1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
118 #include <linux/uaccess.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
137 #include <trace/events/sock.h>
140 #include <net/busy_poll.h>
142 #include <linux/ethtool.h>
144 static DEFINE_MUTEX(proto_list_mutex);
145 static LIST_HEAD(proto_list);
147 static void sock_inuse_add(struct net *net, int val);
150 * sk_ns_capable - General socket capability test
151 * @sk: Socket to use a capability on or through
152 * @user_ns: The user namespace of the capability to use
153 * @cap: The capability to use
155 * Test to see if the opener of the socket had when the socket was
156 * created and the current process has the capability @cap in the user
157 * namespace @user_ns.
159 bool sk_ns_capable(const struct sock *sk,
160 struct user_namespace *user_ns, int cap)
162 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
163 ns_capable(user_ns, cap);
165 EXPORT_SYMBOL(sk_ns_capable);
168 * sk_capable - Socket global capability test
169 * @sk: Socket to use a capability on or through
170 * @cap: The global capability to use
172 * Test to see if the opener of the socket had when the socket was
173 * created and the current process has the capability @cap in all user
176 bool sk_capable(const struct sock *sk, int cap)
178 return sk_ns_capable(sk, &init_user_ns, cap);
180 EXPORT_SYMBOL(sk_capable);
183 * sk_net_capable - Network namespace socket capability test
184 * @sk: Socket to use a capability on or through
185 * @cap: The capability to use
187 * Test to see if the opener of the socket had when the socket was created
188 * and the current process has the capability @cap over the network namespace
189 * the socket is a member of.
191 bool sk_net_capable(const struct sock *sk, int cap)
193 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195 EXPORT_SYMBOL(sk_net_capable);
198 * Each address family might have different locking rules, so we have
199 * one slock key per address family and separate keys for internal and
202 static struct lock_class_key af_family_keys[AF_MAX];
203 static struct lock_class_key af_family_kern_keys[AF_MAX];
204 static struct lock_class_key af_family_slock_keys[AF_MAX];
205 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
208 * Make lock validator output more readable. (we pre-construct these
209 * strings build-time, so that runtime initialization of socket
213 #define _sock_locks(x) \
214 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
215 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
216 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
217 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
218 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
219 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
220 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
221 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
222 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
223 x "27" , x "28" , x "AF_CAN" , \
224 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
225 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
226 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
227 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
228 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
232 static const char *const af_family_key_strings[AF_MAX+1] = {
233 _sock_locks("sk_lock-")
235 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
236 _sock_locks("slock-")
238 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
239 _sock_locks("clock-")
242 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
243 _sock_locks("k-sk_lock-")
245 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-slock-")
248 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
249 _sock_locks("k-clock-")
251 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
252 _sock_locks("rlock-")
254 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
255 _sock_locks("wlock-")
257 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
258 _sock_locks("elock-")
262 * sk_callback_lock and sk queues locking rules are per-address-family,
263 * so split the lock classes by using a per-AF key:
265 static struct lock_class_key af_callback_keys[AF_MAX];
266 static struct lock_class_key af_rlock_keys[AF_MAX];
267 static struct lock_class_key af_wlock_keys[AF_MAX];
268 static struct lock_class_key af_elock_keys[AF_MAX];
269 static struct lock_class_key af_kern_callback_keys[AF_MAX];
271 /* Run time adjustable parameters. */
272 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
273 EXPORT_SYMBOL(sysctl_wmem_max);
274 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
275 EXPORT_SYMBOL(sysctl_rmem_max);
276 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
277 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
279 /* Maximal space eaten by iovec or ancillary data plus some space */
280 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
281 EXPORT_SYMBOL(sysctl_optmem_max);
283 int sysctl_tstamp_allow_data __read_mostly = 1;
285 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
286 EXPORT_SYMBOL_GPL(memalloc_socks_key);
289 * sk_set_memalloc - sets %SOCK_MEMALLOC
290 * @sk: socket to set it on
292 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
293 * It's the responsibility of the admin to adjust min_free_kbytes
294 * to meet the requirements
296 void sk_set_memalloc(struct sock *sk)
298 sock_set_flag(sk, SOCK_MEMALLOC);
299 sk->sk_allocation |= __GFP_MEMALLOC;
300 static_branch_inc(&memalloc_socks_key);
302 EXPORT_SYMBOL_GPL(sk_set_memalloc);
304 void sk_clear_memalloc(struct sock *sk)
306 sock_reset_flag(sk, SOCK_MEMALLOC);
307 sk->sk_allocation &= ~__GFP_MEMALLOC;
308 static_branch_dec(&memalloc_socks_key);
311 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
312 * progress of swapping. SOCK_MEMALLOC may be cleared while
313 * it has rmem allocations due to the last swapfile being deactivated
314 * but there is a risk that the socket is unusable due to exceeding
315 * the rmem limits. Reclaim the reserves and obey rmem limits again.
319 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
321 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
324 unsigned int noreclaim_flag;
326 /* these should have been dropped before queueing */
327 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
329 noreclaim_flag = memalloc_noreclaim_save();
330 ret = sk->sk_backlog_rcv(sk, skb);
331 memalloc_noreclaim_restore(noreclaim_flag);
335 EXPORT_SYMBOL(__sk_backlog_rcv);
337 void sk_error_report(struct sock *sk)
339 sk->sk_error_report(sk);
341 switch (sk->sk_family) {
345 trace_inet_sk_error_report(sk);
351 EXPORT_SYMBOL(sk_error_report);
353 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
355 struct __kernel_sock_timeval tv;
357 if (timeo == MAX_SCHEDULE_TIMEOUT) {
361 tv.tv_sec = timeo / HZ;
362 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
365 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
366 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
367 *(struct old_timeval32 *)optval = tv32;
372 struct __kernel_old_timeval old_tv;
373 old_tv.tv_sec = tv.tv_sec;
374 old_tv.tv_usec = tv.tv_usec;
375 *(struct __kernel_old_timeval *)optval = old_tv;
376 return sizeof(old_tv);
379 *(struct __kernel_sock_timeval *)optval = tv;
382 EXPORT_SYMBOL(sock_get_timeout);
384 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
385 sockptr_t optval, int optlen, bool old_timeval)
387 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
388 struct old_timeval32 tv32;
390 if (optlen < sizeof(tv32))
393 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
395 tv->tv_sec = tv32.tv_sec;
396 tv->tv_usec = tv32.tv_usec;
397 } else if (old_timeval) {
398 struct __kernel_old_timeval old_tv;
400 if (optlen < sizeof(old_tv))
402 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
404 tv->tv_sec = old_tv.tv_sec;
405 tv->tv_usec = old_tv.tv_usec;
407 if (optlen < sizeof(*tv))
409 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
415 EXPORT_SYMBOL(sock_copy_user_timeval);
417 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
420 struct __kernel_sock_timeval tv;
421 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
426 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
430 static int warned __read_mostly;
433 if (warned < 10 && net_ratelimit()) {
435 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
436 __func__, current->comm, task_pid_nr(current));
440 *timeo_p = MAX_SCHEDULE_TIMEOUT;
441 if (tv.tv_sec == 0 && tv.tv_usec == 0)
443 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
444 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
448 static bool sock_needs_netstamp(const struct sock *sk)
450 switch (sk->sk_family) {
459 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
461 if (sk->sk_flags & flags) {
462 sk->sk_flags &= ~flags;
463 if (sock_needs_netstamp(sk) &&
464 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
465 net_disable_timestamp();
470 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
473 struct sk_buff_head *list = &sk->sk_receive_queue;
475 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
476 atomic_inc(&sk->sk_drops);
477 trace_sock_rcvqueue_full(sk, skb);
481 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
482 atomic_inc(&sk->sk_drops);
487 skb_set_owner_r(skb, sk);
489 /* we escape from rcu protected region, make sure we dont leak
494 spin_lock_irqsave(&list->lock, flags);
495 sock_skb_set_dropcount(sk, skb);
496 __skb_queue_tail(list, skb);
497 spin_unlock_irqrestore(&list->lock, flags);
499 if (!sock_flag(sk, SOCK_DEAD))
500 sk->sk_data_ready(sk);
503 EXPORT_SYMBOL(__sock_queue_rcv_skb);
505 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
509 err = sk_filter(sk, skb);
513 return __sock_queue_rcv_skb(sk, skb);
515 EXPORT_SYMBOL(sock_queue_rcv_skb);
517 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
518 const int nested, unsigned int trim_cap, bool refcounted)
520 int rc = NET_RX_SUCCESS;
522 if (sk_filter_trim_cap(sk, skb, trim_cap))
523 goto discard_and_relse;
527 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
528 atomic_inc(&sk->sk_drops);
529 goto discard_and_relse;
532 bh_lock_sock_nested(sk);
535 if (!sock_owned_by_user(sk)) {
537 * trylock + unlock semantics:
539 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
541 rc = sk_backlog_rcv(sk, skb);
543 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
544 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
546 atomic_inc(&sk->sk_drops);
547 goto discard_and_relse;
559 EXPORT_SYMBOL(__sk_receive_skb);
561 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
563 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
565 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
567 struct dst_entry *dst = __sk_dst_get(sk);
569 if (dst && dst->obsolete &&
570 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
571 dst, cookie) == NULL) {
572 sk_tx_queue_clear(sk);
573 sk->sk_dst_pending_confirm = 0;
574 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
581 EXPORT_SYMBOL(__sk_dst_check);
583 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
585 struct dst_entry *dst = sk_dst_get(sk);
587 if (dst && dst->obsolete &&
588 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
589 dst, cookie) == NULL) {
597 EXPORT_SYMBOL(sk_dst_check);
599 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
601 int ret = -ENOPROTOOPT;
602 #ifdef CONFIG_NETDEVICES
603 struct net *net = sock_net(sk);
607 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
614 sk->sk_bound_dev_if = ifindex;
615 if (sk->sk_prot->rehash)
616 sk->sk_prot->rehash(sk);
627 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
633 ret = sock_bindtoindex_locked(sk, ifindex);
639 EXPORT_SYMBOL(sock_bindtoindex);
641 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
643 int ret = -ENOPROTOOPT;
644 #ifdef CONFIG_NETDEVICES
645 struct net *net = sock_net(sk);
646 char devname[IFNAMSIZ];
653 /* Bind this socket to a particular device like "eth0",
654 * as specified in the passed interface name. If the
655 * name is "" or the option length is zero the socket
658 if (optlen > IFNAMSIZ - 1)
659 optlen = IFNAMSIZ - 1;
660 memset(devname, 0, sizeof(devname));
663 if (copy_from_sockptr(devname, optval, optlen))
667 if (devname[0] != '\0') {
668 struct net_device *dev;
671 dev = dev_get_by_name_rcu(net, devname);
673 index = dev->ifindex;
680 return sock_bindtoindex(sk, index, true);
687 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
688 int __user *optlen, int len)
690 int ret = -ENOPROTOOPT;
691 #ifdef CONFIG_NETDEVICES
692 struct net *net = sock_net(sk);
693 char devname[IFNAMSIZ];
695 if (sk->sk_bound_dev_if == 0) {
704 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
708 len = strlen(devname) + 1;
711 if (copy_to_user(optval, devname, len))
716 if (put_user(len, optlen))
727 bool sk_mc_loop(struct sock *sk)
729 if (dev_recursion_level())
733 switch (sk->sk_family) {
735 return inet_sk(sk)->mc_loop;
736 #if IS_ENABLED(CONFIG_IPV6)
738 return inet6_sk(sk)->mc_loop;
744 EXPORT_SYMBOL(sk_mc_loop);
746 void sock_set_reuseaddr(struct sock *sk)
749 sk->sk_reuse = SK_CAN_REUSE;
752 EXPORT_SYMBOL(sock_set_reuseaddr);
754 void sock_set_reuseport(struct sock *sk)
757 sk->sk_reuseport = true;
760 EXPORT_SYMBOL(sock_set_reuseport);
762 void sock_no_linger(struct sock *sk)
765 sk->sk_lingertime = 0;
766 sock_set_flag(sk, SOCK_LINGER);
769 EXPORT_SYMBOL(sock_no_linger);
771 void sock_set_priority(struct sock *sk, u32 priority)
774 sk->sk_priority = priority;
777 EXPORT_SYMBOL(sock_set_priority);
779 void sock_set_sndtimeo(struct sock *sk, s64 secs)
782 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
783 sk->sk_sndtimeo = secs * HZ;
785 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
788 EXPORT_SYMBOL(sock_set_sndtimeo);
790 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
793 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
794 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
795 sock_set_flag(sk, SOCK_RCVTSTAMP);
796 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
798 sock_reset_flag(sk, SOCK_RCVTSTAMP);
799 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
803 void sock_enable_timestamps(struct sock *sk)
806 __sock_set_timestamps(sk, true, false, true);
809 EXPORT_SYMBOL(sock_enable_timestamps);
811 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
814 case SO_TIMESTAMP_OLD:
815 __sock_set_timestamps(sk, valbool, false, false);
817 case SO_TIMESTAMP_NEW:
818 __sock_set_timestamps(sk, valbool, true, false);
820 case SO_TIMESTAMPNS_OLD:
821 __sock_set_timestamps(sk, valbool, false, true);
823 case SO_TIMESTAMPNS_NEW:
824 __sock_set_timestamps(sk, valbool, true, true);
829 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
831 struct net *net = sock_net(sk);
832 struct net_device *dev = NULL;
837 if (sk->sk_bound_dev_if)
838 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
841 pr_err("%s: sock not bind to device\n", __func__);
845 num = ethtool_get_phc_vclocks(dev, &vclock_index);
846 for (i = 0; i < num; i++) {
847 if (*(vclock_index + i) == phc_index) {
859 sk->sk_bind_phc = phc_index;
864 int sock_set_timestamping(struct sock *sk, int optname,
865 struct so_timestamping timestamping)
867 int val = timestamping.flags;
870 if (val & ~SOF_TIMESTAMPING_MASK)
873 if (val & SOF_TIMESTAMPING_OPT_ID &&
874 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
875 if (sk->sk_protocol == IPPROTO_TCP &&
876 sk->sk_type == SOCK_STREAM) {
877 if ((1 << sk->sk_state) &
878 (TCPF_CLOSE | TCPF_LISTEN))
880 sk->sk_tskey = tcp_sk(sk)->snd_una;
886 if (val & SOF_TIMESTAMPING_OPT_STATS &&
887 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
890 if (val & SOF_TIMESTAMPING_BIND_PHC) {
891 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
896 sk->sk_tsflags = val;
897 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
899 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
900 sock_enable_timestamp(sk,
901 SOCK_TIMESTAMPING_RX_SOFTWARE);
903 sock_disable_timestamp(sk,
904 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
908 void sock_set_keepalive(struct sock *sk)
911 if (sk->sk_prot->keepalive)
912 sk->sk_prot->keepalive(sk, true);
913 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
916 EXPORT_SYMBOL(sock_set_keepalive);
918 static void __sock_set_rcvbuf(struct sock *sk, int val)
920 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
921 * as a negative value.
923 val = min_t(int, val, INT_MAX / 2);
924 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
926 /* We double it on the way in to account for "struct sk_buff" etc.
927 * overhead. Applications assume that the SO_RCVBUF setting they make
928 * will allow that much actual data to be received on that socket.
930 * Applications are unaware that "struct sk_buff" and other overheads
931 * allocate from the receive buffer during socket buffer allocation.
933 * And after considering the possible alternatives, returning the value
934 * we actually used in getsockopt is the most desirable behavior.
936 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
939 void sock_set_rcvbuf(struct sock *sk, int val)
942 __sock_set_rcvbuf(sk, val);
945 EXPORT_SYMBOL(sock_set_rcvbuf);
947 static void __sock_set_mark(struct sock *sk, u32 val)
949 if (val != sk->sk_mark) {
955 void sock_set_mark(struct sock *sk, u32 val)
958 __sock_set_mark(sk, val);
961 EXPORT_SYMBOL(sock_set_mark);
963 static void sock_release_reserved_memory(struct sock *sk, int bytes)
965 /* Round down bytes to multiple of pages */
966 bytes &= ~(SK_MEM_QUANTUM - 1);
968 WARN_ON(bytes > sk->sk_reserved_mem);
969 sk->sk_reserved_mem -= bytes;
973 static int sock_reserve_memory(struct sock *sk, int bytes)
979 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg)
985 pages = sk_mem_pages(bytes);
987 /* pre-charge to memcg */
988 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
989 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
993 /* pre-charge to forward_alloc */
994 allocated = sk_memory_allocated_add(sk, pages);
995 /* If the system goes into memory pressure with this
996 * precharge, give up and return error.
998 if (allocated > sk_prot_mem_limits(sk, 1)) {
999 sk_memory_allocated_sub(sk, pages);
1000 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1003 sk->sk_forward_alloc += pages << SK_MEM_QUANTUM_SHIFT;
1005 sk->sk_reserved_mem += pages << SK_MEM_QUANTUM_SHIFT;
1011 * This is meant for all protocols to use and covers goings on
1012 * at the socket level. Everything here is generic.
1015 int sock_setsockopt(struct socket *sock, int level, int optname,
1016 sockptr_t optval, unsigned int optlen)
1018 struct so_timestamping timestamping;
1019 struct sock_txtime sk_txtime;
1020 struct sock *sk = sock->sk;
1027 * Options without arguments
1030 if (optname == SO_BINDTODEVICE)
1031 return sock_setbindtodevice(sk, optval, optlen);
1033 if (optlen < sizeof(int))
1036 if (copy_from_sockptr(&val, optval, sizeof(val)))
1039 valbool = val ? 1 : 0;
1045 if (val && !capable(CAP_NET_ADMIN))
1048 sock_valbool_flag(sk, SOCK_DBG, valbool);
1051 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1054 sk->sk_reuseport = valbool;
1063 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1067 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1070 /* Don't error on this BSD doesn't and if you think
1071 * about it this is right. Otherwise apps have to
1072 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1073 * are treated in BSD as hints
1075 val = min_t(u32, val, sysctl_wmem_max);
1077 /* Ensure val * 2 fits into an int, to prevent max_t()
1078 * from treating it as a negative value.
1080 val = min_t(int, val, INT_MAX / 2);
1081 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1082 WRITE_ONCE(sk->sk_sndbuf,
1083 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1084 /* Wake up sending tasks if we upped the value. */
1085 sk->sk_write_space(sk);
1088 case SO_SNDBUFFORCE:
1089 if (!capable(CAP_NET_ADMIN)) {
1094 /* No negative values (to prevent underflow, as val will be
1102 /* Don't error on this BSD doesn't and if you think
1103 * about it this is right. Otherwise apps have to
1104 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1105 * are treated in BSD as hints
1107 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
1110 case SO_RCVBUFFORCE:
1111 if (!capable(CAP_NET_ADMIN)) {
1116 /* No negative values (to prevent underflow, as val will be
1119 __sock_set_rcvbuf(sk, max(val, 0));
1123 if (sk->sk_prot->keepalive)
1124 sk->sk_prot->keepalive(sk, valbool);
1125 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1129 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1133 sk->sk_no_check_tx = valbool;
1137 if ((val >= 0 && val <= 6) ||
1138 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1139 sk->sk_priority = val;
1145 if (optlen < sizeof(ling)) {
1146 ret = -EINVAL; /* 1003.1g */
1149 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1154 sock_reset_flag(sk, SOCK_LINGER);
1156 #if (BITS_PER_LONG == 32)
1157 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1158 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1161 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1162 sock_set_flag(sk, SOCK_LINGER);
1171 set_bit(SOCK_PASSCRED, &sock->flags);
1173 clear_bit(SOCK_PASSCRED, &sock->flags);
1176 case SO_TIMESTAMP_OLD:
1177 case SO_TIMESTAMP_NEW:
1178 case SO_TIMESTAMPNS_OLD:
1179 case SO_TIMESTAMPNS_NEW:
1180 sock_set_timestamp(sk, optname, valbool);
1183 case SO_TIMESTAMPING_NEW:
1184 case SO_TIMESTAMPING_OLD:
1185 if (optlen == sizeof(timestamping)) {
1186 if (copy_from_sockptr(×tamping, optval,
1187 sizeof(timestamping))) {
1192 memset(×tamping, 0, sizeof(timestamping));
1193 timestamping.flags = val;
1195 ret = sock_set_timestamping(sk, optname, timestamping);
1201 if (sock->ops->set_rcvlowat)
1202 ret = sock->ops->set_rcvlowat(sk, val);
1204 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1207 case SO_RCVTIMEO_OLD:
1208 case SO_RCVTIMEO_NEW:
1209 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1210 optlen, optname == SO_RCVTIMEO_OLD);
1213 case SO_SNDTIMEO_OLD:
1214 case SO_SNDTIMEO_NEW:
1215 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1216 optlen, optname == SO_SNDTIMEO_OLD);
1219 case SO_ATTACH_FILTER: {
1220 struct sock_fprog fprog;
1222 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1224 ret = sk_attach_filter(&fprog, sk);
1229 if (optlen == sizeof(u32)) {
1233 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1236 ret = sk_attach_bpf(ufd, sk);
1240 case SO_ATTACH_REUSEPORT_CBPF: {
1241 struct sock_fprog fprog;
1243 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1245 ret = sk_reuseport_attach_filter(&fprog, sk);
1248 case SO_ATTACH_REUSEPORT_EBPF:
1250 if (optlen == sizeof(u32)) {
1254 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1257 ret = sk_reuseport_attach_bpf(ufd, sk);
1261 case SO_DETACH_REUSEPORT_BPF:
1262 ret = reuseport_detach_prog(sk);
1265 case SO_DETACH_FILTER:
1266 ret = sk_detach_filter(sk);
1269 case SO_LOCK_FILTER:
1270 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1273 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1278 set_bit(SOCK_PASSSEC, &sock->flags);
1280 clear_bit(SOCK_PASSSEC, &sock->flags);
1283 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1288 __sock_set_mark(sk, val);
1292 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1295 case SO_WIFI_STATUS:
1296 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1300 if (sock->ops->set_peek_off)
1301 ret = sock->ops->set_peek_off(sk, val);
1307 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1310 case SO_SELECT_ERR_QUEUE:
1311 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1314 #ifdef CONFIG_NET_RX_BUSY_POLL
1316 /* allow unprivileged users to decrease the value */
1317 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1323 WRITE_ONCE(sk->sk_ll_usec, val);
1326 case SO_PREFER_BUSY_POLL:
1327 if (valbool && !capable(CAP_NET_ADMIN))
1330 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1332 case SO_BUSY_POLL_BUDGET:
1333 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1336 if (val < 0 || val > U16_MAX)
1339 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1344 case SO_MAX_PACING_RATE:
1346 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1348 if (sizeof(ulval) != sizeof(val) &&
1349 optlen >= sizeof(ulval) &&
1350 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1355 cmpxchg(&sk->sk_pacing_status,
1358 sk->sk_max_pacing_rate = ulval;
1359 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1362 case SO_INCOMING_CPU:
1363 WRITE_ONCE(sk->sk_incoming_cpu, val);
1368 dst_negative_advice(sk);
1372 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1373 if (!((sk->sk_type == SOCK_STREAM &&
1374 sk->sk_protocol == IPPROTO_TCP) ||
1375 (sk->sk_type == SOCK_DGRAM &&
1376 sk->sk_protocol == IPPROTO_UDP)))
1378 } else if (sk->sk_family != PF_RDS) {
1382 if (val < 0 || val > 1)
1385 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1390 if (optlen != sizeof(struct sock_txtime)) {
1393 } else if (copy_from_sockptr(&sk_txtime, optval,
1394 sizeof(struct sock_txtime))) {
1397 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1401 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1402 * scheduler has enough safe guards.
1404 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1405 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1409 sock_valbool_flag(sk, SOCK_TXTIME, true);
1410 sk->sk_clockid = sk_txtime.clockid;
1411 sk->sk_txtime_deadline_mode =
1412 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1413 sk->sk_txtime_report_errors =
1414 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1417 case SO_BINDTOIFINDEX:
1418 ret = sock_bindtoindex_locked(sk, val);
1422 if (val & ~SOCK_BUF_LOCK_MASK) {
1426 sk->sk_userlocks = val | (sk->sk_userlocks &
1427 ~SOCK_BUF_LOCK_MASK);
1430 case SO_RESERVE_MEM:
1439 delta = val - sk->sk_reserved_mem;
1441 sock_release_reserved_memory(sk, -delta);
1443 ret = sock_reserve_memory(sk, delta);
1454 EXPORT_SYMBOL(sock_setsockopt);
1456 static const struct cred *sk_get_peer_cred(struct sock *sk)
1458 const struct cred *cred;
1460 spin_lock(&sk->sk_peer_lock);
1461 cred = get_cred(sk->sk_peer_cred);
1462 spin_unlock(&sk->sk_peer_lock);
1467 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1468 struct ucred *ucred)
1470 ucred->pid = pid_vnr(pid);
1471 ucred->uid = ucred->gid = -1;
1473 struct user_namespace *current_ns = current_user_ns();
1475 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1476 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1480 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1482 struct user_namespace *user_ns = current_user_ns();
1485 for (i = 0; i < src->ngroups; i++)
1486 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1492 int sock_getsockopt(struct socket *sock, int level, int optname,
1493 char __user *optval, int __user *optlen)
1495 struct sock *sk = sock->sk;
1500 unsigned long ulval;
1502 struct old_timeval32 tm32;
1503 struct __kernel_old_timeval tm;
1504 struct __kernel_sock_timeval stm;
1505 struct sock_txtime txtime;
1506 struct so_timestamping timestamping;
1509 int lv = sizeof(int);
1512 if (get_user(len, optlen))
1517 memset(&v, 0, sizeof(v));
1521 v.val = sock_flag(sk, SOCK_DBG);
1525 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1529 v.val = sock_flag(sk, SOCK_BROADCAST);
1533 v.val = sk->sk_sndbuf;
1537 v.val = sk->sk_rcvbuf;
1541 v.val = sk->sk_reuse;
1545 v.val = sk->sk_reuseport;
1549 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1553 v.val = sk->sk_type;
1557 v.val = sk->sk_protocol;
1561 v.val = sk->sk_family;
1565 v.val = -sock_error(sk);
1567 v.val = xchg(&sk->sk_err_soft, 0);
1571 v.val = sock_flag(sk, SOCK_URGINLINE);
1575 v.val = sk->sk_no_check_tx;
1579 v.val = sk->sk_priority;
1583 lv = sizeof(v.ling);
1584 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1585 v.ling.l_linger = sk->sk_lingertime / HZ;
1591 case SO_TIMESTAMP_OLD:
1592 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1593 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1594 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1597 case SO_TIMESTAMPNS_OLD:
1598 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1601 case SO_TIMESTAMP_NEW:
1602 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1605 case SO_TIMESTAMPNS_NEW:
1606 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1609 case SO_TIMESTAMPING_OLD:
1610 lv = sizeof(v.timestamping);
1611 v.timestamping.flags = sk->sk_tsflags;
1612 v.timestamping.bind_phc = sk->sk_bind_phc;
1615 case SO_RCVTIMEO_OLD:
1616 case SO_RCVTIMEO_NEW:
1617 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1620 case SO_SNDTIMEO_OLD:
1621 case SO_SNDTIMEO_NEW:
1622 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1626 v.val = sk->sk_rcvlowat;
1634 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1639 struct ucred peercred;
1640 if (len > sizeof(peercred))
1641 len = sizeof(peercred);
1643 spin_lock(&sk->sk_peer_lock);
1644 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1645 spin_unlock(&sk->sk_peer_lock);
1647 if (copy_to_user(optval, &peercred, len))
1654 const struct cred *cred;
1657 cred = sk_get_peer_cred(sk);
1661 n = cred->group_info->ngroups;
1662 if (len < n * sizeof(gid_t)) {
1663 len = n * sizeof(gid_t);
1665 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1667 len = n * sizeof(gid_t);
1669 ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1680 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1685 if (copy_to_user(optval, address, len))
1690 /* Dubious BSD thing... Probably nobody even uses it, but
1691 * the UNIX standard wants it for whatever reason... -DaveM
1694 v.val = sk->sk_state == TCP_LISTEN;
1698 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1702 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1705 v.val = sk->sk_mark;
1709 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1712 case SO_WIFI_STATUS:
1713 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1717 if (!sock->ops->set_peek_off)
1720 v.val = sk->sk_peek_off;
1723 v.val = sock_flag(sk, SOCK_NOFCS);
1726 case SO_BINDTODEVICE:
1727 return sock_getbindtodevice(sk, optval, optlen, len);
1730 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1736 case SO_LOCK_FILTER:
1737 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1740 case SO_BPF_EXTENSIONS:
1741 v.val = bpf_tell_extensions();
1744 case SO_SELECT_ERR_QUEUE:
1745 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1748 #ifdef CONFIG_NET_RX_BUSY_POLL
1750 v.val = sk->sk_ll_usec;
1752 case SO_PREFER_BUSY_POLL:
1753 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1757 case SO_MAX_PACING_RATE:
1758 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1759 lv = sizeof(v.ulval);
1760 v.ulval = sk->sk_max_pacing_rate;
1763 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1767 case SO_INCOMING_CPU:
1768 v.val = READ_ONCE(sk->sk_incoming_cpu);
1773 u32 meminfo[SK_MEMINFO_VARS];
1775 sk_get_meminfo(sk, meminfo);
1777 len = min_t(unsigned int, len, sizeof(meminfo));
1778 if (copy_to_user(optval, &meminfo, len))
1784 #ifdef CONFIG_NET_RX_BUSY_POLL
1785 case SO_INCOMING_NAPI_ID:
1786 v.val = READ_ONCE(sk->sk_napi_id);
1788 /* aggregate non-NAPI IDs down to 0 */
1789 if (v.val < MIN_NAPI_ID)
1799 v.val64 = sock_gen_cookie(sk);
1803 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1807 lv = sizeof(v.txtime);
1808 v.txtime.clockid = sk->sk_clockid;
1809 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1810 SOF_TXTIME_DEADLINE_MODE : 0;
1811 v.txtime.flags |= sk->sk_txtime_report_errors ?
1812 SOF_TXTIME_REPORT_ERRORS : 0;
1815 case SO_BINDTOIFINDEX:
1816 v.val = sk->sk_bound_dev_if;
1819 case SO_NETNS_COOKIE:
1823 v.val64 = sock_net(sk)->net_cookie;
1827 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1830 case SO_RESERVE_MEM:
1831 v.val = sk->sk_reserved_mem;
1835 /* We implement the SO_SNDLOWAT etc to not be settable
1838 return -ENOPROTOOPT;
1843 if (copy_to_user(optval, &v, len))
1846 if (put_user(len, optlen))
1852 * Initialize an sk_lock.
1854 * (We also register the sk_lock with the lock validator.)
1856 static inline void sock_lock_init(struct sock *sk)
1858 if (sk->sk_kern_sock)
1859 sock_lock_init_class_and_name(
1861 af_family_kern_slock_key_strings[sk->sk_family],
1862 af_family_kern_slock_keys + sk->sk_family,
1863 af_family_kern_key_strings[sk->sk_family],
1864 af_family_kern_keys + sk->sk_family);
1866 sock_lock_init_class_and_name(
1868 af_family_slock_key_strings[sk->sk_family],
1869 af_family_slock_keys + sk->sk_family,
1870 af_family_key_strings[sk->sk_family],
1871 af_family_keys + sk->sk_family);
1875 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1876 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1877 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1879 static void sock_copy(struct sock *nsk, const struct sock *osk)
1881 const struct proto *prot = READ_ONCE(osk->sk_prot);
1882 #ifdef CONFIG_SECURITY_NETWORK
1883 void *sptr = nsk->sk_security;
1886 /* If we move sk_tx_queue_mapping out of the private section,
1887 * we must check if sk_tx_queue_clear() is called after
1888 * sock_copy() in sk_clone_lock().
1890 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1891 offsetof(struct sock, sk_dontcopy_begin) ||
1892 offsetof(struct sock, sk_tx_queue_mapping) >=
1893 offsetof(struct sock, sk_dontcopy_end));
1895 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1897 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1898 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1900 #ifdef CONFIG_SECURITY_NETWORK
1901 nsk->sk_security = sptr;
1902 security_sk_clone(osk, nsk);
1906 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1910 struct kmem_cache *slab;
1914 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1917 if (want_init_on_alloc(priority))
1918 sk_prot_clear_nulls(sk, prot->obj_size);
1920 sk = kmalloc(prot->obj_size, priority);
1923 if (security_sk_alloc(sk, family, priority))
1926 if (!try_module_get(prot->owner))
1933 security_sk_free(sk);
1936 kmem_cache_free(slab, sk);
1942 static void sk_prot_free(struct proto *prot, struct sock *sk)
1944 struct kmem_cache *slab;
1945 struct module *owner;
1947 owner = prot->owner;
1950 cgroup_sk_free(&sk->sk_cgrp_data);
1951 mem_cgroup_sk_free(sk);
1952 security_sk_free(sk);
1954 kmem_cache_free(slab, sk);
1961 * sk_alloc - All socket objects are allocated here
1962 * @net: the applicable net namespace
1963 * @family: protocol family
1964 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1965 * @prot: struct proto associated with this new sock instance
1966 * @kern: is this to be a kernel socket?
1968 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1969 struct proto *prot, int kern)
1973 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1975 sk->sk_family = family;
1977 * See comment in struct sock definition to understand
1978 * why we need sk_prot_creator -acme
1980 sk->sk_prot = sk->sk_prot_creator = prot;
1981 sk->sk_kern_sock = kern;
1983 sk->sk_net_refcnt = kern ? 0 : 1;
1984 if (likely(sk->sk_net_refcnt)) {
1986 sock_inuse_add(net, 1);
1989 sock_net_set(sk, net);
1990 refcount_set(&sk->sk_wmem_alloc, 1);
1992 mem_cgroup_sk_alloc(sk);
1993 cgroup_sk_alloc(&sk->sk_cgrp_data);
1994 sock_update_classid(&sk->sk_cgrp_data);
1995 sock_update_netprioidx(&sk->sk_cgrp_data);
1996 sk_tx_queue_clear(sk);
2001 EXPORT_SYMBOL(sk_alloc);
2003 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2004 * grace period. This is the case for UDP sockets and TCP listeners.
2006 static void __sk_destruct(struct rcu_head *head)
2008 struct sock *sk = container_of(head, struct sock, sk_rcu);
2009 struct sk_filter *filter;
2011 if (sk->sk_destruct)
2012 sk->sk_destruct(sk);
2014 filter = rcu_dereference_check(sk->sk_filter,
2015 refcount_read(&sk->sk_wmem_alloc) == 0);
2017 sk_filter_uncharge(sk, filter);
2018 RCU_INIT_POINTER(sk->sk_filter, NULL);
2021 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2023 #ifdef CONFIG_BPF_SYSCALL
2024 bpf_sk_storage_free(sk);
2027 if (atomic_read(&sk->sk_omem_alloc))
2028 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2029 __func__, atomic_read(&sk->sk_omem_alloc));
2031 if (sk->sk_frag.page) {
2032 put_page(sk->sk_frag.page);
2033 sk->sk_frag.page = NULL;
2036 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2037 put_cred(sk->sk_peer_cred);
2038 put_pid(sk->sk_peer_pid);
2040 if (likely(sk->sk_net_refcnt))
2041 put_net(sock_net(sk));
2042 sk_prot_free(sk->sk_prot_creator, sk);
2045 void sk_destruct(struct sock *sk)
2047 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2049 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2050 reuseport_detach_sock(sk);
2051 use_call_rcu = true;
2055 call_rcu(&sk->sk_rcu, __sk_destruct);
2057 __sk_destruct(&sk->sk_rcu);
2060 static void __sk_free(struct sock *sk)
2062 if (likely(sk->sk_net_refcnt))
2063 sock_inuse_add(sock_net(sk), -1);
2065 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2066 sock_diag_broadcast_destroy(sk);
2071 void sk_free(struct sock *sk)
2074 * We subtract one from sk_wmem_alloc and can know if
2075 * some packets are still in some tx queue.
2076 * If not null, sock_wfree() will call __sk_free(sk) later
2078 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2081 EXPORT_SYMBOL(sk_free);
2083 static void sk_init_common(struct sock *sk)
2085 skb_queue_head_init(&sk->sk_receive_queue);
2086 skb_queue_head_init(&sk->sk_write_queue);
2087 skb_queue_head_init(&sk->sk_error_queue);
2089 rwlock_init(&sk->sk_callback_lock);
2090 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2091 af_rlock_keys + sk->sk_family,
2092 af_family_rlock_key_strings[sk->sk_family]);
2093 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2094 af_wlock_keys + sk->sk_family,
2095 af_family_wlock_key_strings[sk->sk_family]);
2096 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2097 af_elock_keys + sk->sk_family,
2098 af_family_elock_key_strings[sk->sk_family]);
2099 lockdep_set_class_and_name(&sk->sk_callback_lock,
2100 af_callback_keys + sk->sk_family,
2101 af_family_clock_key_strings[sk->sk_family]);
2105 * sk_clone_lock - clone a socket, and lock its clone
2106 * @sk: the socket to clone
2107 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2109 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2111 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2113 struct proto *prot = READ_ONCE(sk->sk_prot);
2114 struct sk_filter *filter;
2115 bool is_charged = true;
2118 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2122 sock_copy(newsk, sk);
2124 newsk->sk_prot_creator = prot;
2127 if (likely(newsk->sk_net_refcnt))
2128 get_net(sock_net(newsk));
2129 sk_node_init(&newsk->sk_node);
2130 sock_lock_init(newsk);
2131 bh_lock_sock(newsk);
2132 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2133 newsk->sk_backlog.len = 0;
2135 atomic_set(&newsk->sk_rmem_alloc, 0);
2137 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2138 refcount_set(&newsk->sk_wmem_alloc, 1);
2140 atomic_set(&newsk->sk_omem_alloc, 0);
2141 sk_init_common(newsk);
2143 newsk->sk_dst_cache = NULL;
2144 newsk->sk_dst_pending_confirm = 0;
2145 newsk->sk_wmem_queued = 0;
2146 newsk->sk_forward_alloc = 0;
2147 newsk->sk_reserved_mem = 0;
2148 atomic_set(&newsk->sk_drops, 0);
2149 newsk->sk_send_head = NULL;
2150 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2151 atomic_set(&newsk->sk_zckey, 0);
2153 sock_reset_flag(newsk, SOCK_DONE);
2155 /* sk->sk_memcg will be populated at accept() time */
2156 newsk->sk_memcg = NULL;
2158 cgroup_sk_clone(&newsk->sk_cgrp_data);
2161 filter = rcu_dereference(sk->sk_filter);
2163 /* though it's an empty new sock, the charging may fail
2164 * if sysctl_optmem_max was changed between creation of
2165 * original socket and cloning
2167 is_charged = sk_filter_charge(newsk, filter);
2168 RCU_INIT_POINTER(newsk->sk_filter, filter);
2171 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2172 /* We need to make sure that we don't uncharge the new
2173 * socket if we couldn't charge it in the first place
2174 * as otherwise we uncharge the parent's filter.
2177 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2178 sk_free_unlock_clone(newsk);
2182 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2184 if (bpf_sk_storage_clone(sk, newsk)) {
2185 sk_free_unlock_clone(newsk);
2190 /* Clear sk_user_data if parent had the pointer tagged
2191 * as not suitable for copying when cloning.
2193 if (sk_user_data_is_nocopy(newsk))
2194 newsk->sk_user_data = NULL;
2197 newsk->sk_err_soft = 0;
2198 newsk->sk_priority = 0;
2199 newsk->sk_incoming_cpu = raw_smp_processor_id();
2200 if (likely(newsk->sk_net_refcnt))
2201 sock_inuse_add(sock_net(newsk), 1);
2203 /* Before updating sk_refcnt, we must commit prior changes to memory
2204 * (Documentation/RCU/rculist_nulls.rst for details)
2207 refcount_set(&newsk->sk_refcnt, 2);
2209 /* Increment the counter in the same struct proto as the master
2210 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2211 * is the same as sk->sk_prot->socks, as this field was copied
2214 * This _changes_ the previous behaviour, where
2215 * tcp_create_openreq_child always was incrementing the
2216 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2217 * to be taken into account in all callers. -acme
2219 sk_refcnt_debug_inc(newsk);
2220 sk_set_socket(newsk, NULL);
2221 sk_tx_queue_clear(newsk);
2222 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2224 if (newsk->sk_prot->sockets_allocated)
2225 sk_sockets_allocated_inc(newsk);
2227 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2228 net_enable_timestamp();
2232 EXPORT_SYMBOL_GPL(sk_clone_lock);
2234 void sk_free_unlock_clone(struct sock *sk)
2236 /* It is still raw copy of parent, so invalidate
2237 * destructor and make plain sk_free() */
2238 sk->sk_destruct = NULL;
2242 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2244 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2248 sk_dst_set(sk, dst);
2249 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2250 if (sk->sk_route_caps & NETIF_F_GSO)
2251 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2252 sk->sk_route_caps &= ~sk->sk_route_nocaps;
2253 if (sk_can_gso(sk)) {
2254 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2255 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2257 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2258 sk->sk_gso_max_size = dst->dev->gso_max_size;
2259 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2262 sk->sk_gso_max_segs = max_segs;
2264 EXPORT_SYMBOL_GPL(sk_setup_caps);
2267 * Simple resource managers for sockets.
2272 * Write buffer destructor automatically called from kfree_skb.
2274 void sock_wfree(struct sk_buff *skb)
2276 struct sock *sk = skb->sk;
2277 unsigned int len = skb->truesize;
2279 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2281 * Keep a reference on sk_wmem_alloc, this will be released
2282 * after sk_write_space() call
2284 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2285 sk->sk_write_space(sk);
2289 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2290 * could not do because of in-flight packets
2292 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2295 EXPORT_SYMBOL(sock_wfree);
2297 /* This variant of sock_wfree() is used by TCP,
2298 * since it sets SOCK_USE_WRITE_QUEUE.
2300 void __sock_wfree(struct sk_buff *skb)
2302 struct sock *sk = skb->sk;
2304 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2308 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2313 if (unlikely(!sk_fullsock(sk))) {
2314 skb->destructor = sock_edemux;
2319 skb->destructor = sock_wfree;
2320 skb_set_hash_from_sk(skb, sk);
2322 * We used to take a refcount on sk, but following operation
2323 * is enough to guarantee sk_free() wont free this sock until
2324 * all in-flight packets are completed
2326 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2328 EXPORT_SYMBOL(skb_set_owner_w);
2330 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2332 #ifdef CONFIG_TLS_DEVICE
2333 /* Drivers depend on in-order delivery for crypto offload,
2334 * partial orphan breaks out-of-order-OK logic.
2339 return (skb->destructor == sock_wfree ||
2340 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2343 /* This helper is used by netem, as it can hold packets in its
2344 * delay queue. We want to allow the owner socket to send more
2345 * packets, as if they were already TX completed by a typical driver.
2346 * But we also want to keep skb->sk set because some packet schedulers
2347 * rely on it (sch_fq for example).
2349 void skb_orphan_partial(struct sk_buff *skb)
2351 if (skb_is_tcp_pure_ack(skb))
2354 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2359 EXPORT_SYMBOL(skb_orphan_partial);
2362 * Read buffer destructor automatically called from kfree_skb.
2364 void sock_rfree(struct sk_buff *skb)
2366 struct sock *sk = skb->sk;
2367 unsigned int len = skb->truesize;
2369 atomic_sub(len, &sk->sk_rmem_alloc);
2370 sk_mem_uncharge(sk, len);
2372 EXPORT_SYMBOL(sock_rfree);
2375 * Buffer destructor for skbs that are not used directly in read or write
2376 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2378 void sock_efree(struct sk_buff *skb)
2382 EXPORT_SYMBOL(sock_efree);
2384 /* Buffer destructor for prefetch/receive path where reference count may
2385 * not be held, e.g. for listen sockets.
2388 void sock_pfree(struct sk_buff *skb)
2390 if (sk_is_refcounted(skb->sk))
2391 sock_gen_put(skb->sk);
2393 EXPORT_SYMBOL(sock_pfree);
2394 #endif /* CONFIG_INET */
2396 kuid_t sock_i_uid(struct sock *sk)
2400 read_lock_bh(&sk->sk_callback_lock);
2401 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2402 read_unlock_bh(&sk->sk_callback_lock);
2405 EXPORT_SYMBOL(sock_i_uid);
2407 unsigned long sock_i_ino(struct sock *sk)
2411 read_lock_bh(&sk->sk_callback_lock);
2412 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2413 read_unlock_bh(&sk->sk_callback_lock);
2416 EXPORT_SYMBOL(sock_i_ino);
2419 * Allocate a skb from the socket's send buffer.
2421 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2425 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2426 struct sk_buff *skb = alloc_skb(size, priority);
2429 skb_set_owner_w(skb, sk);
2435 EXPORT_SYMBOL(sock_wmalloc);
2437 static void sock_ofree(struct sk_buff *skb)
2439 struct sock *sk = skb->sk;
2441 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2444 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2447 struct sk_buff *skb;
2449 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2450 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2454 skb = alloc_skb(size, priority);
2458 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2460 skb->destructor = sock_ofree;
2465 * Allocate a memory block from the socket's option memory buffer.
2467 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2469 if ((unsigned int)size <= sysctl_optmem_max &&
2470 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2472 /* First do the add, to avoid the race if kmalloc
2475 atomic_add(size, &sk->sk_omem_alloc);
2476 mem = kmalloc(size, priority);
2479 atomic_sub(size, &sk->sk_omem_alloc);
2483 EXPORT_SYMBOL(sock_kmalloc);
2485 /* Free an option memory block. Note, we actually want the inline
2486 * here as this allows gcc to detect the nullify and fold away the
2487 * condition entirely.
2489 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2492 if (WARN_ON_ONCE(!mem))
2495 kfree_sensitive(mem);
2498 atomic_sub(size, &sk->sk_omem_alloc);
2501 void sock_kfree_s(struct sock *sk, void *mem, int size)
2503 __sock_kfree_s(sk, mem, size, false);
2505 EXPORT_SYMBOL(sock_kfree_s);
2507 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2509 __sock_kfree_s(sk, mem, size, true);
2511 EXPORT_SYMBOL(sock_kzfree_s);
2513 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2514 I think, these locks should be removed for datagram sockets.
2516 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2520 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2524 if (signal_pending(current))
2526 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2527 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2528 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2530 if (sk->sk_shutdown & SEND_SHUTDOWN)
2534 timeo = schedule_timeout(timeo);
2536 finish_wait(sk_sleep(sk), &wait);
2542 * Generic send/receive buffer handlers
2545 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2546 unsigned long data_len, int noblock,
2547 int *errcode, int max_page_order)
2549 struct sk_buff *skb;
2553 timeo = sock_sndtimeo(sk, noblock);
2555 err = sock_error(sk);
2560 if (sk->sk_shutdown & SEND_SHUTDOWN)
2563 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2566 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2567 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2571 if (signal_pending(current))
2573 timeo = sock_wait_for_wmem(sk, timeo);
2575 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2576 errcode, sk->sk_allocation);
2578 skb_set_owner_w(skb, sk);
2582 err = sock_intr_errno(timeo);
2587 EXPORT_SYMBOL(sock_alloc_send_pskb);
2589 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2590 int noblock, int *errcode)
2592 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2594 EXPORT_SYMBOL(sock_alloc_send_skb);
2596 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2597 struct sockcm_cookie *sockc)
2601 switch (cmsg->cmsg_type) {
2603 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2605 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2607 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2609 case SO_TIMESTAMPING_OLD:
2610 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2613 tsflags = *(u32 *)CMSG_DATA(cmsg);
2614 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2617 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2618 sockc->tsflags |= tsflags;
2621 if (!sock_flag(sk, SOCK_TXTIME))
2623 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2625 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2627 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2629 case SCM_CREDENTIALS:
2636 EXPORT_SYMBOL(__sock_cmsg_send);
2638 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2639 struct sockcm_cookie *sockc)
2641 struct cmsghdr *cmsg;
2644 for_each_cmsghdr(cmsg, msg) {
2645 if (!CMSG_OK(msg, cmsg))
2647 if (cmsg->cmsg_level != SOL_SOCKET)
2649 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2655 EXPORT_SYMBOL(sock_cmsg_send);
2657 static void sk_enter_memory_pressure(struct sock *sk)
2659 if (!sk->sk_prot->enter_memory_pressure)
2662 sk->sk_prot->enter_memory_pressure(sk);
2665 static void sk_leave_memory_pressure(struct sock *sk)
2667 if (sk->sk_prot->leave_memory_pressure) {
2668 sk->sk_prot->leave_memory_pressure(sk);
2670 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2672 if (memory_pressure && READ_ONCE(*memory_pressure))
2673 WRITE_ONCE(*memory_pressure, 0);
2677 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2680 * skb_page_frag_refill - check that a page_frag contains enough room
2681 * @sz: minimum size of the fragment we want to get
2682 * @pfrag: pointer to page_frag
2683 * @gfp: priority for memory allocation
2685 * Note: While this allocator tries to use high order pages, there is
2686 * no guarantee that allocations succeed. Therefore, @sz MUST be
2687 * less or equal than PAGE_SIZE.
2689 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2692 if (page_ref_count(pfrag->page) == 1) {
2696 if (pfrag->offset + sz <= pfrag->size)
2698 put_page(pfrag->page);
2702 if (SKB_FRAG_PAGE_ORDER &&
2703 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2704 /* Avoid direct reclaim but allow kswapd to wake */
2705 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2706 __GFP_COMP | __GFP_NOWARN |
2708 SKB_FRAG_PAGE_ORDER);
2709 if (likely(pfrag->page)) {
2710 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2714 pfrag->page = alloc_page(gfp);
2715 if (likely(pfrag->page)) {
2716 pfrag->size = PAGE_SIZE;
2721 EXPORT_SYMBOL(skb_page_frag_refill);
2723 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2725 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2728 sk_enter_memory_pressure(sk);
2729 sk_stream_moderate_sndbuf(sk);
2732 EXPORT_SYMBOL(sk_page_frag_refill);
2734 void __lock_sock(struct sock *sk)
2735 __releases(&sk->sk_lock.slock)
2736 __acquires(&sk->sk_lock.slock)
2741 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2742 TASK_UNINTERRUPTIBLE);
2743 spin_unlock_bh(&sk->sk_lock.slock);
2745 spin_lock_bh(&sk->sk_lock.slock);
2746 if (!sock_owned_by_user(sk))
2749 finish_wait(&sk->sk_lock.wq, &wait);
2752 void __release_sock(struct sock *sk)
2753 __releases(&sk->sk_lock.slock)
2754 __acquires(&sk->sk_lock.slock)
2756 struct sk_buff *skb, *next;
2758 while ((skb = sk->sk_backlog.head) != NULL) {
2759 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2761 spin_unlock_bh(&sk->sk_lock.slock);
2766 WARN_ON_ONCE(skb_dst_is_noref(skb));
2767 skb_mark_not_on_list(skb);
2768 sk_backlog_rcv(sk, skb);
2773 } while (skb != NULL);
2775 spin_lock_bh(&sk->sk_lock.slock);
2779 * Doing the zeroing here guarantee we can not loop forever
2780 * while a wild producer attempts to flood us.
2782 sk->sk_backlog.len = 0;
2785 void __sk_flush_backlog(struct sock *sk)
2787 spin_lock_bh(&sk->sk_lock.slock);
2789 spin_unlock_bh(&sk->sk_lock.slock);
2793 * sk_wait_data - wait for data to arrive at sk_receive_queue
2794 * @sk: sock to wait on
2795 * @timeo: for how long
2796 * @skb: last skb seen on sk_receive_queue
2798 * Now socket state including sk->sk_err is changed only under lock,
2799 * hence we may omit checks after joining wait queue.
2800 * We check receive queue before schedule() only as optimization;
2801 * it is very likely that release_sock() added new data.
2803 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2805 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2808 add_wait_queue(sk_sleep(sk), &wait);
2809 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2810 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2811 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2812 remove_wait_queue(sk_sleep(sk), &wait);
2815 EXPORT_SYMBOL(sk_wait_data);
2818 * __sk_mem_raise_allocated - increase memory_allocated
2820 * @size: memory size to allocate
2821 * @amt: pages to allocate
2822 * @kind: allocation type
2824 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2826 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2828 struct proto *prot = sk->sk_prot;
2829 long allocated = sk_memory_allocated_add(sk, amt);
2830 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2831 bool charged = true;
2834 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2835 gfp_memcg_charge())))
2836 goto suppress_allocation;
2839 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2840 sk_leave_memory_pressure(sk);
2844 /* Under pressure. */
2845 if (allocated > sk_prot_mem_limits(sk, 1))
2846 sk_enter_memory_pressure(sk);
2848 /* Over hard limit. */
2849 if (allocated > sk_prot_mem_limits(sk, 2))
2850 goto suppress_allocation;
2852 /* guarantee minimum buffer size under pressure */
2853 if (kind == SK_MEM_RECV) {
2854 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2857 } else { /* SK_MEM_SEND */
2858 int wmem0 = sk_get_wmem0(sk, prot);
2860 if (sk->sk_type == SOCK_STREAM) {
2861 if (sk->sk_wmem_queued < wmem0)
2863 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2868 if (sk_has_memory_pressure(sk)) {
2871 if (!sk_under_memory_pressure(sk))
2873 alloc = sk_sockets_allocated_read_positive(sk);
2874 if (sk_prot_mem_limits(sk, 2) > alloc *
2875 sk_mem_pages(sk->sk_wmem_queued +
2876 atomic_read(&sk->sk_rmem_alloc) +
2877 sk->sk_forward_alloc))
2881 suppress_allocation:
2883 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2884 sk_stream_moderate_sndbuf(sk);
2886 /* Fail only if socket is _under_ its sndbuf.
2887 * In this case we cannot block, so that we have to fail.
2889 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
2890 /* Force charge with __GFP_NOFAIL */
2891 if (memcg_charge && !charged) {
2892 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2893 gfp_memcg_charge() | __GFP_NOFAIL);
2899 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2900 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2902 sk_memory_allocated_sub(sk, amt);
2904 if (memcg_charge && charged)
2905 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2909 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2912 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2914 * @size: memory size to allocate
2915 * @kind: allocation type
2917 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2918 * rmem allocation. This function assumes that protocols which have
2919 * memory_pressure use sk_wmem_queued as write buffer accounting.
2921 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2923 int ret, amt = sk_mem_pages(size);
2925 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2926 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2928 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2931 EXPORT_SYMBOL(__sk_mem_schedule);
2934 * __sk_mem_reduce_allocated - reclaim memory_allocated
2936 * @amount: number of quanta
2938 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2940 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2942 sk_memory_allocated_sub(sk, amount);
2944 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2945 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2947 if (sk_under_memory_pressure(sk) &&
2948 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2949 sk_leave_memory_pressure(sk);
2951 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2954 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2956 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2958 void __sk_mem_reclaim(struct sock *sk, int amount)
2960 amount >>= SK_MEM_QUANTUM_SHIFT;
2961 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2962 __sk_mem_reduce_allocated(sk, amount);
2964 EXPORT_SYMBOL(__sk_mem_reclaim);
2966 int sk_set_peek_off(struct sock *sk, int val)
2968 sk->sk_peek_off = val;
2971 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2974 * Set of default routines for initialising struct proto_ops when
2975 * the protocol does not support a particular function. In certain
2976 * cases where it makes no sense for a protocol to have a "do nothing"
2977 * function, some default processing is provided.
2980 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2984 EXPORT_SYMBOL(sock_no_bind);
2986 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2991 EXPORT_SYMBOL(sock_no_connect);
2993 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2997 EXPORT_SYMBOL(sock_no_socketpair);
2999 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3004 EXPORT_SYMBOL(sock_no_accept);
3006 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3011 EXPORT_SYMBOL(sock_no_getname);
3013 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3017 EXPORT_SYMBOL(sock_no_ioctl);
3019 int sock_no_listen(struct socket *sock, int backlog)
3023 EXPORT_SYMBOL(sock_no_listen);
3025 int sock_no_shutdown(struct socket *sock, int how)
3029 EXPORT_SYMBOL(sock_no_shutdown);
3031 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3035 EXPORT_SYMBOL(sock_no_sendmsg);
3037 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3041 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3043 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3048 EXPORT_SYMBOL(sock_no_recvmsg);
3050 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3052 /* Mirror missing mmap method error code */
3055 EXPORT_SYMBOL(sock_no_mmap);
3058 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3059 * various sock-based usage counts.
3061 void __receive_sock(struct file *file)
3063 struct socket *sock;
3065 sock = sock_from_file(file);
3067 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3068 sock_update_classid(&sock->sk->sk_cgrp_data);
3072 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3075 struct msghdr msg = {.msg_flags = flags};
3077 char *kaddr = kmap(page);
3078 iov.iov_base = kaddr + offset;
3080 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3084 EXPORT_SYMBOL(sock_no_sendpage);
3086 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3087 int offset, size_t size, int flags)
3090 struct msghdr msg = {.msg_flags = flags};
3092 char *kaddr = kmap(page);
3094 iov.iov_base = kaddr + offset;
3096 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3100 EXPORT_SYMBOL(sock_no_sendpage_locked);
3103 * Default Socket Callbacks
3106 static void sock_def_wakeup(struct sock *sk)
3108 struct socket_wq *wq;
3111 wq = rcu_dereference(sk->sk_wq);
3112 if (skwq_has_sleeper(wq))
3113 wake_up_interruptible_all(&wq->wait);
3117 static void sock_def_error_report(struct sock *sk)
3119 struct socket_wq *wq;
3122 wq = rcu_dereference(sk->sk_wq);
3123 if (skwq_has_sleeper(wq))
3124 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3125 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3129 void sock_def_readable(struct sock *sk)
3131 struct socket_wq *wq;
3134 wq = rcu_dereference(sk->sk_wq);
3135 if (skwq_has_sleeper(wq))
3136 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3137 EPOLLRDNORM | EPOLLRDBAND);
3138 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3142 static void sock_def_write_space(struct sock *sk)
3144 struct socket_wq *wq;
3148 /* Do not wake up a writer until he can make "significant"
3151 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
3152 wq = rcu_dereference(sk->sk_wq);
3153 if (skwq_has_sleeper(wq))
3154 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3155 EPOLLWRNORM | EPOLLWRBAND);
3157 /* Should agree with poll, otherwise some programs break */
3158 if (sock_writeable(sk))
3159 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3165 static void sock_def_destruct(struct sock *sk)
3169 void sk_send_sigurg(struct sock *sk)
3171 if (sk->sk_socket && sk->sk_socket->file)
3172 if (send_sigurg(&sk->sk_socket->file->f_owner))
3173 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3175 EXPORT_SYMBOL(sk_send_sigurg);
3177 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3178 unsigned long expires)
3180 if (!mod_timer(timer, expires))
3183 EXPORT_SYMBOL(sk_reset_timer);
3185 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3187 if (del_timer(timer))
3190 EXPORT_SYMBOL(sk_stop_timer);
3192 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3194 if (del_timer_sync(timer))
3197 EXPORT_SYMBOL(sk_stop_timer_sync);
3199 void sock_init_data(struct socket *sock, struct sock *sk)
3202 sk->sk_send_head = NULL;
3204 timer_setup(&sk->sk_timer, NULL, 0);
3206 sk->sk_allocation = GFP_KERNEL;
3207 sk->sk_rcvbuf = sysctl_rmem_default;
3208 sk->sk_sndbuf = sysctl_wmem_default;
3209 sk->sk_state = TCP_CLOSE;
3210 sk_set_socket(sk, sock);
3212 sock_set_flag(sk, SOCK_ZAPPED);
3215 sk->sk_type = sock->type;
3216 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3218 sk->sk_uid = SOCK_INODE(sock)->i_uid;
3220 RCU_INIT_POINTER(sk->sk_wq, NULL);
3221 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3224 rwlock_init(&sk->sk_callback_lock);
3225 if (sk->sk_kern_sock)
3226 lockdep_set_class_and_name(
3227 &sk->sk_callback_lock,
3228 af_kern_callback_keys + sk->sk_family,
3229 af_family_kern_clock_key_strings[sk->sk_family]);
3231 lockdep_set_class_and_name(
3232 &sk->sk_callback_lock,
3233 af_callback_keys + sk->sk_family,
3234 af_family_clock_key_strings[sk->sk_family]);
3236 sk->sk_state_change = sock_def_wakeup;
3237 sk->sk_data_ready = sock_def_readable;
3238 sk->sk_write_space = sock_def_write_space;
3239 sk->sk_error_report = sock_def_error_report;
3240 sk->sk_destruct = sock_def_destruct;
3242 sk->sk_frag.page = NULL;
3243 sk->sk_frag.offset = 0;
3244 sk->sk_peek_off = -1;
3246 sk->sk_peer_pid = NULL;
3247 sk->sk_peer_cred = NULL;
3248 spin_lock_init(&sk->sk_peer_lock);
3250 sk->sk_write_pending = 0;
3251 sk->sk_rcvlowat = 1;
3252 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3253 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3255 sk->sk_stamp = SK_DEFAULT_STAMP;
3256 #if BITS_PER_LONG==32
3257 seqlock_init(&sk->sk_stamp_seq);
3259 atomic_set(&sk->sk_zckey, 0);
3261 #ifdef CONFIG_NET_RX_BUSY_POLL
3263 sk->sk_ll_usec = sysctl_net_busy_read;
3266 sk->sk_max_pacing_rate = ~0UL;
3267 sk->sk_pacing_rate = ~0UL;
3268 WRITE_ONCE(sk->sk_pacing_shift, 10);
3269 sk->sk_incoming_cpu = -1;
3271 sk_rx_queue_clear(sk);
3273 * Before updating sk_refcnt, we must commit prior changes to memory
3274 * (Documentation/RCU/rculist_nulls.rst for details)
3277 refcount_set(&sk->sk_refcnt, 1);
3278 atomic_set(&sk->sk_drops, 0);
3280 EXPORT_SYMBOL(sock_init_data);
3282 void lock_sock_nested(struct sock *sk, int subclass)
3284 /* The sk_lock has mutex_lock() semantics here. */
3285 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3288 spin_lock_bh(&sk->sk_lock.slock);
3289 if (sk->sk_lock.owned)
3291 sk->sk_lock.owned = 1;
3292 spin_unlock_bh(&sk->sk_lock.slock);
3294 EXPORT_SYMBOL(lock_sock_nested);
3296 void release_sock(struct sock *sk)
3298 spin_lock_bh(&sk->sk_lock.slock);
3299 if (sk->sk_backlog.tail)
3302 /* Warning : release_cb() might need to release sk ownership,
3303 * ie call sock_release_ownership(sk) before us.
3305 if (sk->sk_prot->release_cb)
3306 sk->sk_prot->release_cb(sk);
3308 sock_release_ownership(sk);
3309 if (waitqueue_active(&sk->sk_lock.wq))
3310 wake_up(&sk->sk_lock.wq);
3311 spin_unlock_bh(&sk->sk_lock.slock);
3313 EXPORT_SYMBOL(release_sock);
3315 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3318 spin_lock_bh(&sk->sk_lock.slock);
3320 if (!sk->sk_lock.owned) {
3322 * Fast path return with bottom halves disabled and
3323 * sock::sk_lock.slock held.
3325 * The 'mutex' is not contended and holding
3326 * sock::sk_lock.slock prevents all other lockers to
3327 * proceed so the corresponding unlock_sock_fast() can
3328 * avoid the slow path of release_sock() completely and
3329 * just release slock.
3331 * From a semantical POV this is equivalent to 'acquiring'
3332 * the 'mutex', hence the corresponding lockdep
3333 * mutex_release() has to happen in the fast path of
3334 * unlock_sock_fast().
3340 sk->sk_lock.owned = 1;
3341 __acquire(&sk->sk_lock.slock);
3342 spin_unlock_bh(&sk->sk_lock.slock);
3345 EXPORT_SYMBOL(__lock_sock_fast);
3347 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3348 bool timeval, bool time32)
3350 struct sock *sk = sock->sk;
3351 struct timespec64 ts;
3353 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3354 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3355 if (ts.tv_sec == -1)
3357 if (ts.tv_sec == 0) {
3358 ktime_t kt = ktime_get_real();
3359 sock_write_timestamp(sk, kt);
3360 ts = ktime_to_timespec64(kt);
3366 #ifdef CONFIG_COMPAT_32BIT_TIME
3368 return put_old_timespec32(&ts, userstamp);
3370 #ifdef CONFIG_SPARC64
3371 /* beware of padding in sparc64 timeval */
3372 if (timeval && !in_compat_syscall()) {
3373 struct __kernel_old_timeval __user tv = {
3374 .tv_sec = ts.tv_sec,
3375 .tv_usec = ts.tv_nsec,
3377 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3382 return put_timespec64(&ts, userstamp);
3384 EXPORT_SYMBOL(sock_gettstamp);
3386 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3388 if (!sock_flag(sk, flag)) {
3389 unsigned long previous_flags = sk->sk_flags;
3391 sock_set_flag(sk, flag);
3393 * we just set one of the two flags which require net
3394 * time stamping, but time stamping might have been on
3395 * already because of the other one
3397 if (sock_needs_netstamp(sk) &&
3398 !(previous_flags & SK_FLAGS_TIMESTAMP))
3399 net_enable_timestamp();
3403 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3404 int level, int type)
3406 struct sock_exterr_skb *serr;
3407 struct sk_buff *skb;
3411 skb = sock_dequeue_err_skb(sk);
3417 msg->msg_flags |= MSG_TRUNC;
3420 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3424 sock_recv_timestamp(msg, sk, skb);
3426 serr = SKB_EXT_ERR(skb);
3427 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3429 msg->msg_flags |= MSG_ERRQUEUE;
3437 EXPORT_SYMBOL(sock_recv_errqueue);
3440 * Get a socket option on an socket.
3442 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3443 * asynchronous errors should be reported by getsockopt. We assume
3444 * this means if you specify SO_ERROR (otherwise whats the point of it).
3446 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3447 char __user *optval, int __user *optlen)
3449 struct sock *sk = sock->sk;
3451 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3453 EXPORT_SYMBOL(sock_common_getsockopt);
3455 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3458 struct sock *sk = sock->sk;
3462 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3463 flags & ~MSG_DONTWAIT, &addr_len);
3465 msg->msg_namelen = addr_len;
3468 EXPORT_SYMBOL(sock_common_recvmsg);
3471 * Set socket options on an inet socket.
3473 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3474 sockptr_t optval, unsigned int optlen)
3476 struct sock *sk = sock->sk;
3478 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3480 EXPORT_SYMBOL(sock_common_setsockopt);
3482 void sk_common_release(struct sock *sk)
3484 if (sk->sk_prot->destroy)
3485 sk->sk_prot->destroy(sk);
3488 * Observation: when sk_common_release is called, processes have
3489 * no access to socket. But net still has.
3490 * Step one, detach it from networking:
3492 * A. Remove from hash tables.
3495 sk->sk_prot->unhash(sk);
3498 * In this point socket cannot receive new packets, but it is possible
3499 * that some packets are in flight because some CPU runs receiver and
3500 * did hash table lookup before we unhashed socket. They will achieve
3501 * receive queue and will be purged by socket destructor.
3503 * Also we still have packets pending on receive queue and probably,
3504 * our own packets waiting in device queues. sock_destroy will drain
3505 * receive queue, but transmitted packets will delay socket destruction
3506 * until the last reference will be released.
3511 xfrm_sk_free_policy(sk);
3513 sk_refcnt_debug_release(sk);
3517 EXPORT_SYMBOL(sk_common_release);
3519 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3521 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3523 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3524 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3525 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3526 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3527 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3528 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3529 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3530 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3531 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3534 #ifdef CONFIG_PROC_FS
3535 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3537 int val[PROTO_INUSE_NR];
3540 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3542 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3544 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3546 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3548 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3550 int cpu, idx = prot->inuse_idx;
3553 for_each_possible_cpu(cpu)
3554 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3556 return res >= 0 ? res : 0;
3558 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3560 static void sock_inuse_add(struct net *net, int val)
3562 this_cpu_add(*net->core.sock_inuse, val);
3565 int sock_inuse_get(struct net *net)
3569 for_each_possible_cpu(cpu)
3570 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3575 EXPORT_SYMBOL_GPL(sock_inuse_get);
3577 static int __net_init sock_inuse_init_net(struct net *net)
3579 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3580 if (net->core.prot_inuse == NULL)
3583 net->core.sock_inuse = alloc_percpu(int);
3584 if (net->core.sock_inuse == NULL)
3590 free_percpu(net->core.prot_inuse);
3594 static void __net_exit sock_inuse_exit_net(struct net *net)
3596 free_percpu(net->core.prot_inuse);
3597 free_percpu(net->core.sock_inuse);
3600 static struct pernet_operations net_inuse_ops = {
3601 .init = sock_inuse_init_net,
3602 .exit = sock_inuse_exit_net,
3605 static __init int net_inuse_init(void)
3607 if (register_pernet_subsys(&net_inuse_ops))
3608 panic("Cannot initialize net inuse counters");
3613 core_initcall(net_inuse_init);
3615 static int assign_proto_idx(struct proto *prot)
3617 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3619 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3620 pr_err("PROTO_INUSE_NR exhausted\n");
3624 set_bit(prot->inuse_idx, proto_inuse_idx);
3628 static void release_proto_idx(struct proto *prot)
3630 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3631 clear_bit(prot->inuse_idx, proto_inuse_idx);
3634 static inline int assign_proto_idx(struct proto *prot)
3639 static inline void release_proto_idx(struct proto *prot)
3643 static void sock_inuse_add(struct net *net, int val)
3648 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3652 kfree(twsk_prot->twsk_slab_name);
3653 twsk_prot->twsk_slab_name = NULL;
3654 kmem_cache_destroy(twsk_prot->twsk_slab);
3655 twsk_prot->twsk_slab = NULL;
3658 static int tw_prot_init(const struct proto *prot)
3660 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3665 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3667 if (!twsk_prot->twsk_slab_name)
3670 twsk_prot->twsk_slab =
3671 kmem_cache_create(twsk_prot->twsk_slab_name,
3672 twsk_prot->twsk_obj_size, 0,
3673 SLAB_ACCOUNT | prot->slab_flags,
3675 if (!twsk_prot->twsk_slab) {
3676 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3684 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3688 kfree(rsk_prot->slab_name);
3689 rsk_prot->slab_name = NULL;
3690 kmem_cache_destroy(rsk_prot->slab);
3691 rsk_prot->slab = NULL;
3694 static int req_prot_init(const struct proto *prot)
3696 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3701 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3703 if (!rsk_prot->slab_name)
3706 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3707 rsk_prot->obj_size, 0,
3708 SLAB_ACCOUNT | prot->slab_flags,
3711 if (!rsk_prot->slab) {
3712 pr_crit("%s: Can't create request sock SLAB cache!\n",
3719 int proto_register(struct proto *prot, int alloc_slab)
3724 prot->slab = kmem_cache_create_usercopy(prot->name,
3726 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3728 prot->useroffset, prot->usersize,
3731 if (prot->slab == NULL) {
3732 pr_crit("%s: Can't create sock SLAB cache!\n",
3737 if (req_prot_init(prot))
3738 goto out_free_request_sock_slab;
3740 if (tw_prot_init(prot))
3741 goto out_free_timewait_sock_slab;
3744 mutex_lock(&proto_list_mutex);
3745 ret = assign_proto_idx(prot);
3747 mutex_unlock(&proto_list_mutex);
3748 goto out_free_timewait_sock_slab;
3750 list_add(&prot->node, &proto_list);
3751 mutex_unlock(&proto_list_mutex);
3754 out_free_timewait_sock_slab:
3756 tw_prot_cleanup(prot->twsk_prot);
3757 out_free_request_sock_slab:
3759 req_prot_cleanup(prot->rsk_prot);
3761 kmem_cache_destroy(prot->slab);
3767 EXPORT_SYMBOL(proto_register);
3769 void proto_unregister(struct proto *prot)
3771 mutex_lock(&proto_list_mutex);
3772 release_proto_idx(prot);
3773 list_del(&prot->node);
3774 mutex_unlock(&proto_list_mutex);
3776 kmem_cache_destroy(prot->slab);
3779 req_prot_cleanup(prot->rsk_prot);
3780 tw_prot_cleanup(prot->twsk_prot);
3782 EXPORT_SYMBOL(proto_unregister);
3784 int sock_load_diag_module(int family, int protocol)
3787 if (!sock_is_registered(family))
3790 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3791 NETLINK_SOCK_DIAG, family);
3795 if (family == AF_INET &&
3796 protocol != IPPROTO_RAW &&
3797 protocol < MAX_INET_PROTOS &&
3798 !rcu_access_pointer(inet_protos[protocol]))
3802 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3803 NETLINK_SOCK_DIAG, family, protocol);
3805 EXPORT_SYMBOL(sock_load_diag_module);
3807 #ifdef CONFIG_PROC_FS
3808 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3809 __acquires(proto_list_mutex)
3811 mutex_lock(&proto_list_mutex);
3812 return seq_list_start_head(&proto_list, *pos);
3815 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3817 return seq_list_next(v, &proto_list, pos);
3820 static void proto_seq_stop(struct seq_file *seq, void *v)
3821 __releases(proto_list_mutex)
3823 mutex_unlock(&proto_list_mutex);
3826 static char proto_method_implemented(const void *method)
3828 return method == NULL ? 'n' : 'y';
3830 static long sock_prot_memory_allocated(struct proto *proto)
3832 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3835 static const char *sock_prot_memory_pressure(struct proto *proto)
3837 return proto->memory_pressure != NULL ?
3838 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3841 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3844 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3845 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3848 sock_prot_inuse_get(seq_file_net(seq), proto),
3849 sock_prot_memory_allocated(proto),
3850 sock_prot_memory_pressure(proto),
3852 proto->slab == NULL ? "no" : "yes",
3853 module_name(proto->owner),
3854 proto_method_implemented(proto->close),
3855 proto_method_implemented(proto->connect),
3856 proto_method_implemented(proto->disconnect),
3857 proto_method_implemented(proto->accept),
3858 proto_method_implemented(proto->ioctl),
3859 proto_method_implemented(proto->init),
3860 proto_method_implemented(proto->destroy),
3861 proto_method_implemented(proto->shutdown),
3862 proto_method_implemented(proto->setsockopt),
3863 proto_method_implemented(proto->getsockopt),
3864 proto_method_implemented(proto->sendmsg),
3865 proto_method_implemented(proto->recvmsg),
3866 proto_method_implemented(proto->sendpage),
3867 proto_method_implemented(proto->bind),
3868 proto_method_implemented(proto->backlog_rcv),
3869 proto_method_implemented(proto->hash),
3870 proto_method_implemented(proto->unhash),
3871 proto_method_implemented(proto->get_port),
3872 proto_method_implemented(proto->enter_memory_pressure));
3875 static int proto_seq_show(struct seq_file *seq, void *v)
3877 if (v == &proto_list)
3878 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3887 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3889 proto_seq_printf(seq, list_entry(v, struct proto, node));
3893 static const struct seq_operations proto_seq_ops = {
3894 .start = proto_seq_start,
3895 .next = proto_seq_next,
3896 .stop = proto_seq_stop,
3897 .show = proto_seq_show,
3900 static __net_init int proto_init_net(struct net *net)
3902 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3903 sizeof(struct seq_net_private)))
3909 static __net_exit void proto_exit_net(struct net *net)
3911 remove_proc_entry("protocols", net->proc_net);
3915 static __net_initdata struct pernet_operations proto_net_ops = {
3916 .init = proto_init_net,
3917 .exit = proto_exit_net,
3920 static int __init proto_init(void)
3922 return register_pernet_subsys(&proto_net_ops);
3925 subsys_initcall(proto_init);
3927 #endif /* PROC_FS */
3929 #ifdef CONFIG_NET_RX_BUSY_POLL
3930 bool sk_busy_loop_end(void *p, unsigned long start_time)
3932 struct sock *sk = p;
3934 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3935 sk_busy_loop_timeout(sk, start_time);
3937 EXPORT_SYMBOL(sk_busy_loop_end);
3938 #endif /* CONFIG_NET_RX_BUSY_POLL */
3940 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3942 if (!sk->sk_prot->bind_add)
3944 return sk->sk_prot->bind_add(sk, addr, addr_len);
3946 EXPORT_SYMBOL(sock_bind_add);