1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
118 #include <linux/uaccess.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
137 #include <trace/events/sock.h>
140 #include <net/busy_poll.h>
142 static DEFINE_MUTEX(proto_list_mutex);
143 static LIST_HEAD(proto_list);
145 static void sock_inuse_add(struct net *net, int val);
148 * sk_ns_capable - General socket capability test
149 * @sk: Socket to use a capability on or through
150 * @user_ns: The user namespace of the capability to use
151 * @cap: The capability to use
153 * Test to see if the opener of the socket had when the socket was
154 * created and the current process has the capability @cap in the user
155 * namespace @user_ns.
157 bool sk_ns_capable(const struct sock *sk,
158 struct user_namespace *user_ns, int cap)
160 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
161 ns_capable(user_ns, cap);
163 EXPORT_SYMBOL(sk_ns_capable);
166 * sk_capable - Socket global capability test
167 * @sk: Socket to use a capability on or through
168 * @cap: The global capability to use
170 * Test to see if the opener of the socket had when the socket was
171 * created and the current process has the capability @cap in all user
174 bool sk_capable(const struct sock *sk, int cap)
176 return sk_ns_capable(sk, &init_user_ns, cap);
178 EXPORT_SYMBOL(sk_capable);
181 * sk_net_capable - Network namespace socket capability test
182 * @sk: Socket to use a capability on or through
183 * @cap: The capability to use
185 * Test to see if the opener of the socket had when the socket was created
186 * and the current process has the capability @cap over the network namespace
187 * the socket is a member of.
189 bool sk_net_capable(const struct sock *sk, int cap)
191 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193 EXPORT_SYMBOL(sk_net_capable);
196 * Each address family might have different locking rules, so we have
197 * one slock key per address family and separate keys for internal and
200 static struct lock_class_key af_family_keys[AF_MAX];
201 static struct lock_class_key af_family_kern_keys[AF_MAX];
202 static struct lock_class_key af_family_slock_keys[AF_MAX];
203 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
211 #define _sock_locks(x) \
212 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
213 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
214 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
215 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
216 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
217 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
218 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
219 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
220 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
221 x "27" , x "28" , x "AF_CAN" , \
222 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
223 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
224 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
225 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
226 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
229 static const char *const af_family_key_strings[AF_MAX+1] = {
230 _sock_locks("sk_lock-")
232 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
233 _sock_locks("slock-")
235 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
236 _sock_locks("clock-")
239 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
240 _sock_locks("k-sk_lock-")
242 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("k-slock-")
245 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-clock-")
248 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
249 _sock_locks("rlock-")
251 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
252 _sock_locks("wlock-")
254 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
255 _sock_locks("elock-")
259 * sk_callback_lock and sk queues locking rules are per-address-family,
260 * so split the lock classes by using a per-AF key:
262 static struct lock_class_key af_callback_keys[AF_MAX];
263 static struct lock_class_key af_rlock_keys[AF_MAX];
264 static struct lock_class_key af_wlock_keys[AF_MAX];
265 static struct lock_class_key af_elock_keys[AF_MAX];
266 static struct lock_class_key af_kern_callback_keys[AF_MAX];
268 /* Run time adjustable parameters. */
269 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
270 EXPORT_SYMBOL(sysctl_wmem_max);
271 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
272 EXPORT_SYMBOL(sysctl_rmem_max);
273 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
274 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
276 /* Maximal space eaten by iovec or ancillary data plus some space */
277 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
278 EXPORT_SYMBOL(sysctl_optmem_max);
280 int sysctl_tstamp_allow_data __read_mostly = 1;
282 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
283 EXPORT_SYMBOL_GPL(memalloc_socks_key);
286 * sk_set_memalloc - sets %SOCK_MEMALLOC
287 * @sk: socket to set it on
289 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
290 * It's the responsibility of the admin to adjust min_free_kbytes
291 * to meet the requirements
293 void sk_set_memalloc(struct sock *sk)
295 sock_set_flag(sk, SOCK_MEMALLOC);
296 sk->sk_allocation |= __GFP_MEMALLOC;
297 static_branch_inc(&memalloc_socks_key);
299 EXPORT_SYMBOL_GPL(sk_set_memalloc);
301 void sk_clear_memalloc(struct sock *sk)
303 sock_reset_flag(sk, SOCK_MEMALLOC);
304 sk->sk_allocation &= ~__GFP_MEMALLOC;
305 static_branch_dec(&memalloc_socks_key);
308 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
309 * progress of swapping. SOCK_MEMALLOC may be cleared while
310 * it has rmem allocations due to the last swapfile being deactivated
311 * but there is a risk that the socket is unusable due to exceeding
312 * the rmem limits. Reclaim the reserves and obey rmem limits again.
316 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
318 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
321 unsigned int noreclaim_flag;
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
326 noreclaim_flag = memalloc_noreclaim_save();
327 ret = sk->sk_backlog_rcv(sk, skb);
328 memalloc_noreclaim_restore(noreclaim_flag);
332 EXPORT_SYMBOL(__sk_backlog_rcv);
334 void sk_error_report(struct sock *sk)
336 sk->sk_error_report(sk);
338 switch (sk->sk_family) {
342 trace_inet_sk_error_report(sk);
348 EXPORT_SYMBOL(sk_error_report);
350 static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
352 struct __kernel_sock_timeval tv;
354 if (timeo == MAX_SCHEDULE_TIMEOUT) {
358 tv.tv_sec = timeo / HZ;
359 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
362 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
363 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
364 *(struct old_timeval32 *)optval = tv32;
369 struct __kernel_old_timeval old_tv;
370 old_tv.tv_sec = tv.tv_sec;
371 old_tv.tv_usec = tv.tv_usec;
372 *(struct __kernel_old_timeval *)optval = old_tv;
373 return sizeof(old_tv);
376 *(struct __kernel_sock_timeval *)optval = tv;
380 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
383 struct __kernel_sock_timeval tv;
385 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
386 struct old_timeval32 tv32;
388 if (optlen < sizeof(tv32))
391 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
393 tv.tv_sec = tv32.tv_sec;
394 tv.tv_usec = tv32.tv_usec;
395 } else if (old_timeval) {
396 struct __kernel_old_timeval old_tv;
398 if (optlen < sizeof(old_tv))
400 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
402 tv.tv_sec = old_tv.tv_sec;
403 tv.tv_usec = old_tv.tv_usec;
405 if (optlen < sizeof(tv))
407 if (copy_from_sockptr(&tv, optval, sizeof(tv)))
410 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
414 static int warned __read_mostly;
417 if (warned < 10 && net_ratelimit()) {
419 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
420 __func__, current->comm, task_pid_nr(current));
424 *timeo_p = MAX_SCHEDULE_TIMEOUT;
425 if (tv.tv_sec == 0 && tv.tv_usec == 0)
427 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
428 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
432 static bool sock_needs_netstamp(const struct sock *sk)
434 switch (sk->sk_family) {
443 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
445 if (sk->sk_flags & flags) {
446 sk->sk_flags &= ~flags;
447 if (sock_needs_netstamp(sk) &&
448 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
449 net_disable_timestamp();
454 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
457 struct sk_buff_head *list = &sk->sk_receive_queue;
459 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
460 atomic_inc(&sk->sk_drops);
461 trace_sock_rcvqueue_full(sk, skb);
465 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
466 atomic_inc(&sk->sk_drops);
471 skb_set_owner_r(skb, sk);
473 /* we escape from rcu protected region, make sure we dont leak
478 spin_lock_irqsave(&list->lock, flags);
479 sock_skb_set_dropcount(sk, skb);
480 __skb_queue_tail(list, skb);
481 spin_unlock_irqrestore(&list->lock, flags);
483 if (!sock_flag(sk, SOCK_DEAD))
484 sk->sk_data_ready(sk);
487 EXPORT_SYMBOL(__sock_queue_rcv_skb);
489 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
493 err = sk_filter(sk, skb);
497 return __sock_queue_rcv_skb(sk, skb);
499 EXPORT_SYMBOL(sock_queue_rcv_skb);
501 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
502 const int nested, unsigned int trim_cap, bool refcounted)
504 int rc = NET_RX_SUCCESS;
506 if (sk_filter_trim_cap(sk, skb, trim_cap))
507 goto discard_and_relse;
511 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
512 atomic_inc(&sk->sk_drops);
513 goto discard_and_relse;
516 bh_lock_sock_nested(sk);
519 if (!sock_owned_by_user(sk)) {
521 * trylock + unlock semantics:
523 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
525 rc = sk_backlog_rcv(sk, skb);
527 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
528 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
530 atomic_inc(&sk->sk_drops);
531 goto discard_and_relse;
543 EXPORT_SYMBOL(__sk_receive_skb);
545 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
547 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
549 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
551 struct dst_entry *dst = __sk_dst_get(sk);
553 if (dst && dst->obsolete &&
554 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
555 dst, cookie) == NULL) {
556 sk_tx_queue_clear(sk);
557 sk->sk_dst_pending_confirm = 0;
558 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
565 EXPORT_SYMBOL(__sk_dst_check);
567 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
569 struct dst_entry *dst = sk_dst_get(sk);
571 if (dst && dst->obsolete &&
572 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
573 dst, cookie) == NULL) {
581 EXPORT_SYMBOL(sk_dst_check);
583 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
585 int ret = -ENOPROTOOPT;
586 #ifdef CONFIG_NETDEVICES
587 struct net *net = sock_net(sk);
591 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
598 sk->sk_bound_dev_if = ifindex;
599 if (sk->sk_prot->rehash)
600 sk->sk_prot->rehash(sk);
611 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
617 ret = sock_bindtoindex_locked(sk, ifindex);
623 EXPORT_SYMBOL(sock_bindtoindex);
625 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
627 int ret = -ENOPROTOOPT;
628 #ifdef CONFIG_NETDEVICES
629 struct net *net = sock_net(sk);
630 char devname[IFNAMSIZ];
637 /* Bind this socket to a particular device like "eth0",
638 * as specified in the passed interface name. If the
639 * name is "" or the option length is zero the socket
642 if (optlen > IFNAMSIZ - 1)
643 optlen = IFNAMSIZ - 1;
644 memset(devname, 0, sizeof(devname));
647 if (copy_from_sockptr(devname, optval, optlen))
651 if (devname[0] != '\0') {
652 struct net_device *dev;
655 dev = dev_get_by_name_rcu(net, devname);
657 index = dev->ifindex;
664 return sock_bindtoindex(sk, index, true);
671 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
672 int __user *optlen, int len)
674 int ret = -ENOPROTOOPT;
675 #ifdef CONFIG_NETDEVICES
676 struct net *net = sock_net(sk);
677 char devname[IFNAMSIZ];
679 if (sk->sk_bound_dev_if == 0) {
688 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
692 len = strlen(devname) + 1;
695 if (copy_to_user(optval, devname, len))
700 if (put_user(len, optlen))
711 bool sk_mc_loop(struct sock *sk)
713 if (dev_recursion_level())
717 switch (sk->sk_family) {
719 return inet_sk(sk)->mc_loop;
720 #if IS_ENABLED(CONFIG_IPV6)
722 return inet6_sk(sk)->mc_loop;
728 EXPORT_SYMBOL(sk_mc_loop);
730 void sock_set_reuseaddr(struct sock *sk)
733 sk->sk_reuse = SK_CAN_REUSE;
736 EXPORT_SYMBOL(sock_set_reuseaddr);
738 void sock_set_reuseport(struct sock *sk)
741 sk->sk_reuseport = true;
744 EXPORT_SYMBOL(sock_set_reuseport);
746 void sock_no_linger(struct sock *sk)
749 sk->sk_lingertime = 0;
750 sock_set_flag(sk, SOCK_LINGER);
753 EXPORT_SYMBOL(sock_no_linger);
755 void sock_set_priority(struct sock *sk, u32 priority)
758 sk->sk_priority = priority;
761 EXPORT_SYMBOL(sock_set_priority);
763 void sock_set_sndtimeo(struct sock *sk, s64 secs)
766 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
767 sk->sk_sndtimeo = secs * HZ;
769 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
772 EXPORT_SYMBOL(sock_set_sndtimeo);
774 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
777 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
778 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
779 sock_set_flag(sk, SOCK_RCVTSTAMP);
780 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
782 sock_reset_flag(sk, SOCK_RCVTSTAMP);
783 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
787 void sock_enable_timestamps(struct sock *sk)
790 __sock_set_timestamps(sk, true, false, true);
793 EXPORT_SYMBOL(sock_enable_timestamps);
795 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
798 case SO_TIMESTAMP_OLD:
799 __sock_set_timestamps(sk, valbool, false, false);
801 case SO_TIMESTAMP_NEW:
802 __sock_set_timestamps(sk, valbool, true, false);
804 case SO_TIMESTAMPNS_OLD:
805 __sock_set_timestamps(sk, valbool, false, true);
807 case SO_TIMESTAMPNS_NEW:
808 __sock_set_timestamps(sk, valbool, true, true);
813 int sock_set_timestamping(struct sock *sk, int optname, int val)
815 if (val & ~SOF_TIMESTAMPING_MASK)
818 if (val & SOF_TIMESTAMPING_OPT_ID &&
819 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
820 if (sk->sk_protocol == IPPROTO_TCP &&
821 sk->sk_type == SOCK_STREAM) {
822 if ((1 << sk->sk_state) &
823 (TCPF_CLOSE | TCPF_LISTEN))
825 sk->sk_tskey = tcp_sk(sk)->snd_una;
831 if (val & SOF_TIMESTAMPING_OPT_STATS &&
832 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
835 sk->sk_tsflags = val;
836 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
838 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
839 sock_enable_timestamp(sk,
840 SOCK_TIMESTAMPING_RX_SOFTWARE);
842 sock_disable_timestamp(sk,
843 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
847 void sock_set_keepalive(struct sock *sk)
850 if (sk->sk_prot->keepalive)
851 sk->sk_prot->keepalive(sk, true);
852 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
855 EXPORT_SYMBOL(sock_set_keepalive);
857 static void __sock_set_rcvbuf(struct sock *sk, int val)
859 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
860 * as a negative value.
862 val = min_t(int, val, INT_MAX / 2);
863 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
865 /* We double it on the way in to account for "struct sk_buff" etc.
866 * overhead. Applications assume that the SO_RCVBUF setting they make
867 * will allow that much actual data to be received on that socket.
869 * Applications are unaware that "struct sk_buff" and other overheads
870 * allocate from the receive buffer during socket buffer allocation.
872 * And after considering the possible alternatives, returning the value
873 * we actually used in getsockopt is the most desirable behavior.
875 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
878 void sock_set_rcvbuf(struct sock *sk, int val)
881 __sock_set_rcvbuf(sk, val);
884 EXPORT_SYMBOL(sock_set_rcvbuf);
886 static void __sock_set_mark(struct sock *sk, u32 val)
888 if (val != sk->sk_mark) {
894 void sock_set_mark(struct sock *sk, u32 val)
897 __sock_set_mark(sk, val);
900 EXPORT_SYMBOL(sock_set_mark);
903 * This is meant for all protocols to use and covers goings on
904 * at the socket level. Everything here is generic.
907 int sock_setsockopt(struct socket *sock, int level, int optname,
908 sockptr_t optval, unsigned int optlen)
910 struct sock_txtime sk_txtime;
911 struct sock *sk = sock->sk;
918 * Options without arguments
921 if (optname == SO_BINDTODEVICE)
922 return sock_setbindtodevice(sk, optval, optlen);
924 if (optlen < sizeof(int))
927 if (copy_from_sockptr(&val, optval, sizeof(val)))
930 valbool = val ? 1 : 0;
936 if (val && !capable(CAP_NET_ADMIN))
939 sock_valbool_flag(sk, SOCK_DBG, valbool);
942 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
945 sk->sk_reuseport = valbool;
954 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
958 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
961 /* Don't error on this BSD doesn't and if you think
962 * about it this is right. Otherwise apps have to
963 * play 'guess the biggest size' games. RCVBUF/SNDBUF
964 * are treated in BSD as hints
966 val = min_t(u32, val, sysctl_wmem_max);
968 /* Ensure val * 2 fits into an int, to prevent max_t()
969 * from treating it as a negative value.
971 val = min_t(int, val, INT_MAX / 2);
972 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
973 WRITE_ONCE(sk->sk_sndbuf,
974 max_t(int, val * 2, SOCK_MIN_SNDBUF));
975 /* Wake up sending tasks if we upped the value. */
976 sk->sk_write_space(sk);
980 if (!capable(CAP_NET_ADMIN)) {
985 /* No negative values (to prevent underflow, as val will be
993 /* Don't error on this BSD doesn't and if you think
994 * about it this is right. Otherwise apps have to
995 * play 'guess the biggest size' games. RCVBUF/SNDBUF
996 * are treated in BSD as hints
998 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
1001 case SO_RCVBUFFORCE:
1002 if (!capable(CAP_NET_ADMIN)) {
1007 /* No negative values (to prevent underflow, as val will be
1010 __sock_set_rcvbuf(sk, max(val, 0));
1014 if (sk->sk_prot->keepalive)
1015 sk->sk_prot->keepalive(sk, valbool);
1016 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1020 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1024 sk->sk_no_check_tx = valbool;
1028 if ((val >= 0 && val <= 6) ||
1029 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1030 sk->sk_priority = val;
1036 if (optlen < sizeof(ling)) {
1037 ret = -EINVAL; /* 1003.1g */
1040 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1045 sock_reset_flag(sk, SOCK_LINGER);
1047 #if (BITS_PER_LONG == 32)
1048 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1049 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1052 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1053 sock_set_flag(sk, SOCK_LINGER);
1062 set_bit(SOCK_PASSCRED, &sock->flags);
1064 clear_bit(SOCK_PASSCRED, &sock->flags);
1067 case SO_TIMESTAMP_OLD:
1068 case SO_TIMESTAMP_NEW:
1069 case SO_TIMESTAMPNS_OLD:
1070 case SO_TIMESTAMPNS_NEW:
1071 sock_set_timestamp(sk, valbool, optname);
1074 case SO_TIMESTAMPING_NEW:
1075 case SO_TIMESTAMPING_OLD:
1076 ret = sock_set_timestamping(sk, optname, val);
1082 if (sock->ops->set_rcvlowat)
1083 ret = sock->ops->set_rcvlowat(sk, val);
1085 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1088 case SO_RCVTIMEO_OLD:
1089 case SO_RCVTIMEO_NEW:
1090 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1091 optlen, optname == SO_RCVTIMEO_OLD);
1094 case SO_SNDTIMEO_OLD:
1095 case SO_SNDTIMEO_NEW:
1096 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1097 optlen, optname == SO_SNDTIMEO_OLD);
1100 case SO_ATTACH_FILTER: {
1101 struct sock_fprog fprog;
1103 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1105 ret = sk_attach_filter(&fprog, sk);
1110 if (optlen == sizeof(u32)) {
1114 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1117 ret = sk_attach_bpf(ufd, sk);
1121 case SO_ATTACH_REUSEPORT_CBPF: {
1122 struct sock_fprog fprog;
1124 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1126 ret = sk_reuseport_attach_filter(&fprog, sk);
1129 case SO_ATTACH_REUSEPORT_EBPF:
1131 if (optlen == sizeof(u32)) {
1135 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1138 ret = sk_reuseport_attach_bpf(ufd, sk);
1142 case SO_DETACH_REUSEPORT_BPF:
1143 ret = reuseport_detach_prog(sk);
1146 case SO_DETACH_FILTER:
1147 ret = sk_detach_filter(sk);
1150 case SO_LOCK_FILTER:
1151 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1154 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1159 set_bit(SOCK_PASSSEC, &sock->flags);
1161 clear_bit(SOCK_PASSSEC, &sock->flags);
1164 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1169 __sock_set_mark(sk, val);
1173 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1176 case SO_WIFI_STATUS:
1177 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1181 if (sock->ops->set_peek_off)
1182 ret = sock->ops->set_peek_off(sk, val);
1188 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1191 case SO_SELECT_ERR_QUEUE:
1192 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1195 #ifdef CONFIG_NET_RX_BUSY_POLL
1197 /* allow unprivileged users to decrease the value */
1198 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1204 sk->sk_ll_usec = val;
1207 case SO_PREFER_BUSY_POLL:
1208 if (valbool && !capable(CAP_NET_ADMIN))
1211 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1213 case SO_BUSY_POLL_BUDGET:
1214 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1217 if (val < 0 || val > U16_MAX)
1220 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1225 case SO_MAX_PACING_RATE:
1227 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1229 if (sizeof(ulval) != sizeof(val) &&
1230 optlen >= sizeof(ulval) &&
1231 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1236 cmpxchg(&sk->sk_pacing_status,
1239 sk->sk_max_pacing_rate = ulval;
1240 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1243 case SO_INCOMING_CPU:
1244 WRITE_ONCE(sk->sk_incoming_cpu, val);
1249 dst_negative_advice(sk);
1253 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1254 if (!((sk->sk_type == SOCK_STREAM &&
1255 sk->sk_protocol == IPPROTO_TCP) ||
1256 (sk->sk_type == SOCK_DGRAM &&
1257 sk->sk_protocol == IPPROTO_UDP)))
1259 } else if (sk->sk_family != PF_RDS) {
1263 if (val < 0 || val > 1)
1266 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1271 if (optlen != sizeof(struct sock_txtime)) {
1274 } else if (copy_from_sockptr(&sk_txtime, optval,
1275 sizeof(struct sock_txtime))) {
1278 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1282 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1283 * scheduler has enough safe guards.
1285 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1286 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1290 sock_valbool_flag(sk, SOCK_TXTIME, true);
1291 sk->sk_clockid = sk_txtime.clockid;
1292 sk->sk_txtime_deadline_mode =
1293 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1294 sk->sk_txtime_report_errors =
1295 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1298 case SO_BINDTOIFINDEX:
1299 ret = sock_bindtoindex_locked(sk, val);
1309 EXPORT_SYMBOL(sock_setsockopt);
1312 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1313 struct ucred *ucred)
1315 ucred->pid = pid_vnr(pid);
1316 ucred->uid = ucred->gid = -1;
1318 struct user_namespace *current_ns = current_user_ns();
1320 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1321 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1325 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1327 struct user_namespace *user_ns = current_user_ns();
1330 for (i = 0; i < src->ngroups; i++)
1331 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1337 int sock_getsockopt(struct socket *sock, int level, int optname,
1338 char __user *optval, int __user *optlen)
1340 struct sock *sk = sock->sk;
1345 unsigned long ulval;
1347 struct old_timeval32 tm32;
1348 struct __kernel_old_timeval tm;
1349 struct __kernel_sock_timeval stm;
1350 struct sock_txtime txtime;
1353 int lv = sizeof(int);
1356 if (get_user(len, optlen))
1361 memset(&v, 0, sizeof(v));
1365 v.val = sock_flag(sk, SOCK_DBG);
1369 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1373 v.val = sock_flag(sk, SOCK_BROADCAST);
1377 v.val = sk->sk_sndbuf;
1381 v.val = sk->sk_rcvbuf;
1385 v.val = sk->sk_reuse;
1389 v.val = sk->sk_reuseport;
1393 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1397 v.val = sk->sk_type;
1401 v.val = sk->sk_protocol;
1405 v.val = sk->sk_family;
1409 v.val = -sock_error(sk);
1411 v.val = xchg(&sk->sk_err_soft, 0);
1415 v.val = sock_flag(sk, SOCK_URGINLINE);
1419 v.val = sk->sk_no_check_tx;
1423 v.val = sk->sk_priority;
1427 lv = sizeof(v.ling);
1428 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1429 v.ling.l_linger = sk->sk_lingertime / HZ;
1435 case SO_TIMESTAMP_OLD:
1436 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1437 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1438 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1441 case SO_TIMESTAMPNS_OLD:
1442 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1445 case SO_TIMESTAMP_NEW:
1446 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1449 case SO_TIMESTAMPNS_NEW:
1450 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1453 case SO_TIMESTAMPING_OLD:
1454 v.val = sk->sk_tsflags;
1457 case SO_RCVTIMEO_OLD:
1458 case SO_RCVTIMEO_NEW:
1459 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1462 case SO_SNDTIMEO_OLD:
1463 case SO_SNDTIMEO_NEW:
1464 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1468 v.val = sk->sk_rcvlowat;
1476 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1481 struct ucred peercred;
1482 if (len > sizeof(peercred))
1483 len = sizeof(peercred);
1484 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1485 if (copy_to_user(optval, &peercred, len))
1494 if (!sk->sk_peer_cred)
1497 n = sk->sk_peer_cred->group_info->ngroups;
1498 if (len < n * sizeof(gid_t)) {
1499 len = n * sizeof(gid_t);
1500 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1502 len = n * sizeof(gid_t);
1504 ret = groups_to_user((gid_t __user *)optval,
1505 sk->sk_peer_cred->group_info);
1515 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1520 if (copy_to_user(optval, address, len))
1525 /* Dubious BSD thing... Probably nobody even uses it, but
1526 * the UNIX standard wants it for whatever reason... -DaveM
1529 v.val = sk->sk_state == TCP_LISTEN;
1533 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1537 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1540 v.val = sk->sk_mark;
1544 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1547 case SO_WIFI_STATUS:
1548 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1552 if (!sock->ops->set_peek_off)
1555 v.val = sk->sk_peek_off;
1558 v.val = sock_flag(sk, SOCK_NOFCS);
1561 case SO_BINDTODEVICE:
1562 return sock_getbindtodevice(sk, optval, optlen, len);
1565 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1571 case SO_LOCK_FILTER:
1572 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1575 case SO_BPF_EXTENSIONS:
1576 v.val = bpf_tell_extensions();
1579 case SO_SELECT_ERR_QUEUE:
1580 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1583 #ifdef CONFIG_NET_RX_BUSY_POLL
1585 v.val = sk->sk_ll_usec;
1587 case SO_PREFER_BUSY_POLL:
1588 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1592 case SO_MAX_PACING_RATE:
1593 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1594 lv = sizeof(v.ulval);
1595 v.ulval = sk->sk_max_pacing_rate;
1598 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1602 case SO_INCOMING_CPU:
1603 v.val = READ_ONCE(sk->sk_incoming_cpu);
1608 u32 meminfo[SK_MEMINFO_VARS];
1610 sk_get_meminfo(sk, meminfo);
1612 len = min_t(unsigned int, len, sizeof(meminfo));
1613 if (copy_to_user(optval, &meminfo, len))
1619 #ifdef CONFIG_NET_RX_BUSY_POLL
1620 case SO_INCOMING_NAPI_ID:
1621 v.val = READ_ONCE(sk->sk_napi_id);
1623 /* aggregate non-NAPI IDs down to 0 */
1624 if (v.val < MIN_NAPI_ID)
1634 v.val64 = sock_gen_cookie(sk);
1638 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1642 lv = sizeof(v.txtime);
1643 v.txtime.clockid = sk->sk_clockid;
1644 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1645 SOF_TXTIME_DEADLINE_MODE : 0;
1646 v.txtime.flags |= sk->sk_txtime_report_errors ?
1647 SOF_TXTIME_REPORT_ERRORS : 0;
1650 case SO_BINDTOIFINDEX:
1651 v.val = sk->sk_bound_dev_if;
1654 case SO_NETNS_COOKIE:
1658 v.val64 = sock_net(sk)->net_cookie;
1662 /* We implement the SO_SNDLOWAT etc to not be settable
1665 return -ENOPROTOOPT;
1670 if (copy_to_user(optval, &v, len))
1673 if (put_user(len, optlen))
1679 * Initialize an sk_lock.
1681 * (We also register the sk_lock with the lock validator.)
1683 static inline void sock_lock_init(struct sock *sk)
1685 if (sk->sk_kern_sock)
1686 sock_lock_init_class_and_name(
1688 af_family_kern_slock_key_strings[sk->sk_family],
1689 af_family_kern_slock_keys + sk->sk_family,
1690 af_family_kern_key_strings[sk->sk_family],
1691 af_family_kern_keys + sk->sk_family);
1693 sock_lock_init_class_and_name(
1695 af_family_slock_key_strings[sk->sk_family],
1696 af_family_slock_keys + sk->sk_family,
1697 af_family_key_strings[sk->sk_family],
1698 af_family_keys + sk->sk_family);
1702 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1703 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1704 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1706 static void sock_copy(struct sock *nsk, const struct sock *osk)
1708 const struct proto *prot = READ_ONCE(osk->sk_prot);
1709 #ifdef CONFIG_SECURITY_NETWORK
1710 void *sptr = nsk->sk_security;
1713 /* If we move sk_tx_queue_mapping out of the private section,
1714 * we must check if sk_tx_queue_clear() is called after
1715 * sock_copy() in sk_clone_lock().
1717 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1718 offsetof(struct sock, sk_dontcopy_begin) ||
1719 offsetof(struct sock, sk_tx_queue_mapping) >=
1720 offsetof(struct sock, sk_dontcopy_end));
1722 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1724 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1725 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1727 #ifdef CONFIG_SECURITY_NETWORK
1728 nsk->sk_security = sptr;
1729 security_sk_clone(osk, nsk);
1733 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1737 struct kmem_cache *slab;
1741 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1744 if (want_init_on_alloc(priority))
1745 sk_prot_clear_nulls(sk, prot->obj_size);
1747 sk = kmalloc(prot->obj_size, priority);
1750 if (security_sk_alloc(sk, family, priority))
1753 if (!try_module_get(prot->owner))
1760 security_sk_free(sk);
1763 kmem_cache_free(slab, sk);
1769 static void sk_prot_free(struct proto *prot, struct sock *sk)
1771 struct kmem_cache *slab;
1772 struct module *owner;
1774 owner = prot->owner;
1777 cgroup_sk_free(&sk->sk_cgrp_data);
1778 mem_cgroup_sk_free(sk);
1779 security_sk_free(sk);
1781 kmem_cache_free(slab, sk);
1788 * sk_alloc - All socket objects are allocated here
1789 * @net: the applicable net namespace
1790 * @family: protocol family
1791 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1792 * @prot: struct proto associated with this new sock instance
1793 * @kern: is this to be a kernel socket?
1795 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1796 struct proto *prot, int kern)
1800 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1802 sk->sk_family = family;
1804 * See comment in struct sock definition to understand
1805 * why we need sk_prot_creator -acme
1807 sk->sk_prot = sk->sk_prot_creator = prot;
1808 sk->sk_kern_sock = kern;
1810 sk->sk_net_refcnt = kern ? 0 : 1;
1811 if (likely(sk->sk_net_refcnt)) {
1813 sock_inuse_add(net, 1);
1816 sock_net_set(sk, net);
1817 refcount_set(&sk->sk_wmem_alloc, 1);
1819 mem_cgroup_sk_alloc(sk);
1820 cgroup_sk_alloc(&sk->sk_cgrp_data);
1821 sock_update_classid(&sk->sk_cgrp_data);
1822 sock_update_netprioidx(&sk->sk_cgrp_data);
1823 sk_tx_queue_clear(sk);
1828 EXPORT_SYMBOL(sk_alloc);
1830 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1831 * grace period. This is the case for UDP sockets and TCP listeners.
1833 static void __sk_destruct(struct rcu_head *head)
1835 struct sock *sk = container_of(head, struct sock, sk_rcu);
1836 struct sk_filter *filter;
1838 if (sk->sk_destruct)
1839 sk->sk_destruct(sk);
1841 filter = rcu_dereference_check(sk->sk_filter,
1842 refcount_read(&sk->sk_wmem_alloc) == 0);
1844 sk_filter_uncharge(sk, filter);
1845 RCU_INIT_POINTER(sk->sk_filter, NULL);
1848 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1850 #ifdef CONFIG_BPF_SYSCALL
1851 bpf_sk_storage_free(sk);
1854 if (atomic_read(&sk->sk_omem_alloc))
1855 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1856 __func__, atomic_read(&sk->sk_omem_alloc));
1858 if (sk->sk_frag.page) {
1859 put_page(sk->sk_frag.page);
1860 sk->sk_frag.page = NULL;
1863 if (sk->sk_peer_cred)
1864 put_cred(sk->sk_peer_cred);
1865 put_pid(sk->sk_peer_pid);
1866 if (likely(sk->sk_net_refcnt))
1867 put_net(sock_net(sk));
1868 sk_prot_free(sk->sk_prot_creator, sk);
1871 void sk_destruct(struct sock *sk)
1873 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1875 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1876 reuseport_detach_sock(sk);
1877 use_call_rcu = true;
1881 call_rcu(&sk->sk_rcu, __sk_destruct);
1883 __sk_destruct(&sk->sk_rcu);
1886 static void __sk_free(struct sock *sk)
1888 if (likely(sk->sk_net_refcnt))
1889 sock_inuse_add(sock_net(sk), -1);
1891 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1892 sock_diag_broadcast_destroy(sk);
1897 void sk_free(struct sock *sk)
1900 * We subtract one from sk_wmem_alloc and can know if
1901 * some packets are still in some tx queue.
1902 * If not null, sock_wfree() will call __sk_free(sk) later
1904 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1907 EXPORT_SYMBOL(sk_free);
1909 static void sk_init_common(struct sock *sk)
1911 skb_queue_head_init(&sk->sk_receive_queue);
1912 skb_queue_head_init(&sk->sk_write_queue);
1913 skb_queue_head_init(&sk->sk_error_queue);
1915 rwlock_init(&sk->sk_callback_lock);
1916 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1917 af_rlock_keys + sk->sk_family,
1918 af_family_rlock_key_strings[sk->sk_family]);
1919 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1920 af_wlock_keys + sk->sk_family,
1921 af_family_wlock_key_strings[sk->sk_family]);
1922 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1923 af_elock_keys + sk->sk_family,
1924 af_family_elock_key_strings[sk->sk_family]);
1925 lockdep_set_class_and_name(&sk->sk_callback_lock,
1926 af_callback_keys + sk->sk_family,
1927 af_family_clock_key_strings[sk->sk_family]);
1931 * sk_clone_lock - clone a socket, and lock its clone
1932 * @sk: the socket to clone
1933 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1935 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1937 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1939 struct proto *prot = READ_ONCE(sk->sk_prot);
1940 struct sk_filter *filter;
1941 bool is_charged = true;
1944 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
1948 sock_copy(newsk, sk);
1950 newsk->sk_prot_creator = prot;
1953 if (likely(newsk->sk_net_refcnt))
1954 get_net(sock_net(newsk));
1955 sk_node_init(&newsk->sk_node);
1956 sock_lock_init(newsk);
1957 bh_lock_sock(newsk);
1958 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1959 newsk->sk_backlog.len = 0;
1961 atomic_set(&newsk->sk_rmem_alloc, 0);
1963 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
1964 refcount_set(&newsk->sk_wmem_alloc, 1);
1966 atomic_set(&newsk->sk_omem_alloc, 0);
1967 sk_init_common(newsk);
1969 newsk->sk_dst_cache = NULL;
1970 newsk->sk_dst_pending_confirm = 0;
1971 newsk->sk_wmem_queued = 0;
1972 newsk->sk_forward_alloc = 0;
1973 atomic_set(&newsk->sk_drops, 0);
1974 newsk->sk_send_head = NULL;
1975 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1976 atomic_set(&newsk->sk_zckey, 0);
1978 sock_reset_flag(newsk, SOCK_DONE);
1980 /* sk->sk_memcg will be populated at accept() time */
1981 newsk->sk_memcg = NULL;
1983 cgroup_sk_clone(&newsk->sk_cgrp_data);
1986 filter = rcu_dereference(sk->sk_filter);
1988 /* though it's an empty new sock, the charging may fail
1989 * if sysctl_optmem_max was changed between creation of
1990 * original socket and cloning
1992 is_charged = sk_filter_charge(newsk, filter);
1993 RCU_INIT_POINTER(newsk->sk_filter, filter);
1996 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1997 /* We need to make sure that we don't uncharge the new
1998 * socket if we couldn't charge it in the first place
1999 * as otherwise we uncharge the parent's filter.
2002 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2003 sk_free_unlock_clone(newsk);
2007 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2009 if (bpf_sk_storage_clone(sk, newsk)) {
2010 sk_free_unlock_clone(newsk);
2015 /* Clear sk_user_data if parent had the pointer tagged
2016 * as not suitable for copying when cloning.
2018 if (sk_user_data_is_nocopy(newsk))
2019 newsk->sk_user_data = NULL;
2022 newsk->sk_err_soft = 0;
2023 newsk->sk_priority = 0;
2024 newsk->sk_incoming_cpu = raw_smp_processor_id();
2025 if (likely(newsk->sk_net_refcnt))
2026 sock_inuse_add(sock_net(newsk), 1);
2028 /* Before updating sk_refcnt, we must commit prior changes to memory
2029 * (Documentation/RCU/rculist_nulls.rst for details)
2032 refcount_set(&newsk->sk_refcnt, 2);
2034 /* Increment the counter in the same struct proto as the master
2035 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2036 * is the same as sk->sk_prot->socks, as this field was copied
2039 * This _changes_ the previous behaviour, where
2040 * tcp_create_openreq_child always was incrementing the
2041 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2042 * to be taken into account in all callers. -acme
2044 sk_refcnt_debug_inc(newsk);
2045 sk_set_socket(newsk, NULL);
2046 sk_tx_queue_clear(newsk);
2047 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2049 if (newsk->sk_prot->sockets_allocated)
2050 sk_sockets_allocated_inc(newsk);
2052 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2053 net_enable_timestamp();
2057 EXPORT_SYMBOL_GPL(sk_clone_lock);
2059 void sk_free_unlock_clone(struct sock *sk)
2061 /* It is still raw copy of parent, so invalidate
2062 * destructor and make plain sk_free() */
2063 sk->sk_destruct = NULL;
2067 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2069 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2073 sk_dst_set(sk, dst);
2074 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2075 if (sk->sk_route_caps & NETIF_F_GSO)
2076 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2077 sk->sk_route_caps &= ~sk->sk_route_nocaps;
2078 if (sk_can_gso(sk)) {
2079 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2080 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2082 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2083 sk->sk_gso_max_size = dst->dev->gso_max_size;
2084 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2087 sk->sk_gso_max_segs = max_segs;
2089 EXPORT_SYMBOL_GPL(sk_setup_caps);
2092 * Simple resource managers for sockets.
2097 * Write buffer destructor automatically called from kfree_skb.
2099 void sock_wfree(struct sk_buff *skb)
2101 struct sock *sk = skb->sk;
2102 unsigned int len = skb->truesize;
2104 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2106 * Keep a reference on sk_wmem_alloc, this will be released
2107 * after sk_write_space() call
2109 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2110 sk->sk_write_space(sk);
2114 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2115 * could not do because of in-flight packets
2117 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2120 EXPORT_SYMBOL(sock_wfree);
2122 /* This variant of sock_wfree() is used by TCP,
2123 * since it sets SOCK_USE_WRITE_QUEUE.
2125 void __sock_wfree(struct sk_buff *skb)
2127 struct sock *sk = skb->sk;
2129 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2133 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2138 if (unlikely(!sk_fullsock(sk))) {
2139 skb->destructor = sock_edemux;
2144 skb->destructor = sock_wfree;
2145 skb_set_hash_from_sk(skb, sk);
2147 * We used to take a refcount on sk, but following operation
2148 * is enough to guarantee sk_free() wont free this sock until
2149 * all in-flight packets are completed
2151 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2153 EXPORT_SYMBOL(skb_set_owner_w);
2155 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2157 #ifdef CONFIG_TLS_DEVICE
2158 /* Drivers depend on in-order delivery for crypto offload,
2159 * partial orphan breaks out-of-order-OK logic.
2164 return (skb->destructor == sock_wfree ||
2165 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2168 /* This helper is used by netem, as it can hold packets in its
2169 * delay queue. We want to allow the owner socket to send more
2170 * packets, as if they were already TX completed by a typical driver.
2171 * But we also want to keep skb->sk set because some packet schedulers
2172 * rely on it (sch_fq for example).
2174 void skb_orphan_partial(struct sk_buff *skb)
2176 if (skb_is_tcp_pure_ack(skb))
2179 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2184 EXPORT_SYMBOL(skb_orphan_partial);
2187 * Read buffer destructor automatically called from kfree_skb.
2189 void sock_rfree(struct sk_buff *skb)
2191 struct sock *sk = skb->sk;
2192 unsigned int len = skb->truesize;
2194 atomic_sub(len, &sk->sk_rmem_alloc);
2195 sk_mem_uncharge(sk, len);
2197 EXPORT_SYMBOL(sock_rfree);
2200 * Buffer destructor for skbs that are not used directly in read or write
2201 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2203 void sock_efree(struct sk_buff *skb)
2207 EXPORT_SYMBOL(sock_efree);
2209 /* Buffer destructor for prefetch/receive path where reference count may
2210 * not be held, e.g. for listen sockets.
2213 void sock_pfree(struct sk_buff *skb)
2215 if (sk_is_refcounted(skb->sk))
2216 sock_gen_put(skb->sk);
2218 EXPORT_SYMBOL(sock_pfree);
2219 #endif /* CONFIG_INET */
2221 kuid_t sock_i_uid(struct sock *sk)
2225 read_lock_bh(&sk->sk_callback_lock);
2226 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2227 read_unlock_bh(&sk->sk_callback_lock);
2230 EXPORT_SYMBOL(sock_i_uid);
2232 unsigned long sock_i_ino(struct sock *sk)
2236 read_lock_bh(&sk->sk_callback_lock);
2237 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2238 read_unlock_bh(&sk->sk_callback_lock);
2241 EXPORT_SYMBOL(sock_i_ino);
2244 * Allocate a skb from the socket's send buffer.
2246 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2250 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2251 struct sk_buff *skb = alloc_skb(size, priority);
2254 skb_set_owner_w(skb, sk);
2260 EXPORT_SYMBOL(sock_wmalloc);
2262 static void sock_ofree(struct sk_buff *skb)
2264 struct sock *sk = skb->sk;
2266 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2269 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2272 struct sk_buff *skb;
2274 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2275 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2279 skb = alloc_skb(size, priority);
2283 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2285 skb->destructor = sock_ofree;
2290 * Allocate a memory block from the socket's option memory buffer.
2292 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2294 if ((unsigned int)size <= sysctl_optmem_max &&
2295 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2297 /* First do the add, to avoid the race if kmalloc
2300 atomic_add(size, &sk->sk_omem_alloc);
2301 mem = kmalloc(size, priority);
2304 atomic_sub(size, &sk->sk_omem_alloc);
2308 EXPORT_SYMBOL(sock_kmalloc);
2310 /* Free an option memory block. Note, we actually want the inline
2311 * here as this allows gcc to detect the nullify and fold away the
2312 * condition entirely.
2314 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2317 if (WARN_ON_ONCE(!mem))
2320 kfree_sensitive(mem);
2323 atomic_sub(size, &sk->sk_omem_alloc);
2326 void sock_kfree_s(struct sock *sk, void *mem, int size)
2328 __sock_kfree_s(sk, mem, size, false);
2330 EXPORT_SYMBOL(sock_kfree_s);
2332 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2334 __sock_kfree_s(sk, mem, size, true);
2336 EXPORT_SYMBOL(sock_kzfree_s);
2338 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2339 I think, these locks should be removed for datagram sockets.
2341 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2345 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2349 if (signal_pending(current))
2351 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2352 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2353 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2355 if (sk->sk_shutdown & SEND_SHUTDOWN)
2359 timeo = schedule_timeout(timeo);
2361 finish_wait(sk_sleep(sk), &wait);
2367 * Generic send/receive buffer handlers
2370 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2371 unsigned long data_len, int noblock,
2372 int *errcode, int max_page_order)
2374 struct sk_buff *skb;
2378 timeo = sock_sndtimeo(sk, noblock);
2380 err = sock_error(sk);
2385 if (sk->sk_shutdown & SEND_SHUTDOWN)
2388 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2391 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2392 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2396 if (signal_pending(current))
2398 timeo = sock_wait_for_wmem(sk, timeo);
2400 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2401 errcode, sk->sk_allocation);
2403 skb_set_owner_w(skb, sk);
2407 err = sock_intr_errno(timeo);
2412 EXPORT_SYMBOL(sock_alloc_send_pskb);
2414 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2415 int noblock, int *errcode)
2417 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2419 EXPORT_SYMBOL(sock_alloc_send_skb);
2421 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2422 struct sockcm_cookie *sockc)
2426 switch (cmsg->cmsg_type) {
2428 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2430 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2432 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2434 case SO_TIMESTAMPING_OLD:
2435 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2438 tsflags = *(u32 *)CMSG_DATA(cmsg);
2439 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2442 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2443 sockc->tsflags |= tsflags;
2446 if (!sock_flag(sk, SOCK_TXTIME))
2448 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2450 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2452 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2454 case SCM_CREDENTIALS:
2461 EXPORT_SYMBOL(__sock_cmsg_send);
2463 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2464 struct sockcm_cookie *sockc)
2466 struct cmsghdr *cmsg;
2469 for_each_cmsghdr(cmsg, msg) {
2470 if (!CMSG_OK(msg, cmsg))
2472 if (cmsg->cmsg_level != SOL_SOCKET)
2474 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2480 EXPORT_SYMBOL(sock_cmsg_send);
2482 static void sk_enter_memory_pressure(struct sock *sk)
2484 if (!sk->sk_prot->enter_memory_pressure)
2487 sk->sk_prot->enter_memory_pressure(sk);
2490 static void sk_leave_memory_pressure(struct sock *sk)
2492 if (sk->sk_prot->leave_memory_pressure) {
2493 sk->sk_prot->leave_memory_pressure(sk);
2495 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2497 if (memory_pressure && READ_ONCE(*memory_pressure))
2498 WRITE_ONCE(*memory_pressure, 0);
2502 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2503 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2506 * skb_page_frag_refill - check that a page_frag contains enough room
2507 * @sz: minimum size of the fragment we want to get
2508 * @pfrag: pointer to page_frag
2509 * @gfp: priority for memory allocation
2511 * Note: While this allocator tries to use high order pages, there is
2512 * no guarantee that allocations succeed. Therefore, @sz MUST be
2513 * less or equal than PAGE_SIZE.
2515 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2518 if (page_ref_count(pfrag->page) == 1) {
2522 if (pfrag->offset + sz <= pfrag->size)
2524 put_page(pfrag->page);
2528 if (SKB_FRAG_PAGE_ORDER &&
2529 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2530 /* Avoid direct reclaim but allow kswapd to wake */
2531 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2532 __GFP_COMP | __GFP_NOWARN |
2534 SKB_FRAG_PAGE_ORDER);
2535 if (likely(pfrag->page)) {
2536 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2540 pfrag->page = alloc_page(gfp);
2541 if (likely(pfrag->page)) {
2542 pfrag->size = PAGE_SIZE;
2547 EXPORT_SYMBOL(skb_page_frag_refill);
2549 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2551 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2554 sk_enter_memory_pressure(sk);
2555 sk_stream_moderate_sndbuf(sk);
2558 EXPORT_SYMBOL(sk_page_frag_refill);
2560 void __lock_sock(struct sock *sk)
2561 __releases(&sk->sk_lock.slock)
2562 __acquires(&sk->sk_lock.slock)
2567 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2568 TASK_UNINTERRUPTIBLE);
2569 spin_unlock_bh(&sk->sk_lock.slock);
2571 spin_lock_bh(&sk->sk_lock.slock);
2572 if (!sock_owned_by_user(sk))
2575 finish_wait(&sk->sk_lock.wq, &wait);
2578 void __release_sock(struct sock *sk)
2579 __releases(&sk->sk_lock.slock)
2580 __acquires(&sk->sk_lock.slock)
2582 struct sk_buff *skb, *next;
2584 while ((skb = sk->sk_backlog.head) != NULL) {
2585 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2587 spin_unlock_bh(&sk->sk_lock.slock);
2592 WARN_ON_ONCE(skb_dst_is_noref(skb));
2593 skb_mark_not_on_list(skb);
2594 sk_backlog_rcv(sk, skb);
2599 } while (skb != NULL);
2601 spin_lock_bh(&sk->sk_lock.slock);
2605 * Doing the zeroing here guarantee we can not loop forever
2606 * while a wild producer attempts to flood us.
2608 sk->sk_backlog.len = 0;
2611 void __sk_flush_backlog(struct sock *sk)
2613 spin_lock_bh(&sk->sk_lock.slock);
2615 spin_unlock_bh(&sk->sk_lock.slock);
2619 * sk_wait_data - wait for data to arrive at sk_receive_queue
2620 * @sk: sock to wait on
2621 * @timeo: for how long
2622 * @skb: last skb seen on sk_receive_queue
2624 * Now socket state including sk->sk_err is changed only under lock,
2625 * hence we may omit checks after joining wait queue.
2626 * We check receive queue before schedule() only as optimization;
2627 * it is very likely that release_sock() added new data.
2629 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2631 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2634 add_wait_queue(sk_sleep(sk), &wait);
2635 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2636 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2637 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2638 remove_wait_queue(sk_sleep(sk), &wait);
2641 EXPORT_SYMBOL(sk_wait_data);
2644 * __sk_mem_raise_allocated - increase memory_allocated
2646 * @size: memory size to allocate
2647 * @amt: pages to allocate
2648 * @kind: allocation type
2650 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2652 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2654 struct proto *prot = sk->sk_prot;
2655 long allocated = sk_memory_allocated_add(sk, amt);
2656 bool charged = true;
2658 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2659 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2660 goto suppress_allocation;
2663 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2664 sk_leave_memory_pressure(sk);
2668 /* Under pressure. */
2669 if (allocated > sk_prot_mem_limits(sk, 1))
2670 sk_enter_memory_pressure(sk);
2672 /* Over hard limit. */
2673 if (allocated > sk_prot_mem_limits(sk, 2))
2674 goto suppress_allocation;
2676 /* guarantee minimum buffer size under pressure */
2677 if (kind == SK_MEM_RECV) {
2678 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2681 } else { /* SK_MEM_SEND */
2682 int wmem0 = sk_get_wmem0(sk, prot);
2684 if (sk->sk_type == SOCK_STREAM) {
2685 if (sk->sk_wmem_queued < wmem0)
2687 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2692 if (sk_has_memory_pressure(sk)) {
2695 if (!sk_under_memory_pressure(sk))
2697 alloc = sk_sockets_allocated_read_positive(sk);
2698 if (sk_prot_mem_limits(sk, 2) > alloc *
2699 sk_mem_pages(sk->sk_wmem_queued +
2700 atomic_read(&sk->sk_rmem_alloc) +
2701 sk->sk_forward_alloc))
2705 suppress_allocation:
2707 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2708 sk_stream_moderate_sndbuf(sk);
2710 /* Fail only if socket is _under_ its sndbuf.
2711 * In this case we cannot block, so that we have to fail.
2713 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2717 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2718 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2720 sk_memory_allocated_sub(sk, amt);
2722 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2723 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2727 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2730 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2732 * @size: memory size to allocate
2733 * @kind: allocation type
2735 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2736 * rmem allocation. This function assumes that protocols which have
2737 * memory_pressure use sk_wmem_queued as write buffer accounting.
2739 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2741 int ret, amt = sk_mem_pages(size);
2743 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2744 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2746 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2749 EXPORT_SYMBOL(__sk_mem_schedule);
2752 * __sk_mem_reduce_allocated - reclaim memory_allocated
2754 * @amount: number of quanta
2756 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2758 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2760 sk_memory_allocated_sub(sk, amount);
2762 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2763 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2765 if (sk_under_memory_pressure(sk) &&
2766 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2767 sk_leave_memory_pressure(sk);
2769 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2772 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2774 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2776 void __sk_mem_reclaim(struct sock *sk, int amount)
2778 amount >>= SK_MEM_QUANTUM_SHIFT;
2779 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2780 __sk_mem_reduce_allocated(sk, amount);
2782 EXPORT_SYMBOL(__sk_mem_reclaim);
2784 int sk_set_peek_off(struct sock *sk, int val)
2786 sk->sk_peek_off = val;
2789 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2792 * Set of default routines for initialising struct proto_ops when
2793 * the protocol does not support a particular function. In certain
2794 * cases where it makes no sense for a protocol to have a "do nothing"
2795 * function, some default processing is provided.
2798 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2802 EXPORT_SYMBOL(sock_no_bind);
2804 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2809 EXPORT_SYMBOL(sock_no_connect);
2811 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2815 EXPORT_SYMBOL(sock_no_socketpair);
2817 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2822 EXPORT_SYMBOL(sock_no_accept);
2824 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2829 EXPORT_SYMBOL(sock_no_getname);
2831 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2835 EXPORT_SYMBOL(sock_no_ioctl);
2837 int sock_no_listen(struct socket *sock, int backlog)
2841 EXPORT_SYMBOL(sock_no_listen);
2843 int sock_no_shutdown(struct socket *sock, int how)
2847 EXPORT_SYMBOL(sock_no_shutdown);
2849 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2853 EXPORT_SYMBOL(sock_no_sendmsg);
2855 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2859 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2861 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2866 EXPORT_SYMBOL(sock_no_recvmsg);
2868 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2870 /* Mirror missing mmap method error code */
2873 EXPORT_SYMBOL(sock_no_mmap);
2876 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2877 * various sock-based usage counts.
2879 void __receive_sock(struct file *file)
2881 struct socket *sock;
2883 sock = sock_from_file(file);
2885 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2886 sock_update_classid(&sock->sk->sk_cgrp_data);
2890 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2893 struct msghdr msg = {.msg_flags = flags};
2895 char *kaddr = kmap(page);
2896 iov.iov_base = kaddr + offset;
2898 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2902 EXPORT_SYMBOL(sock_no_sendpage);
2904 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2905 int offset, size_t size, int flags)
2908 struct msghdr msg = {.msg_flags = flags};
2910 char *kaddr = kmap(page);
2912 iov.iov_base = kaddr + offset;
2914 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2918 EXPORT_SYMBOL(sock_no_sendpage_locked);
2921 * Default Socket Callbacks
2924 static void sock_def_wakeup(struct sock *sk)
2926 struct socket_wq *wq;
2929 wq = rcu_dereference(sk->sk_wq);
2930 if (skwq_has_sleeper(wq))
2931 wake_up_interruptible_all(&wq->wait);
2935 static void sock_def_error_report(struct sock *sk)
2937 struct socket_wq *wq;
2940 wq = rcu_dereference(sk->sk_wq);
2941 if (skwq_has_sleeper(wq))
2942 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2943 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2947 void sock_def_readable(struct sock *sk)
2949 struct socket_wq *wq;
2952 wq = rcu_dereference(sk->sk_wq);
2953 if (skwq_has_sleeper(wq))
2954 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2955 EPOLLRDNORM | EPOLLRDBAND);
2956 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2960 static void sock_def_write_space(struct sock *sk)
2962 struct socket_wq *wq;
2966 /* Do not wake up a writer until he can make "significant"
2969 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2970 wq = rcu_dereference(sk->sk_wq);
2971 if (skwq_has_sleeper(wq))
2972 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2973 EPOLLWRNORM | EPOLLWRBAND);
2975 /* Should agree with poll, otherwise some programs break */
2976 if (sock_writeable(sk))
2977 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2983 static void sock_def_destruct(struct sock *sk)
2987 void sk_send_sigurg(struct sock *sk)
2989 if (sk->sk_socket && sk->sk_socket->file)
2990 if (send_sigurg(&sk->sk_socket->file->f_owner))
2991 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2993 EXPORT_SYMBOL(sk_send_sigurg);
2995 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2996 unsigned long expires)
2998 if (!mod_timer(timer, expires))
3001 EXPORT_SYMBOL(sk_reset_timer);
3003 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3005 if (del_timer(timer))
3008 EXPORT_SYMBOL(sk_stop_timer);
3010 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3012 if (del_timer_sync(timer))
3015 EXPORT_SYMBOL(sk_stop_timer_sync);
3017 void sock_init_data(struct socket *sock, struct sock *sk)
3020 sk->sk_send_head = NULL;
3022 timer_setup(&sk->sk_timer, NULL, 0);
3024 sk->sk_allocation = GFP_KERNEL;
3025 sk->sk_rcvbuf = sysctl_rmem_default;
3026 sk->sk_sndbuf = sysctl_wmem_default;
3027 sk->sk_state = TCP_CLOSE;
3028 sk_set_socket(sk, sock);
3030 sock_set_flag(sk, SOCK_ZAPPED);
3033 sk->sk_type = sock->type;
3034 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3036 sk->sk_uid = SOCK_INODE(sock)->i_uid;
3038 RCU_INIT_POINTER(sk->sk_wq, NULL);
3039 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3042 rwlock_init(&sk->sk_callback_lock);
3043 if (sk->sk_kern_sock)
3044 lockdep_set_class_and_name(
3045 &sk->sk_callback_lock,
3046 af_kern_callback_keys + sk->sk_family,
3047 af_family_kern_clock_key_strings[sk->sk_family]);
3049 lockdep_set_class_and_name(
3050 &sk->sk_callback_lock,
3051 af_callback_keys + sk->sk_family,
3052 af_family_clock_key_strings[sk->sk_family]);
3054 sk->sk_state_change = sock_def_wakeup;
3055 sk->sk_data_ready = sock_def_readable;
3056 sk->sk_write_space = sock_def_write_space;
3057 sk->sk_error_report = sock_def_error_report;
3058 sk->sk_destruct = sock_def_destruct;
3060 sk->sk_frag.page = NULL;
3061 sk->sk_frag.offset = 0;
3062 sk->sk_peek_off = -1;
3064 sk->sk_peer_pid = NULL;
3065 sk->sk_peer_cred = NULL;
3066 sk->sk_write_pending = 0;
3067 sk->sk_rcvlowat = 1;
3068 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3069 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3071 sk->sk_stamp = SK_DEFAULT_STAMP;
3072 #if BITS_PER_LONG==32
3073 seqlock_init(&sk->sk_stamp_seq);
3075 atomic_set(&sk->sk_zckey, 0);
3077 #ifdef CONFIG_NET_RX_BUSY_POLL
3079 sk->sk_ll_usec = sysctl_net_busy_read;
3082 sk->sk_max_pacing_rate = ~0UL;
3083 sk->sk_pacing_rate = ~0UL;
3084 WRITE_ONCE(sk->sk_pacing_shift, 10);
3085 sk->sk_incoming_cpu = -1;
3087 sk_rx_queue_clear(sk);
3089 * Before updating sk_refcnt, we must commit prior changes to memory
3090 * (Documentation/RCU/rculist_nulls.rst for details)
3093 refcount_set(&sk->sk_refcnt, 1);
3094 atomic_set(&sk->sk_drops, 0);
3096 EXPORT_SYMBOL(sock_init_data);
3098 void lock_sock_nested(struct sock *sk, int subclass)
3101 spin_lock_bh(&sk->sk_lock.slock);
3102 if (sk->sk_lock.owned)
3104 sk->sk_lock.owned = 1;
3105 spin_unlock(&sk->sk_lock.slock);
3107 * The sk_lock has mutex_lock() semantics here:
3109 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3112 EXPORT_SYMBOL(lock_sock_nested);
3114 void release_sock(struct sock *sk)
3116 spin_lock_bh(&sk->sk_lock.slock);
3117 if (sk->sk_backlog.tail)
3120 /* Warning : release_cb() might need to release sk ownership,
3121 * ie call sock_release_ownership(sk) before us.
3123 if (sk->sk_prot->release_cb)
3124 sk->sk_prot->release_cb(sk);
3126 sock_release_ownership(sk);
3127 if (waitqueue_active(&sk->sk_lock.wq))
3128 wake_up(&sk->sk_lock.wq);
3129 spin_unlock_bh(&sk->sk_lock.slock);
3131 EXPORT_SYMBOL(release_sock);
3134 * lock_sock_fast - fast version of lock_sock
3137 * This version should be used for very small section, where process wont block
3138 * return false if fast path is taken:
3140 * sk_lock.slock locked, owned = 0, BH disabled
3142 * return true if slow path is taken:
3144 * sk_lock.slock unlocked, owned = 1, BH enabled
3146 bool lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3149 spin_lock_bh(&sk->sk_lock.slock);
3151 if (!sk->sk_lock.owned)
3153 * Note : We must disable BH
3158 sk->sk_lock.owned = 1;
3159 spin_unlock(&sk->sk_lock.slock);
3161 * The sk_lock has mutex_lock() semantics here:
3163 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3164 __acquire(&sk->sk_lock.slock);
3168 EXPORT_SYMBOL(lock_sock_fast);
3170 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3171 bool timeval, bool time32)
3173 struct sock *sk = sock->sk;
3174 struct timespec64 ts;
3176 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3177 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3178 if (ts.tv_sec == -1)
3180 if (ts.tv_sec == 0) {
3181 ktime_t kt = ktime_get_real();
3182 sock_write_timestamp(sk, kt);
3183 ts = ktime_to_timespec64(kt);
3189 #ifdef CONFIG_COMPAT_32BIT_TIME
3191 return put_old_timespec32(&ts, userstamp);
3193 #ifdef CONFIG_SPARC64
3194 /* beware of padding in sparc64 timeval */
3195 if (timeval && !in_compat_syscall()) {
3196 struct __kernel_old_timeval __user tv = {
3197 .tv_sec = ts.tv_sec,
3198 .tv_usec = ts.tv_nsec,
3200 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3205 return put_timespec64(&ts, userstamp);
3207 EXPORT_SYMBOL(sock_gettstamp);
3209 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3211 if (!sock_flag(sk, flag)) {
3212 unsigned long previous_flags = sk->sk_flags;
3214 sock_set_flag(sk, flag);
3216 * we just set one of the two flags which require net
3217 * time stamping, but time stamping might have been on
3218 * already because of the other one
3220 if (sock_needs_netstamp(sk) &&
3221 !(previous_flags & SK_FLAGS_TIMESTAMP))
3222 net_enable_timestamp();
3226 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3227 int level, int type)
3229 struct sock_exterr_skb *serr;
3230 struct sk_buff *skb;
3234 skb = sock_dequeue_err_skb(sk);
3240 msg->msg_flags |= MSG_TRUNC;
3243 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3247 sock_recv_timestamp(msg, sk, skb);
3249 serr = SKB_EXT_ERR(skb);
3250 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3252 msg->msg_flags |= MSG_ERRQUEUE;
3260 EXPORT_SYMBOL(sock_recv_errqueue);
3263 * Get a socket option on an socket.
3265 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3266 * asynchronous errors should be reported by getsockopt. We assume
3267 * this means if you specify SO_ERROR (otherwise whats the point of it).
3269 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3270 char __user *optval, int __user *optlen)
3272 struct sock *sk = sock->sk;
3274 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3276 EXPORT_SYMBOL(sock_common_getsockopt);
3278 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3281 struct sock *sk = sock->sk;
3285 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3286 flags & ~MSG_DONTWAIT, &addr_len);
3288 msg->msg_namelen = addr_len;
3291 EXPORT_SYMBOL(sock_common_recvmsg);
3294 * Set socket options on an inet socket.
3296 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3297 sockptr_t optval, unsigned int optlen)
3299 struct sock *sk = sock->sk;
3301 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3303 EXPORT_SYMBOL(sock_common_setsockopt);
3305 void sk_common_release(struct sock *sk)
3307 if (sk->sk_prot->destroy)
3308 sk->sk_prot->destroy(sk);
3311 * Observation: when sk_common_release is called, processes have
3312 * no access to socket. But net still has.
3313 * Step one, detach it from networking:
3315 * A. Remove from hash tables.
3318 sk->sk_prot->unhash(sk);
3321 * In this point socket cannot receive new packets, but it is possible
3322 * that some packets are in flight because some CPU runs receiver and
3323 * did hash table lookup before we unhashed socket. They will achieve
3324 * receive queue and will be purged by socket destructor.
3326 * Also we still have packets pending on receive queue and probably,
3327 * our own packets waiting in device queues. sock_destroy will drain
3328 * receive queue, but transmitted packets will delay socket destruction
3329 * until the last reference will be released.
3334 xfrm_sk_free_policy(sk);
3336 sk_refcnt_debug_release(sk);
3340 EXPORT_SYMBOL(sk_common_release);
3342 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3344 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3346 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3347 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3348 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3349 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3350 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3351 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3352 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3353 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3354 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3357 #ifdef CONFIG_PROC_FS
3358 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3360 int val[PROTO_INUSE_NR];
3363 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3365 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3367 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3369 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3371 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3373 int cpu, idx = prot->inuse_idx;
3376 for_each_possible_cpu(cpu)
3377 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3379 return res >= 0 ? res : 0;
3381 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3383 static void sock_inuse_add(struct net *net, int val)
3385 this_cpu_add(*net->core.sock_inuse, val);
3388 int sock_inuse_get(struct net *net)
3392 for_each_possible_cpu(cpu)
3393 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3398 EXPORT_SYMBOL_GPL(sock_inuse_get);
3400 static int __net_init sock_inuse_init_net(struct net *net)
3402 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3403 if (net->core.prot_inuse == NULL)
3406 net->core.sock_inuse = alloc_percpu(int);
3407 if (net->core.sock_inuse == NULL)
3413 free_percpu(net->core.prot_inuse);
3417 static void __net_exit sock_inuse_exit_net(struct net *net)
3419 free_percpu(net->core.prot_inuse);
3420 free_percpu(net->core.sock_inuse);
3423 static struct pernet_operations net_inuse_ops = {
3424 .init = sock_inuse_init_net,
3425 .exit = sock_inuse_exit_net,
3428 static __init int net_inuse_init(void)
3430 if (register_pernet_subsys(&net_inuse_ops))
3431 panic("Cannot initialize net inuse counters");
3436 core_initcall(net_inuse_init);
3438 static int assign_proto_idx(struct proto *prot)
3440 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3442 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3443 pr_err("PROTO_INUSE_NR exhausted\n");
3447 set_bit(prot->inuse_idx, proto_inuse_idx);
3451 static void release_proto_idx(struct proto *prot)
3453 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3454 clear_bit(prot->inuse_idx, proto_inuse_idx);
3457 static inline int assign_proto_idx(struct proto *prot)
3462 static inline void release_proto_idx(struct proto *prot)
3466 static void sock_inuse_add(struct net *net, int val)
3471 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3475 kfree(twsk_prot->twsk_slab_name);
3476 twsk_prot->twsk_slab_name = NULL;
3477 kmem_cache_destroy(twsk_prot->twsk_slab);
3478 twsk_prot->twsk_slab = NULL;
3481 static int tw_prot_init(const struct proto *prot)
3483 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3488 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3490 if (!twsk_prot->twsk_slab_name)
3493 twsk_prot->twsk_slab =
3494 kmem_cache_create(twsk_prot->twsk_slab_name,
3495 twsk_prot->twsk_obj_size, 0,
3496 SLAB_ACCOUNT | prot->slab_flags,
3498 if (!twsk_prot->twsk_slab) {
3499 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3507 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3511 kfree(rsk_prot->slab_name);
3512 rsk_prot->slab_name = NULL;
3513 kmem_cache_destroy(rsk_prot->slab);
3514 rsk_prot->slab = NULL;
3517 static int req_prot_init(const struct proto *prot)
3519 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3524 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3526 if (!rsk_prot->slab_name)
3529 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3530 rsk_prot->obj_size, 0,
3531 SLAB_ACCOUNT | prot->slab_flags,
3534 if (!rsk_prot->slab) {
3535 pr_crit("%s: Can't create request sock SLAB cache!\n",
3542 int proto_register(struct proto *prot, int alloc_slab)
3547 prot->slab = kmem_cache_create_usercopy(prot->name,
3549 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3551 prot->useroffset, prot->usersize,
3554 if (prot->slab == NULL) {
3555 pr_crit("%s: Can't create sock SLAB cache!\n",
3560 if (req_prot_init(prot))
3561 goto out_free_request_sock_slab;
3563 if (tw_prot_init(prot))
3564 goto out_free_timewait_sock_slab;
3567 mutex_lock(&proto_list_mutex);
3568 ret = assign_proto_idx(prot);
3570 mutex_unlock(&proto_list_mutex);
3571 goto out_free_timewait_sock_slab;
3573 list_add(&prot->node, &proto_list);
3574 mutex_unlock(&proto_list_mutex);
3577 out_free_timewait_sock_slab:
3579 tw_prot_cleanup(prot->twsk_prot);
3580 out_free_request_sock_slab:
3582 req_prot_cleanup(prot->rsk_prot);
3584 kmem_cache_destroy(prot->slab);
3590 EXPORT_SYMBOL(proto_register);
3592 void proto_unregister(struct proto *prot)
3594 mutex_lock(&proto_list_mutex);
3595 release_proto_idx(prot);
3596 list_del(&prot->node);
3597 mutex_unlock(&proto_list_mutex);
3599 kmem_cache_destroy(prot->slab);
3602 req_prot_cleanup(prot->rsk_prot);
3603 tw_prot_cleanup(prot->twsk_prot);
3605 EXPORT_SYMBOL(proto_unregister);
3607 int sock_load_diag_module(int family, int protocol)
3610 if (!sock_is_registered(family))
3613 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3614 NETLINK_SOCK_DIAG, family);
3618 if (family == AF_INET &&
3619 protocol != IPPROTO_RAW &&
3620 protocol < MAX_INET_PROTOS &&
3621 !rcu_access_pointer(inet_protos[protocol]))
3625 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3626 NETLINK_SOCK_DIAG, family, protocol);
3628 EXPORT_SYMBOL(sock_load_diag_module);
3630 #ifdef CONFIG_PROC_FS
3631 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3632 __acquires(proto_list_mutex)
3634 mutex_lock(&proto_list_mutex);
3635 return seq_list_start_head(&proto_list, *pos);
3638 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3640 return seq_list_next(v, &proto_list, pos);
3643 static void proto_seq_stop(struct seq_file *seq, void *v)
3644 __releases(proto_list_mutex)
3646 mutex_unlock(&proto_list_mutex);
3649 static char proto_method_implemented(const void *method)
3651 return method == NULL ? 'n' : 'y';
3653 static long sock_prot_memory_allocated(struct proto *proto)
3655 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3658 static const char *sock_prot_memory_pressure(struct proto *proto)
3660 return proto->memory_pressure != NULL ?
3661 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3664 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3667 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3668 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3671 sock_prot_inuse_get(seq_file_net(seq), proto),
3672 sock_prot_memory_allocated(proto),
3673 sock_prot_memory_pressure(proto),
3675 proto->slab == NULL ? "no" : "yes",
3676 module_name(proto->owner),
3677 proto_method_implemented(proto->close),
3678 proto_method_implemented(proto->connect),
3679 proto_method_implemented(proto->disconnect),
3680 proto_method_implemented(proto->accept),
3681 proto_method_implemented(proto->ioctl),
3682 proto_method_implemented(proto->init),
3683 proto_method_implemented(proto->destroy),
3684 proto_method_implemented(proto->shutdown),
3685 proto_method_implemented(proto->setsockopt),
3686 proto_method_implemented(proto->getsockopt),
3687 proto_method_implemented(proto->sendmsg),
3688 proto_method_implemented(proto->recvmsg),
3689 proto_method_implemented(proto->sendpage),
3690 proto_method_implemented(proto->bind),
3691 proto_method_implemented(proto->backlog_rcv),
3692 proto_method_implemented(proto->hash),
3693 proto_method_implemented(proto->unhash),
3694 proto_method_implemented(proto->get_port),
3695 proto_method_implemented(proto->enter_memory_pressure));
3698 static int proto_seq_show(struct seq_file *seq, void *v)
3700 if (v == &proto_list)
3701 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3710 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3712 proto_seq_printf(seq, list_entry(v, struct proto, node));
3716 static const struct seq_operations proto_seq_ops = {
3717 .start = proto_seq_start,
3718 .next = proto_seq_next,
3719 .stop = proto_seq_stop,
3720 .show = proto_seq_show,
3723 static __net_init int proto_init_net(struct net *net)
3725 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3726 sizeof(struct seq_net_private)))
3732 static __net_exit void proto_exit_net(struct net *net)
3734 remove_proc_entry("protocols", net->proc_net);
3738 static __net_initdata struct pernet_operations proto_net_ops = {
3739 .init = proto_init_net,
3740 .exit = proto_exit_net,
3743 static int __init proto_init(void)
3745 return register_pernet_subsys(&proto_net_ops);
3748 subsys_initcall(proto_init);
3750 #endif /* PROC_FS */
3752 #ifdef CONFIG_NET_RX_BUSY_POLL
3753 bool sk_busy_loop_end(void *p, unsigned long start_time)
3755 struct sock *sk = p;
3757 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3758 sk_busy_loop_timeout(sk, start_time);
3760 EXPORT_SYMBOL(sk_busy_loop_end);
3761 #endif /* CONFIG_NET_RX_BUSY_POLL */
3763 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3765 if (!sk->sk_prot->bind_add)
3767 return sk->sk_prot->bind_add(sk, addr, addr_len);
3769 EXPORT_SYMBOL(sock_bind_add);