sockets.
2347 * Write buffer destructor automatically called from kfree_skb.
2349 void sock_wfree(struct sk_buff *skb)
2351 struct sock *sk = skb->sk;
2352 unsigned int len = skb->truesize;
2355 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2356 if (sock_flag(sk, SOCK_RCU_FREE) &&
2357 sk->sk_write_space == sock_def_write_space) {
2359 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2360 sock_def_write_space_wfree(sk);
2368 * Keep a reference on sk_wmem_alloc, this will be released
2369 * after sk_write_space() call
2371 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2372 sk->sk_write_space(sk);
2376 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2377 * could not do because of in-flight packets
2379 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2382 EXPORT_SYMBOL(sock_wfree);
2384 /* This variant of sock_wfree() is used by TCP,
2385 * since it sets SOCK_USE_WRITE_QUEUE.
2387 void __sock_wfree(struct sk_buff *skb)
2389 struct sock *sk = skb->sk;
2391 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2395 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2400 if (unlikely(!sk_fullsock(sk))) {
2401 skb->destructor = sock_edemux;
2406 skb->destructor = sock_wfree;
2407 skb_set_hash_from_sk(skb, sk);
2409 * We used to take a refcount on sk, but following operation
2410 * is enough to guarantee sk_free() wont free this sock until
2411 * all in-flight packets are completed
2413 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2415 EXPORT_SYMBOL(skb_set_owner_w);
2417 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2419 #ifdef CONFIG_TLS_DEVICE
2420 /* Drivers depend on in-order delivery for crypto offload,
2421 * partial orphan breaks out-of-order-OK logic.
2426 return (skb->destructor == sock_wfree ||
2427 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2430 /* This helper is used by netem, as it can hold packets in its
2431 * delay queue. We want to allow the owner socket to send more
2432 * packets, as if they were already TX completed by a typical driver.
2433 * But we also want to keep skb->sk set because some packet schedulers
2434 * rely on it (sch_fq for example).
2436 void skb_orphan_partial(struct sk_buff *skb)
2438 if (skb_is_tcp_pure_ack(skb))
2441 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2446 EXPORT_SYMBOL(skb_orphan_partial);
2449 * Read buffer destructor automatically called from kfree_skb.
2451 void sock_rfree(struct sk_buff *skb)
2453 struct sock *sk = skb->sk;
2454 unsigned int len = skb->truesize;
2456 atomic_sub(len, &sk->sk_rmem_alloc);
2457 sk_mem_uncharge(sk, len);
2459 EXPORT_SYMBOL(sock_rfree);
2462 * Buffer destructor for skbs that are not used directly in read or write
2463 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2465 void sock_efree(struct sk_buff *skb)
2469 EXPORT_SYMBOL(sock_efree);
2471 /* Buffer destructor for prefetch/receive path where reference count may
2472 * not be held, e.g. for listen sockets.
2475 void sock_pfree(struct sk_buff *skb)
2477 if (sk_is_refcounted(skb->sk))
2478 sock_gen_put(skb->sk);
2480 EXPORT_SYMBOL(sock_pfree);
2481 #endif /* CONFIG_INET */
2483 kuid_t sock_i_uid(struct sock *sk)
2487 read_lock_bh(&sk->sk_callback_lock);
2488 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2489 read_unlock_bh(&sk->sk_callback_lock);
2492 EXPORT_SYMBOL(sock_i_uid);
2494 unsigned long sock_i_ino(struct sock *sk)
2498 read_lock_bh(&sk->sk_callback_lock);
2499 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2500 read_unlock_bh(&sk->sk_callback_lock);
2503 EXPORT_SYMBOL(sock_i_ino);
2506 * Allocate a skb from the socket's send buffer.
2508 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2512 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2513 struct sk_buff *skb = alloc_skb(size, priority);
2516 skb_set_owner_w(skb, sk);
2522 EXPORT_SYMBOL(sock_wmalloc);
2524 static void sock_ofree(struct sk_buff *skb)
2526 struct sock *sk = skb->sk;
2528 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2531 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2534 struct sk_buff *skb;
2536 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2537 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2541 skb = alloc_skb(size, priority);
2545 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2547 skb->destructor = sock_ofree;
2552 * Allocate a memory block from the socket's option memory buffer.
2554 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2556 if ((unsigned int)size <= sysctl_optmem_max &&
2557 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2559 /* First do the add, to avoid the race if kmalloc
2562 atomic_add(size, &sk->sk_omem_alloc);
2563 mem = kmalloc(size, priority);
2566 atomic_sub(size, &sk->sk_omem_alloc);
2570 EXPORT_SYMBOL(sock_kmalloc);
2572 /* Free an option memory block. Note, we actually want the inline
2573 * here as this allows gcc to detect the nullify and fold away the
2574 * condition entirely.
2576 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2579 if (WARN_ON_ONCE(!mem))
2582 kfree_sensitive(mem);
2585 atomic_sub(size, &sk->sk_omem_alloc);
2588 void sock_kfree_s(struct sock *sk, void *mem, int size)
2590 __sock_kfree_s(sk, mem, size, false);
2592 EXPORT_SYMBOL(sock_kfree_s);
2594 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2596 __sock_kfree_s(sk, mem, size, true);
2598 EXPORT_SYMBOL(sock_kzfree_s);
2600 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2601 I think, these locks should be removed for datagram sockets.
2603 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2607 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2611 if (signal_pending(current))
2613 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2614 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2615 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2617 if (sk->sk_shutdown & SEND_SHUTDOWN)
2621 timeo = schedule_timeout(timeo);
2623 finish_wait(sk_sleep(sk), &wait);
2629 * Generic send/receive buffer handlers
2632 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2633 unsigned long data_len, int noblock,
2634 int *errcode, int max_page_order)
2636 struct sk_buff *skb;
2640 timeo = sock_sndtimeo(sk, noblock);
2642 err = sock_error(sk);
2647 if (sk->sk_shutdown & SEND_SHUTDOWN)
2650 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2653 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2654 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2658 if (signal_pending(current))
2660 timeo = sock_wait_for_wmem(sk, timeo);
2662 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2663 errcode, sk->sk_allocation);
2665 skb_set_owner_w(skb, sk);
2669 err = sock_intr_errno(timeo);
2674 EXPORT_SYMBOL(sock_alloc_send_pskb);
2676 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2677 struct sockcm_cookie *sockc)
2681 switch (cmsg->cmsg_type) {
2683 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2684 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2686 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2688 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2690 case SO_TIMESTAMPING_OLD:
2691 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2694 tsflags = *(u32 *)CMSG_DATA(cmsg);
2695 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2698 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2699 sockc->tsflags |= tsflags;
2702 if (!sock_flag(sk, SOCK_TXTIME))
2704 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2706 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2708 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2710 case SCM_CREDENTIALS:
2717 EXPORT_SYMBOL(__sock_cmsg_send);
2719 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2720 struct sockcm_cookie *sockc)
2722 struct cmsghdr *cmsg;
2725 for_each_cmsghdr(cmsg, msg) {
2726 if (!CMSG_OK(msg, cmsg))
2728 if (cmsg->cmsg_level != SOL_SOCKET)
2730 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2736 EXPORT_SYMBOL(sock_cmsg_send);
2738 static void sk_enter_memory_pressure(struct sock *sk)
2740 if (!sk->sk_prot->enter_memory_pressure)
2743 sk->sk_prot->enter_memory_pressure(sk);
2746 static void sk_leave_memory_pressure(struct sock *sk)
2748 if (sk->sk_prot->leave_memory_pressure) {
2749 sk->sk_prot->leave_memory_pressure(sk);
2751 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2753 if (memory_pressure && READ_ONCE(*memory_pressure))
2754 WRITE_ONCE(*memory_pressure, 0);
2758 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2761 * skb_page_frag_refill - check that a page_frag contains enough room
2762 * @sz: minimum size of the fragment we want to get
2763 * @pfrag: pointer to page_frag
2764 * @gfp: priority for memory allocation
2766 * Note: While this allocator tries to use high order pages, there is
2767 * no guarantee that allocations succeed. Therefore, @sz MUST be
2768 * less or equal than PAGE_SIZE.
2770 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2773 if (page_ref_count(pfrag->page) == 1) {
2777 if (pfrag->offset + sz <= pfrag->size)
2779 put_page(pfrag->page);
2783 if (SKB_FRAG_PAGE_ORDER &&
2784 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2785 /* Avoid direct reclaim but allow kswapd to wake */
2786 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2787 __GFP_COMP | __GFP_NOWARN |
2789 SKB_FRAG_PAGE_ORDER);
2790 if (likely(pfrag->page)) {
2791 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2795 pfrag->page = alloc_page(gfp);
2796 if (likely(pfrag->page)) {
2797 pfrag->size = PAGE_SIZE;
2802 EXPORT_SYMBOL(skb_page_frag_refill);
2804 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2806 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2809 sk_enter_memory_pressure(sk);
2810 sk_stream_moderate_sndbuf(sk);
2813 EXPORT_SYMBOL(sk_page_frag_refill);
2815 void __lock_sock(struct sock *sk)
2816 __releases(&sk->sk_lock.slock)
2817 __acquires(&sk->sk_lock.slock)
2822 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2823 TASK_UNINTERRUPTIBLE);
2824 spin_unlock_bh(&sk->sk_lock.slock);
2826 spin_lock_bh(&sk->sk_lock.slock);
2827 if (!sock_owned_by_user(sk))
2830 finish_wait(&sk->sk_lock.wq, &wait);
2833 void __release_sock(struct sock *sk)
2834 __releases(&sk->sk_lock.slock)
2835 __acquires(&sk->sk_lock.slock)
2837 struct sk_buff *skb, *next;
2839 while ((skb = sk->sk_backlog.head) != NULL) {
2840 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2842 spin_unlock_bh(&sk->sk_lock.slock);
2847 WARN_ON_ONCE(skb_dst_is_noref(skb));
2848 skb_mark_not_on_list(skb);
2849 sk_backlog_rcv(sk, skb);
2854 } while (skb != NULL);
2856 spin_lock_bh(&sk->sk_lock.slock);
2860 * Doing the zeroing here guarantee we can not loop forever
2861 * while a wild producer attempts to flood us.
2863 sk->sk_backlog.len = 0;
2866 void __sk_flush_backlog(struct sock *sk)
2868 spin_lock_bh(&sk->sk_lock.slock);
2870 spin_unlock_bh(&sk->sk_lock.slock);
2874 * sk_wait_data - wait for data to arrive at sk_receive_queue
2875 * @sk: sock to wait on
2876 * @timeo: for how long
2877 * @skb: last skb seen on sk_receive_queue
2879 * Now socket state including sk->sk_err is changed only under lock,
2880 * hence we may omit checks after joining wait queue.
2881 * We check receive queue before schedule() only as optimization;
2882 * it is very likely that release_sock() added new data.
2884 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2886 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2889 add_wait_queue(sk_sleep(sk), &wait);
2890 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2891 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2892 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2893 remove_wait_queue(sk_sleep(sk), &wait);
2896 EXPORT_SYMBOL(sk_wait_data);
2899 * __sk_mem_raise_allocated - increase memory_allocated
2901 * @size: memory size to allocate
2902 * @amt: pages to allocate
2903 * @kind: allocation type
2905 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2907 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2909 struct proto *prot = sk->sk_prot;
2910 long allocated = sk_memory_allocated_add(sk, amt);
2911 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2912 bool charged = true;
2915 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2916 gfp_memcg_charge())))
2917 goto suppress_allocation;
2920 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2921 sk_leave_memory_pressure(sk);
2925 /* Under pressure. */
2926 if (allocated > sk_prot_mem_limits(sk, 1))
2927 sk_enter_memory_pressure(sk);
2929 /* Over hard limit. */
2930 if (allocated > sk_prot_mem_limits(sk, 2))
2931 goto suppress_allocation;
2933 /* guarantee minimum buffer size under pressure */
2934 if (kind == SK_MEM_RECV) {
2935 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2938 } else { /* SK_MEM_SEND */
2939 int wmem0 = sk_get_wmem0(sk, prot);
2941 if (sk->sk_type == SOCK_STREAM) {
2942 if (sk->sk_wmem_queued < wmem0)
2944 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2949 if (sk_has_memory_pressure(sk)) {
2952 if (!sk_under_memory_pressure(sk))
2954 alloc = sk_sockets_allocated_read_positive(sk);
2955 if (sk_prot_mem_limits(sk, 2) > alloc *
2956 sk_mem_pages(sk->sk_wmem_queued +
2957 atomic_read(&sk->sk_rmem_alloc) +
2958 sk->sk_forward_alloc))
2962 suppress_allocation:
2964 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2965 sk_stream_moderate_sndbuf(sk);
2967 /* Fail only if socket is _under_ its sndbuf.
2968 * In this case we cannot block, so that we have to fail.
2970 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
2971 /* Force charge with __GFP_NOFAIL */
2972 if (memcg_charge && !charged) {
2973 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2974 gfp_memcg_charge() | __GFP_NOFAIL);
2980 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2981 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2983 sk_memory_allocated_sub(sk, amt);
2985 if (memcg_charge && charged)
2986 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2990 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2993 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2995 * @size: memory size to allocate
2996 * @kind: allocation type
2998 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2999 * rmem allocation. This function assumes that protocols which have
3000 * memory_pressure use sk_wmem_queued as write buffer accounting.
3002 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3004 int ret, amt = sk_mem_pages(size);
3006 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
3007 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3009 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
3012 EXPORT_SYMBOL(__sk_mem_schedule);
3015 * __sk_mem_reduce_allocated - reclaim memory_allocated
3017 * @amount: number of quanta
3019 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3021 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3023 sk_memory_allocated_sub(sk, amount);
3025 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3026 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3028 if (sk_under_memory_pressure(sk) &&
3029 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3030 sk_leave_memory_pressure(sk);
3032 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
3035 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3037 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
3039 void __sk_mem_reclaim(struct sock *sk, int amount)
3041 amount >>= SK_MEM_QUANTUM_SHIFT;
3042 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
3043 __sk_mem_reduce_allocated(sk, amount);
3045 EXPORT_SYMBOL(__sk_mem_reclaim);
3047 int sk_set_peek_off(struct sock *sk, int val)
3049 sk->sk_peek_off = val;
3052 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3055 * Set of default routines for initialising struct proto_ops when
3056 * the protocol does not support a particular function. In certain
3057 * cases where it makes no sense for a protocol to have a "do nothing"
3058 * function, some default processing is provided.
3061 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3065 EXPORT_SYMBOL(sock_no_bind);
3067 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3072 EXPORT_SYMBOL(sock_no_connect);
3074 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3078 EXPORT_SYMBOL(sock_no_socketpair);
3080 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3085 EXPORT_SYMBOL(sock_no_accept);
3087 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3092 EXPORT_SYMBOL(sock_no_getname);
3094 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3098 EXPORT_SYMBOL(sock_no_ioctl);
3100 int sock_no_listen(struct socket *sock, int backlog)
3104 EXPORT_SYMBOL(sock_no_listen);
3106 int sock_no_shutdown(struct socket *sock, int how)
3110 EXPORT_SYMBOL(sock_no_shutdown);
3112 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3116 EXPORT_SYMBOL(sock_no_sendmsg);
3118 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3122 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3124 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3129 EXPORT_SYMBOL(sock_no_recvmsg);
3131 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3133 /* Mirror missing mmap method error code */
3136 EXPORT_SYMBOL(sock_no_mmap);
3139 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3140 * various sock-based usage counts.
3142 void __receive_sock(struct file *file)
3144 struct socket *sock;
3146 sock = sock_from_file(file);
3148 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3149 sock_update_classid(&sock->sk->sk_cgrp_data);
3153 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3156 struct msghdr msg = {.msg_flags = flags};
3158 char *kaddr = kmap(page);
3159 iov.iov_base = kaddr + offset;
3161 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3165 EXPORT_SYMBOL(sock_no_sendpage);
3167 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3168 int offset, size_t size, int flags)
3171 struct msghdr msg = {.msg_flags = flags};
3173 char *kaddr = kmap(page);
3175 iov.iov_base = kaddr + offset;
3177 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3181 EXPORT_SYMBOL(sock_no_sendpage_locked);
3184 * Default Socket Callbacks
3187 static void sock_def_wakeup(struct sock *sk)
3189 struct socket_wq *wq;
3192 wq = rcu_dereference(sk->sk_wq);
3193 if (skwq_has_sleeper(wq))
3194 wake_up_interruptible_all(&wq->wait);
3198 static void sock_def_error_report(struct sock *sk)
3200 struct socket_wq *wq;
3203 wq = rcu_dereference(sk->sk_wq);
3204 if (skwq_has_sleeper(wq))
3205 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3206 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3210 void sock_def_readable(struct sock *sk)
3212 struct socket_wq *wq;
3215 wq = rcu_dereference(sk->sk_wq);
3216 if (skwq_has_sleeper(wq))
3217 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3218 EPOLLRDNORM | EPOLLRDBAND);
3219 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3223 static void sock_def_write_space(struct sock *sk)
3225 struct socket_wq *wq;
3229 /* Do not wake up a writer until he can make "significant"
3232 if (sock_writeable(sk)) {
3233 wq = rcu_dereference(sk->sk_wq);
3234 if (skwq_has_sleeper(wq))
3235 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3236 EPOLLWRNORM | EPOLLWRBAND);
3238 /* Should agree with poll, otherwise some programs break */
3239 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3245 /* An optimised version of sock_def_write_space(), should only be called
3246 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3249 static void sock_def_write_space_wfree(struct sock *sk)
3251 /* Do not wake up a writer until he can make "significant"
3254 if (sock_writeable(sk)) {
3255 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3257 /* rely on refcount_sub from sock_wfree() */
3258 smp_mb__after_atomic();
3259 if (wq && waitqueue_active(&wq->wait))
3260 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3261 EPOLLWRNORM | EPOLLWRBAND);
3263 /* Should agree with poll, otherwise some programs break */
3264 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3268 static void sock_def_destruct(struct sock *sk)
3272 void sk_send_sigurg(struct sock *sk)
3274 if (sk->sk_socket && sk->sk_socket->file)
3275 if (send_sigurg(&sk->sk_socket->file->f_owner))
3276 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3278 EXPORT_SYMBOL(sk_send_sigurg);
3280 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3281 unsigned long expires)
3283 if (!mod_timer(timer, expires))
3286 EXPORT_SYMBOL(sk_reset_timer);
3288 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3290 if (del_timer(timer))
3293 EXPORT_SYMBOL(sk_stop_timer);
3295 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3297 if (del_timer_sync(timer))
3300 EXPORT_SYMBOL(sk_stop_timer_sync);
3302 void sock_init_data(struct socket *sock, struct sock *sk)
3305 sk->sk_send_head = NULL;
3307 timer_setup(&sk->sk_timer, NULL, 0);
3309 sk->sk_allocation = GFP_KERNEL;
3310 sk->sk_rcvbuf = sysctl_rmem_default;
3311 sk->sk_sndbuf = sysctl_wmem_default;
3312 sk->sk_state = TCP_CLOSE;
3313 sk_set_socket(sk, sock);
3315 sock_set_flag(sk, SOCK_ZAPPED);
3318 sk->sk_type = sock->type;
3319 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3321 sk->sk_uid = SOCK_INODE(sock)->i_uid;
3323 RCU_INIT_POINTER(sk->sk_wq, NULL);
3324 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3327 rwlock_init(&sk->sk_callback_lock);
3328 if (sk->sk_kern_sock)
3329 lockdep_set_class_and_name(
3330 &sk->sk_callback_lock,
3331 af_kern_callback_keys + sk->sk_family,
3332 af_family_kern_clock_key_strings[sk->sk_family]);
3334 lockdep_set_class_and_name(
3335 &sk->sk_callback_lock,
3336 af_callback_keys + sk->sk_family,
3337 af_family_clock_key_strings[sk->sk_family]);
3339 sk->sk_state_change = sock_def_wakeup;
3340 sk->sk_data_ready = sock_def_readable;
3341 sk->sk_write_space = sock_def_write_space;
3342 sk->sk_error_report = sock_def_error_report;
3343 sk->sk_destruct = sock_def_destruct;
3345 sk->sk_frag.page = NULL;
3346 sk->sk_frag.offset = 0;
3347 sk->sk_peek_off = -1;
3349 sk->sk_peer_pid = NULL;
3350 sk->sk_peer_cred = NULL;
3351 spin_lock_init(&sk->sk_peer_lock);
3353 sk->sk_write_pending = 0;
3354 sk->sk_rcvlowat = 1;
3355 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3356 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3358 sk->sk_stamp = SK_DEFAULT_STAMP;
3359 #if BITS_PER_LONG==32
3360 seqlock_init(&sk->sk_stamp_seq);
3362 atomic_set(&sk->sk_zckey, 0);
3364 #ifdef CONFIG_NET_RX_BUSY_POLL
3366 sk->sk_ll_usec = sysctl_net_busy_read;
3369 sk->sk_max_pacing_rate = ~0UL;
3370 sk->sk_pacing_rate = ~0UL;
3371 WRITE_ONCE(sk->sk_pacing_shift, 10);
3372 sk->sk_incoming_cpu = -1;
3373 sk->sk_txrehash = SOCK_TXREHASH_DEFAULT;
3375 sk_rx_queue_clear(sk);
3377 * Before updating sk_refcnt, we must commit prior changes to memory
3378 * (Documentation/RCU/rculist_nulls.rst for details)
3381 refcount_set(&sk->sk_refcnt, 1);
3382 atomic_set(&sk->sk_drops, 0);
3384 EXPORT_SYMBOL(sock_init_data);
3386 void lock_sock_nested(struct sock *sk, int subclass)
3388 /* The sk_lock has mutex_lock() semantics here. */
3389 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3392 spin_lock_bh(&sk->sk_lock.slock);
3393 if (sock_owned_by_user_nocheck(sk))
3395 sk->sk_lock.owned = 1;
3396 spin_unlock_bh(&sk->sk_lock.slock);
3398 EXPORT_SYMBOL(lock_sock_nested);
3400 void release_sock(struct sock *sk)
3402 spin_lock_bh(&sk->sk_lock.slock);
3403 if (sk->sk_backlog.tail)
3406 /* Warning : release_cb() might need to release sk ownership,
3407 * ie call sock_release_ownership(sk) before us.
3409 if (sk->sk_prot->release_cb)
3410 sk->sk_prot->release_cb(sk);
3412 sock_release_ownership(sk);
3413 if (waitqueue_active(&sk->sk_lock.wq))
3414 wake_up(&sk->sk_lock.wq);
3415 spin_unlock_bh(&sk->sk_lock.slock);
3417 EXPORT_SYMBOL(release_sock);
3419 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3422 spin_lock_bh(&sk->sk_lock.slock);
3424 if (!sock_owned_by_user_nocheck(sk)) {
3426 * Fast path return with bottom halves disabled and
3427 * sock::sk_lock.slock held.
3429 * The 'mutex' is not contended and holding
3430 * sock::sk_lock.slock prevents all other lockers to
3431 * proceed so the corresponding unlock_sock_fast() can
3432 * avoid the slow path of release_sock() completely and
3433 * just release slock.
3435 * From a semantical POV this is equivalent to 'acquiring'
3436 * the 'mutex', hence the corresponding lockdep
3437 * mutex_release() has to happen in the fast path of
3438 * unlock_sock_fast().
3444 sk->sk_lock.owned = 1;
3445 __acquire(&sk->sk_lock.slock);
3446 spin_unlock_bh(&sk->sk_lock.slock);
3449 EXPORT_SYMBOL(__lock_sock_fast);
3451 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3452 bool timeval, bool time32)
3454 struct sock *sk = sock->sk;
3455 struct timespec64 ts;
3457 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3458 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3459 if (ts.tv_sec == -1)
3461 if (ts.tv_sec == 0) {
3462 ktime_t kt = ktime_get_real();
3463 sock_write_timestamp(sk, kt);
3464 ts = ktime_to_timespec64(kt);
3470 #ifdef CONFIG_COMPAT_32BIT_TIME
3472 return put_old_timespec32(&ts, userstamp);
3474 #ifdef CONFIG_SPARC64
3475 /* beware of padding in sparc64 timeval */
3476 if (timeval && !in_compat_syscall()) {
3477 struct __kernel_old_timeval __user tv = {
3478 .tv_sec = ts.tv_sec,
3479 .tv_usec = ts.tv_nsec,
3481 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3486 return put_timespec64(&ts, userstamp);
3488 EXPORT_SYMBOL(sock_gettstamp);
3490 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3492 if (!sock_flag(sk, flag)) {
3493 unsigned long previous_flags = sk->sk_flags;
3495 sock_set_flag(sk, flag);
3497 * we just set one of the two flags which require net
3498 * time stamping, but time stamping might have been on
3499 * already because of the other one
3501 if (sock_needs_netstamp(sk) &&
3502 !(previous_flags & SK_FLAGS_TIMESTAMP))
3503 net_enable_timestamp();
3507 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3508 int level, int type)
3510 struct sock_exterr_skb *serr;
3511 struct sk_buff *skb;
3515 skb = sock_dequeue_err_skb(sk);
3521 msg->msg_flags |= MSG_TRUNC;
3524 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3528 sock_recv_timestamp(msg, sk, skb);
3530 serr = SKB_EXT_ERR(skb);
3531 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3533 msg->msg_flags |= MSG_ERRQUEUE;
3541 EXPORT_SYMBOL(sock_recv_errqueue);
3544 * Get a socket option on an socket.
3546 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3547 * asynchronous errors should be reported by getsockopt. We assume
3548 * this means if you specify SO_ERROR (otherwise whats the point of it).
3550 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3551 char __user *optval, int __user *optlen)
3553 struct sock *sk = sock->sk;
3555 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3557 EXPORT_SYMBOL(sock_common_getsockopt);
3559 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3562 struct sock *sk = sock->sk;
3566 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3568 msg->msg_namelen = addr_len;
3571 EXPORT_SYMBOL(sock_common_recvmsg);
3574 * Set socket options on an inet socket.
3576 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3577 sockptr_t optval, unsigned int optlen)
3579 struct sock *sk = sock->sk;
3581 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3583 EXPORT_SYMBOL(sock_common_setsockopt);
3585 void sk_common_release(struct sock *sk)
3587 if (sk->sk_prot->destroy)
3588 sk->sk_prot->destroy(sk);
3591 * Observation: when sk_common_release is called, processes have
3592 * no access to socket. But net still has.
3593 * Step one, detach it from networking:
3595 * A. Remove from hash tables.
3598 sk->sk_prot->unhash(sk);
3601 * In this point socket cannot receive new packets, but it is possible
3602 * that some packets are in flight because some CPU runs receiver and
3603 * did hash table lookup before we unhashed socket. They will achieve
3604 * receive queue and will be purged by socket destructor.
3606 * Also we still have packets pending on receive queue and probably,
3607 * our own packets waiting in device queues. sock_destroy will drain
3608 * receive queue, but transmitted packets will delay socket destruction
3609 * until the last reference will be released.
3614 xfrm_sk_free_policy(sk);
3616 sk_refcnt_debug_release(sk);
3620 EXPORT_SYMBOL(sk_common_release);
3622 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3624 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3626 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3627 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3628 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3629 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3630 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3631 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3632 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3633 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3634 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3637 #ifdef CONFIG_PROC_FS
3638 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3640 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3642 int cpu, idx = prot->inuse_idx;
3645 for_each_possible_cpu(cpu)
3646 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3648 return res >= 0 ? res : 0;
3650 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3652 int sock_inuse_get(struct net *net)
3656 for_each_possible_cpu(cpu)
3657 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3662 EXPORT_SYMBOL_GPL(sock_inuse_get);
3664 static int __net_init sock_inuse_init_net(struct net *net)
3666 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3667 if (net->core.prot_inuse == NULL)
3672 static void __net_exit sock_inuse_exit_net(struct net *net)
3674 free_percpu(net->core.prot_inuse);
3677 static struct pernet_operations net_inuse_ops = {
3678 .init = sock_inuse_init_net,
3679 .exit = sock_inuse_exit_net,
3682 static __init int net_inuse_init(void)
3684 if (register_pernet_subsys(&net_inuse_ops))
3685 panic("Cannot initialize net inuse counters");
3690 core_initcall(net_inuse_init);
3692 static int assign_proto_idx(struct proto *prot)
3694 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3696 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3697 pr_err("PROTO_INUSE_NR exhausted\n");
3701 set_bit(prot->inuse_idx, proto_inuse_idx);
3705 static void release_proto_idx(struct proto *prot)
3707 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3708 clear_bit(prot->inuse_idx, proto_inuse_idx);
3711 static inline int assign_proto_idx(struct proto *prot)
3716 static inline void release_proto_idx(struct proto *prot)
3722 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3726 kfree(twsk_prot->twsk_slab_name);
3727 twsk_prot->twsk_slab_name = NULL;
3728 kmem_cache_destroy(twsk_prot->twsk_slab);
3729 twsk_prot->twsk_slab = NULL;
3732 static int tw_prot_init(const struct proto *prot)
3734 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3739 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3741 if (!twsk_prot->twsk_slab_name)
3744 twsk_prot->twsk_slab =
3745 kmem_cache_create(twsk_prot->twsk_slab_name,
3746 twsk_prot->twsk_obj_size, 0,
3747 SLAB_ACCOUNT | prot->slab_flags,
3749 if (!twsk_prot->twsk_slab) {
3750 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3758 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3762 kfree(rsk_prot->slab_name);
3763 rsk_prot->slab_name = NULL;
3764 kmem_cache_destroy(rsk_prot->slab);
3765 rsk_prot->slab = NULL;
3768 static int req_prot_init(const struct proto *prot)
3770 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3775 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3777 if (!rsk_prot->slab_name)
3780 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3781 rsk_prot->obj_size, 0,
3782 SLAB_ACCOUNT | prot->slab_flags,
3785 if (!rsk_prot->slab) {
3786 pr_crit("%s: Can't create request sock SLAB cache!\n",
3793 int proto_register(struct proto *prot, int alloc_slab)
3797 if (prot->memory_allocated && !prot->sysctl_mem) {
3798 pr_err("%s: missing sysctl_mem\n", prot->name);
3802 prot->slab = kmem_cache_create_usercopy(prot->name,
3804 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3806 prot->useroffset, prot->usersize,
3809 if (prot->slab == NULL) {
3810 pr_crit("%s: Can't create sock SLAB cache!\n",
3815 if (req_prot_init(prot))
3816 goto out_free_request_sock_slab;
3818 if (tw_prot_init(prot))
3819 goto out_free_timewait_sock_slab;
3822 mutex_lock(&proto_list_mutex);
3823 ret = assign_proto_idx(prot);
3825 mutex_unlock(&proto_list_mutex);
3826 goto out_free_timewait_sock_slab;
3828 list_add(&prot->node, &proto_list);
3829 mutex_unlock(&proto_list_mutex);
3832 out_free_timewait_sock_slab:
3834 tw_prot_cleanup(prot->twsk_prot);
3835 out_free_request_sock_slab:
3837 req_prot_cleanup(prot->rsk_prot);
3839 kmem_cache_destroy(prot->slab);
3845 EXPORT_SYMBOL(proto_register);
3847 void proto_unregister(struct proto *prot)
3849 mutex_lock(&proto_list_mutex);
3850 release_proto_idx(prot);
3851 list_del(&prot->node);
3852 mutex_unlock(&proto_list_mutex);
3854 kmem_cache_destroy(prot->slab);
3857 req_prot_cleanup(prot->rsk_prot);
3858 tw_prot_cleanup(prot->twsk_prot);
3860 EXPORT_SYMBOL(proto_unregister);
3862 int sock_load_diag_module(int family, int protocol)
3865 if (!sock_is_registered(family))
3868 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3869 NETLINK_SOCK_DIAG, family);
3873 if (family == AF_INET &&
3874 protocol != IPPROTO_RAW &&
3875 protocol < MAX_INET_PROTOS &&
3876 !rcu_access_pointer(inet_protos[protocol]))
3880 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3881 NETLINK_SOCK_DIAG, family, protocol);
3883 EXPORT_SYMBOL(sock_load_diag_module);
3885 #ifdef CONFIG_PROC_FS
3886 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3887 __acquires(proto_list_mutex)
3889 mutex_lock(&proto_list_mutex);
3890 return seq_list_start_head(&proto_list, *pos);
3893 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3895 return seq_list_next(v, &proto_list, pos);
3898 static void proto_seq_stop(struct seq_file *seq, void *v)
3899 __releases(proto_list_mutex)
3901 mutex_unlock(&proto_list_mutex);
3904 static char proto_method_implemented(const void *method)
3906 return method == NULL ? 'n' : 'y';
3908 static long sock_prot_memory_allocated(struct proto *proto)
3910 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3913 static const char *sock_prot_memory_pressure(struct proto *proto)
3915 return proto->memory_pressure != NULL ?
3916 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3919 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3922 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3923 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3926 sock_prot_inuse_get(seq_file_net(seq), proto),
3927 sock_prot_memory_allocated(proto),
3928 sock_prot_memory_pressure(proto),
3930 proto->slab == NULL ? "no" : "yes",
3931 module_name(proto->owner),
3932 proto_method_implemented(proto->close),
3933 proto_method_implemented(proto->connect),
3934 proto_method_implemented(proto->disconnect),
3935 proto_method_implemented(proto->accept),
3936 proto_method_implemented(proto->ioctl),
3937 proto_method_implemented(proto->init),
3938 proto_method_implemented(proto->destroy),
3939 proto_method_implemented(proto->shutdown),
3940 proto_method_implemented(proto->setsockopt),
3941 proto_method_implemented(proto->getsockopt),
3942 proto_method_implemented(proto->sendmsg),
3943 proto_method_implemented(proto->recvmsg),
3944 proto_method_implemented(proto->sendpage),
3945 proto_method_implemented(proto->bind),
3946 proto_method_implemented(proto->backlog_rcv),
3947 proto_method_implemented(proto->hash),
3948 proto_method_implemented(proto->unhash),
3949 proto_method_implemented(proto->get_port),
3950 proto_method_implemented(proto->enter_memory_pressure));
3953 static int proto_seq_show(struct seq_file *seq, void *v)
3955 if (v == &proto_list)
3956 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3965 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3967 proto_seq_printf(seq, list_entry(v, struct proto, node));
3971 static const struct seq_operations proto_seq_ops = {
3972 .start = proto_seq_start,
3973 .next = proto_seq_next,
3974 .stop = proto_seq_stop,
3975 .show = proto_seq_show,
3978 static __net_init int proto_init_net(struct net *net)
3980 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3981 sizeof(struct seq_net_private)))
3987 static __net_exit void proto_exit_net(struct net *net)
3989 remove_proc_entry("protocols", net->proc_net);
3993 static __net_initdata struct pernet_operations proto_net_ops = {
3994 .init = proto_init_net,
3995 .exit = proto_exit_net,
3998 static int __init proto_init(void)
4000 return register_pernet_subsys(&proto_net_ops);
4003 subsys_initcall(proto_init);
4005 #endif /* PROC_FS */
4007 #ifdef CONFIG_NET_RX_BUSY_POLL
4008 bool sk_busy_loop_end(void *p, unsigned long start_time)
4010 struct sock *sk = p;
4012 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
4013 sk_busy_loop_timeout(sk, start_time);
4015 EXPORT_SYMBOL(sk_busy_loop_end);
4016 #endif /* CONFIG_NET_RX_BUSY_POLL */
4018 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4020 if (!sk->sk_prot->bind_add)
4022 return sk->sk_prot->bind_add(sk, addr, addr_len);
4024 EXPORT_SYMBOL(sock_bind_add);