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 static 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;
383 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
386 struct __kernel_sock_timeval tv;
388 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
389 struct old_timeval32 tv32;
391 if (optlen < sizeof(tv32))
394 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
396 tv.tv_sec = tv32.tv_sec;
397 tv.tv_usec = tv32.tv_usec;
398 } else if (old_timeval) {
399 struct __kernel_old_timeval old_tv;
401 if (optlen < sizeof(old_tv))
403 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
405 tv.tv_sec = old_tv.tv_sec;
406 tv.tv_usec = old_tv.tv_usec;
408 if (optlen < sizeof(tv))
410 if (copy_from_sockptr(&tv, optval, sizeof(tv)))
413 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
417 static int warned __read_mostly;
420 if (warned < 10 && net_ratelimit()) {
422 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
423 __func__, current->comm, task_pid_nr(current));
427 *timeo_p = MAX_SCHEDULE_TIMEOUT;
428 if (tv.tv_sec == 0 && tv.tv_usec == 0)
430 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
431 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
435 static bool sock_needs_netstamp(const struct sock *sk)
437 switch (sk->sk_family) {
446 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
448 if (sk->sk_flags & flags) {
449 sk->sk_flags &= ~flags;
450 if (sock_needs_netstamp(sk) &&
451 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
452 net_disable_timestamp();
457 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
460 struct sk_buff_head *list = &sk->sk_receive_queue;
462 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
463 atomic_inc(&sk->sk_drops);
464 trace_sock_rcvqueue_full(sk, skb);
468 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
469 atomic_inc(&sk->sk_drops);
474 skb_set_owner_r(skb, sk);
476 /* we escape from rcu protected region, make sure we dont leak
481 spin_lock_irqsave(&list->lock, flags);
482 sock_skb_set_dropcount(sk, skb);
483 __skb_queue_tail(list, skb);
484 spin_unlock_irqrestore(&list->lock, flags);
486 if (!sock_flag(sk, SOCK_DEAD))
487 sk->sk_data_ready(sk);
490 EXPORT_SYMBOL(__sock_queue_rcv_skb);
492 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
496 err = sk_filter(sk, skb);
500 return __sock_queue_rcv_skb(sk, skb);
502 EXPORT_SYMBOL(sock_queue_rcv_skb);
504 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
505 const int nested, unsigned int trim_cap, bool refcounted)
507 int rc = NET_RX_SUCCESS;
509 if (sk_filter_trim_cap(sk, skb, trim_cap))
510 goto discard_and_relse;
514 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
515 atomic_inc(&sk->sk_drops);
516 goto discard_and_relse;
519 bh_lock_sock_nested(sk);
522 if (!sock_owned_by_user(sk)) {
524 * trylock + unlock semantics:
526 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
528 rc = sk_backlog_rcv(sk, skb);
530 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
531 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
533 atomic_inc(&sk->sk_drops);
534 goto discard_and_relse;
546 EXPORT_SYMBOL(__sk_receive_skb);
548 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
550 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
552 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
554 struct dst_entry *dst = __sk_dst_get(sk);
556 if (dst && dst->obsolete &&
557 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
558 dst, cookie) == NULL) {
559 sk_tx_queue_clear(sk);
560 sk->sk_dst_pending_confirm = 0;
561 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
568 EXPORT_SYMBOL(__sk_dst_check);
570 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
572 struct dst_entry *dst = sk_dst_get(sk);
574 if (dst && dst->obsolete &&
575 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
576 dst, cookie) == NULL) {
584 EXPORT_SYMBOL(sk_dst_check);
586 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
588 int ret = -ENOPROTOOPT;
589 #ifdef CONFIG_NETDEVICES
590 struct net *net = sock_net(sk);
594 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
601 sk->sk_bound_dev_if = ifindex;
602 if (sk->sk_prot->rehash)
603 sk->sk_prot->rehash(sk);
614 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
620 ret = sock_bindtoindex_locked(sk, ifindex);
626 EXPORT_SYMBOL(sock_bindtoindex);
628 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
630 int ret = -ENOPROTOOPT;
631 #ifdef CONFIG_NETDEVICES
632 struct net *net = sock_net(sk);
633 char devname[IFNAMSIZ];
640 /* Bind this socket to a particular device like "eth0",
641 * as specified in the passed interface name. If the
642 * name is "" or the option length is zero the socket
645 if (optlen > IFNAMSIZ - 1)
646 optlen = IFNAMSIZ - 1;
647 memset(devname, 0, sizeof(devname));
650 if (copy_from_sockptr(devname, optval, optlen))
654 if (devname[0] != '\0') {
655 struct net_device *dev;
658 dev = dev_get_by_name_rcu(net, devname);
660 index = dev->ifindex;
667 return sock_bindtoindex(sk, index, true);
674 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
675 int __user *optlen, int len)
677 int ret = -ENOPROTOOPT;
678 #ifdef CONFIG_NETDEVICES
679 struct net *net = sock_net(sk);
680 char devname[IFNAMSIZ];
682 if (sk->sk_bound_dev_if == 0) {
691 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
695 len = strlen(devname) + 1;
698 if (copy_to_user(optval, devname, len))
703 if (put_user(len, optlen))
714 bool sk_mc_loop(struct sock *sk)
716 if (dev_recursion_level())
720 switch (sk->sk_family) {
722 return inet_sk(sk)->mc_loop;
723 #if IS_ENABLED(CONFIG_IPV6)
725 return inet6_sk(sk)->mc_loop;
731 EXPORT_SYMBOL(sk_mc_loop);
733 void sock_set_reuseaddr(struct sock *sk)
736 sk->sk_reuse = SK_CAN_REUSE;
739 EXPORT_SYMBOL(sock_set_reuseaddr);
741 void sock_set_reuseport(struct sock *sk)
744 sk->sk_reuseport = true;
747 EXPORT_SYMBOL(sock_set_reuseport);
749 void sock_no_linger(struct sock *sk)
752 sk->sk_lingertime = 0;
753 sock_set_flag(sk, SOCK_LINGER);
756 EXPORT_SYMBOL(sock_no_linger);
758 void sock_set_priority(struct sock *sk, u32 priority)
761 sk->sk_priority = priority;
764 EXPORT_SYMBOL(sock_set_priority);
766 void sock_set_sndtimeo(struct sock *sk, s64 secs)
769 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
770 sk->sk_sndtimeo = secs * HZ;
772 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
775 EXPORT_SYMBOL(sock_set_sndtimeo);
777 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
780 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
781 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
782 sock_set_flag(sk, SOCK_RCVTSTAMP);
783 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
785 sock_reset_flag(sk, SOCK_RCVTSTAMP);
786 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
790 void sock_enable_timestamps(struct sock *sk)
793 __sock_set_timestamps(sk, true, false, true);
796 EXPORT_SYMBOL(sock_enable_timestamps);
798 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
801 case SO_TIMESTAMP_OLD:
802 __sock_set_timestamps(sk, valbool, false, false);
804 case SO_TIMESTAMP_NEW:
805 __sock_set_timestamps(sk, valbool, true, false);
807 case SO_TIMESTAMPNS_OLD:
808 __sock_set_timestamps(sk, valbool, false, true);
810 case SO_TIMESTAMPNS_NEW:
811 __sock_set_timestamps(sk, valbool, true, true);
816 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
818 struct net *net = sock_net(sk);
819 struct net_device *dev = NULL;
824 if (sk->sk_bound_dev_if)
825 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
828 pr_err("%s: sock not bind to device\n", __func__);
832 num = ethtool_get_phc_vclocks(dev, &vclock_index);
833 for (i = 0; i < num; i++) {
834 if (*(vclock_index + i) == phc_index) {
846 sk->sk_bind_phc = phc_index;
851 int sock_set_timestamping(struct sock *sk, int optname,
852 struct so_timestamping timestamping)
854 int val = timestamping.flags;
857 if (val & ~SOF_TIMESTAMPING_MASK)
860 if (val & SOF_TIMESTAMPING_OPT_ID &&
861 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
862 if (sk->sk_protocol == IPPROTO_TCP &&
863 sk->sk_type == SOCK_STREAM) {
864 if ((1 << sk->sk_state) &
865 (TCPF_CLOSE | TCPF_LISTEN))
867 sk->sk_tskey = tcp_sk(sk)->snd_una;
873 if (val & SOF_TIMESTAMPING_OPT_STATS &&
874 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
877 if (val & SOF_TIMESTAMPING_BIND_PHC) {
878 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
883 sk->sk_tsflags = val;
884 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
886 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
887 sock_enable_timestamp(sk,
888 SOCK_TIMESTAMPING_RX_SOFTWARE);
890 sock_disable_timestamp(sk,
891 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
895 void sock_set_keepalive(struct sock *sk)
898 if (sk->sk_prot->keepalive)
899 sk->sk_prot->keepalive(sk, true);
900 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
903 EXPORT_SYMBOL(sock_set_keepalive);
905 static void __sock_set_rcvbuf(struct sock *sk, int val)
907 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
908 * as a negative value.
910 val = min_t(int, val, INT_MAX / 2);
911 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
913 /* We double it on the way in to account for "struct sk_buff" etc.
914 * overhead. Applications assume that the SO_RCVBUF setting they make
915 * will allow that much actual data to be received on that socket.
917 * Applications are unaware that "struct sk_buff" and other overheads
918 * allocate from the receive buffer during socket buffer allocation.
920 * And after considering the possible alternatives, returning the value
921 * we actually used in getsockopt is the most desirable behavior.
923 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
926 void sock_set_rcvbuf(struct sock *sk, int val)
929 __sock_set_rcvbuf(sk, val);
932 EXPORT_SYMBOL(sock_set_rcvbuf);
934 static void __sock_set_mark(struct sock *sk, u32 val)
936 if (val != sk->sk_mark) {
942 void sock_set_mark(struct sock *sk, u32 val)
945 __sock_set_mark(sk, val);
948 EXPORT_SYMBOL(sock_set_mark);
950 static void sock_release_reserved_memory(struct sock *sk, int bytes)
952 /* Round down bytes to multiple of pages */
953 bytes &= ~(SK_MEM_QUANTUM - 1);
955 WARN_ON(bytes > sk->sk_reserved_mem);
956 sk->sk_reserved_mem -= bytes;
960 static int sock_reserve_memory(struct sock *sk, int bytes)
966 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg)
972 pages = sk_mem_pages(bytes);
974 /* pre-charge to memcg */
975 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
976 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
980 /* pre-charge to forward_alloc */
981 allocated = sk_memory_allocated_add(sk, pages);
982 /* If the system goes into memory pressure with this
983 * precharge, give up and return error.
985 if (allocated > sk_prot_mem_limits(sk, 1)) {
986 sk_memory_allocated_sub(sk, pages);
987 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
990 sk->sk_forward_alloc += pages << SK_MEM_QUANTUM_SHIFT;
992 sk->sk_reserved_mem += pages << SK_MEM_QUANTUM_SHIFT;
998 * This is meant for all protocols to use and covers goings on
999 * at the socket level. Everything here is generic.
1002 int sock_setsockopt(struct socket *sock, int level, int optname,
1003 sockptr_t optval, unsigned int optlen)
1005 struct so_timestamping timestamping;
1006 struct sock_txtime sk_txtime;
1007 struct sock *sk = sock->sk;
1014 * Options without arguments
1017 if (optname == SO_BINDTODEVICE)
1018 return sock_setbindtodevice(sk, optval, optlen);
1020 if (optlen < sizeof(int))
1023 if (copy_from_sockptr(&val, optval, sizeof(val)))
1026 valbool = val ? 1 : 0;
1032 if (val && !capable(CAP_NET_ADMIN))
1035 sock_valbool_flag(sk, SOCK_DBG, valbool);
1038 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1041 sk->sk_reuseport = valbool;
1050 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1054 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1057 /* Don't error on this BSD doesn't and if you think
1058 * about it this is right. Otherwise apps have to
1059 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1060 * are treated in BSD as hints
1062 val = min_t(u32, val, sysctl_wmem_max);
1064 /* Ensure val * 2 fits into an int, to prevent max_t()
1065 * from treating it as a negative value.
1067 val = min_t(int, val, INT_MAX / 2);
1068 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1069 WRITE_ONCE(sk->sk_sndbuf,
1070 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1071 /* Wake up sending tasks if we upped the value. */
1072 sk->sk_write_space(sk);
1075 case SO_SNDBUFFORCE:
1076 if (!capable(CAP_NET_ADMIN)) {
1081 /* No negative values (to prevent underflow, as val will be
1089 /* Don't error on this BSD doesn't and if you think
1090 * about it this is right. Otherwise apps have to
1091 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1092 * are treated in BSD as hints
1094 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
1097 case SO_RCVBUFFORCE:
1098 if (!capable(CAP_NET_ADMIN)) {
1103 /* No negative values (to prevent underflow, as val will be
1106 __sock_set_rcvbuf(sk, max(val, 0));
1110 if (sk->sk_prot->keepalive)
1111 sk->sk_prot->keepalive(sk, valbool);
1112 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1116 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1120 sk->sk_no_check_tx = valbool;
1124 if ((val >= 0 && val <= 6) ||
1125 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1126 sk->sk_priority = val;
1132 if (optlen < sizeof(ling)) {
1133 ret = -EINVAL; /* 1003.1g */
1136 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1141 sock_reset_flag(sk, SOCK_LINGER);
1143 #if (BITS_PER_LONG == 32)
1144 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1145 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1148 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1149 sock_set_flag(sk, SOCK_LINGER);
1158 set_bit(SOCK_PASSCRED, &sock->flags);
1160 clear_bit(SOCK_PASSCRED, &sock->flags);
1163 case SO_TIMESTAMP_OLD:
1164 case SO_TIMESTAMP_NEW:
1165 case SO_TIMESTAMPNS_OLD:
1166 case SO_TIMESTAMPNS_NEW:
1167 sock_set_timestamp(sk, optname, valbool);
1170 case SO_TIMESTAMPING_NEW:
1171 case SO_TIMESTAMPING_OLD:
1172 if (optlen == sizeof(timestamping)) {
1173 if (copy_from_sockptr(×tamping, optval,
1174 sizeof(timestamping))) {
1179 memset(×tamping, 0, sizeof(timestamping));
1180 timestamping.flags = val;
1182 ret = sock_set_timestamping(sk, optname, timestamping);
1188 if (sock->ops->set_rcvlowat)
1189 ret = sock->ops->set_rcvlowat(sk, val);
1191 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1194 case SO_RCVTIMEO_OLD:
1195 case SO_RCVTIMEO_NEW:
1196 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1197 optlen, optname == SO_RCVTIMEO_OLD);
1200 case SO_SNDTIMEO_OLD:
1201 case SO_SNDTIMEO_NEW:
1202 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1203 optlen, optname == SO_SNDTIMEO_OLD);
1206 case SO_ATTACH_FILTER: {
1207 struct sock_fprog fprog;
1209 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1211 ret = sk_attach_filter(&fprog, sk);
1216 if (optlen == sizeof(u32)) {
1220 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1223 ret = sk_attach_bpf(ufd, sk);
1227 case SO_ATTACH_REUSEPORT_CBPF: {
1228 struct sock_fprog fprog;
1230 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1232 ret = sk_reuseport_attach_filter(&fprog, sk);
1235 case SO_ATTACH_REUSEPORT_EBPF:
1237 if (optlen == sizeof(u32)) {
1241 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1244 ret = sk_reuseport_attach_bpf(ufd, sk);
1248 case SO_DETACH_REUSEPORT_BPF:
1249 ret = reuseport_detach_prog(sk);
1252 case SO_DETACH_FILTER:
1253 ret = sk_detach_filter(sk);
1256 case SO_LOCK_FILTER:
1257 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1260 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1265 set_bit(SOCK_PASSSEC, &sock->flags);
1267 clear_bit(SOCK_PASSSEC, &sock->flags);
1270 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1275 __sock_set_mark(sk, val);
1279 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1282 case SO_WIFI_STATUS:
1283 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1287 if (sock->ops->set_peek_off)
1288 ret = sock->ops->set_peek_off(sk, val);
1294 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1297 case SO_SELECT_ERR_QUEUE:
1298 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1301 #ifdef CONFIG_NET_RX_BUSY_POLL
1303 /* allow unprivileged users to decrease the value */
1304 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1310 WRITE_ONCE(sk->sk_ll_usec, val);
1313 case SO_PREFER_BUSY_POLL:
1314 if (valbool && !capable(CAP_NET_ADMIN))
1317 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1319 case SO_BUSY_POLL_BUDGET:
1320 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1323 if (val < 0 || val > U16_MAX)
1326 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1331 case SO_MAX_PACING_RATE:
1333 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1335 if (sizeof(ulval) != sizeof(val) &&
1336 optlen >= sizeof(ulval) &&
1337 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1342 cmpxchg(&sk->sk_pacing_status,
1345 sk->sk_max_pacing_rate = ulval;
1346 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1349 case SO_INCOMING_CPU:
1350 WRITE_ONCE(sk->sk_incoming_cpu, val);
1355 dst_negative_advice(sk);
1359 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1360 if (!((sk->sk_type == SOCK_STREAM &&
1361 sk->sk_protocol == IPPROTO_TCP) ||
1362 (sk->sk_type == SOCK_DGRAM &&
1363 sk->sk_protocol == IPPROTO_UDP)))
1365 } else if (sk->sk_family != PF_RDS) {
1369 if (val < 0 || val > 1)
1372 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1377 if (optlen != sizeof(struct sock_txtime)) {
1380 } else if (copy_from_sockptr(&sk_txtime, optval,
1381 sizeof(struct sock_txtime))) {
1384 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1388 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1389 * scheduler has enough safe guards.
1391 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1392 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1396 sock_valbool_flag(sk, SOCK_TXTIME, true);
1397 sk->sk_clockid = sk_txtime.clockid;
1398 sk->sk_txtime_deadline_mode =
1399 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1400 sk->sk_txtime_report_errors =
1401 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1404 case SO_BINDTOIFINDEX:
1405 ret = sock_bindtoindex_locked(sk, val);
1409 if (val & ~SOCK_BUF_LOCK_MASK) {
1413 sk->sk_userlocks = val | (sk->sk_userlocks &
1414 ~SOCK_BUF_LOCK_MASK);
1417 case SO_RESERVE_MEM:
1426 delta = val - sk->sk_reserved_mem;
1428 sock_release_reserved_memory(sk, -delta);
1430 ret = sock_reserve_memory(sk, delta);
1441 EXPORT_SYMBOL(sock_setsockopt);
1443 static const struct cred *sk_get_peer_cred(struct sock *sk)
1445 const struct cred *cred;
1447 spin_lock(&sk->sk_peer_lock);
1448 cred = get_cred(sk->sk_peer_cred);
1449 spin_unlock(&sk->sk_peer_lock);
1454 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1455 struct ucred *ucred)
1457 ucred->pid = pid_vnr(pid);
1458 ucred->uid = ucred->gid = -1;
1460 struct user_namespace *current_ns = current_user_ns();
1462 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1463 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1467 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1469 struct user_namespace *user_ns = current_user_ns();
1472 for (i = 0; i < src->ngroups; i++)
1473 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1479 int sock_getsockopt(struct socket *sock, int level, int optname,
1480 char __user *optval, int __user *optlen)
1482 struct sock *sk = sock->sk;
1487 unsigned long ulval;
1489 struct old_timeval32 tm32;
1490 struct __kernel_old_timeval tm;
1491 struct __kernel_sock_timeval stm;
1492 struct sock_txtime txtime;
1493 struct so_timestamping timestamping;
1496 int lv = sizeof(int);
1499 if (get_user(len, optlen))
1504 memset(&v, 0, sizeof(v));
1508 v.val = sock_flag(sk, SOCK_DBG);
1512 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1516 v.val = sock_flag(sk, SOCK_BROADCAST);
1520 v.val = sk->sk_sndbuf;
1524 v.val = sk->sk_rcvbuf;
1528 v.val = sk->sk_reuse;
1532 v.val = sk->sk_reuseport;
1536 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1540 v.val = sk->sk_type;
1544 v.val = sk->sk_protocol;
1548 v.val = sk->sk_family;
1552 v.val = -sock_error(sk);
1554 v.val = xchg(&sk->sk_err_soft, 0);
1558 v.val = sock_flag(sk, SOCK_URGINLINE);
1562 v.val = sk->sk_no_check_tx;
1566 v.val = sk->sk_priority;
1570 lv = sizeof(v.ling);
1571 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1572 v.ling.l_linger = sk->sk_lingertime / HZ;
1578 case SO_TIMESTAMP_OLD:
1579 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1580 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1581 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1584 case SO_TIMESTAMPNS_OLD:
1585 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1588 case SO_TIMESTAMP_NEW:
1589 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1592 case SO_TIMESTAMPNS_NEW:
1593 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1596 case SO_TIMESTAMPING_OLD:
1597 lv = sizeof(v.timestamping);
1598 v.timestamping.flags = sk->sk_tsflags;
1599 v.timestamping.bind_phc = sk->sk_bind_phc;
1602 case SO_RCVTIMEO_OLD:
1603 case SO_RCVTIMEO_NEW:
1604 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1607 case SO_SNDTIMEO_OLD:
1608 case SO_SNDTIMEO_NEW:
1609 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1613 v.val = sk->sk_rcvlowat;
1621 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1626 struct ucred peercred;
1627 if (len > sizeof(peercred))
1628 len = sizeof(peercred);
1630 spin_lock(&sk->sk_peer_lock);
1631 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1632 spin_unlock(&sk->sk_peer_lock);
1634 if (copy_to_user(optval, &peercred, len))
1641 const struct cred *cred;
1644 cred = sk_get_peer_cred(sk);
1648 n = cred->group_info->ngroups;
1649 if (len < n * sizeof(gid_t)) {
1650 len = n * sizeof(gid_t);
1652 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1654 len = n * sizeof(gid_t);
1656 ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1667 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1672 if (copy_to_user(optval, address, len))
1677 /* Dubious BSD thing... Probably nobody even uses it, but
1678 * the UNIX standard wants it for whatever reason... -DaveM
1681 v.val = sk->sk_state == TCP_LISTEN;
1685 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1689 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1692 v.val = sk->sk_mark;
1696 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1699 case SO_WIFI_STATUS:
1700 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1704 if (!sock->ops->set_peek_off)
1707 v.val = sk->sk_peek_off;
1710 v.val = sock_flag(sk, SOCK_NOFCS);
1713 case SO_BINDTODEVICE:
1714 return sock_getbindtodevice(sk, optval, optlen, len);
1717 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1723 case SO_LOCK_FILTER:
1724 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1727 case SO_BPF_EXTENSIONS:
1728 v.val = bpf_tell_extensions();
1731 case SO_SELECT_ERR_QUEUE:
1732 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1735 #ifdef CONFIG_NET_RX_BUSY_POLL
1737 v.val = sk->sk_ll_usec;
1739 case SO_PREFER_BUSY_POLL:
1740 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1744 case SO_MAX_PACING_RATE:
1745 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1746 lv = sizeof(v.ulval);
1747 v.ulval = sk->sk_max_pacing_rate;
1750 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1754 case SO_INCOMING_CPU:
1755 v.val = READ_ONCE(sk->sk_incoming_cpu);
1760 u32 meminfo[SK_MEMINFO_VARS];
1762 sk_get_meminfo(sk, meminfo);
1764 len = min_t(unsigned int, len, sizeof(meminfo));
1765 if (copy_to_user(optval, &meminfo, len))
1771 #ifdef CONFIG_NET_RX_BUSY_POLL
1772 case SO_INCOMING_NAPI_ID:
1773 v.val = READ_ONCE(sk->sk_napi_id);
1775 /* aggregate non-NAPI IDs down to 0 */
1776 if (v.val < MIN_NAPI_ID)
1786 v.val64 = sock_gen_cookie(sk);
1790 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1794 lv = sizeof(v.txtime);
1795 v.txtime.clockid = sk->sk_clockid;
1796 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1797 SOF_TXTIME_DEADLINE_MODE : 0;
1798 v.txtime.flags |= sk->sk_txtime_report_errors ?
1799 SOF_TXTIME_REPORT_ERRORS : 0;
1802 case SO_BINDTOIFINDEX:
1803 v.val = sk->sk_bound_dev_if;
1806 case SO_NETNS_COOKIE:
1810 v.val64 = sock_net(sk)->net_cookie;
1814 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1817 case SO_RESERVE_MEM:
1818 v.val = sk->sk_reserved_mem;
1822 /* We implement the SO_SNDLOWAT etc to not be settable
1825 return -ENOPROTOOPT;
1830 if (copy_to_user(optval, &v, len))
1833 if (put_user(len, optlen))
1839 * Initialize an sk_lock.
1841 * (We also register the sk_lock with the lock validator.)
1843 static inline void sock_lock_init(struct sock *sk)
1845 if (sk->sk_kern_sock)
1846 sock_lock_init_class_and_name(
1848 af_family_kern_slock_key_strings[sk->sk_family],
1849 af_family_kern_slock_keys + sk->sk_family,
1850 af_family_kern_key_strings[sk->sk_family],
1851 af_family_kern_keys + sk->sk_family);
1853 sock_lock_init_class_and_name(
1855 af_family_slock_key_strings[sk->sk_family],
1856 af_family_slock_keys + sk->sk_family,
1857 af_family_key_strings[sk->sk_family],
1858 af_family_keys + sk->sk_family);
1862 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1863 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1864 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1866 static void sock_copy(struct sock *nsk, const struct sock *osk)
1868 const struct proto *prot = READ_ONCE(osk->sk_prot);
1869 #ifdef CONFIG_SECURITY_NETWORK
1870 void *sptr = nsk->sk_security;
1873 /* If we move sk_tx_queue_mapping out of the private section,
1874 * we must check if sk_tx_queue_clear() is called after
1875 * sock_copy() in sk_clone_lock().
1877 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1878 offsetof(struct sock, sk_dontcopy_begin) ||
1879 offsetof(struct sock, sk_tx_queue_mapping) >=
1880 offsetof(struct sock, sk_dontcopy_end));
1882 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1884 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1885 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1887 #ifdef CONFIG_SECURITY_NETWORK
1888 nsk->sk_security = sptr;
1889 security_sk_clone(osk, nsk);
1893 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1897 struct kmem_cache *slab;
1901 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1904 if (want_init_on_alloc(priority))
1905 sk_prot_clear_nulls(sk, prot->obj_size);
1907 sk = kmalloc(prot->obj_size, priority);
1910 if (security_sk_alloc(sk, family, priority))
1913 if (!try_module_get(prot->owner))
1920 security_sk_free(sk);
1923 kmem_cache_free(slab, sk);
1929 static void sk_prot_free(struct proto *prot, struct sock *sk)
1931 struct kmem_cache *slab;
1932 struct module *owner;
1934 owner = prot->owner;
1937 cgroup_sk_free(&sk->sk_cgrp_data);
1938 mem_cgroup_sk_free(sk);
1939 security_sk_free(sk);
1941 kmem_cache_free(slab, sk);
1948 * sk_alloc - All socket objects are allocated here
1949 * @net: the applicable net namespace
1950 * @family: protocol family
1951 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1952 * @prot: struct proto associated with this new sock instance
1953 * @kern: is this to be a kernel socket?
1955 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1956 struct proto *prot, int kern)
1960 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1962 sk->sk_family = family;
1964 * See comment in struct sock definition to understand
1965 * why we need sk_prot_creator -acme
1967 sk->sk_prot = sk->sk_prot_creator = prot;
1968 sk->sk_kern_sock = kern;
1970 sk->sk_net_refcnt = kern ? 0 : 1;
1971 if (likely(sk->sk_net_refcnt)) {
1973 sock_inuse_add(net, 1);
1976 sock_net_set(sk, net);
1977 refcount_set(&sk->sk_wmem_alloc, 1);
1979 mem_cgroup_sk_alloc(sk);
1980 cgroup_sk_alloc(&sk->sk_cgrp_data);
1981 sock_update_classid(&sk->sk_cgrp_data);
1982 sock_update_netprioidx(&sk->sk_cgrp_data);
1983 sk_tx_queue_clear(sk);
1988 EXPORT_SYMBOL(sk_alloc);
1990 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1991 * grace period. This is the case for UDP sockets and TCP listeners.
1993 static void __sk_destruct(struct rcu_head *head)
1995 struct sock *sk = container_of(head, struct sock, sk_rcu);
1996 struct sk_filter *filter;
1998 if (sk->sk_destruct)
1999 sk->sk_destruct(sk);
2001 filter = rcu_dereference_check(sk->sk_filter,
2002 refcount_read(&sk->sk_wmem_alloc) == 0);
2004 sk_filter_uncharge(sk, filter);
2005 RCU_INIT_POINTER(sk->sk_filter, NULL);
2008 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2010 #ifdef CONFIG_BPF_SYSCALL
2011 bpf_sk_storage_free(sk);
2014 if (atomic_read(&sk->sk_omem_alloc))
2015 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2016 __func__, atomic_read(&sk->sk_omem_alloc));
2018 if (sk->sk_frag.page) {
2019 put_page(sk->sk_frag.page);
2020 sk->sk_frag.page = NULL;
2023 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2024 put_cred(sk->sk_peer_cred);
2025 put_pid(sk->sk_peer_pid);
2027 if (likely(sk->sk_net_refcnt))
2028 put_net(sock_net(sk));
2029 sk_prot_free(sk->sk_prot_creator, sk);
2032 void sk_destruct(struct sock *sk)
2034 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2036 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2037 reuseport_detach_sock(sk);
2038 use_call_rcu = true;
2042 call_rcu(&sk->sk_rcu, __sk_destruct);
2044 __sk_destruct(&sk->sk_rcu);
2047 static void __sk_free(struct sock *sk)
2049 if (likely(sk->sk_net_refcnt))
2050 sock_inuse_add(sock_net(sk), -1);
2052 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2053 sock_diag_broadcast_destroy(sk);
2058 void sk_free(struct sock *sk)
2061 * We subtract one from sk_wmem_alloc and can know if
2062 * some packets are still in some tx queue.
2063 * If not null, sock_wfree() will call __sk_free(sk) later
2065 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2068 EXPORT_SYMBOL(sk_free);
2070 static void sk_init_common(struct sock *sk)
2072 skb_queue_head_init(&sk->sk_receive_queue);
2073 skb_queue_head_init(&sk->sk_write_queue);
2074 skb_queue_head_init(&sk->sk_error_queue);
2076 rwlock_init(&sk->sk_callback_lock);
2077 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2078 af_rlock_keys + sk->sk_family,
2079 af_family_rlock_key_strings[sk->sk_family]);
2080 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2081 af_wlock_keys + sk->sk_family,
2082 af_family_wlock_key_strings[sk->sk_family]);
2083 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2084 af_elock_keys + sk->sk_family,
2085 af_family_elock_key_strings[sk->sk_family]);
2086 lockdep_set_class_and_name(&sk->sk_callback_lock,
2087 af_callback_keys + sk->sk_family,
2088 af_family_clock_key_strings[sk->sk_family]);
2092 * sk_clone_lock - clone a socket, and lock its clone
2093 * @sk: the socket to clone
2094 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2096 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2098 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2100 struct proto *prot = READ_ONCE(sk->sk_prot);
2101 struct sk_filter *filter;
2102 bool is_charged = true;
2105 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2109 sock_copy(newsk, sk);
2111 newsk->sk_prot_creator = prot;
2114 if (likely(newsk->sk_net_refcnt))
2115 get_net(sock_net(newsk));
2116 sk_node_init(&newsk->sk_node);
2117 sock_lock_init(newsk);
2118 bh_lock_sock(newsk);
2119 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2120 newsk->sk_backlog.len = 0;
2122 atomic_set(&newsk->sk_rmem_alloc, 0);
2124 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2125 refcount_set(&newsk->sk_wmem_alloc, 1);
2127 atomic_set(&newsk->sk_omem_alloc, 0);
2128 sk_init_common(newsk);
2130 newsk->sk_dst_cache = NULL;
2131 newsk->sk_dst_pending_confirm = 0;
2132 newsk->sk_wmem_queued = 0;
2133 newsk->sk_forward_alloc = 0;
2134 newsk->sk_reserved_mem = 0;
2135 atomic_set(&newsk->sk_drops, 0);
2136 newsk->sk_send_head = NULL;
2137 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2138 atomic_set(&newsk->sk_zckey, 0);
2140 sock_reset_flag(newsk, SOCK_DONE);
2142 /* sk->sk_memcg will be populated at accept() time */
2143 newsk->sk_memcg = NULL;
2145 cgroup_sk_clone(&newsk->sk_cgrp_data);
2148 filter = rcu_dereference(sk->sk_filter);
2150 /* though it's an empty new sock, the charging may fail
2151 * if sysctl_optmem_max was changed between creation of
2152 * original socket and cloning
2154 is_charged = sk_filter_charge(newsk, filter);
2155 RCU_INIT_POINTER(newsk->sk_filter, filter);
2158 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2159 /* We need to make sure that we don't uncharge the new
2160 * socket if we couldn't charge it in the first place
2161 * as otherwise we uncharge the parent's filter.
2164 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2165 sk_free_unlock_clone(newsk);
2169 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2171 if (bpf_sk_storage_clone(sk, newsk)) {
2172 sk_free_unlock_clone(newsk);
2177 /* Clear sk_user_data if parent had the pointer tagged
2178 * as not suitable for copying when cloning.
2180 if (sk_user_data_is_nocopy(newsk))
2181 newsk->sk_user_data = NULL;
2184 newsk->sk_err_soft = 0;
2185 newsk->sk_priority = 0;
2186 newsk->sk_incoming_cpu = raw_smp_processor_id();
2187 if (likely(newsk->sk_net_refcnt))
2188 sock_inuse_add(sock_net(newsk), 1);
2190 /* Before updating sk_refcnt, we must commit prior changes to memory
2191 * (Documentation/RCU/rculist_nulls.rst for details)
2194 refcount_set(&newsk->sk_refcnt, 2);
2196 /* Increment the counter in the same struct proto as the master
2197 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2198 * is the same as sk->sk_prot->socks, as this field was copied
2201 * This _changes_ the previous behaviour, where
2202 * tcp_create_openreq_child always was incrementing the
2203 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2204 * to be taken into account in all callers. -acme
2206 sk_refcnt_debug_inc(newsk);
2207 sk_set_socket(newsk, NULL);
2208 sk_tx_queue_clear(newsk);
2209 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2211 if (newsk->sk_prot->sockets_allocated)
2212 sk_sockets_allocated_inc(newsk);
2214 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2215 net_enable_timestamp();
2219 EXPORT_SYMBOL_GPL(sk_clone_lock);
2221 void sk_free_unlock_clone(struct sock *sk)
2223 /* It is still raw copy of parent, so invalidate
2224 * destructor and make plain sk_free() */
2225 sk->sk_destruct = NULL;
2229 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2231 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2235 sk_dst_set(sk, dst);
2236 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2237 if (sk->sk_route_caps & NETIF_F_GSO)
2238 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2239 sk->sk_route_caps &= ~sk->sk_route_nocaps;
2240 if (sk_can_gso(sk)) {
2241 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2242 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2244 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2245 sk->sk_gso_max_size = dst->dev->gso_max_size;
2246 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2249 sk->sk_gso_max_segs = max_segs;
2251 EXPORT_SYMBOL_GPL(sk_setup_caps);
2254 * Simple resource managers for sockets.
2259 * Write buffer destructor automatically called from kfree_skb.
2261 void sock_wfree(struct sk_buff *skb)
2263 struct sock *sk = skb->sk;
2264 unsigned int len = skb->truesize;
2266 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2268 * Keep a reference on sk_wmem_alloc, this will be released
2269 * after sk_write_space() call
2271 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2272 sk->sk_write_space(sk);
2276 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2277 * could not do because of in-flight packets
2279 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2282 EXPORT_SYMBOL(sock_wfree);
2284 /* This variant of sock_wfree() is used by TCP,
2285 * since it sets SOCK_USE_WRITE_QUEUE.
2287 void __sock_wfree(struct sk_buff *skb)
2289 struct sock *sk = skb->sk;
2291 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2295 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2300 if (unlikely(!sk_fullsock(sk))) {
2301 skb->destructor = sock_edemux;
2306 skb->destructor = sock_wfree;
2307 skb_set_hash_from_sk(skb, sk);
2309 * We used to take a refcount on sk, but following operation
2310 * is enough to guarantee sk_free() wont free this sock until
2311 * all in-flight packets are completed
2313 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2315 EXPORT_SYMBOL(skb_set_owner_w);
2317 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2319 #ifdef CONFIG_TLS_DEVICE
2320 /* Drivers depend on in-order delivery for crypto offload,
2321 * partial orphan breaks out-of-order-OK logic.
2326 return (skb->destructor == sock_wfree ||
2327 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2330 /* This helper is used by netem, as it can hold packets in its
2331 * delay queue. We want to allow the owner socket to send more
2332 * packets, as if they were already TX completed by a typical driver.
2333 * But we also want to keep skb->sk set because some packet schedulers
2334 * rely on it (sch_fq for example).
2336 void skb_orphan_partial(struct sk_buff *skb)
2338 if (skb_is_tcp_pure_ack(skb))
2341 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2346 EXPORT_SYMBOL(skb_orphan_partial);
2349 * Read buffer destructor automatically called from kfree_skb.
2351 void sock_rfree(struct sk_buff *skb)
2353 struct sock *sk = skb->sk;
2354 unsigned int len = skb->truesize;
2356 atomic_sub(len, &sk->sk_rmem_alloc);
2357 sk_mem_uncharge(sk, len);
2359 EXPORT_SYMBOL(sock_rfree);
2362 * Buffer destructor for skbs that are not used directly in read or write
2363 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2365 void sock_efree(struct sk_buff *skb)
2369 EXPORT_SYMBOL(sock_efree);
2371 /* Buffer destructor for prefetch/receive path where reference count may
2372 * not be held, e.g. for listen sockets.
2375 void sock_pfree(struct sk_buff *skb)
2377 if (sk_is_refcounted(skb->sk))
2378 sock_gen_put(skb->sk);
2380 EXPORT_SYMBOL(sock_pfree);
2381 #endif /* CONFIG_INET */
2383 kuid_t sock_i_uid(struct sock *sk)
2387 read_lock_bh(&sk->sk_callback_lock);
2388 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2389 read_unlock_bh(&sk->sk_callback_lock);
2392 EXPORT_SYMBOL(sock_i_uid);
2394 unsigned long sock_i_ino(struct sock *sk)
2398 read_lock_bh(&sk->sk_callback_lock);
2399 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2400 read_unlock_bh(&sk->sk_callback_lock);
2403 EXPORT_SYMBOL(sock_i_ino);
2406 * Allocate a skb from the socket's send buffer.
2408 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2412 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2413 struct sk_buff *skb = alloc_skb(size, priority);
2416 skb_set_owner_w(skb, sk);
2422 EXPORT_SYMBOL(sock_wmalloc);
2424 static void sock_ofree(struct sk_buff *skb)
2426 struct sock *sk = skb->sk;
2428 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2431 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2434 struct sk_buff *skb;
2436 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2437 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2441 skb = alloc_skb(size, priority);
2445 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2447 skb->destructor = sock_ofree;
2452 * Allocate a memory block from the socket's option memory buffer.
2454 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2456 if ((unsigned int)size <= sysctl_optmem_max &&
2457 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2459 /* First do the add, to avoid the race if kmalloc
2462 atomic_add(size, &sk->sk_omem_alloc);
2463 mem = kmalloc(size, priority);
2466 atomic_sub(size, &sk->sk_omem_alloc);
2470 EXPORT_SYMBOL(sock_kmalloc);
2472 /* Free an option memory block. Note, we actually want the inline
2473 * here as this allows gcc to detect the nullify and fold away the
2474 * condition entirely.
2476 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2479 if (WARN_ON_ONCE(!mem))
2482 kfree_sensitive(mem);
2485 atomic_sub(size, &sk->sk_omem_alloc);
2488 void sock_kfree_s(struct sock *sk, void *mem, int size)
2490 __sock_kfree_s(sk, mem, size, false);
2492 EXPORT_SYMBOL(sock_kfree_s);
2494 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2496 __sock_kfree_s(sk, mem, size, true);
2498 EXPORT_SYMBOL(sock_kzfree_s);
2500 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2501 I think, these locks should be removed for datagram sockets.
2503 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2507 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2511 if (signal_pending(current))
2513 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2514 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2515 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2517 if (sk->sk_shutdown & SEND_SHUTDOWN)
2521 timeo = schedule_timeout(timeo);
2523 finish_wait(sk_sleep(sk), &wait);
2529 * Generic send/receive buffer handlers
2532 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2533 unsigned long data_len, int noblock,
2534 int *errcode, int max_page_order)
2536 struct sk_buff *skb;
2540 timeo = sock_sndtimeo(sk, noblock);
2542 err = sock_error(sk);
2547 if (sk->sk_shutdown & SEND_SHUTDOWN)
2550 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2553 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2554 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2558 if (signal_pending(current))
2560 timeo = sock_wait_for_wmem(sk, timeo);
2562 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2563 errcode, sk->sk_allocation);
2565 skb_set_owner_w(skb, sk);
2569 err = sock_intr_errno(timeo);
2574 EXPORT_SYMBOL(sock_alloc_send_pskb);
2576 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2577 int noblock, int *errcode)
2579 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2581 EXPORT_SYMBOL(sock_alloc_send_skb);
2583 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2584 struct sockcm_cookie *sockc)
2588 switch (cmsg->cmsg_type) {
2590 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2592 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2594 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2596 case SO_TIMESTAMPING_OLD:
2597 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2600 tsflags = *(u32 *)CMSG_DATA(cmsg);
2601 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2604 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2605 sockc->tsflags |= tsflags;
2608 if (!sock_flag(sk, SOCK_TXTIME))
2610 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2612 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2614 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2616 case SCM_CREDENTIALS:
2623 EXPORT_SYMBOL(__sock_cmsg_send);
2625 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2626 struct sockcm_cookie *sockc)
2628 struct cmsghdr *cmsg;
2631 for_each_cmsghdr(cmsg, msg) {
2632 if (!CMSG_OK(msg, cmsg))
2634 if (cmsg->cmsg_level != SOL_SOCKET)
2636 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2642 EXPORT_SYMBOL(sock_cmsg_send);
2644 static void sk_enter_memory_pressure(struct sock *sk)
2646 if (!sk->sk_prot->enter_memory_pressure)
2649 sk->sk_prot->enter_memory_pressure(sk);
2652 static void sk_leave_memory_pressure(struct sock *sk)
2654 if (sk->sk_prot->leave_memory_pressure) {
2655 sk->sk_prot->leave_memory_pressure(sk);
2657 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2659 if (memory_pressure && READ_ONCE(*memory_pressure))
2660 WRITE_ONCE(*memory_pressure, 0);
2664 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2667 * skb_page_frag_refill - check that a page_frag contains enough room
2668 * @sz: minimum size of the fragment we want to get
2669 * @pfrag: pointer to page_frag
2670 * @gfp: priority for memory allocation
2672 * Note: While this allocator tries to use high order pages, there is
2673 * no guarantee that allocations succeed. Therefore, @sz MUST be
2674 * less or equal than PAGE_SIZE.
2676 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2679 if (page_ref_count(pfrag->page) == 1) {
2683 if (pfrag->offset + sz <= pfrag->size)
2685 put_page(pfrag->page);
2689 if (SKB_FRAG_PAGE_ORDER &&
2690 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2691 /* Avoid direct reclaim but allow kswapd to wake */
2692 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2693 __GFP_COMP | __GFP_NOWARN |
2695 SKB_FRAG_PAGE_ORDER);
2696 if (likely(pfrag->page)) {
2697 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2701 pfrag->page = alloc_page(gfp);
2702 if (likely(pfrag->page)) {
2703 pfrag->size = PAGE_SIZE;
2708 EXPORT_SYMBOL(skb_page_frag_refill);
2710 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2712 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2715 sk_enter_memory_pressure(sk);
2716 sk_stream_moderate_sndbuf(sk);
2719 EXPORT_SYMBOL(sk_page_frag_refill);
2721 void __lock_sock(struct sock *sk)
2722 __releases(&sk->sk_lock.slock)
2723 __acquires(&sk->sk_lock.slock)
2728 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2729 TASK_UNINTERRUPTIBLE);
2730 spin_unlock_bh(&sk->sk_lock.slock);
2732 spin_lock_bh(&sk->sk_lock.slock);
2733 if (!sock_owned_by_user(sk))
2736 finish_wait(&sk->sk_lock.wq, &wait);
2739 void __release_sock(struct sock *sk)
2740 __releases(&sk->sk_lock.slock)
2741 __acquires(&sk->sk_lock.slock)
2743 struct sk_buff *skb, *next;
2745 while ((skb = sk->sk_backlog.head) != NULL) {
2746 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2748 spin_unlock_bh(&sk->sk_lock.slock);
2753 WARN_ON_ONCE(skb_dst_is_noref(skb));
2754 skb_mark_not_on_list(skb);
2755 sk_backlog_rcv(sk, skb);
2760 } while (skb != NULL);
2762 spin_lock_bh(&sk->sk_lock.slock);
2766 * Doing the zeroing here guarantee we can not loop forever
2767 * while a wild producer attempts to flood us.
2769 sk->sk_backlog.len = 0;
2772 void __sk_flush_backlog(struct sock *sk)
2774 spin_lock_bh(&sk->sk_lock.slock);
2776 spin_unlock_bh(&sk->sk_lock.slock);
2780 * sk_wait_data - wait for data to arrive at sk_receive_queue
2781 * @sk: sock to wait on
2782 * @timeo: for how long
2783 * @skb: last skb seen on sk_receive_queue
2785 * Now socket state including sk->sk_err is changed only under lock,
2786 * hence we may omit checks after joining wait queue.
2787 * We check receive queue before schedule() only as optimization;
2788 * it is very likely that release_sock() added new data.
2790 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2792 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2795 add_wait_queue(sk_sleep(sk), &wait);
2796 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2797 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2798 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2799 remove_wait_queue(sk_sleep(sk), &wait);
2802 EXPORT_SYMBOL(sk_wait_data);
2805 * __sk_mem_raise_allocated - increase memory_allocated
2807 * @size: memory size to allocate
2808 * @amt: pages to allocate
2809 * @kind: allocation type
2811 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2813 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2815 struct proto *prot = sk->sk_prot;
2816 long allocated = sk_memory_allocated_add(sk, amt);
2817 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2818 bool charged = true;
2821 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2822 gfp_memcg_charge())))
2823 goto suppress_allocation;
2826 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2827 sk_leave_memory_pressure(sk);
2831 /* Under pressure. */
2832 if (allocated > sk_prot_mem_limits(sk, 1))
2833 sk_enter_memory_pressure(sk);
2835 /* Over hard limit. */
2836 if (allocated > sk_prot_mem_limits(sk, 2))
2837 goto suppress_allocation;
2839 /* guarantee minimum buffer size under pressure */
2840 if (kind == SK_MEM_RECV) {
2841 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2844 } else { /* SK_MEM_SEND */
2845 int wmem0 = sk_get_wmem0(sk, prot);
2847 if (sk->sk_type == SOCK_STREAM) {
2848 if (sk->sk_wmem_queued < wmem0)
2850 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2855 if (sk_has_memory_pressure(sk)) {
2858 if (!sk_under_memory_pressure(sk))
2860 alloc = sk_sockets_allocated_read_positive(sk);
2861 if (sk_prot_mem_limits(sk, 2) > alloc *
2862 sk_mem_pages(sk->sk_wmem_queued +
2863 atomic_read(&sk->sk_rmem_alloc) +
2864 sk->sk_forward_alloc))
2868 suppress_allocation:
2870 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2871 sk_stream_moderate_sndbuf(sk);
2873 /* Fail only if socket is _under_ its sndbuf.
2874 * In this case we cannot block, so that we have to fail.
2876 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
2877 /* Force charge with __GFP_NOFAIL */
2878 if (memcg_charge && !charged) {
2879 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2880 gfp_memcg_charge() | __GFP_NOFAIL);
2886 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2887 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2889 sk_memory_allocated_sub(sk, amt);
2891 if (memcg_charge && charged)
2892 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2896 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2899 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2901 * @size: memory size to allocate
2902 * @kind: allocation type
2904 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2905 * rmem allocation. This function assumes that protocols which have
2906 * memory_pressure use sk_wmem_queued as write buffer accounting.
2908 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2910 int ret, amt = sk_mem_pages(size);
2912 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2913 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2915 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2918 EXPORT_SYMBOL(__sk_mem_schedule);
2921 * __sk_mem_reduce_allocated - reclaim memory_allocated
2923 * @amount: number of quanta
2925 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2927 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2929 sk_memory_allocated_sub(sk, amount);
2931 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2932 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2934 if (sk_under_memory_pressure(sk) &&
2935 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2936 sk_leave_memory_pressure(sk);
2938 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2941 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2943 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2945 void __sk_mem_reclaim(struct sock *sk, int amount)
2947 amount >>= SK_MEM_QUANTUM_SHIFT;
2948 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2949 __sk_mem_reduce_allocated(sk, amount);
2951 EXPORT_SYMBOL(__sk_mem_reclaim);
2953 int sk_set_peek_off(struct sock *sk, int val)
2955 sk->sk_peek_off = val;
2958 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2961 * Set of default routines for initialising struct proto_ops when
2962 * the protocol does not support a particular function. In certain
2963 * cases where it makes no sense for a protocol to have a "do nothing"
2964 * function, some default processing is provided.
2967 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2971 EXPORT_SYMBOL(sock_no_bind);
2973 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2978 EXPORT_SYMBOL(sock_no_connect);
2980 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2984 EXPORT_SYMBOL(sock_no_socketpair);
2986 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2991 EXPORT_SYMBOL(sock_no_accept);
2993 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2998 EXPORT_SYMBOL(sock_no_getname);
3000 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3004 EXPORT_SYMBOL(sock_no_ioctl);
3006 int sock_no_listen(struct socket *sock, int backlog)
3010 EXPORT_SYMBOL(sock_no_listen);
3012 int sock_no_shutdown(struct socket *sock, int how)
3016 EXPORT_SYMBOL(sock_no_shutdown);
3018 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3022 EXPORT_SYMBOL(sock_no_sendmsg);
3024 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3028 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3030 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3035 EXPORT_SYMBOL(sock_no_recvmsg);
3037 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3039 /* Mirror missing mmap method error code */
3042 EXPORT_SYMBOL(sock_no_mmap);
3045 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3046 * various sock-based usage counts.
3048 void __receive_sock(struct file *file)
3050 struct socket *sock;
3052 sock = sock_from_file(file);
3054 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3055 sock_update_classid(&sock->sk->sk_cgrp_data);
3059 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3062 struct msghdr msg = {.msg_flags = flags};
3064 char *kaddr = kmap(page);
3065 iov.iov_base = kaddr + offset;
3067 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3071 EXPORT_SYMBOL(sock_no_sendpage);
3073 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3074 int offset, size_t size, int flags)
3077 struct msghdr msg = {.msg_flags = flags};
3079 char *kaddr = kmap(page);
3081 iov.iov_base = kaddr + offset;
3083 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3087 EXPORT_SYMBOL(sock_no_sendpage_locked);
3090 * Default Socket Callbacks
3093 static void sock_def_wakeup(struct sock *sk)
3095 struct socket_wq *wq;
3098 wq = rcu_dereference(sk->sk_wq);
3099 if (skwq_has_sleeper(wq))
3100 wake_up_interruptible_all(&wq->wait);
3104 static void sock_def_error_report(struct sock *sk)
3106 struct socket_wq *wq;
3109 wq = rcu_dereference(sk->sk_wq);
3110 if (skwq_has_sleeper(wq))
3111 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3112 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3116 void sock_def_readable(struct sock *sk)
3118 struct socket_wq *wq;
3121 wq = rcu_dereference(sk->sk_wq);
3122 if (skwq_has_sleeper(wq))
3123 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3124 EPOLLRDNORM | EPOLLRDBAND);
3125 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3129 static void sock_def_write_space(struct sock *sk)
3131 struct socket_wq *wq;
3135 /* Do not wake up a writer until he can make "significant"
3138 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
3139 wq = rcu_dereference(sk->sk_wq);
3140 if (skwq_has_sleeper(wq))
3141 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3142 EPOLLWRNORM | EPOLLWRBAND);
3144 /* Should agree with poll, otherwise some programs break */
3145 if (sock_writeable(sk))
3146 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3152 static void sock_def_destruct(struct sock *sk)
3156 void sk_send_sigurg(struct sock *sk)
3158 if (sk->sk_socket && sk->sk_socket->file)
3159 if (send_sigurg(&sk->sk_socket->file->f_owner))
3160 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3162 EXPORT_SYMBOL(sk_send_sigurg);
3164 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3165 unsigned long expires)
3167 if (!mod_timer(timer, expires))
3170 EXPORT_SYMBOL(sk_reset_timer);
3172 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3174 if (del_timer(timer))
3177 EXPORT_SYMBOL(sk_stop_timer);
3179 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3181 if (del_timer_sync(timer))
3184 EXPORT_SYMBOL(sk_stop_timer_sync);
3186 void sock_init_data(struct socket *sock, struct sock *sk)
3189 sk->sk_send_head = NULL;
3191 timer_setup(&sk->sk_timer, NULL, 0);
3193 sk->sk_allocation = GFP_KERNEL;
3194 sk->sk_rcvbuf = sysctl_rmem_default;
3195 sk->sk_sndbuf = sysctl_wmem_default;
3196 sk->sk_state = TCP_CLOSE;
3197 sk_set_socket(sk, sock);
3199 sock_set_flag(sk, SOCK_ZAPPED);
3202 sk->sk_type = sock->type;
3203 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3205 sk->sk_uid = SOCK_INODE(sock)->i_uid;
3207 RCU_INIT_POINTER(sk->sk_wq, NULL);
3208 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3211 rwlock_init(&sk->sk_callback_lock);
3212 if (sk->sk_kern_sock)
3213 lockdep_set_class_and_name(
3214 &sk->sk_callback_lock,
3215 af_kern_callback_keys + sk->sk_family,
3216 af_family_kern_clock_key_strings[sk->sk_family]);
3218 lockdep_set_class_and_name(
3219 &sk->sk_callback_lock,
3220 af_callback_keys + sk->sk_family,
3221 af_family_clock_key_strings[sk->sk_family]);
3223 sk->sk_state_change = sock_def_wakeup;
3224 sk->sk_data_ready = sock_def_readable;
3225 sk->sk_write_space = sock_def_write_space;
3226 sk->sk_error_report = sock_def_error_report;
3227 sk->sk_destruct = sock_def_destruct;
3229 sk->sk_frag.page = NULL;
3230 sk->sk_frag.offset = 0;
3231 sk->sk_peek_off = -1;
3233 sk->sk_peer_pid = NULL;
3234 sk->sk_peer_cred = NULL;
3235 spin_lock_init(&sk->sk_peer_lock);
3237 sk->sk_write_pending = 0;
3238 sk->sk_rcvlowat = 1;
3239 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3240 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3242 sk->sk_stamp = SK_DEFAULT_STAMP;
3243 #if BITS_PER_LONG==32
3244 seqlock_init(&sk->sk_stamp_seq);
3246 atomic_set(&sk->sk_zckey, 0);
3248 #ifdef CONFIG_NET_RX_BUSY_POLL
3250 sk->sk_ll_usec = sysctl_net_busy_read;
3253 sk->sk_max_pacing_rate = ~0UL;
3254 sk->sk_pacing_rate = ~0UL;
3255 WRITE_ONCE(sk->sk_pacing_shift, 10);
3256 sk->sk_incoming_cpu = -1;
3258 sk_rx_queue_clear(sk);
3260 * Before updating sk_refcnt, we must commit prior changes to memory
3261 * (Documentation/RCU/rculist_nulls.rst for details)
3264 refcount_set(&sk->sk_refcnt, 1);
3265 atomic_set(&sk->sk_drops, 0);
3267 EXPORT_SYMBOL(sock_init_data);
3269 void lock_sock_nested(struct sock *sk, int subclass)
3271 /* The sk_lock has mutex_lock() semantics here. */
3272 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3275 spin_lock_bh(&sk->sk_lock.slock);
3276 if (sk->sk_lock.owned)
3278 sk->sk_lock.owned = 1;
3279 spin_unlock_bh(&sk->sk_lock.slock);
3281 EXPORT_SYMBOL(lock_sock_nested);
3283 void release_sock(struct sock *sk)
3285 spin_lock_bh(&sk->sk_lock.slock);
3286 if (sk->sk_backlog.tail)
3289 /* Warning : release_cb() might need to release sk ownership,
3290 * ie call sock_release_ownership(sk) before us.
3292 if (sk->sk_prot->release_cb)
3293 sk->sk_prot->release_cb(sk);
3295 sock_release_ownership(sk);
3296 if (waitqueue_active(&sk->sk_lock.wq))
3297 wake_up(&sk->sk_lock.wq);
3298 spin_unlock_bh(&sk->sk_lock.slock);
3300 EXPORT_SYMBOL(release_sock);
3302 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3305 spin_lock_bh(&sk->sk_lock.slock);
3307 if (!sk->sk_lock.owned) {
3309 * Fast path return with bottom halves disabled and
3310 * sock::sk_lock.slock held.
3312 * The 'mutex' is not contended and holding
3313 * sock::sk_lock.slock prevents all other lockers to
3314 * proceed so the corresponding unlock_sock_fast() can
3315 * avoid the slow path of release_sock() completely and
3316 * just release slock.
3318 * From a semantical POV this is equivalent to 'acquiring'
3319 * the 'mutex', hence the corresponding lockdep
3320 * mutex_release() has to happen in the fast path of
3321 * unlock_sock_fast().
3327 sk->sk_lock.owned = 1;
3328 __acquire(&sk->sk_lock.slock);
3329 spin_unlock_bh(&sk->sk_lock.slock);
3332 EXPORT_SYMBOL(__lock_sock_fast);
3334 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3335 bool timeval, bool time32)
3337 struct sock *sk = sock->sk;
3338 struct timespec64 ts;
3340 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3341 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3342 if (ts.tv_sec == -1)
3344 if (ts.tv_sec == 0) {
3345 ktime_t kt = ktime_get_real();
3346 sock_write_timestamp(sk, kt);
3347 ts = ktime_to_timespec64(kt);
3353 #ifdef CONFIG_COMPAT_32BIT_TIME
3355 return put_old_timespec32(&ts, userstamp);
3357 #ifdef CONFIG_SPARC64
3358 /* beware of padding in sparc64 timeval */
3359 if (timeval && !in_compat_syscall()) {
3360 struct __kernel_old_timeval __user tv = {
3361 .tv_sec = ts.tv_sec,
3362 .tv_usec = ts.tv_nsec,
3364 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3369 return put_timespec64(&ts, userstamp);
3371 EXPORT_SYMBOL(sock_gettstamp);
3373 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3375 if (!sock_flag(sk, flag)) {
3376 unsigned long previous_flags = sk->sk_flags;
3378 sock_set_flag(sk, flag);
3380 * we just set one of the two flags which require net
3381 * time stamping, but time stamping might have been on
3382 * already because of the other one
3384 if (sock_needs_netstamp(sk) &&
3385 !(previous_flags & SK_FLAGS_TIMESTAMP))
3386 net_enable_timestamp();
3390 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3391 int level, int type)
3393 struct sock_exterr_skb *serr;
3394 struct sk_buff *skb;
3398 skb = sock_dequeue_err_skb(sk);
3404 msg->msg_flags |= MSG_TRUNC;
3407 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3411 sock_recv_timestamp(msg, sk, skb);
3413 serr = SKB_EXT_ERR(skb);
3414 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3416 msg->msg_flags |= MSG_ERRQUEUE;
3424 EXPORT_SYMBOL(sock_recv_errqueue);
3427 * Get a socket option on an socket.
3429 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3430 * asynchronous errors should be reported by getsockopt. We assume
3431 * this means if you specify SO_ERROR (otherwise whats the point of it).
3433 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3434 char __user *optval, int __user *optlen)
3436 struct sock *sk = sock->sk;
3438 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3440 EXPORT_SYMBOL(sock_common_getsockopt);
3442 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3445 struct sock *sk = sock->sk;
3449 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3450 flags & ~MSG_DONTWAIT, &addr_len);
3452 msg->msg_namelen = addr_len;
3455 EXPORT_SYMBOL(sock_common_recvmsg);
3458 * Set socket options on an inet socket.
3460 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3461 sockptr_t optval, unsigned int optlen)
3463 struct sock *sk = sock->sk;
3465 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3467 EXPORT_SYMBOL(sock_common_setsockopt);
3469 void sk_common_release(struct sock *sk)
3471 if (sk->sk_prot->destroy)
3472 sk->sk_prot->destroy(sk);
3475 * Observation: when sk_common_release is called, processes have
3476 * no access to socket. But net still has.
3477 * Step one, detach it from networking:
3479 * A. Remove from hash tables.
3482 sk->sk_prot->unhash(sk);
3485 * In this point socket cannot receive new packets, but it is possible
3486 * that some packets are in flight because some CPU runs receiver and
3487 * did hash table lookup before we unhashed socket. They will achieve
3488 * receive queue and will be purged by socket destructor.
3490 * Also we still have packets pending on receive queue and probably,
3491 * our own packets waiting in device queues. sock_destroy will drain
3492 * receive queue, but transmitted packets will delay socket destruction
3493 * until the last reference will be released.
3498 xfrm_sk_free_policy(sk);
3500 sk_refcnt_debug_release(sk);
3504 EXPORT_SYMBOL(sk_common_release);
3506 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3508 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3510 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3511 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3512 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3513 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3514 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3515 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3516 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3517 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3518 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3521 #ifdef CONFIG_PROC_FS
3522 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3524 int val[PROTO_INUSE_NR];
3527 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3529 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3531 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3533 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3535 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3537 int cpu, idx = prot->inuse_idx;
3540 for_each_possible_cpu(cpu)
3541 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3543 return res >= 0 ? res : 0;
3545 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3547 static void sock_inuse_add(struct net *net, int val)
3549 this_cpu_add(*net->core.sock_inuse, val);
3552 int sock_inuse_get(struct net *net)
3556 for_each_possible_cpu(cpu)
3557 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3562 EXPORT_SYMBOL_GPL(sock_inuse_get);
3564 static int __net_init sock_inuse_init_net(struct net *net)
3566 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3567 if (net->core.prot_inuse == NULL)
3570 net->core.sock_inuse = alloc_percpu(int);
3571 if (net->core.sock_inuse == NULL)
3577 free_percpu(net->core.prot_inuse);
3581 static void __net_exit sock_inuse_exit_net(struct net *net)
3583 free_percpu(net->core.prot_inuse);
3584 free_percpu(net->core.sock_inuse);
3587 static struct pernet_operations net_inuse_ops = {
3588 .init = sock_inuse_init_net,
3589 .exit = sock_inuse_exit_net,
3592 static __init int net_inuse_init(void)
3594 if (register_pernet_subsys(&net_inuse_ops))
3595 panic("Cannot initialize net inuse counters");
3600 core_initcall(net_inuse_init);
3602 static int assign_proto_idx(struct proto *prot)
3604 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3606 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3607 pr_err("PROTO_INUSE_NR exhausted\n");
3611 set_bit(prot->inuse_idx, proto_inuse_idx);
3615 static void release_proto_idx(struct proto *prot)
3617 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3618 clear_bit(prot->inuse_idx, proto_inuse_idx);
3621 static inline int assign_proto_idx(struct proto *prot)
3626 static inline void release_proto_idx(struct proto *prot)
3630 static void sock_inuse_add(struct net *net, int val)
3635 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3639 kfree(twsk_prot->twsk_slab_name);
3640 twsk_prot->twsk_slab_name = NULL;
3641 kmem_cache_destroy(twsk_prot->twsk_slab);
3642 twsk_prot->twsk_slab = NULL;
3645 static int tw_prot_init(const struct proto *prot)
3647 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3652 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3654 if (!twsk_prot->twsk_slab_name)
3657 twsk_prot->twsk_slab =
3658 kmem_cache_create(twsk_prot->twsk_slab_name,
3659 twsk_prot->twsk_obj_size, 0,
3660 SLAB_ACCOUNT | prot->slab_flags,
3662 if (!twsk_prot->twsk_slab) {
3663 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3671 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3675 kfree(rsk_prot->slab_name);
3676 rsk_prot->slab_name = NULL;
3677 kmem_cache_destroy(rsk_prot->slab);
3678 rsk_prot->slab = NULL;
3681 static int req_prot_init(const struct proto *prot)
3683 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3688 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3690 if (!rsk_prot->slab_name)
3693 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3694 rsk_prot->obj_size, 0,
3695 SLAB_ACCOUNT | prot->slab_flags,
3698 if (!rsk_prot->slab) {
3699 pr_crit("%s: Can't create request sock SLAB cache!\n",
3706 int proto_register(struct proto *prot, int alloc_slab)
3711 prot->slab = kmem_cache_create_usercopy(prot->name,
3713 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3715 prot->useroffset, prot->usersize,
3718 if (prot->slab == NULL) {
3719 pr_crit("%s: Can't create sock SLAB cache!\n",
3724 if (req_prot_init(prot))
3725 goto out_free_request_sock_slab;
3727 if (tw_prot_init(prot))
3728 goto out_free_timewait_sock_slab;
3731 mutex_lock(&proto_list_mutex);
3732 ret = assign_proto_idx(prot);
3734 mutex_unlock(&proto_list_mutex);
3735 goto out_free_timewait_sock_slab;
3737 list_add(&prot->node, &proto_list);
3738 mutex_unlock(&proto_list_mutex);
3741 out_free_timewait_sock_slab:
3743 tw_prot_cleanup(prot->twsk_prot);
3744 out_free_request_sock_slab:
3746 req_prot_cleanup(prot->rsk_prot);
3748 kmem_cache_destroy(prot->slab);
3754 EXPORT_SYMBOL(proto_register);
3756 void proto_unregister(struct proto *prot)
3758 mutex_lock(&proto_list_mutex);
3759 release_proto_idx(prot);
3760 list_del(&prot->node);
3761 mutex_unlock(&proto_list_mutex);
3763 kmem_cache_destroy(prot->slab);
3766 req_prot_cleanup(prot->rsk_prot);
3767 tw_prot_cleanup(prot->twsk_prot);
3769 EXPORT_SYMBOL(proto_unregister);
3771 int sock_load_diag_module(int family, int protocol)
3774 if (!sock_is_registered(family))
3777 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3778 NETLINK_SOCK_DIAG, family);
3782 if (family == AF_INET &&
3783 protocol != IPPROTO_RAW &&
3784 protocol < MAX_INET_PROTOS &&
3785 !rcu_access_pointer(inet_protos[protocol]))
3789 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3790 NETLINK_SOCK_DIAG, family, protocol);
3792 EXPORT_SYMBOL(sock_load_diag_module);
3794 #ifdef CONFIG_PROC_FS
3795 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3796 __acquires(proto_list_mutex)
3798 mutex_lock(&proto_list_mutex);
3799 return seq_list_start_head(&proto_list, *pos);
3802 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3804 return seq_list_next(v, &proto_list, pos);
3807 static void proto_seq_stop(struct seq_file *seq, void *v)
3808 __releases(proto_list_mutex)
3810 mutex_unlock(&proto_list_mutex);
3813 static char proto_method_implemented(const void *method)
3815 return method == NULL ? 'n' : 'y';
3817 static long sock_prot_memory_allocated(struct proto *proto)
3819 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3822 static const char *sock_prot_memory_pressure(struct proto *proto)
3824 return proto->memory_pressure != NULL ?
3825 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3828 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3831 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3832 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3835 sock_prot_inuse_get(seq_file_net(seq), proto),
3836 sock_prot_memory_allocated(proto),
3837 sock_prot_memory_pressure(proto),
3839 proto->slab == NULL ? "no" : "yes",
3840 module_name(proto->owner),
3841 proto_method_implemented(proto->close),
3842 proto_method_implemented(proto->connect),
3843 proto_method_implemented(proto->disconnect),
3844 proto_method_implemented(proto->accept),
3845 proto_method_implemented(proto->ioctl),
3846 proto_method_implemented(proto->init),
3847 proto_method_implemented(proto->destroy),
3848 proto_method_implemented(proto->shutdown),
3849 proto_method_implemented(proto->setsockopt),
3850 proto_method_implemented(proto->getsockopt),
3851 proto_method_implemented(proto->sendmsg),
3852 proto_method_implemented(proto->recvmsg),
3853 proto_method_implemented(proto->sendpage),
3854 proto_method_implemented(proto->bind),
3855 proto_method_implemented(proto->backlog_rcv),
3856 proto_method_implemented(proto->hash),
3857 proto_method_implemented(proto->unhash),
3858 proto_method_implemented(proto->get_port),
3859 proto_method_implemented(proto->enter_memory_pressure));
3862 static int proto_seq_show(struct seq_file *seq, void *v)
3864 if (v == &proto_list)
3865 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3874 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3876 proto_seq_printf(seq, list_entry(v, struct proto, node));
3880 static const struct seq_operations proto_seq_ops = {
3881 .start = proto_seq_start,
3882 .next = proto_seq_next,
3883 .stop = proto_seq_stop,
3884 .show = proto_seq_show,
3887 static __net_init int proto_init_net(struct net *net)
3889 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3890 sizeof(struct seq_net_private)))
3896 static __net_exit void proto_exit_net(struct net *net)
3898 remove_proc_entry("protocols", net->proc_net);
3902 static __net_initdata struct pernet_operations proto_net_ops = {
3903 .init = proto_init_net,
3904 .exit = proto_exit_net,
3907 static int __init proto_init(void)
3909 return register_pernet_subsys(&proto_net_ops);
3912 subsys_initcall(proto_init);
3914 #endif /* PROC_FS */
3916 #ifdef CONFIG_NET_RX_BUSY_POLL
3917 bool sk_busy_loop_end(void *p, unsigned long start_time)
3919 struct sock *sk = p;
3921 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3922 sk_busy_loop_timeout(sk, start_time);
3924 EXPORT_SYMBOL(sk_busy_loop_end);
3925 #endif /* CONFIG_NET_RX_BUSY_POLL */
3927 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3929 if (!sk->sk_prot->bind_add)
3931 return sk->sk_prot->bind_add(sk, addr, addr_len);
3933 EXPORT_SYMBOL(sock_bind_add);