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>
117 #include <linux/mroute.h>
118 #include <linux/mroute6.h>
119 #include <linux/icmpv6.h>
121 #include <linux/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
138 #include <net/bpf_sk_storage.h>
140 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
144 #include <net/phonet/phonet.h>
146 #include <linux/ethtool.h>
150 static DEFINE_MUTEX(proto_list_mutex);
151 static LIST_HEAD(proto_list);
153 static void sock_def_write_space_wfree(struct sock *sk);
154 static void sock_def_write_space(struct sock *sk);
157 * sk_ns_capable - General socket capability test
158 * @sk: Socket to use a capability on or through
159 * @user_ns: The user namespace of the capability to use
160 * @cap: The capability to use
162 * Test to see if the opener of the socket had when the socket was
163 * created and the current process has the capability @cap in the user
164 * namespace @user_ns.
166 bool sk_ns_capable(const struct sock *sk,
167 struct user_namespace *user_ns, int cap)
169 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
170 ns_capable(user_ns, cap);
172 EXPORT_SYMBOL(sk_ns_capable);
175 * sk_capable - Socket global capability test
176 * @sk: Socket to use a capability on or through
177 * @cap: The global capability to use
179 * Test to see if the opener of the socket had when the socket was
180 * created and the current process has the capability @cap in all user
183 bool sk_capable(const struct sock *sk, int cap)
185 return sk_ns_capable(sk, &init_user_ns, cap);
187 EXPORT_SYMBOL(sk_capable);
190 * sk_net_capable - Network namespace socket capability test
191 * @sk: Socket to use a capability on or through
192 * @cap: The capability to use
194 * Test to see if the opener of the socket had when the socket was created
195 * and the current process has the capability @cap over the network namespace
196 * the socket is a member of.
198 bool sk_net_capable(const struct sock *sk, int cap)
200 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
202 EXPORT_SYMBOL(sk_net_capable);
205 * Each address family might have different locking rules, so we have
206 * one slock key per address family and separate keys for internal and
209 static struct lock_class_key af_family_keys[AF_MAX];
210 static struct lock_class_key af_family_kern_keys[AF_MAX];
211 static struct lock_class_key af_family_slock_keys[AF_MAX];
212 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
215 * Make lock validator output more readable. (we pre-construct these
216 * strings build-time, so that runtime initialization of socket
220 #define _sock_locks(x) \
221 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
222 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
223 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
224 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
225 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
226 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
227 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
228 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
229 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
230 x "27" , x "28" , x "AF_CAN" , \
231 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
232 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
233 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
234 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
235 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
239 static const char *const af_family_key_strings[AF_MAX+1] = {
240 _sock_locks("sk_lock-")
242 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("slock-")
245 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("clock-")
249 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
250 _sock_locks("k-sk_lock-")
252 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
253 _sock_locks("k-slock-")
255 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
256 _sock_locks("k-clock-")
258 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
259 _sock_locks("rlock-")
261 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
262 _sock_locks("wlock-")
264 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
265 _sock_locks("elock-")
269 * sk_callback_lock and sk queues locking rules are per-address-family,
270 * so split the lock classes by using a per-AF key:
272 static struct lock_class_key af_callback_keys[AF_MAX];
273 static struct lock_class_key af_rlock_keys[AF_MAX];
274 static struct lock_class_key af_wlock_keys[AF_MAX];
275 static struct lock_class_key af_elock_keys[AF_MAX];
276 static struct lock_class_key af_kern_callback_keys[AF_MAX];
278 /* Run time adjustable parameters. */
279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
280 EXPORT_SYMBOL(sysctl_wmem_max);
281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
282 EXPORT_SYMBOL(sysctl_rmem_max);
283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
286 /* Maximal space eaten by iovec or ancillary data plus some space */
287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
288 EXPORT_SYMBOL(sysctl_optmem_max);
290 int sysctl_tstamp_allow_data __read_mostly = 1;
292 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
293 EXPORT_SYMBOL_GPL(memalloc_socks_key);
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
303 void sk_set_memalloc(struct sock *sk)
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
307 static_branch_inc(&memalloc_socks_key);
309 EXPORT_SYMBOL_GPL(sk_set_memalloc);
311 void sk_clear_memalloc(struct sock *sk)
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
315 static_branch_dec(&memalloc_socks_key);
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
326 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
331 unsigned int noreclaim_flag;
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
336 noreclaim_flag = memalloc_noreclaim_save();
337 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
341 memalloc_noreclaim_restore(noreclaim_flag);
345 EXPORT_SYMBOL(__sk_backlog_rcv);
347 void sk_error_report(struct sock *sk)
349 sk->sk_error_report(sk);
351 switch (sk->sk_family) {
355 trace_inet_sk_error_report(sk);
361 EXPORT_SYMBOL(sk_error_report);
363 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
365 struct __kernel_sock_timeval tv;
367 if (timeo == MAX_SCHEDULE_TIMEOUT) {
371 tv.tv_sec = timeo / HZ;
372 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
375 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
376 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
377 *(struct old_timeval32 *)optval = tv32;
382 struct __kernel_old_timeval old_tv;
383 old_tv.tv_sec = tv.tv_sec;
384 old_tv.tv_usec = tv.tv_usec;
385 *(struct __kernel_old_timeval *)optval = old_tv;
386 return sizeof(old_tv);
389 *(struct __kernel_sock_timeval *)optval = tv;
392 EXPORT_SYMBOL(sock_get_timeout);
394 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
395 sockptr_t optval, int optlen, bool old_timeval)
397 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
398 struct old_timeval32 tv32;
400 if (optlen < sizeof(tv32))
403 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
405 tv->tv_sec = tv32.tv_sec;
406 tv->tv_usec = tv32.tv_usec;
407 } else if (old_timeval) {
408 struct __kernel_old_timeval old_tv;
410 if (optlen < sizeof(old_tv))
412 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
414 tv->tv_sec = old_tv.tv_sec;
415 tv->tv_usec = old_tv.tv_usec;
417 if (optlen < sizeof(*tv))
419 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
425 EXPORT_SYMBOL(sock_copy_user_timeval);
427 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
430 struct __kernel_sock_timeval tv;
431 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
437 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
441 static int warned __read_mostly;
443 WRITE_ONCE(*timeo_p, 0);
444 if (warned < 10 && net_ratelimit()) {
446 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
447 __func__, current->comm, task_pid_nr(current));
451 val = MAX_SCHEDULE_TIMEOUT;
452 if ((tv.tv_sec || tv.tv_usec) &&
453 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
454 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
456 WRITE_ONCE(*timeo_p, val);
460 static bool sock_needs_netstamp(const struct sock *sk)
462 switch (sk->sk_family) {
471 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
473 if (sk->sk_flags & flags) {
474 sk->sk_flags &= ~flags;
475 if (sock_needs_netstamp(sk) &&
476 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
477 net_disable_timestamp();
482 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
485 struct sk_buff_head *list = &sk->sk_receive_queue;
487 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
488 atomic_inc(&sk->sk_drops);
489 trace_sock_rcvqueue_full(sk, skb);
493 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
494 atomic_inc(&sk->sk_drops);
499 skb_set_owner_r(skb, sk);
501 /* we escape from rcu protected region, make sure we dont leak
506 spin_lock_irqsave(&list->lock, flags);
507 sock_skb_set_dropcount(sk, skb);
508 __skb_queue_tail(list, skb);
509 spin_unlock_irqrestore(&list->lock, flags);
511 if (!sock_flag(sk, SOCK_DEAD))
512 sk->sk_data_ready(sk);
515 EXPORT_SYMBOL(__sock_queue_rcv_skb);
517 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
518 enum skb_drop_reason *reason)
520 enum skb_drop_reason drop_reason;
523 err = sk_filter(sk, skb);
525 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
528 err = __sock_queue_rcv_skb(sk, skb);
531 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
534 drop_reason = SKB_DROP_REASON_PROTO_MEM;
537 drop_reason = SKB_NOT_DROPPED_YET;
542 *reason = drop_reason;
545 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
547 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
548 const int nested, unsigned int trim_cap, bool refcounted)
550 int rc = NET_RX_SUCCESS;
552 if (sk_filter_trim_cap(sk, skb, trim_cap))
553 goto discard_and_relse;
557 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
558 atomic_inc(&sk->sk_drops);
559 goto discard_and_relse;
562 bh_lock_sock_nested(sk);
565 if (!sock_owned_by_user(sk)) {
567 * trylock + unlock semantics:
569 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
571 rc = sk_backlog_rcv(sk, skb);
573 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
574 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
576 atomic_inc(&sk->sk_drops);
577 goto discard_and_relse;
589 EXPORT_SYMBOL(__sk_receive_skb);
591 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
593 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
595 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
597 struct dst_entry *dst = __sk_dst_get(sk);
599 if (dst && dst->obsolete &&
600 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
601 dst, cookie) == NULL) {
602 sk_tx_queue_clear(sk);
603 sk->sk_dst_pending_confirm = 0;
604 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
611 EXPORT_SYMBOL(__sk_dst_check);
613 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
615 struct dst_entry *dst = sk_dst_get(sk);
617 if (dst && dst->obsolete &&
618 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
619 dst, cookie) == NULL) {
627 EXPORT_SYMBOL(sk_dst_check);
629 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
631 int ret = -ENOPROTOOPT;
632 #ifdef CONFIG_NETDEVICES
633 struct net *net = sock_net(sk);
637 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
644 /* Paired with all READ_ONCE() done locklessly. */
645 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
647 if (sk->sk_prot->rehash)
648 sk->sk_prot->rehash(sk);
659 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
665 ret = sock_bindtoindex_locked(sk, ifindex);
671 EXPORT_SYMBOL(sock_bindtoindex);
673 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
675 int ret = -ENOPROTOOPT;
676 #ifdef CONFIG_NETDEVICES
677 struct net *net = sock_net(sk);
678 char devname[IFNAMSIZ];
685 /* Bind this socket to a particular device like "eth0",
686 * as specified in the passed interface name. If the
687 * name is "" or the option length is zero the socket
690 if (optlen > IFNAMSIZ - 1)
691 optlen = IFNAMSIZ - 1;
692 memset(devname, 0, sizeof(devname));
695 if (copy_from_sockptr(devname, optval, optlen))
699 if (devname[0] != '\0') {
700 struct net_device *dev;
703 dev = dev_get_by_name_rcu(net, devname);
705 index = dev->ifindex;
712 sockopt_lock_sock(sk);
713 ret = sock_bindtoindex_locked(sk, index);
714 sockopt_release_sock(sk);
721 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
722 sockptr_t optlen, int len)
724 int ret = -ENOPROTOOPT;
725 #ifdef CONFIG_NETDEVICES
726 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
727 struct net *net = sock_net(sk);
728 char devname[IFNAMSIZ];
730 if (bound_dev_if == 0) {
739 ret = netdev_get_name(net, devname, bound_dev_if);
743 len = strlen(devname) + 1;
746 if (copy_to_sockptr(optval, devname, len))
751 if (copy_to_sockptr(optlen, &len, sizeof(int)))
762 bool sk_mc_loop(struct sock *sk)
764 if (dev_recursion_level())
768 switch (sk->sk_family) {
770 return inet_test_bit(MC_LOOP, sk);
771 #if IS_ENABLED(CONFIG_IPV6)
773 return inet6_sk(sk)->mc_loop;
779 EXPORT_SYMBOL(sk_mc_loop);
781 void sock_set_reuseaddr(struct sock *sk)
784 sk->sk_reuse = SK_CAN_REUSE;
787 EXPORT_SYMBOL(sock_set_reuseaddr);
789 void sock_set_reuseport(struct sock *sk)
792 sk->sk_reuseport = true;
795 EXPORT_SYMBOL(sock_set_reuseport);
797 void sock_no_linger(struct sock *sk)
800 WRITE_ONCE(sk->sk_lingertime, 0);
801 sock_set_flag(sk, SOCK_LINGER);
804 EXPORT_SYMBOL(sock_no_linger);
806 void sock_set_priority(struct sock *sk, u32 priority)
809 WRITE_ONCE(sk->sk_priority, priority);
812 EXPORT_SYMBOL(sock_set_priority);
814 void sock_set_sndtimeo(struct sock *sk, s64 secs)
817 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
818 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
820 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
823 EXPORT_SYMBOL(sock_set_sndtimeo);
825 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
828 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
829 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
830 sock_set_flag(sk, SOCK_RCVTSTAMP);
831 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
833 sock_reset_flag(sk, SOCK_RCVTSTAMP);
834 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
838 void sock_enable_timestamps(struct sock *sk)
841 __sock_set_timestamps(sk, true, false, true);
844 EXPORT_SYMBOL(sock_enable_timestamps);
846 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
849 case SO_TIMESTAMP_OLD:
850 __sock_set_timestamps(sk, valbool, false, false);
852 case SO_TIMESTAMP_NEW:
853 __sock_set_timestamps(sk, valbool, true, false);
855 case SO_TIMESTAMPNS_OLD:
856 __sock_set_timestamps(sk, valbool, false, true);
858 case SO_TIMESTAMPNS_NEW:
859 __sock_set_timestamps(sk, valbool, true, true);
864 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
866 struct net *net = sock_net(sk);
867 struct net_device *dev = NULL;
872 if (sk->sk_bound_dev_if)
873 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
876 pr_err("%s: sock not bind to device\n", __func__);
880 num = ethtool_get_phc_vclocks(dev, &vclock_index);
883 for (i = 0; i < num; i++) {
884 if (*(vclock_index + i) == phc_index) {
896 sk->sk_bind_phc = phc_index;
901 int sock_set_timestamping(struct sock *sk, int optname,
902 struct so_timestamping timestamping)
904 int val = timestamping.flags;
907 if (val & ~SOF_TIMESTAMPING_MASK)
910 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
911 !(val & SOF_TIMESTAMPING_OPT_ID))
914 if (val & SOF_TIMESTAMPING_OPT_ID &&
915 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
917 if ((1 << sk->sk_state) &
918 (TCPF_CLOSE | TCPF_LISTEN))
920 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
921 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
923 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
925 atomic_set(&sk->sk_tskey, 0);
929 if (val & SOF_TIMESTAMPING_OPT_STATS &&
930 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
933 if (val & SOF_TIMESTAMPING_BIND_PHC) {
934 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
939 sk->sk_tsflags = val;
940 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
942 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
943 sock_enable_timestamp(sk,
944 SOCK_TIMESTAMPING_RX_SOFTWARE);
946 sock_disable_timestamp(sk,
947 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
951 void sock_set_keepalive(struct sock *sk)
954 if (sk->sk_prot->keepalive)
955 sk->sk_prot->keepalive(sk, true);
956 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
959 EXPORT_SYMBOL(sock_set_keepalive);
961 static void __sock_set_rcvbuf(struct sock *sk, int val)
963 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
964 * as a negative value.
966 val = min_t(int, val, INT_MAX / 2);
967 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
969 /* We double it on the way in to account for "struct sk_buff" etc.
970 * overhead. Applications assume that the SO_RCVBUF setting they make
971 * will allow that much actual data to be received on that socket.
973 * Applications are unaware that "struct sk_buff" and other overheads
974 * allocate from the receive buffer during socket buffer allocation.
976 * And after considering the possible alternatives, returning the value
977 * we actually used in getsockopt is the most desirable behavior.
979 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
982 void sock_set_rcvbuf(struct sock *sk, int val)
985 __sock_set_rcvbuf(sk, val);
988 EXPORT_SYMBOL(sock_set_rcvbuf);
990 static void __sock_set_mark(struct sock *sk, u32 val)
992 if (val != sk->sk_mark) {
993 WRITE_ONCE(sk->sk_mark, val);
998 void sock_set_mark(struct sock *sk, u32 val)
1001 __sock_set_mark(sk, val);
1004 EXPORT_SYMBOL(sock_set_mark);
1006 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1008 /* Round down bytes to multiple of pages */
1009 bytes = round_down(bytes, PAGE_SIZE);
1011 WARN_ON(bytes > sk->sk_reserved_mem);
1012 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1016 static int sock_reserve_memory(struct sock *sk, int bytes)
1022 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1028 pages = sk_mem_pages(bytes);
1030 /* pre-charge to memcg */
1031 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1032 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1036 /* pre-charge to forward_alloc */
1037 sk_memory_allocated_add(sk, pages);
1038 allocated = sk_memory_allocated(sk);
1039 /* If the system goes into memory pressure with this
1040 * precharge, give up and return error.
1042 if (allocated > sk_prot_mem_limits(sk, 1)) {
1043 sk_memory_allocated_sub(sk, pages);
1044 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1047 sk->sk_forward_alloc += pages << PAGE_SHIFT;
1049 WRITE_ONCE(sk->sk_reserved_mem,
1050 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1055 void sockopt_lock_sock(struct sock *sk)
1057 /* When current->bpf_ctx is set, the setsockopt is called from
1058 * a bpf prog. bpf has ensured the sk lock has been
1059 * acquired before calling setsockopt().
1061 if (has_current_bpf_ctx())
1066 EXPORT_SYMBOL(sockopt_lock_sock);
1068 void sockopt_release_sock(struct sock *sk)
1070 if (has_current_bpf_ctx())
1075 EXPORT_SYMBOL(sockopt_release_sock);
1077 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1079 return has_current_bpf_ctx() || ns_capable(ns, cap);
1081 EXPORT_SYMBOL(sockopt_ns_capable);
1083 bool sockopt_capable(int cap)
1085 return has_current_bpf_ctx() || capable(cap);
1087 EXPORT_SYMBOL(sockopt_capable);
1090 * This is meant for all protocols to use and covers goings on
1091 * at the socket level. Everything here is generic.
1094 int sk_setsockopt(struct sock *sk, int level, int optname,
1095 sockptr_t optval, unsigned int optlen)
1097 struct so_timestamping timestamping;
1098 struct socket *sock = sk->sk_socket;
1099 struct sock_txtime sk_txtime;
1106 * Options without arguments
1109 if (optname == SO_BINDTODEVICE)
1110 return sock_setbindtodevice(sk, optval, optlen);
1112 if (optlen < sizeof(int))
1115 if (copy_from_sockptr(&val, optval, sizeof(val)))
1118 valbool = val ? 1 : 0;
1120 sockopt_lock_sock(sk);
1124 if (val && !sockopt_capable(CAP_NET_ADMIN))
1127 sock_valbool_flag(sk, SOCK_DBG, valbool);
1130 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1133 sk->sk_reuseport = valbool;
1142 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1146 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1149 /* Don't error on this BSD doesn't and if you think
1150 * about it this is right. Otherwise apps have to
1151 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1152 * are treated in BSD as hints
1154 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1156 /* Ensure val * 2 fits into an int, to prevent max_t()
1157 * from treating it as a negative value.
1159 val = min_t(int, val, INT_MAX / 2);
1160 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1161 WRITE_ONCE(sk->sk_sndbuf,
1162 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1163 /* Wake up sending tasks if we upped the value. */
1164 sk->sk_write_space(sk);
1167 case SO_SNDBUFFORCE:
1168 if (!sockopt_capable(CAP_NET_ADMIN)) {
1173 /* No negative values (to prevent underflow, as val will be
1181 /* Don't error on this BSD doesn't and if you think
1182 * about it this is right. Otherwise apps have to
1183 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1184 * are treated in BSD as hints
1186 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1189 case SO_RCVBUFFORCE:
1190 if (!sockopt_capable(CAP_NET_ADMIN)) {
1195 /* No negative values (to prevent underflow, as val will be
1198 __sock_set_rcvbuf(sk, max(val, 0));
1202 if (sk->sk_prot->keepalive)
1203 sk->sk_prot->keepalive(sk, valbool);
1204 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1208 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1212 sk->sk_no_check_tx = valbool;
1216 if ((val >= 0 && val <= 6) ||
1217 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1218 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1219 WRITE_ONCE(sk->sk_priority, val);
1225 if (optlen < sizeof(ling)) {
1226 ret = -EINVAL; /* 1003.1g */
1229 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1233 if (!ling.l_onoff) {
1234 sock_reset_flag(sk, SOCK_LINGER);
1236 unsigned long t_sec = ling.l_linger;
1238 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1239 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1241 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1242 sock_set_flag(sk, SOCK_LINGER);
1250 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1254 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1257 case SO_TIMESTAMP_OLD:
1258 case SO_TIMESTAMP_NEW:
1259 case SO_TIMESTAMPNS_OLD:
1260 case SO_TIMESTAMPNS_NEW:
1261 sock_set_timestamp(sk, optname, valbool);
1264 case SO_TIMESTAMPING_NEW:
1265 case SO_TIMESTAMPING_OLD:
1266 if (optlen == sizeof(timestamping)) {
1267 if (copy_from_sockptr(×tamping, optval,
1268 sizeof(timestamping))) {
1273 memset(×tamping, 0, sizeof(timestamping));
1274 timestamping.flags = val;
1276 ret = sock_set_timestamping(sk, optname, timestamping);
1281 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1286 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1288 ret = set_rcvlowat(sk, val);
1290 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1293 case SO_RCVTIMEO_OLD:
1294 case SO_RCVTIMEO_NEW:
1295 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1296 optlen, optname == SO_RCVTIMEO_OLD);
1299 case SO_SNDTIMEO_OLD:
1300 case SO_SNDTIMEO_NEW:
1301 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1302 optlen, optname == SO_SNDTIMEO_OLD);
1305 case SO_ATTACH_FILTER: {
1306 struct sock_fprog fprog;
1308 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1310 ret = sk_attach_filter(&fprog, sk);
1315 if (optlen == sizeof(u32)) {
1319 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1322 ret = sk_attach_bpf(ufd, sk);
1326 case SO_ATTACH_REUSEPORT_CBPF: {
1327 struct sock_fprog fprog;
1329 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1331 ret = sk_reuseport_attach_filter(&fprog, sk);
1334 case SO_ATTACH_REUSEPORT_EBPF:
1336 if (optlen == sizeof(u32)) {
1340 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1343 ret = sk_reuseport_attach_bpf(ufd, sk);
1347 case SO_DETACH_REUSEPORT_BPF:
1348 ret = reuseport_detach_prog(sk);
1351 case SO_DETACH_FILTER:
1352 ret = sk_detach_filter(sk);
1355 case SO_LOCK_FILTER:
1356 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1359 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1363 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1366 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1367 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1372 __sock_set_mark(sk, val);
1375 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1379 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1382 case SO_WIFI_STATUS:
1383 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1388 int (*set_peek_off)(struct sock *sk, int val);
1390 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1392 ret = set_peek_off(sk, val);
1399 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1402 case SO_SELECT_ERR_QUEUE:
1403 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1406 #ifdef CONFIG_NET_RX_BUSY_POLL
1411 WRITE_ONCE(sk->sk_ll_usec, val);
1413 case SO_PREFER_BUSY_POLL:
1414 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1417 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1419 case SO_BUSY_POLL_BUDGET:
1420 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) {
1423 if (val < 0 || val > U16_MAX)
1426 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1431 case SO_MAX_PACING_RATE:
1433 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1435 if (sizeof(ulval) != sizeof(val) &&
1436 optlen >= sizeof(ulval) &&
1437 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1442 cmpxchg(&sk->sk_pacing_status,
1445 /* Pairs with READ_ONCE() from sk_getsockopt() */
1446 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1447 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1450 case SO_INCOMING_CPU:
1451 reuseport_update_incoming_cpu(sk, val);
1456 dst_negative_advice(sk);
1460 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1461 if (!(sk_is_tcp(sk) ||
1462 (sk->sk_type == SOCK_DGRAM &&
1463 sk->sk_protocol == IPPROTO_UDP)))
1465 } else if (sk->sk_family != PF_RDS) {
1469 if (val < 0 || val > 1)
1472 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1477 if (optlen != sizeof(struct sock_txtime)) {
1480 } else if (copy_from_sockptr(&sk_txtime, optval,
1481 sizeof(struct sock_txtime))) {
1484 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1488 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1489 * scheduler has enough safe guards.
1491 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1492 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1496 sock_valbool_flag(sk, SOCK_TXTIME, true);
1497 sk->sk_clockid = sk_txtime.clockid;
1498 sk->sk_txtime_deadline_mode =
1499 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1500 sk->sk_txtime_report_errors =
1501 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1504 case SO_BINDTOIFINDEX:
1505 ret = sock_bindtoindex_locked(sk, val);
1509 if (val & ~SOCK_BUF_LOCK_MASK) {
1513 sk->sk_userlocks = val | (sk->sk_userlocks &
1514 ~SOCK_BUF_LOCK_MASK);
1517 case SO_RESERVE_MEM:
1526 delta = val - sk->sk_reserved_mem;
1528 sock_release_reserved_memory(sk, -delta);
1530 ret = sock_reserve_memory(sk, delta);
1535 if (val < -1 || val > 1) {
1539 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1540 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1541 /* Paired with READ_ONCE() in tcp_rtx_synack()
1542 * and sk_getsockopt().
1544 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1551 sockopt_release_sock(sk);
1555 int sock_setsockopt(struct socket *sock, int level, int optname,
1556 sockptr_t optval, unsigned int optlen)
1558 return sk_setsockopt(sock->sk, level, optname,
1561 EXPORT_SYMBOL(sock_setsockopt);
1563 static const struct cred *sk_get_peer_cred(struct sock *sk)
1565 const struct cred *cred;
1567 spin_lock(&sk->sk_peer_lock);
1568 cred = get_cred(sk->sk_peer_cred);
1569 spin_unlock(&sk->sk_peer_lock);
1574 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1575 struct ucred *ucred)
1577 ucred->pid = pid_vnr(pid);
1578 ucred->uid = ucred->gid = -1;
1580 struct user_namespace *current_ns = current_user_ns();
1582 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1583 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1587 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1589 struct user_namespace *user_ns = current_user_ns();
1592 for (i = 0; i < src->ngroups; i++) {
1593 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1595 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1602 int sk_getsockopt(struct sock *sk, int level, int optname,
1603 sockptr_t optval, sockptr_t optlen)
1605 struct socket *sock = sk->sk_socket;
1610 unsigned long ulval;
1612 struct old_timeval32 tm32;
1613 struct __kernel_old_timeval tm;
1614 struct __kernel_sock_timeval stm;
1615 struct sock_txtime txtime;
1616 struct so_timestamping timestamping;
1619 int lv = sizeof(int);
1622 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1627 memset(&v, 0, sizeof(v));
1631 v.val = sock_flag(sk, SOCK_DBG);
1635 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1639 v.val = sock_flag(sk, SOCK_BROADCAST);
1643 v.val = READ_ONCE(sk->sk_sndbuf);
1647 v.val = READ_ONCE(sk->sk_rcvbuf);
1651 v.val = sk->sk_reuse;
1655 v.val = sk->sk_reuseport;
1659 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1663 v.val = sk->sk_type;
1667 v.val = sk->sk_protocol;
1671 v.val = sk->sk_family;
1675 v.val = -sock_error(sk);
1677 v.val = xchg(&sk->sk_err_soft, 0);
1681 v.val = sock_flag(sk, SOCK_URGINLINE);
1685 v.val = sk->sk_no_check_tx;
1689 v.val = READ_ONCE(sk->sk_priority);
1693 lv = sizeof(v.ling);
1694 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1695 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1701 case SO_TIMESTAMP_OLD:
1702 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1703 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1704 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1707 case SO_TIMESTAMPNS_OLD:
1708 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1711 case SO_TIMESTAMP_NEW:
1712 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1715 case SO_TIMESTAMPNS_NEW:
1716 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1719 case SO_TIMESTAMPING_OLD:
1720 lv = sizeof(v.timestamping);
1721 v.timestamping.flags = sk->sk_tsflags;
1722 v.timestamping.bind_phc = sk->sk_bind_phc;
1725 case SO_RCVTIMEO_OLD:
1726 case SO_RCVTIMEO_NEW:
1727 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1728 SO_RCVTIMEO_OLD == optname);
1731 case SO_SNDTIMEO_OLD:
1732 case SO_SNDTIMEO_NEW:
1733 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1734 SO_SNDTIMEO_OLD == optname);
1738 v.val = READ_ONCE(sk->sk_rcvlowat);
1746 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1750 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1755 struct ucred peercred;
1756 if (len > sizeof(peercred))
1757 len = sizeof(peercred);
1759 spin_lock(&sk->sk_peer_lock);
1760 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1761 spin_unlock(&sk->sk_peer_lock);
1763 if (copy_to_sockptr(optval, &peercred, len))
1770 struct pid *peer_pid;
1771 struct file *pidfd_file = NULL;
1774 if (len > sizeof(pidfd))
1775 len = sizeof(pidfd);
1777 spin_lock(&sk->sk_peer_lock);
1778 peer_pid = get_pid(sk->sk_peer_pid);
1779 spin_unlock(&sk->sk_peer_lock);
1784 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1789 if (copy_to_sockptr(optval, &pidfd, len) ||
1790 copy_to_sockptr(optlen, &len, sizeof(int))) {
1791 put_unused_fd(pidfd);
1797 fd_install(pidfd, pidfd_file);
1803 const struct cred *cred;
1806 cred = sk_get_peer_cred(sk);
1810 n = cred->group_info->ngroups;
1811 if (len < n * sizeof(gid_t)) {
1812 len = n * sizeof(gid_t);
1814 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1816 len = n * sizeof(gid_t);
1818 ret = groups_to_user(optval, cred->group_info);
1827 struct sockaddr_storage address;
1829 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1834 if (copy_to_sockptr(optval, &address, len))
1839 /* Dubious BSD thing... Probably nobody even uses it, but
1840 * the UNIX standard wants it for whatever reason... -DaveM
1843 v.val = sk->sk_state == TCP_LISTEN;
1847 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1851 return security_socket_getpeersec_stream(sock,
1852 optval, optlen, len);
1855 v.val = READ_ONCE(sk->sk_mark);
1859 v.val = sock_flag(sk, SOCK_RCVMARK);
1863 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1866 case SO_WIFI_STATUS:
1867 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1871 if (!READ_ONCE(sock->ops)->set_peek_off)
1874 v.val = READ_ONCE(sk->sk_peek_off);
1877 v.val = sock_flag(sk, SOCK_NOFCS);
1880 case SO_BINDTODEVICE:
1881 return sock_getbindtodevice(sk, optval, optlen, len);
1884 len = sk_get_filter(sk, optval, len);
1890 case SO_LOCK_FILTER:
1891 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1894 case SO_BPF_EXTENSIONS:
1895 v.val = bpf_tell_extensions();
1898 case SO_SELECT_ERR_QUEUE:
1899 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1902 #ifdef CONFIG_NET_RX_BUSY_POLL
1904 v.val = READ_ONCE(sk->sk_ll_usec);
1906 case SO_PREFER_BUSY_POLL:
1907 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1911 case SO_MAX_PACING_RATE:
1912 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1913 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1914 lv = sizeof(v.ulval);
1915 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1918 v.val = min_t(unsigned long, ~0U,
1919 READ_ONCE(sk->sk_max_pacing_rate));
1923 case SO_INCOMING_CPU:
1924 v.val = READ_ONCE(sk->sk_incoming_cpu);
1929 u32 meminfo[SK_MEMINFO_VARS];
1931 sk_get_meminfo(sk, meminfo);
1933 len = min_t(unsigned int, len, sizeof(meminfo));
1934 if (copy_to_sockptr(optval, &meminfo, len))
1940 #ifdef CONFIG_NET_RX_BUSY_POLL
1941 case SO_INCOMING_NAPI_ID:
1942 v.val = READ_ONCE(sk->sk_napi_id);
1944 /* aggregate non-NAPI IDs down to 0 */
1945 if (v.val < MIN_NAPI_ID)
1955 v.val64 = sock_gen_cookie(sk);
1959 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1963 lv = sizeof(v.txtime);
1964 v.txtime.clockid = sk->sk_clockid;
1965 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1966 SOF_TXTIME_DEADLINE_MODE : 0;
1967 v.txtime.flags |= sk->sk_txtime_report_errors ?
1968 SOF_TXTIME_REPORT_ERRORS : 0;
1971 case SO_BINDTOIFINDEX:
1972 v.val = READ_ONCE(sk->sk_bound_dev_if);
1975 case SO_NETNS_COOKIE:
1979 v.val64 = sock_net(sk)->net_cookie;
1983 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1986 case SO_RESERVE_MEM:
1987 v.val = READ_ONCE(sk->sk_reserved_mem);
1991 /* Paired with WRITE_ONCE() in sk_setsockopt() */
1992 v.val = READ_ONCE(sk->sk_txrehash);
1996 /* We implement the SO_SNDLOWAT etc to not be settable
1999 return -ENOPROTOOPT;
2004 if (copy_to_sockptr(optval, &v, len))
2007 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2012 int sock_getsockopt(struct socket *sock, int level, int optname,
2013 char __user *optval, int __user *optlen)
2015 return sk_getsockopt(sock->sk, level, optname,
2016 USER_SOCKPTR(optval),
2017 USER_SOCKPTR(optlen));
2021 * Initialize an sk_lock.
2023 * (We also register the sk_lock with the lock validator.)
2025 static inline void sock_lock_init(struct sock *sk)
2027 if (sk->sk_kern_sock)
2028 sock_lock_init_class_and_name(
2030 af_family_kern_slock_key_strings[sk->sk_family],
2031 af_family_kern_slock_keys + sk->sk_family,
2032 af_family_kern_key_strings[sk->sk_family],
2033 af_family_kern_keys + sk->sk_family);
2035 sock_lock_init_class_and_name(
2037 af_family_slock_key_strings[sk->sk_family],
2038 af_family_slock_keys + sk->sk_family,
2039 af_family_key_strings[sk->sk_family],
2040 af_family_keys + sk->sk_family);
2044 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2045 * even temporarly, because of RCU lookups. sk_node should also be left as is.
2046 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2048 static void sock_copy(struct sock *nsk, const struct sock *osk)
2050 const struct proto *prot = READ_ONCE(osk->sk_prot);
2051 #ifdef CONFIG_SECURITY_NETWORK
2052 void *sptr = nsk->sk_security;
2055 /* If we move sk_tx_queue_mapping out of the private section,
2056 * we must check if sk_tx_queue_clear() is called after
2057 * sock_copy() in sk_clone_lock().
2059 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2060 offsetof(struct sock, sk_dontcopy_begin) ||
2061 offsetof(struct sock, sk_tx_queue_mapping) >=
2062 offsetof(struct sock, sk_dontcopy_end));
2064 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2066 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2067 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2069 #ifdef CONFIG_SECURITY_NETWORK
2070 nsk->sk_security = sptr;
2071 security_sk_clone(osk, nsk);
2075 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2079 struct kmem_cache *slab;
2083 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2086 if (want_init_on_alloc(priority))
2087 sk_prot_clear_nulls(sk, prot->obj_size);
2089 sk = kmalloc(prot->obj_size, priority);
2092 if (security_sk_alloc(sk, family, priority))
2095 if (!try_module_get(prot->owner))
2102 security_sk_free(sk);
2105 kmem_cache_free(slab, sk);
2111 static void sk_prot_free(struct proto *prot, struct sock *sk)
2113 struct kmem_cache *slab;
2114 struct module *owner;
2116 owner = prot->owner;
2119 cgroup_sk_free(&sk->sk_cgrp_data);
2120 mem_cgroup_sk_free(sk);
2121 security_sk_free(sk);
2123 kmem_cache_free(slab, sk);
2130 * sk_alloc - All socket objects are allocated here
2131 * @net: the applicable net namespace
2132 * @family: protocol family
2133 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2134 * @prot: struct proto associated with this new sock instance
2135 * @kern: is this to be a kernel socket?
2137 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2138 struct proto *prot, int kern)
2142 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2144 sk->sk_family = family;
2146 * See comment in struct sock definition to understand
2147 * why we need sk_prot_creator -acme
2149 sk->sk_prot = sk->sk_prot_creator = prot;
2150 sk->sk_kern_sock = kern;
2152 sk->sk_net_refcnt = kern ? 0 : 1;
2153 if (likely(sk->sk_net_refcnt)) {
2154 get_net_track(net, &sk->ns_tracker, priority);
2155 sock_inuse_add(net, 1);
2157 __netns_tracker_alloc(net, &sk->ns_tracker,
2161 sock_net_set(sk, net);
2162 refcount_set(&sk->sk_wmem_alloc, 1);
2164 mem_cgroup_sk_alloc(sk);
2165 cgroup_sk_alloc(&sk->sk_cgrp_data);
2166 sock_update_classid(&sk->sk_cgrp_data);
2167 sock_update_netprioidx(&sk->sk_cgrp_data);
2168 sk_tx_queue_clear(sk);
2173 EXPORT_SYMBOL(sk_alloc);
2175 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2176 * grace period. This is the case for UDP sockets and TCP listeners.
2178 static void __sk_destruct(struct rcu_head *head)
2180 struct sock *sk = container_of(head, struct sock, sk_rcu);
2181 struct sk_filter *filter;
2183 if (sk->sk_destruct)
2184 sk->sk_destruct(sk);
2186 filter = rcu_dereference_check(sk->sk_filter,
2187 refcount_read(&sk->sk_wmem_alloc) == 0);
2189 sk_filter_uncharge(sk, filter);
2190 RCU_INIT_POINTER(sk->sk_filter, NULL);
2193 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2195 #ifdef CONFIG_BPF_SYSCALL
2196 bpf_sk_storage_free(sk);
2199 if (atomic_read(&sk->sk_omem_alloc))
2200 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2201 __func__, atomic_read(&sk->sk_omem_alloc));
2203 if (sk->sk_frag.page) {
2204 put_page(sk->sk_frag.page);
2205 sk->sk_frag.page = NULL;
2208 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2209 put_cred(sk->sk_peer_cred);
2210 put_pid(sk->sk_peer_pid);
2212 if (likely(sk->sk_net_refcnt))
2213 put_net_track(sock_net(sk), &sk->ns_tracker);
2215 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2217 sk_prot_free(sk->sk_prot_creator, sk);
2220 void sk_destruct(struct sock *sk)
2222 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2224 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2225 reuseport_detach_sock(sk);
2226 use_call_rcu = true;
2230 call_rcu(&sk->sk_rcu, __sk_destruct);
2232 __sk_destruct(&sk->sk_rcu);
2235 static void __sk_free(struct sock *sk)
2237 if (likely(sk->sk_net_refcnt))
2238 sock_inuse_add(sock_net(sk), -1);
2240 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2241 sock_diag_broadcast_destroy(sk);
2246 void sk_free(struct sock *sk)
2249 * We subtract one from sk_wmem_alloc and can know if
2250 * some packets are still in some tx queue.
2251 * If not null, sock_wfree() will call __sk_free(sk) later
2253 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2256 EXPORT_SYMBOL(sk_free);
2258 static void sk_init_common(struct sock *sk)
2260 skb_queue_head_init(&sk->sk_receive_queue);
2261 skb_queue_head_init(&sk->sk_write_queue);
2262 skb_queue_head_init(&sk->sk_error_queue);
2264 rwlock_init(&sk->sk_callback_lock);
2265 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2266 af_rlock_keys + sk->sk_family,
2267 af_family_rlock_key_strings[sk->sk_family]);
2268 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2269 af_wlock_keys + sk->sk_family,
2270 af_family_wlock_key_strings[sk->sk_family]);
2271 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2272 af_elock_keys + sk->sk_family,
2273 af_family_elock_key_strings[sk->sk_family]);
2274 lockdep_set_class_and_name(&sk->sk_callback_lock,
2275 af_callback_keys + sk->sk_family,
2276 af_family_clock_key_strings[sk->sk_family]);
2280 * sk_clone_lock - clone a socket, and lock its clone
2281 * @sk: the socket to clone
2282 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2284 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2286 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2288 struct proto *prot = READ_ONCE(sk->sk_prot);
2289 struct sk_filter *filter;
2290 bool is_charged = true;
2293 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2297 sock_copy(newsk, sk);
2299 newsk->sk_prot_creator = prot;
2302 if (likely(newsk->sk_net_refcnt)) {
2303 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2304 sock_inuse_add(sock_net(newsk), 1);
2306 /* Kernel sockets are not elevating the struct net refcount.
2307 * Instead, use a tracker to more easily detect if a layer
2308 * is not properly dismantling its kernel sockets at netns
2311 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2314 sk_node_init(&newsk->sk_node);
2315 sock_lock_init(newsk);
2316 bh_lock_sock(newsk);
2317 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2318 newsk->sk_backlog.len = 0;
2320 atomic_set(&newsk->sk_rmem_alloc, 0);
2322 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2323 refcount_set(&newsk->sk_wmem_alloc, 1);
2325 atomic_set(&newsk->sk_omem_alloc, 0);
2326 sk_init_common(newsk);
2328 newsk->sk_dst_cache = NULL;
2329 newsk->sk_dst_pending_confirm = 0;
2330 newsk->sk_wmem_queued = 0;
2331 newsk->sk_forward_alloc = 0;
2332 newsk->sk_reserved_mem = 0;
2333 atomic_set(&newsk->sk_drops, 0);
2334 newsk->sk_send_head = NULL;
2335 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2336 atomic_set(&newsk->sk_zckey, 0);
2338 sock_reset_flag(newsk, SOCK_DONE);
2340 /* sk->sk_memcg will be populated at accept() time */
2341 newsk->sk_memcg = NULL;
2343 cgroup_sk_clone(&newsk->sk_cgrp_data);
2346 filter = rcu_dereference(sk->sk_filter);
2348 /* though it's an empty new sock, the charging may fail
2349 * if sysctl_optmem_max was changed between creation of
2350 * original socket and cloning
2352 is_charged = sk_filter_charge(newsk, filter);
2353 RCU_INIT_POINTER(newsk->sk_filter, filter);
2356 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2357 /* We need to make sure that we don't uncharge the new
2358 * socket if we couldn't charge it in the first place
2359 * as otherwise we uncharge the parent's filter.
2362 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2363 sk_free_unlock_clone(newsk);
2367 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2369 if (bpf_sk_storage_clone(sk, newsk)) {
2370 sk_free_unlock_clone(newsk);
2375 /* Clear sk_user_data if parent had the pointer tagged
2376 * as not suitable for copying when cloning.
2378 if (sk_user_data_is_nocopy(newsk))
2379 newsk->sk_user_data = NULL;
2382 newsk->sk_err_soft = 0;
2383 newsk->sk_priority = 0;
2384 newsk->sk_incoming_cpu = raw_smp_processor_id();
2386 /* Before updating sk_refcnt, we must commit prior changes to memory
2387 * (Documentation/RCU/rculist_nulls.rst for details)
2390 refcount_set(&newsk->sk_refcnt, 2);
2392 sk_set_socket(newsk, NULL);
2393 sk_tx_queue_clear(newsk);
2394 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2396 if (newsk->sk_prot->sockets_allocated)
2397 sk_sockets_allocated_inc(newsk);
2399 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2400 net_enable_timestamp();
2404 EXPORT_SYMBOL_GPL(sk_clone_lock);
2406 void sk_free_unlock_clone(struct sock *sk)
2408 /* It is still raw copy of parent, so invalidate
2409 * destructor and make plain sk_free() */
2410 sk->sk_destruct = NULL;
2414 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2416 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2418 bool is_ipv6 = false;
2421 #if IS_ENABLED(CONFIG_IPV6)
2422 is_ipv6 = (sk->sk_family == AF_INET6 &&
2423 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2425 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2426 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2427 READ_ONCE(dst->dev->gso_ipv4_max_size);
2428 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2429 max_size = GSO_LEGACY_MAX_SIZE;
2431 return max_size - (MAX_TCP_HEADER + 1);
2434 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2438 sk->sk_route_caps = dst->dev->features;
2440 sk->sk_route_caps |= NETIF_F_GSO;
2441 if (sk->sk_route_caps & NETIF_F_GSO)
2442 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2443 if (unlikely(sk->sk_gso_disabled))
2444 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2445 if (sk_can_gso(sk)) {
2446 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2447 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2449 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2450 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2451 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2452 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2455 sk->sk_gso_max_segs = max_segs;
2456 sk_dst_set(sk, dst);
2458 EXPORT_SYMBOL_GPL(sk_setup_caps);
2461 * Simple resource managers for sockets.
2466 * Write buffer destructor automatically called from kfree_skb.
2468 void sock_wfree(struct sk_buff *skb)
2470 struct sock *sk = skb->sk;
2471 unsigned int len = skb->truesize;
2474 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2475 if (sock_flag(sk, SOCK_RCU_FREE) &&
2476 sk->sk_write_space == sock_def_write_space) {
2478 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2479 sock_def_write_space_wfree(sk);
2487 * Keep a reference on sk_wmem_alloc, this will be released
2488 * after sk_write_space() call
2490 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2491 sk->sk_write_space(sk);
2495 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2496 * could not do because of in-flight packets
2498 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2501 EXPORT_SYMBOL(sock_wfree);
2503 /* This variant of sock_wfree() is used by TCP,
2504 * since it sets SOCK_USE_WRITE_QUEUE.
2506 void __sock_wfree(struct sk_buff *skb)
2508 struct sock *sk = skb->sk;
2510 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2514 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2519 if (unlikely(!sk_fullsock(sk))) {
2520 skb->destructor = sock_edemux;
2525 skb->destructor = sock_wfree;
2526 skb_set_hash_from_sk(skb, sk);
2528 * We used to take a refcount on sk, but following operation
2529 * is enough to guarantee sk_free() wont free this sock until
2530 * all in-flight packets are completed
2532 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2534 EXPORT_SYMBOL(skb_set_owner_w);
2536 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2538 #ifdef CONFIG_TLS_DEVICE
2539 /* Drivers depend on in-order delivery for crypto offload,
2540 * partial orphan breaks out-of-order-OK logic.
2545 return (skb->destructor == sock_wfree ||
2546 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2549 /* This helper is used by netem, as it can hold packets in its
2550 * delay queue. We want to allow the owner socket to send more
2551 * packets, as if they were already TX completed by a typical driver.
2552 * But we also want to keep skb->sk set because some packet schedulers
2553 * rely on it (sch_fq for example).
2555 void skb_orphan_partial(struct sk_buff *skb)
2557 if (skb_is_tcp_pure_ack(skb))
2560 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2565 EXPORT_SYMBOL(skb_orphan_partial);
2568 * Read buffer destructor automatically called from kfree_skb.
2570 void sock_rfree(struct sk_buff *skb)
2572 struct sock *sk = skb->sk;
2573 unsigned int len = skb->truesize;
2575 atomic_sub(len, &sk->sk_rmem_alloc);
2576 sk_mem_uncharge(sk, len);
2578 EXPORT_SYMBOL(sock_rfree);
2581 * Buffer destructor for skbs that are not used directly in read or write
2582 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2584 void sock_efree(struct sk_buff *skb)
2588 EXPORT_SYMBOL(sock_efree);
2590 /* Buffer destructor for prefetch/receive path where reference count may
2591 * not be held, e.g. for listen sockets.
2594 void sock_pfree(struct sk_buff *skb)
2596 if (sk_is_refcounted(skb->sk))
2597 sock_gen_put(skb->sk);
2599 EXPORT_SYMBOL(sock_pfree);
2600 #endif /* CONFIG_INET */
2602 kuid_t sock_i_uid(struct sock *sk)
2606 read_lock_bh(&sk->sk_callback_lock);
2607 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2608 read_unlock_bh(&sk->sk_callback_lock);
2611 EXPORT_SYMBOL(sock_i_uid);
2613 unsigned long __sock_i_ino(struct sock *sk)
2617 read_lock(&sk->sk_callback_lock);
2618 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2619 read_unlock(&sk->sk_callback_lock);
2622 EXPORT_SYMBOL(__sock_i_ino);
2624 unsigned long sock_i_ino(struct sock *sk)
2629 ino = __sock_i_ino(sk);
2633 EXPORT_SYMBOL(sock_i_ino);
2636 * Allocate a skb from the socket's send buffer.
2638 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2642 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2643 struct sk_buff *skb = alloc_skb(size, priority);
2646 skb_set_owner_w(skb, sk);
2652 EXPORT_SYMBOL(sock_wmalloc);
2654 static void sock_ofree(struct sk_buff *skb)
2656 struct sock *sk = skb->sk;
2658 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2661 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2664 struct sk_buff *skb;
2666 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2667 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2668 READ_ONCE(sysctl_optmem_max))
2671 skb = alloc_skb(size, priority);
2675 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2677 skb->destructor = sock_ofree;
2682 * Allocate a memory block from the socket's option memory buffer.
2684 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2686 int optmem_max = READ_ONCE(sysctl_optmem_max);
2688 if ((unsigned int)size <= optmem_max &&
2689 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2691 /* First do the add, to avoid the race if kmalloc
2694 atomic_add(size, &sk->sk_omem_alloc);
2695 mem = kmalloc(size, priority);
2698 atomic_sub(size, &sk->sk_omem_alloc);
2702 EXPORT_SYMBOL(sock_kmalloc);
2704 /* Free an option memory block. Note, we actually want the inline
2705 * here as this allows gcc to detect the nullify and fold away the
2706 * condition entirely.
2708 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2711 if (WARN_ON_ONCE(!mem))
2714 kfree_sensitive(mem);
2717 atomic_sub(size, &sk->sk_omem_alloc);
2720 void sock_kfree_s(struct sock *sk, void *mem, int size)
2722 __sock_kfree_s(sk, mem, size, false);
2724 EXPORT_SYMBOL(sock_kfree_s);
2726 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2728 __sock_kfree_s(sk, mem, size, true);
2730 EXPORT_SYMBOL(sock_kzfree_s);
2732 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2733 I think, these locks should be removed for datagram sockets.
2735 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2739 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2743 if (signal_pending(current))
2745 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2746 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2747 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2749 if (sk->sk_shutdown & SEND_SHUTDOWN)
2753 timeo = schedule_timeout(timeo);
2755 finish_wait(sk_sleep(sk), &wait);
2761 * Generic send/receive buffer handlers
2764 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2765 unsigned long data_len, int noblock,
2766 int *errcode, int max_page_order)
2768 struct sk_buff *skb;
2772 timeo = sock_sndtimeo(sk, noblock);
2774 err = sock_error(sk);
2779 if (sk->sk_shutdown & SEND_SHUTDOWN)
2782 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2785 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2786 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2790 if (signal_pending(current))
2792 timeo = sock_wait_for_wmem(sk, timeo);
2794 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2795 errcode, sk->sk_allocation);
2797 skb_set_owner_w(skb, sk);
2801 err = sock_intr_errno(timeo);
2806 EXPORT_SYMBOL(sock_alloc_send_pskb);
2808 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2809 struct sockcm_cookie *sockc)
2813 switch (cmsg->cmsg_type) {
2815 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2816 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2818 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2820 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2822 case SO_TIMESTAMPING_OLD:
2823 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2826 tsflags = *(u32 *)CMSG_DATA(cmsg);
2827 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2830 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2831 sockc->tsflags |= tsflags;
2834 if (!sock_flag(sk, SOCK_TXTIME))
2836 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2838 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2840 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2842 case SCM_CREDENTIALS:
2849 EXPORT_SYMBOL(__sock_cmsg_send);
2851 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2852 struct sockcm_cookie *sockc)
2854 struct cmsghdr *cmsg;
2857 for_each_cmsghdr(cmsg, msg) {
2858 if (!CMSG_OK(msg, cmsg))
2860 if (cmsg->cmsg_level != SOL_SOCKET)
2862 ret = __sock_cmsg_send(sk, cmsg, sockc);
2868 EXPORT_SYMBOL(sock_cmsg_send);
2870 static void sk_enter_memory_pressure(struct sock *sk)
2872 if (!sk->sk_prot->enter_memory_pressure)
2875 sk->sk_prot->enter_memory_pressure(sk);
2878 static void sk_leave_memory_pressure(struct sock *sk)
2880 if (sk->sk_prot->leave_memory_pressure) {
2881 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2882 tcp_leave_memory_pressure, sk);
2884 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2886 if (memory_pressure && READ_ONCE(*memory_pressure))
2887 WRITE_ONCE(*memory_pressure, 0);
2891 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2894 * skb_page_frag_refill - check that a page_frag contains enough room
2895 * @sz: minimum size of the fragment we want to get
2896 * @pfrag: pointer to page_frag
2897 * @gfp: priority for memory allocation
2899 * Note: While this allocator tries to use high order pages, there is
2900 * no guarantee that allocations succeed. Therefore, @sz MUST be
2901 * less or equal than PAGE_SIZE.
2903 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2906 if (page_ref_count(pfrag->page) == 1) {
2910 if (pfrag->offset + sz <= pfrag->size)
2912 put_page(pfrag->page);
2916 if (SKB_FRAG_PAGE_ORDER &&
2917 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2918 /* Avoid direct reclaim but allow kswapd to wake */
2919 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2920 __GFP_COMP | __GFP_NOWARN |
2922 SKB_FRAG_PAGE_ORDER);
2923 if (likely(pfrag->page)) {
2924 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2928 pfrag->page = alloc_page(gfp);
2929 if (likely(pfrag->page)) {
2930 pfrag->size = PAGE_SIZE;
2935 EXPORT_SYMBOL(skb_page_frag_refill);
2937 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2939 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2942 sk_enter_memory_pressure(sk);
2943 sk_stream_moderate_sndbuf(sk);
2946 EXPORT_SYMBOL(sk_page_frag_refill);
2948 void __lock_sock(struct sock *sk)
2949 __releases(&sk->sk_lock.slock)
2950 __acquires(&sk->sk_lock.slock)
2955 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2956 TASK_UNINTERRUPTIBLE);
2957 spin_unlock_bh(&sk->sk_lock.slock);
2959 spin_lock_bh(&sk->sk_lock.slock);
2960 if (!sock_owned_by_user(sk))
2963 finish_wait(&sk->sk_lock.wq, &wait);
2966 void __release_sock(struct sock *sk)
2967 __releases(&sk->sk_lock.slock)
2968 __acquires(&sk->sk_lock.slock)
2970 struct sk_buff *skb, *next;
2972 while ((skb = sk->sk_backlog.head) != NULL) {
2973 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2975 spin_unlock_bh(&sk->sk_lock.slock);
2980 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2981 skb_mark_not_on_list(skb);
2982 sk_backlog_rcv(sk, skb);
2987 } while (skb != NULL);
2989 spin_lock_bh(&sk->sk_lock.slock);
2993 * Doing the zeroing here guarantee we can not loop forever
2994 * while a wild producer attempts to flood us.
2996 sk->sk_backlog.len = 0;
2999 void __sk_flush_backlog(struct sock *sk)
3001 spin_lock_bh(&sk->sk_lock.slock);
3003 spin_unlock_bh(&sk->sk_lock.slock);
3005 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3008 * sk_wait_data - wait for data to arrive at sk_receive_queue
3009 * @sk: sock to wait on
3010 * @timeo: for how long
3011 * @skb: last skb seen on sk_receive_queue
3013 * Now socket state including sk->sk_err is changed only under lock,
3014 * hence we may omit checks after joining wait queue.
3015 * We check receive queue before schedule() only as optimization;
3016 * it is very likely that release_sock() added new data.
3018 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3020 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3023 add_wait_queue(sk_sleep(sk), &wait);
3024 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3025 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3026 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3027 remove_wait_queue(sk_sleep(sk), &wait);
3030 EXPORT_SYMBOL(sk_wait_data);
3033 * __sk_mem_raise_allocated - increase memory_allocated
3035 * @size: memory size to allocate
3036 * @amt: pages to allocate
3037 * @kind: allocation type
3039 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
3041 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3043 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
3044 struct proto *prot = sk->sk_prot;
3045 bool charged = true;
3048 sk_memory_allocated_add(sk, amt);
3049 allocated = sk_memory_allocated(sk);
3051 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3052 gfp_memcg_charge())))
3053 goto suppress_allocation;
3056 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3057 sk_leave_memory_pressure(sk);
3061 /* Under pressure. */
3062 if (allocated > sk_prot_mem_limits(sk, 1))
3063 sk_enter_memory_pressure(sk);
3065 /* Over hard limit. */
3066 if (allocated > sk_prot_mem_limits(sk, 2))
3067 goto suppress_allocation;
3069 /* guarantee minimum buffer size under pressure */
3070 if (kind == SK_MEM_RECV) {
3071 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3074 } else { /* SK_MEM_SEND */
3075 int wmem0 = sk_get_wmem0(sk, prot);
3077 if (sk->sk_type == SOCK_STREAM) {
3078 if (sk->sk_wmem_queued < wmem0)
3080 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3085 if (sk_has_memory_pressure(sk)) {
3088 if (!sk_under_memory_pressure(sk))
3090 alloc = sk_sockets_allocated_read_positive(sk);
3091 if (sk_prot_mem_limits(sk, 2) > alloc *
3092 sk_mem_pages(sk->sk_wmem_queued +
3093 atomic_read(&sk->sk_rmem_alloc) +
3094 sk->sk_forward_alloc))
3098 suppress_allocation:
3100 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3101 sk_stream_moderate_sndbuf(sk);
3103 /* Fail only if socket is _under_ its sndbuf.
3104 * In this case we cannot block, so that we have to fail.
3106 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3107 /* Force charge with __GFP_NOFAIL */
3108 if (memcg_charge && !charged) {
3109 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3110 gfp_memcg_charge() | __GFP_NOFAIL);
3116 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3117 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3119 sk_memory_allocated_sub(sk, amt);
3121 if (memcg_charge && charged)
3122 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
3128 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3130 * @size: memory size to allocate
3131 * @kind: allocation type
3133 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3134 * rmem allocation. This function assumes that protocols which have
3135 * memory_pressure use sk_wmem_queued as write buffer accounting.
3137 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3139 int ret, amt = sk_mem_pages(size);
3141 sk->sk_forward_alloc += amt << PAGE_SHIFT;
3142 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3144 sk->sk_forward_alloc -= amt << PAGE_SHIFT;
3147 EXPORT_SYMBOL(__sk_mem_schedule);
3150 * __sk_mem_reduce_allocated - reclaim memory_allocated
3152 * @amount: number of quanta
3154 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3156 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3158 sk_memory_allocated_sub(sk, amount);
3160 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3161 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3163 if (sk_under_global_memory_pressure(sk) &&
3164 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3165 sk_leave_memory_pressure(sk);
3169 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3171 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3173 void __sk_mem_reclaim(struct sock *sk, int amount)
3175 amount >>= PAGE_SHIFT;
3176 sk->sk_forward_alloc -= amount << PAGE_SHIFT;
3177 __sk_mem_reduce_allocated(sk, amount);
3179 EXPORT_SYMBOL(__sk_mem_reclaim);
3181 int sk_set_peek_off(struct sock *sk, int val)
3183 WRITE_ONCE(sk->sk_peek_off, val);
3186 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3189 * Set of default routines for initialising struct proto_ops when
3190 * the protocol does not support a particular function. In certain
3191 * cases where it makes no sense for a protocol to have a "do nothing"
3192 * function, some default processing is provided.
3195 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3199 EXPORT_SYMBOL(sock_no_bind);
3201 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3206 EXPORT_SYMBOL(sock_no_connect);
3208 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3212 EXPORT_SYMBOL(sock_no_socketpair);
3214 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3219 EXPORT_SYMBOL(sock_no_accept);
3221 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3226 EXPORT_SYMBOL(sock_no_getname);
3228 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3232 EXPORT_SYMBOL(sock_no_ioctl);
3234 int sock_no_listen(struct socket *sock, int backlog)
3238 EXPORT_SYMBOL(sock_no_listen);
3240 int sock_no_shutdown(struct socket *sock, int how)
3244 EXPORT_SYMBOL(sock_no_shutdown);
3246 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3250 EXPORT_SYMBOL(sock_no_sendmsg);
3252 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3256 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3258 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3263 EXPORT_SYMBOL(sock_no_recvmsg);
3265 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3267 /* Mirror missing mmap method error code */
3270 EXPORT_SYMBOL(sock_no_mmap);
3273 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3274 * various sock-based usage counts.
3276 void __receive_sock(struct file *file)
3278 struct socket *sock;
3280 sock = sock_from_file(file);
3282 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3283 sock_update_classid(&sock->sk->sk_cgrp_data);
3288 * Default Socket Callbacks
3291 static void sock_def_wakeup(struct sock *sk)
3293 struct socket_wq *wq;
3296 wq = rcu_dereference(sk->sk_wq);
3297 if (skwq_has_sleeper(wq))
3298 wake_up_interruptible_all(&wq->wait);
3302 static void sock_def_error_report(struct sock *sk)
3304 struct socket_wq *wq;
3307 wq = rcu_dereference(sk->sk_wq);
3308 if (skwq_has_sleeper(wq))
3309 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3310 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3314 void sock_def_readable(struct sock *sk)
3316 struct socket_wq *wq;
3318 trace_sk_data_ready(sk);
3321 wq = rcu_dereference(sk->sk_wq);
3322 if (skwq_has_sleeper(wq))
3323 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3324 EPOLLRDNORM | EPOLLRDBAND);
3325 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3329 static void sock_def_write_space(struct sock *sk)
3331 struct socket_wq *wq;
3335 /* Do not wake up a writer until he can make "significant"
3338 if (sock_writeable(sk)) {
3339 wq = rcu_dereference(sk->sk_wq);
3340 if (skwq_has_sleeper(wq))
3341 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3342 EPOLLWRNORM | EPOLLWRBAND);
3344 /* Should agree with poll, otherwise some programs break */
3345 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3351 /* An optimised version of sock_def_write_space(), should only be called
3352 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3355 static void sock_def_write_space_wfree(struct sock *sk)
3357 /* Do not wake up a writer until he can make "significant"
3360 if (sock_writeable(sk)) {
3361 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3363 /* rely on refcount_sub from sock_wfree() */
3364 smp_mb__after_atomic();
3365 if (wq && waitqueue_active(&wq->wait))
3366 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3367 EPOLLWRNORM | EPOLLWRBAND);
3369 /* Should agree with poll, otherwise some programs break */
3370 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3374 static void sock_def_destruct(struct sock *sk)
3378 void sk_send_sigurg(struct sock *sk)
3380 if (sk->sk_socket && sk->sk_socket->file)
3381 if (send_sigurg(&sk->sk_socket->file->f_owner))
3382 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3384 EXPORT_SYMBOL(sk_send_sigurg);
3386 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3387 unsigned long expires)
3389 if (!mod_timer(timer, expires))
3392 EXPORT_SYMBOL(sk_reset_timer);
3394 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3396 if (del_timer(timer))
3399 EXPORT_SYMBOL(sk_stop_timer);
3401 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3403 if (del_timer_sync(timer))
3406 EXPORT_SYMBOL(sk_stop_timer_sync);
3408 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3411 sk->sk_send_head = NULL;
3413 timer_setup(&sk->sk_timer, NULL, 0);
3415 sk->sk_allocation = GFP_KERNEL;
3416 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3417 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3418 sk->sk_state = TCP_CLOSE;
3419 sk->sk_use_task_frag = true;
3420 sk_set_socket(sk, sock);
3422 sock_set_flag(sk, SOCK_ZAPPED);
3425 sk->sk_type = sock->type;
3426 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3429 RCU_INIT_POINTER(sk->sk_wq, NULL);
3433 rwlock_init(&sk->sk_callback_lock);
3434 if (sk->sk_kern_sock)
3435 lockdep_set_class_and_name(
3436 &sk->sk_callback_lock,
3437 af_kern_callback_keys + sk->sk_family,
3438 af_family_kern_clock_key_strings[sk->sk_family]);
3440 lockdep_set_class_and_name(
3441 &sk->sk_callback_lock,
3442 af_callback_keys + sk->sk_family,
3443 af_family_clock_key_strings[sk->sk_family]);
3445 sk->sk_state_change = sock_def_wakeup;
3446 sk->sk_data_ready = sock_def_readable;
3447 sk->sk_write_space = sock_def_write_space;
3448 sk->sk_error_report = sock_def_error_report;
3449 sk->sk_destruct = sock_def_destruct;
3451 sk->sk_frag.page = NULL;
3452 sk->sk_frag.offset = 0;
3453 sk->sk_peek_off = -1;
3455 sk->sk_peer_pid = NULL;
3456 sk->sk_peer_cred = NULL;
3457 spin_lock_init(&sk->sk_peer_lock);
3459 sk->sk_write_pending = 0;
3460 sk->sk_rcvlowat = 1;
3461 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3462 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3464 sk->sk_stamp = SK_DEFAULT_STAMP;
3465 #if BITS_PER_LONG==32
3466 seqlock_init(&sk->sk_stamp_seq);
3468 atomic_set(&sk->sk_zckey, 0);
3470 #ifdef CONFIG_NET_RX_BUSY_POLL
3472 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3475 sk->sk_max_pacing_rate = ~0UL;
3476 sk->sk_pacing_rate = ~0UL;
3477 WRITE_ONCE(sk->sk_pacing_shift, 10);
3478 sk->sk_incoming_cpu = -1;
3480 sk_rx_queue_clear(sk);
3482 * Before updating sk_refcnt, we must commit prior changes to memory
3483 * (Documentation/RCU/rculist_nulls.rst for details)
3486 refcount_set(&sk->sk_refcnt, 1);
3487 atomic_set(&sk->sk_drops, 0);
3489 EXPORT_SYMBOL(sock_init_data_uid);
3491 void sock_init_data(struct socket *sock, struct sock *sk)
3494 SOCK_INODE(sock)->i_uid :
3495 make_kuid(sock_net(sk)->user_ns, 0);
3497 sock_init_data_uid(sock, sk, uid);
3499 EXPORT_SYMBOL(sock_init_data);
3501 void lock_sock_nested(struct sock *sk, int subclass)
3503 /* The sk_lock has mutex_lock() semantics here. */
3504 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3507 spin_lock_bh(&sk->sk_lock.slock);
3508 if (sock_owned_by_user_nocheck(sk))
3510 sk->sk_lock.owned = 1;
3511 spin_unlock_bh(&sk->sk_lock.slock);
3513 EXPORT_SYMBOL(lock_sock_nested);
3515 void release_sock(struct sock *sk)
3517 spin_lock_bh(&sk->sk_lock.slock);
3518 if (sk->sk_backlog.tail)
3521 /* Warning : release_cb() might need to release sk ownership,
3522 * ie call sock_release_ownership(sk) before us.
3524 if (sk->sk_prot->release_cb)
3525 sk->sk_prot->release_cb(sk);
3527 sock_release_ownership(sk);
3528 if (waitqueue_active(&sk->sk_lock.wq))
3529 wake_up(&sk->sk_lock.wq);
3530 spin_unlock_bh(&sk->sk_lock.slock);
3532 EXPORT_SYMBOL(release_sock);
3534 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3537 spin_lock_bh(&sk->sk_lock.slock);
3539 if (!sock_owned_by_user_nocheck(sk)) {
3541 * Fast path return with bottom halves disabled and
3542 * sock::sk_lock.slock held.
3544 * The 'mutex' is not contended and holding
3545 * sock::sk_lock.slock prevents all other lockers to
3546 * proceed so the corresponding unlock_sock_fast() can
3547 * avoid the slow path of release_sock() completely and
3548 * just release slock.
3550 * From a semantical POV this is equivalent to 'acquiring'
3551 * the 'mutex', hence the corresponding lockdep
3552 * mutex_release() has to happen in the fast path of
3553 * unlock_sock_fast().
3559 sk->sk_lock.owned = 1;
3560 __acquire(&sk->sk_lock.slock);
3561 spin_unlock_bh(&sk->sk_lock.slock);
3564 EXPORT_SYMBOL(__lock_sock_fast);
3566 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3567 bool timeval, bool time32)
3569 struct sock *sk = sock->sk;
3570 struct timespec64 ts;
3572 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3573 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3574 if (ts.tv_sec == -1)
3576 if (ts.tv_sec == 0) {
3577 ktime_t kt = ktime_get_real();
3578 sock_write_timestamp(sk, kt);
3579 ts = ktime_to_timespec64(kt);
3585 #ifdef CONFIG_COMPAT_32BIT_TIME
3587 return put_old_timespec32(&ts, userstamp);
3589 #ifdef CONFIG_SPARC64
3590 /* beware of padding in sparc64 timeval */
3591 if (timeval && !in_compat_syscall()) {
3592 struct __kernel_old_timeval __user tv = {
3593 .tv_sec = ts.tv_sec,
3594 .tv_usec = ts.tv_nsec,
3596 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3601 return put_timespec64(&ts, userstamp);
3603 EXPORT_SYMBOL(sock_gettstamp);
3605 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3607 if (!sock_flag(sk, flag)) {
3608 unsigned long previous_flags = sk->sk_flags;
3610 sock_set_flag(sk, flag);
3612 * we just set one of the two flags which require net
3613 * time stamping, but time stamping might have been on
3614 * already because of the other one
3616 if (sock_needs_netstamp(sk) &&
3617 !(previous_flags & SK_FLAGS_TIMESTAMP))
3618 net_enable_timestamp();
3622 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3623 int level, int type)
3625 struct sock_exterr_skb *serr;
3626 struct sk_buff *skb;
3630 skb = sock_dequeue_err_skb(sk);
3636 msg->msg_flags |= MSG_TRUNC;
3639 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3643 sock_recv_timestamp(msg, sk, skb);
3645 serr = SKB_EXT_ERR(skb);
3646 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3648 msg->msg_flags |= MSG_ERRQUEUE;
3656 EXPORT_SYMBOL(sock_recv_errqueue);
3659 * Get a socket option on an socket.
3661 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3662 * asynchronous errors should be reported by getsockopt. We assume
3663 * this means if you specify SO_ERROR (otherwise whats the point of it).
3665 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3666 char __user *optval, int __user *optlen)
3668 struct sock *sk = sock->sk;
3670 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3671 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3673 EXPORT_SYMBOL(sock_common_getsockopt);
3675 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3678 struct sock *sk = sock->sk;
3682 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3684 msg->msg_namelen = addr_len;
3687 EXPORT_SYMBOL(sock_common_recvmsg);
3690 * Set socket options on an inet socket.
3692 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3693 sockptr_t optval, unsigned int optlen)
3695 struct sock *sk = sock->sk;
3697 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3698 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3700 EXPORT_SYMBOL(sock_common_setsockopt);
3702 void sk_common_release(struct sock *sk)
3704 if (sk->sk_prot->destroy)
3705 sk->sk_prot->destroy(sk);
3708 * Observation: when sk_common_release is called, processes have
3709 * no access to socket. But net still has.
3710 * Step one, detach it from networking:
3712 * A. Remove from hash tables.
3715 sk->sk_prot->unhash(sk);
3718 * In this point socket cannot receive new packets, but it is possible
3719 * that some packets are in flight because some CPU runs receiver and
3720 * did hash table lookup before we unhashed socket. They will achieve
3721 * receive queue and will be purged by socket destructor.
3723 * Also we still have packets pending on receive queue and probably,
3724 * our own packets waiting in device queues. sock_destroy will drain
3725 * receive queue, but transmitted packets will delay socket destruction
3726 * until the last reference will be released.
3731 xfrm_sk_free_policy(sk);
3735 EXPORT_SYMBOL(sk_common_release);
3737 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3739 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3741 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3742 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3743 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3744 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3745 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3746 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3747 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3748 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3749 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3752 #ifdef CONFIG_PROC_FS
3753 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3755 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3757 int cpu, idx = prot->inuse_idx;
3760 for_each_possible_cpu(cpu)
3761 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3763 return res >= 0 ? res : 0;
3765 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3767 int sock_inuse_get(struct net *net)
3771 for_each_possible_cpu(cpu)
3772 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3777 EXPORT_SYMBOL_GPL(sock_inuse_get);
3779 static int __net_init sock_inuse_init_net(struct net *net)
3781 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3782 if (net->core.prot_inuse == NULL)
3787 static void __net_exit sock_inuse_exit_net(struct net *net)
3789 free_percpu(net->core.prot_inuse);
3792 static struct pernet_operations net_inuse_ops = {
3793 .init = sock_inuse_init_net,
3794 .exit = sock_inuse_exit_net,
3797 static __init int net_inuse_init(void)
3799 if (register_pernet_subsys(&net_inuse_ops))
3800 panic("Cannot initialize net inuse counters");
3805 core_initcall(net_inuse_init);
3807 static int assign_proto_idx(struct proto *prot)
3809 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3811 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3812 pr_err("PROTO_INUSE_NR exhausted\n");
3816 set_bit(prot->inuse_idx, proto_inuse_idx);
3820 static void release_proto_idx(struct proto *prot)
3822 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3823 clear_bit(prot->inuse_idx, proto_inuse_idx);
3826 static inline int assign_proto_idx(struct proto *prot)
3831 static inline void release_proto_idx(struct proto *prot)
3837 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3841 kfree(twsk_prot->twsk_slab_name);
3842 twsk_prot->twsk_slab_name = NULL;
3843 kmem_cache_destroy(twsk_prot->twsk_slab);
3844 twsk_prot->twsk_slab = NULL;
3847 static int tw_prot_init(const struct proto *prot)
3849 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3854 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3856 if (!twsk_prot->twsk_slab_name)
3859 twsk_prot->twsk_slab =
3860 kmem_cache_create(twsk_prot->twsk_slab_name,
3861 twsk_prot->twsk_obj_size, 0,
3862 SLAB_ACCOUNT | prot->slab_flags,
3864 if (!twsk_prot->twsk_slab) {
3865 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3873 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3877 kfree(rsk_prot->slab_name);
3878 rsk_prot->slab_name = NULL;
3879 kmem_cache_destroy(rsk_prot->slab);
3880 rsk_prot->slab = NULL;
3883 static int req_prot_init(const struct proto *prot)
3885 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3890 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3892 if (!rsk_prot->slab_name)
3895 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3896 rsk_prot->obj_size, 0,
3897 SLAB_ACCOUNT | prot->slab_flags,
3900 if (!rsk_prot->slab) {
3901 pr_crit("%s: Can't create request sock SLAB cache!\n",
3908 int proto_register(struct proto *prot, int alloc_slab)
3912 if (prot->memory_allocated && !prot->sysctl_mem) {
3913 pr_err("%s: missing sysctl_mem\n", prot->name);
3916 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3917 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3921 prot->slab = kmem_cache_create_usercopy(prot->name,
3923 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3925 prot->useroffset, prot->usersize,
3928 if (prot->slab == NULL) {
3929 pr_crit("%s: Can't create sock SLAB cache!\n",
3934 if (req_prot_init(prot))
3935 goto out_free_request_sock_slab;
3937 if (tw_prot_init(prot))
3938 goto out_free_timewait_sock_slab;
3941 mutex_lock(&proto_list_mutex);
3942 ret = assign_proto_idx(prot);
3944 mutex_unlock(&proto_list_mutex);
3945 goto out_free_timewait_sock_slab;
3947 list_add(&prot->node, &proto_list);
3948 mutex_unlock(&proto_list_mutex);
3951 out_free_timewait_sock_slab:
3953 tw_prot_cleanup(prot->twsk_prot);
3954 out_free_request_sock_slab:
3956 req_prot_cleanup(prot->rsk_prot);
3958 kmem_cache_destroy(prot->slab);
3964 EXPORT_SYMBOL(proto_register);
3966 void proto_unregister(struct proto *prot)
3968 mutex_lock(&proto_list_mutex);
3969 release_proto_idx(prot);
3970 list_del(&prot->node);
3971 mutex_unlock(&proto_list_mutex);
3973 kmem_cache_destroy(prot->slab);
3976 req_prot_cleanup(prot->rsk_prot);
3977 tw_prot_cleanup(prot->twsk_prot);
3979 EXPORT_SYMBOL(proto_unregister);
3981 int sock_load_diag_module(int family, int protocol)
3984 if (!sock_is_registered(family))
3987 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3988 NETLINK_SOCK_DIAG, family);
3992 if (family == AF_INET &&
3993 protocol != IPPROTO_RAW &&
3994 protocol < MAX_INET_PROTOS &&
3995 !rcu_access_pointer(inet_protos[protocol]))
3999 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4000 NETLINK_SOCK_DIAG, family, protocol);
4002 EXPORT_SYMBOL(sock_load_diag_module);
4004 #ifdef CONFIG_PROC_FS
4005 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4006 __acquires(proto_list_mutex)
4008 mutex_lock(&proto_list_mutex);
4009 return seq_list_start_head(&proto_list, *pos);
4012 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4014 return seq_list_next(v, &proto_list, pos);
4017 static void proto_seq_stop(struct seq_file *seq, void *v)
4018 __releases(proto_list_mutex)
4020 mutex_unlock(&proto_list_mutex);
4023 static char proto_method_implemented(const void *method)
4025 return method == NULL ? 'n' : 'y';
4027 static long sock_prot_memory_allocated(struct proto *proto)
4029 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4032 static const char *sock_prot_memory_pressure(struct proto *proto)
4034 return proto->memory_pressure != NULL ?
4035 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4038 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4041 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4042 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4045 sock_prot_inuse_get(seq_file_net(seq), proto),
4046 sock_prot_memory_allocated(proto),
4047 sock_prot_memory_pressure(proto),
4049 proto->slab == NULL ? "no" : "yes",
4050 module_name(proto->owner),
4051 proto_method_implemented(proto->close),
4052 proto_method_implemented(proto->connect),
4053 proto_method_implemented(proto->disconnect),
4054 proto_method_implemented(proto->accept),
4055 proto_method_implemented(proto->ioctl),
4056 proto_method_implemented(proto->init),
4057 proto_method_implemented(proto->destroy),
4058 proto_method_implemented(proto->shutdown),
4059 proto_method_implemented(proto->setsockopt),
4060 proto_method_implemented(proto->getsockopt),
4061 proto_method_implemented(proto->sendmsg),
4062 proto_method_implemented(proto->recvmsg),
4063 proto_method_implemented(proto->bind),
4064 proto_method_implemented(proto->backlog_rcv),
4065 proto_method_implemented(proto->hash),
4066 proto_method_implemented(proto->unhash),
4067 proto_method_implemented(proto->get_port),
4068 proto_method_implemented(proto->enter_memory_pressure));
4071 static int proto_seq_show(struct seq_file *seq, void *v)
4073 if (v == &proto_list)
4074 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4083 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4085 proto_seq_printf(seq, list_entry(v, struct proto, node));
4089 static const struct seq_operations proto_seq_ops = {
4090 .start = proto_seq_start,
4091 .next = proto_seq_next,
4092 .stop = proto_seq_stop,
4093 .show = proto_seq_show,
4096 static __net_init int proto_init_net(struct net *net)
4098 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4099 sizeof(struct seq_net_private)))
4105 static __net_exit void proto_exit_net(struct net *net)
4107 remove_proc_entry("protocols", net->proc_net);
4111 static __net_initdata struct pernet_operations proto_net_ops = {
4112 .init = proto_init_net,
4113 .exit = proto_exit_net,
4116 static int __init proto_init(void)
4118 return register_pernet_subsys(&proto_net_ops);
4121 subsys_initcall(proto_init);
4123 #endif /* PROC_FS */
4125 #ifdef CONFIG_NET_RX_BUSY_POLL
4126 bool sk_busy_loop_end(void *p, unsigned long start_time)
4128 struct sock *sk = p;
4130 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
4131 sk_busy_loop_timeout(sk, start_time);
4133 EXPORT_SYMBOL(sk_busy_loop_end);
4134 #endif /* CONFIG_NET_RX_BUSY_POLL */
4136 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4138 if (!sk->sk_prot->bind_add)
4140 return sk->sk_prot->bind_add(sk, addr, addr_len);
4142 EXPORT_SYMBOL(sock_bind_add);
4144 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4145 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4146 void __user *arg, void *karg, size_t size)
4150 if (copy_from_user(karg, arg, size))
4153 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4157 if (copy_to_user(arg, karg, size))
4162 EXPORT_SYMBOL(sock_ioctl_inout);
4164 /* This is the most common ioctl prep function, where the result (4 bytes) is
4165 * copied back to userspace if the ioctl() returns successfully. No input is
4166 * copied from userspace as input argument.
4168 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4172 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4176 return put_user(karg, (int __user *)arg);
4179 /* A wrapper around sock ioctls, which copies the data from userspace
4180 * (depending on the protocol/ioctl), and copies back the result to userspace.
4181 * The main motivation for this function is to pass kernel memory to the
4182 * protocol ioctl callbacks, instead of userspace memory.
4184 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4188 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4189 rc = ipmr_sk_ioctl(sk, cmd, arg);
4190 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4191 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4192 else if (sk_is_phonet(sk))
4193 rc = phonet_sk_ioctl(sk, cmd, arg);
4195 /* If ioctl was processed, returns its value */
4199 /* Otherwise call the default handler */
4200 return sock_ioctl_out(sk, cmd, arg);
4202 EXPORT_SYMBOL(sk_ioctl);