1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
118 #include <linux/uaccess.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
137 #include <trace/events/sock.h>
140 #include <net/busy_poll.h>
142 #include <linux/ethtool.h>
144 static DEFINE_MUTEX(proto_list_mutex);
145 static LIST_HEAD(proto_list);
148 * sk_ns_capable - General socket capability test
149 * @sk: Socket to use a capability on or through
150 * @user_ns: The user namespace of the capability to use
151 * @cap: The capability to use
153 * Test to see if the opener of the socket had when the socket was
154 * created and the current process has the capability @cap in the user
155 * namespace @user_ns.
157 bool sk_ns_capable(const struct sock *sk,
158 struct user_namespace *user_ns, int cap)
160 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
161 ns_capable(user_ns, cap);
163 EXPORT_SYMBOL(sk_ns_capable);
166 * sk_capable - Socket global capability test
167 * @sk: Socket to use a capability on or through
168 * @cap: The global capability to use
170 * Test to see if the opener of the socket had when the socket was
171 * created and the current process has the capability @cap in all user
174 bool sk_capable(const struct sock *sk, int cap)
176 return sk_ns_capable(sk, &init_user_ns, cap);
178 EXPORT_SYMBOL(sk_capable);
181 * sk_net_capable - Network namespace socket capability test
182 * @sk: Socket to use a capability on or through
183 * @cap: The capability to use
185 * Test to see if the opener of the socket had when the socket was created
186 * and the current process has the capability @cap over the network namespace
187 * the socket is a member of.
189 bool sk_net_capable(const struct sock *sk, int cap)
191 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193 EXPORT_SYMBOL(sk_net_capable);
196 * Each address family might have different locking rules, so we have
197 * one slock key per address family and separate keys for internal and
200 static struct lock_class_key af_family_keys[AF_MAX];
201 static struct lock_class_key af_family_kern_keys[AF_MAX];
202 static struct lock_class_key af_family_slock_keys[AF_MAX];
203 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
211 #define _sock_locks(x) \
212 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
213 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
214 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
215 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
216 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
217 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
218 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
219 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
220 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
221 x "27" , x "28" , x "AF_CAN" , \
222 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
223 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
224 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
225 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
226 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
230 static const char *const af_family_key_strings[AF_MAX+1] = {
231 _sock_locks("sk_lock-")
233 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
234 _sock_locks("slock-")
236 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
237 _sock_locks("clock-")
240 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
241 _sock_locks("k-sk_lock-")
243 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
244 _sock_locks("k-slock-")
246 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
247 _sock_locks("k-clock-")
249 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
250 _sock_locks("rlock-")
252 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
253 _sock_locks("wlock-")
255 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
256 _sock_locks("elock-")
260 * sk_callback_lock and sk queues locking rules are per-address-family,
261 * so split the lock classes by using a per-AF key:
263 static struct lock_class_key af_callback_keys[AF_MAX];
264 static struct lock_class_key af_rlock_keys[AF_MAX];
265 static struct lock_class_key af_wlock_keys[AF_MAX];
266 static struct lock_class_key af_elock_keys[AF_MAX];
267 static struct lock_class_key af_kern_callback_keys[AF_MAX];
269 /* Run time adjustable parameters. */
270 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
271 EXPORT_SYMBOL(sysctl_wmem_max);
272 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
273 EXPORT_SYMBOL(sysctl_rmem_max);
274 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
275 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
277 /* Maximal space eaten by iovec or ancillary data plus some space */
278 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
279 EXPORT_SYMBOL(sysctl_optmem_max);
281 int sysctl_tstamp_allow_data __read_mostly = 1;
283 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
284 EXPORT_SYMBOL_GPL(memalloc_socks_key);
287 * sk_set_memalloc - sets %SOCK_MEMALLOC
288 * @sk: socket to set it on
290 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
291 * It's the responsibility of the admin to adjust min_free_kbytes
292 * to meet the requirements
294 void sk_set_memalloc(struct sock *sk)
296 sock_set_flag(sk, SOCK_MEMALLOC);
297 sk->sk_allocation |= __GFP_MEMALLOC;
298 static_branch_inc(&memalloc_socks_key);
300 EXPORT_SYMBOL_GPL(sk_set_memalloc);
302 void sk_clear_memalloc(struct sock *sk)
304 sock_reset_flag(sk, SOCK_MEMALLOC);
305 sk->sk_allocation &= ~__GFP_MEMALLOC;
306 static_branch_dec(&memalloc_socks_key);
309 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
310 * progress of swapping. SOCK_MEMALLOC may be cleared while
311 * it has rmem allocations due to the last swapfile being deactivated
312 * but there is a risk that the socket is unusable due to exceeding
313 * the rmem limits. Reclaim the reserves and obey rmem limits again.
317 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
319 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
322 unsigned int noreclaim_flag;
324 /* these should have been dropped before queueing */
325 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
327 noreclaim_flag = memalloc_noreclaim_save();
328 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
332 memalloc_noreclaim_restore(noreclaim_flag);
336 EXPORT_SYMBOL(__sk_backlog_rcv);
338 void sk_error_report(struct sock *sk)
340 sk->sk_error_report(sk);
342 switch (sk->sk_family) {
346 trace_inet_sk_error_report(sk);
352 EXPORT_SYMBOL(sk_error_report);
354 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
356 struct __kernel_sock_timeval tv;
358 if (timeo == MAX_SCHEDULE_TIMEOUT) {
362 tv.tv_sec = timeo / HZ;
363 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
366 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
367 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
368 *(struct old_timeval32 *)optval = tv32;
373 struct __kernel_old_timeval old_tv;
374 old_tv.tv_sec = tv.tv_sec;
375 old_tv.tv_usec = tv.tv_usec;
376 *(struct __kernel_old_timeval *)optval = old_tv;
377 return sizeof(old_tv);
380 *(struct __kernel_sock_timeval *)optval = tv;
383 EXPORT_SYMBOL(sock_get_timeout);
385 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
386 sockptr_t optval, int optlen, bool old_timeval)
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)))
416 EXPORT_SYMBOL(sock_copy_user_timeval);
418 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
421 struct __kernel_sock_timeval tv;
422 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
427 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
431 static int warned __read_mostly;
434 if (warned < 10 && net_ratelimit()) {
436 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
437 __func__, current->comm, task_pid_nr(current));
441 *timeo_p = MAX_SCHEDULE_TIMEOUT;
442 if (tv.tv_sec == 0 && tv.tv_usec == 0)
444 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
445 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
449 static bool sock_needs_netstamp(const struct sock *sk)
451 switch (sk->sk_family) {
460 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
462 if (sk->sk_flags & flags) {
463 sk->sk_flags &= ~flags;
464 if (sock_needs_netstamp(sk) &&
465 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
466 net_disable_timestamp();
471 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
474 struct sk_buff_head *list = &sk->sk_receive_queue;
476 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
477 atomic_inc(&sk->sk_drops);
478 trace_sock_rcvqueue_full(sk, skb);
482 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
483 atomic_inc(&sk->sk_drops);
488 skb_set_owner_r(skb, sk);
490 /* we escape from rcu protected region, make sure we dont leak
495 spin_lock_irqsave(&list->lock, flags);
496 sock_skb_set_dropcount(sk, skb);
497 __skb_queue_tail(list, skb);
498 spin_unlock_irqrestore(&list->lock, flags);
500 if (!sock_flag(sk, SOCK_DEAD))
501 sk->sk_data_ready(sk);
504 EXPORT_SYMBOL(__sock_queue_rcv_skb);
506 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
510 err = sk_filter(sk, skb);
514 return __sock_queue_rcv_skb(sk, skb);
516 EXPORT_SYMBOL(sock_queue_rcv_skb);
518 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
519 const int nested, unsigned int trim_cap, bool refcounted)
521 int rc = NET_RX_SUCCESS;
523 if (sk_filter_trim_cap(sk, skb, trim_cap))
524 goto discard_and_relse;
528 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
529 atomic_inc(&sk->sk_drops);
530 goto discard_and_relse;
533 bh_lock_sock_nested(sk);
536 if (!sock_owned_by_user(sk)) {
538 * trylock + unlock semantics:
540 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
542 rc = sk_backlog_rcv(sk, skb);
544 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
545 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
547 atomic_inc(&sk->sk_drops);
548 goto discard_and_relse;
560 EXPORT_SYMBOL(__sk_receive_skb);
562 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
564 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
566 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
568 struct dst_entry *dst = __sk_dst_get(sk);
570 if (dst && dst->obsolete &&
571 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
572 dst, cookie) == NULL) {
573 sk_tx_queue_clear(sk);
574 sk->sk_dst_pending_confirm = 0;
575 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
582 EXPORT_SYMBOL(__sk_dst_check);
584 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
586 struct dst_entry *dst = sk_dst_get(sk);
588 if (dst && dst->obsolete &&
589 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
590 dst, cookie) == NULL) {
598 EXPORT_SYMBOL(sk_dst_check);
600 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
602 int ret = -ENOPROTOOPT;
603 #ifdef CONFIG_NETDEVICES
604 struct net *net = sock_net(sk);
608 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
615 sk->sk_bound_dev_if = ifindex;
616 if (sk->sk_prot->rehash)
617 sk->sk_prot->rehash(sk);
628 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
634 ret = sock_bindtoindex_locked(sk, ifindex);
640 EXPORT_SYMBOL(sock_bindtoindex);
642 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
644 int ret = -ENOPROTOOPT;
645 #ifdef CONFIG_NETDEVICES
646 struct net *net = sock_net(sk);
647 char devname[IFNAMSIZ];
654 /* Bind this socket to a particular device like "eth0",
655 * as specified in the passed interface name. If the
656 * name is "" or the option length is zero the socket
659 if (optlen > IFNAMSIZ - 1)
660 optlen = IFNAMSIZ - 1;
661 memset(devname, 0, sizeof(devname));
664 if (copy_from_sockptr(devname, optval, optlen))
668 if (devname[0] != '\0') {
669 struct net_device *dev;
672 dev = dev_get_by_name_rcu(net, devname);
674 index = dev->ifindex;
681 return sock_bindtoindex(sk, index, true);
688 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
689 int __user *optlen, int len)
691 int ret = -ENOPROTOOPT;
692 #ifdef CONFIG_NETDEVICES
693 struct net *net = sock_net(sk);
694 char devname[IFNAMSIZ];
696 if (sk->sk_bound_dev_if == 0) {
705 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
709 len = strlen(devname) + 1;
712 if (copy_to_user(optval, devname, len))
717 if (put_user(len, optlen))
728 bool sk_mc_loop(struct sock *sk)
730 if (dev_recursion_level())
734 switch (sk->sk_family) {
736 return inet_sk(sk)->mc_loop;
737 #if IS_ENABLED(CONFIG_IPV6)
739 return inet6_sk(sk)->mc_loop;
745 EXPORT_SYMBOL(sk_mc_loop);
747 void sock_set_reuseaddr(struct sock *sk)
750 sk->sk_reuse = SK_CAN_REUSE;
753 EXPORT_SYMBOL(sock_set_reuseaddr);
755 void sock_set_reuseport(struct sock *sk)
758 sk->sk_reuseport = true;
761 EXPORT_SYMBOL(sock_set_reuseport);
763 void sock_no_linger(struct sock *sk)
766 sk->sk_lingertime = 0;
767 sock_set_flag(sk, SOCK_LINGER);
770 EXPORT_SYMBOL(sock_no_linger);
772 void sock_set_priority(struct sock *sk, u32 priority)
775 sk->sk_priority = priority;
778 EXPORT_SYMBOL(sock_set_priority);
780 void sock_set_sndtimeo(struct sock *sk, s64 secs)
783 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
784 sk->sk_sndtimeo = secs * HZ;
786 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
789 EXPORT_SYMBOL(sock_set_sndtimeo);
791 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
794 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
795 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
796 sock_set_flag(sk, SOCK_RCVTSTAMP);
797 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
799 sock_reset_flag(sk, SOCK_RCVTSTAMP);
800 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
804 void sock_enable_timestamps(struct sock *sk)
807 __sock_set_timestamps(sk, true, false, true);
810 EXPORT_SYMBOL(sock_enable_timestamps);
812 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
815 case SO_TIMESTAMP_OLD:
816 __sock_set_timestamps(sk, valbool, false, false);
818 case SO_TIMESTAMP_NEW:
819 __sock_set_timestamps(sk, valbool, true, false);
821 case SO_TIMESTAMPNS_OLD:
822 __sock_set_timestamps(sk, valbool, false, true);
824 case SO_TIMESTAMPNS_NEW:
825 __sock_set_timestamps(sk, valbool, true, true);
830 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
832 struct net *net = sock_net(sk);
833 struct net_device *dev = NULL;
838 if (sk->sk_bound_dev_if)
839 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
842 pr_err("%s: sock not bind to device\n", __func__);
846 num = ethtool_get_phc_vclocks(dev, &vclock_index);
847 for (i = 0; i < num; i++) {
848 if (*(vclock_index + i) == phc_index) {
860 sk->sk_bind_phc = phc_index;
865 int sock_set_timestamping(struct sock *sk, int optname,
866 struct so_timestamping timestamping)
868 int val = timestamping.flags;
871 if (val & ~SOF_TIMESTAMPING_MASK)
874 if (val & SOF_TIMESTAMPING_OPT_ID &&
875 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
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 || !sk_has_account(sk))
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_RAW) ||
1139 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1140 sk->sk_priority = val;
1146 if (optlen < sizeof(ling)) {
1147 ret = -EINVAL; /* 1003.1g */
1150 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1155 sock_reset_flag(sk, SOCK_LINGER);
1157 #if (BITS_PER_LONG == 32)
1158 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1159 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1162 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1163 sock_set_flag(sk, SOCK_LINGER);
1172 set_bit(SOCK_PASSCRED, &sock->flags);
1174 clear_bit(SOCK_PASSCRED, &sock->flags);
1177 case SO_TIMESTAMP_OLD:
1178 case SO_TIMESTAMP_NEW:
1179 case SO_TIMESTAMPNS_OLD:
1180 case SO_TIMESTAMPNS_NEW:
1181 sock_set_timestamp(sk, optname, valbool);
1184 case SO_TIMESTAMPING_NEW:
1185 case SO_TIMESTAMPING_OLD:
1186 if (optlen == sizeof(timestamping)) {
1187 if (copy_from_sockptr(×tamping, optval,
1188 sizeof(timestamping))) {
1193 memset(×tamping, 0, sizeof(timestamping));
1194 timestamping.flags = val;
1196 ret = sock_set_timestamping(sk, optname, timestamping);
1202 if (sock->ops->set_rcvlowat)
1203 ret = sock->ops->set_rcvlowat(sk, val);
1205 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1208 case SO_RCVTIMEO_OLD:
1209 case SO_RCVTIMEO_NEW:
1210 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1211 optlen, optname == SO_RCVTIMEO_OLD);
1214 case SO_SNDTIMEO_OLD:
1215 case SO_SNDTIMEO_NEW:
1216 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1217 optlen, optname == SO_SNDTIMEO_OLD);
1220 case SO_ATTACH_FILTER: {
1221 struct sock_fprog fprog;
1223 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1225 ret = sk_attach_filter(&fprog, sk);
1230 if (optlen == sizeof(u32)) {
1234 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1237 ret = sk_attach_bpf(ufd, sk);
1241 case SO_ATTACH_REUSEPORT_CBPF: {
1242 struct sock_fprog fprog;
1244 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1246 ret = sk_reuseport_attach_filter(&fprog, sk);
1249 case SO_ATTACH_REUSEPORT_EBPF:
1251 if (optlen == sizeof(u32)) {
1255 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1258 ret = sk_reuseport_attach_bpf(ufd, sk);
1262 case SO_DETACH_REUSEPORT_BPF:
1263 ret = reuseport_detach_prog(sk);
1266 case SO_DETACH_FILTER:
1267 ret = sk_detach_filter(sk);
1270 case SO_LOCK_FILTER:
1271 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1274 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1279 set_bit(SOCK_PASSSEC, &sock->flags);
1281 clear_bit(SOCK_PASSSEC, &sock->flags);
1284 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1285 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1290 __sock_set_mark(sk, val);
1294 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1297 case SO_WIFI_STATUS:
1298 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1302 if (sock->ops->set_peek_off)
1303 ret = sock->ops->set_peek_off(sk, val);
1309 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1312 case SO_SELECT_ERR_QUEUE:
1313 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1316 #ifdef CONFIG_NET_RX_BUSY_POLL
1318 /* allow unprivileged users to decrease the value */
1319 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1325 WRITE_ONCE(sk->sk_ll_usec, val);
1328 case SO_PREFER_BUSY_POLL:
1329 if (valbool && !capable(CAP_NET_ADMIN))
1332 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1334 case SO_BUSY_POLL_BUDGET:
1335 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1338 if (val < 0 || val > U16_MAX)
1341 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1346 case SO_MAX_PACING_RATE:
1348 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1350 if (sizeof(ulval) != sizeof(val) &&
1351 optlen >= sizeof(ulval) &&
1352 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1357 cmpxchg(&sk->sk_pacing_status,
1360 sk->sk_max_pacing_rate = ulval;
1361 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1364 case SO_INCOMING_CPU:
1365 WRITE_ONCE(sk->sk_incoming_cpu, val);
1370 dst_negative_advice(sk);
1374 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1375 if (!(sk_is_tcp(sk) ||
1376 (sk->sk_type == SOCK_DGRAM &&
1377 sk->sk_protocol == IPPROTO_UDP)))
1379 } else if (sk->sk_family != PF_RDS) {
1383 if (val < 0 || val > 1)
1386 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1391 if (optlen != sizeof(struct sock_txtime)) {
1394 } else if (copy_from_sockptr(&sk_txtime, optval,
1395 sizeof(struct sock_txtime))) {
1398 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1402 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1403 * scheduler has enough safe guards.
1405 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1406 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1410 sock_valbool_flag(sk, SOCK_TXTIME, true);
1411 sk->sk_clockid = sk_txtime.clockid;
1412 sk->sk_txtime_deadline_mode =
1413 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1414 sk->sk_txtime_report_errors =
1415 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1418 case SO_BINDTOIFINDEX:
1419 ret = sock_bindtoindex_locked(sk, val);
1423 if (val & ~SOCK_BUF_LOCK_MASK) {
1427 sk->sk_userlocks = val | (sk->sk_userlocks &
1428 ~SOCK_BUF_LOCK_MASK);
1431 case SO_RESERVE_MEM:
1440 delta = val - sk->sk_reserved_mem;
1442 sock_release_reserved_memory(sk, -delta);
1444 ret = sock_reserve_memory(sk, delta);
1455 EXPORT_SYMBOL(sock_setsockopt);
1457 static const struct cred *sk_get_peer_cred(struct sock *sk)
1459 const struct cred *cred;
1461 spin_lock(&sk->sk_peer_lock);
1462 cred = get_cred(sk->sk_peer_cred);
1463 spin_unlock(&sk->sk_peer_lock);
1468 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1469 struct ucred *ucred)
1471 ucred->pid = pid_vnr(pid);
1472 ucred->uid = ucred->gid = -1;
1474 struct user_namespace *current_ns = current_user_ns();
1476 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1477 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1481 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1483 struct user_namespace *user_ns = current_user_ns();
1486 for (i = 0; i < src->ngroups; i++)
1487 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1493 int sock_getsockopt(struct socket *sock, int level, int optname,
1494 char __user *optval, int __user *optlen)
1496 struct sock *sk = sock->sk;
1501 unsigned long ulval;
1503 struct old_timeval32 tm32;
1504 struct __kernel_old_timeval tm;
1505 struct __kernel_sock_timeval stm;
1506 struct sock_txtime txtime;
1507 struct so_timestamping timestamping;
1510 int lv = sizeof(int);
1513 if (get_user(len, optlen))
1518 memset(&v, 0, sizeof(v));
1522 v.val = sock_flag(sk, SOCK_DBG);
1526 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1530 v.val = sock_flag(sk, SOCK_BROADCAST);
1534 v.val = sk->sk_sndbuf;
1538 v.val = sk->sk_rcvbuf;
1542 v.val = sk->sk_reuse;
1546 v.val = sk->sk_reuseport;
1550 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1554 v.val = sk->sk_type;
1558 v.val = sk->sk_protocol;
1562 v.val = sk->sk_family;
1566 v.val = -sock_error(sk);
1568 v.val = xchg(&sk->sk_err_soft, 0);
1572 v.val = sock_flag(sk, SOCK_URGINLINE);
1576 v.val = sk->sk_no_check_tx;
1580 v.val = sk->sk_priority;
1584 lv = sizeof(v.ling);
1585 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1586 v.ling.l_linger = sk->sk_lingertime / HZ;
1592 case SO_TIMESTAMP_OLD:
1593 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1594 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1595 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1598 case SO_TIMESTAMPNS_OLD:
1599 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1602 case SO_TIMESTAMP_NEW:
1603 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1606 case SO_TIMESTAMPNS_NEW:
1607 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1610 case SO_TIMESTAMPING_OLD:
1611 lv = sizeof(v.timestamping);
1612 v.timestamping.flags = sk->sk_tsflags;
1613 v.timestamping.bind_phc = sk->sk_bind_phc;
1616 case SO_RCVTIMEO_OLD:
1617 case SO_RCVTIMEO_NEW:
1618 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1621 case SO_SNDTIMEO_OLD:
1622 case SO_SNDTIMEO_NEW:
1623 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1627 v.val = sk->sk_rcvlowat;
1635 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1640 struct ucred peercred;
1641 if (len > sizeof(peercred))
1642 len = sizeof(peercred);
1644 spin_lock(&sk->sk_peer_lock);
1645 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1646 spin_unlock(&sk->sk_peer_lock);
1648 if (copy_to_user(optval, &peercred, len))
1655 const struct cred *cred;
1658 cred = sk_get_peer_cred(sk);
1662 n = cred->group_info->ngroups;
1663 if (len < n * sizeof(gid_t)) {
1664 len = n * sizeof(gid_t);
1666 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1668 len = n * sizeof(gid_t);
1670 ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1681 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1686 if (copy_to_user(optval, address, len))
1691 /* Dubious BSD thing... Probably nobody even uses it, but
1692 * the UNIX standard wants it for whatever reason... -DaveM
1695 v.val = sk->sk_state == TCP_LISTEN;
1699 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1703 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1706 v.val = sk->sk_mark;
1710 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1713 case SO_WIFI_STATUS:
1714 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1718 if (!sock->ops->set_peek_off)
1721 v.val = sk->sk_peek_off;
1724 v.val = sock_flag(sk, SOCK_NOFCS);
1727 case SO_BINDTODEVICE:
1728 return sock_getbindtodevice(sk, optval, optlen, len);
1731 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1737 case SO_LOCK_FILTER:
1738 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1741 case SO_BPF_EXTENSIONS:
1742 v.val = bpf_tell_extensions();
1745 case SO_SELECT_ERR_QUEUE:
1746 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1749 #ifdef CONFIG_NET_RX_BUSY_POLL
1751 v.val = sk->sk_ll_usec;
1753 case SO_PREFER_BUSY_POLL:
1754 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1758 case SO_MAX_PACING_RATE:
1759 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1760 lv = sizeof(v.ulval);
1761 v.ulval = sk->sk_max_pacing_rate;
1764 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1768 case SO_INCOMING_CPU:
1769 v.val = READ_ONCE(sk->sk_incoming_cpu);
1774 u32 meminfo[SK_MEMINFO_VARS];
1776 sk_get_meminfo(sk, meminfo);
1778 len = min_t(unsigned int, len, sizeof(meminfo));
1779 if (copy_to_user(optval, &meminfo, len))
1785 #ifdef CONFIG_NET_RX_BUSY_POLL
1786 case SO_INCOMING_NAPI_ID:
1787 v.val = READ_ONCE(sk->sk_napi_id);
1789 /* aggregate non-NAPI IDs down to 0 */
1790 if (v.val < MIN_NAPI_ID)
1800 v.val64 = sock_gen_cookie(sk);
1804 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1808 lv = sizeof(v.txtime);
1809 v.txtime.clockid = sk->sk_clockid;
1810 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1811 SOF_TXTIME_DEADLINE_MODE : 0;
1812 v.txtime.flags |= sk->sk_txtime_report_errors ?
1813 SOF_TXTIME_REPORT_ERRORS : 0;
1816 case SO_BINDTOIFINDEX:
1817 v.val = sk->sk_bound_dev_if;
1820 case SO_NETNS_COOKIE:
1824 v.val64 = sock_net(sk)->net_cookie;
1828 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1831 case SO_RESERVE_MEM:
1832 v.val = sk->sk_reserved_mem;
1836 /* We implement the SO_SNDLOWAT etc to not be settable
1839 return -ENOPROTOOPT;
1844 if (copy_to_user(optval, &v, len))
1847 if (put_user(len, optlen))
1853 * Initialize an sk_lock.
1855 * (We also register the sk_lock with the lock validator.)
1857 static inline void sock_lock_init(struct sock *sk)
1859 if (sk->sk_kern_sock)
1860 sock_lock_init_class_and_name(
1862 af_family_kern_slock_key_strings[sk->sk_family],
1863 af_family_kern_slock_keys + sk->sk_family,
1864 af_family_kern_key_strings[sk->sk_family],
1865 af_family_kern_keys + sk->sk_family);
1867 sock_lock_init_class_and_name(
1869 af_family_slock_key_strings[sk->sk_family],
1870 af_family_slock_keys + sk->sk_family,
1871 af_family_key_strings[sk->sk_family],
1872 af_family_keys + sk->sk_family);
1876 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1877 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1878 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1880 static void sock_copy(struct sock *nsk, const struct sock *osk)
1882 const struct proto *prot = READ_ONCE(osk->sk_prot);
1883 #ifdef CONFIG_SECURITY_NETWORK
1884 void *sptr = nsk->sk_security;
1887 /* If we move sk_tx_queue_mapping out of the private section,
1888 * we must check if sk_tx_queue_clear() is called after
1889 * sock_copy() in sk_clone_lock().
1891 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1892 offsetof(struct sock, sk_dontcopy_begin) ||
1893 offsetof(struct sock, sk_tx_queue_mapping) >=
1894 offsetof(struct sock, sk_dontcopy_end));
1896 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1898 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1899 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1901 #ifdef CONFIG_SECURITY_NETWORK
1902 nsk->sk_security = sptr;
1903 security_sk_clone(osk, nsk);
1907 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1911 struct kmem_cache *slab;
1915 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1918 if (want_init_on_alloc(priority))
1919 sk_prot_clear_nulls(sk, prot->obj_size);
1921 sk = kmalloc(prot->obj_size, priority);
1924 if (security_sk_alloc(sk, family, priority))
1927 if (!try_module_get(prot->owner))
1934 security_sk_free(sk);
1937 kmem_cache_free(slab, sk);
1943 static void sk_prot_free(struct proto *prot, struct sock *sk)
1945 struct kmem_cache *slab;
1946 struct module *owner;
1948 owner = prot->owner;
1951 cgroup_sk_free(&sk->sk_cgrp_data);
1952 mem_cgroup_sk_free(sk);
1953 security_sk_free(sk);
1955 kmem_cache_free(slab, sk);
1962 * sk_alloc - All socket objects are allocated here
1963 * @net: the applicable net namespace
1964 * @family: protocol family
1965 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1966 * @prot: struct proto associated with this new sock instance
1967 * @kern: is this to be a kernel socket?
1969 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1970 struct proto *prot, int kern)
1974 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1976 sk->sk_family = family;
1978 * See comment in struct sock definition to understand
1979 * why we need sk_prot_creator -acme
1981 sk->sk_prot = sk->sk_prot_creator = prot;
1982 sk->sk_kern_sock = kern;
1984 sk->sk_net_refcnt = kern ? 0 : 1;
1985 if (likely(sk->sk_net_refcnt)) {
1986 get_net_track(net, &sk->ns_tracker, priority);
1987 sock_inuse_add(net, 1);
1990 sock_net_set(sk, net);
1991 refcount_set(&sk->sk_wmem_alloc, 1);
1993 mem_cgroup_sk_alloc(sk);
1994 cgroup_sk_alloc(&sk->sk_cgrp_data);
1995 sock_update_classid(&sk->sk_cgrp_data);
1996 sock_update_netprioidx(&sk->sk_cgrp_data);
1997 sk_tx_queue_clear(sk);
2002 EXPORT_SYMBOL(sk_alloc);
2004 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2005 * grace period. This is the case for UDP sockets and TCP listeners.
2007 static void __sk_destruct(struct rcu_head *head)
2009 struct sock *sk = container_of(head, struct sock, sk_rcu);
2010 struct sk_filter *filter;
2012 if (sk->sk_destruct)
2013 sk->sk_destruct(sk);
2015 filter = rcu_dereference_check(sk->sk_filter,
2016 refcount_read(&sk->sk_wmem_alloc) == 0);
2018 sk_filter_uncharge(sk, filter);
2019 RCU_INIT_POINTER(sk->sk_filter, NULL);
2022 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2024 #ifdef CONFIG_BPF_SYSCALL
2025 bpf_sk_storage_free(sk);
2028 if (atomic_read(&sk->sk_omem_alloc))
2029 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2030 __func__, atomic_read(&sk->sk_omem_alloc));
2032 if (sk->sk_frag.page) {
2033 put_page(sk->sk_frag.page);
2034 sk->sk_frag.page = NULL;
2037 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2038 put_cred(sk->sk_peer_cred);
2039 put_pid(sk->sk_peer_pid);
2041 if (likely(sk->sk_net_refcnt))
2042 put_net_track(sock_net(sk), &sk->ns_tracker);
2043 sk_prot_free(sk->sk_prot_creator, sk);
2046 void sk_destruct(struct sock *sk)
2048 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2050 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2051 reuseport_detach_sock(sk);
2052 use_call_rcu = true;
2056 call_rcu(&sk->sk_rcu, __sk_destruct);
2058 __sk_destruct(&sk->sk_rcu);
2061 static void __sk_free(struct sock *sk)
2063 if (likely(sk->sk_net_refcnt))
2064 sock_inuse_add(sock_net(sk), -1);
2066 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2067 sock_diag_broadcast_destroy(sk);
2072 void sk_free(struct sock *sk)
2075 * We subtract one from sk_wmem_alloc and can know if
2076 * some packets are still in some tx queue.
2077 * If not null, sock_wfree() will call __sk_free(sk) later
2079 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2082 EXPORT_SYMBOL(sk_free);
2084 static void sk_init_common(struct sock *sk)
2086 skb_queue_head_init(&sk->sk_receive_queue);
2087 skb_queue_head_init(&sk->sk_write_queue);
2088 skb_queue_head_init(&sk->sk_error_queue);
2090 rwlock_init(&sk->sk_callback_lock);
2091 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2092 af_rlock_keys + sk->sk_family,
2093 af_family_rlock_key_strings[sk->sk_family]);
2094 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2095 af_wlock_keys + sk->sk_family,
2096 af_family_wlock_key_strings[sk->sk_family]);
2097 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2098 af_elock_keys + sk->sk_family,
2099 af_family_elock_key_strings[sk->sk_family]);
2100 lockdep_set_class_and_name(&sk->sk_callback_lock,
2101 af_callback_keys + sk->sk_family,
2102 af_family_clock_key_strings[sk->sk_family]);
2106 * sk_clone_lock - clone a socket, and lock its clone
2107 * @sk: the socket to clone
2108 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2110 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2112 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2114 struct proto *prot = READ_ONCE(sk->sk_prot);
2115 struct sk_filter *filter;
2116 bool is_charged = true;
2119 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2123 sock_copy(newsk, sk);
2125 newsk->sk_prot_creator = prot;
2128 if (likely(newsk->sk_net_refcnt)) {
2129 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2130 sock_inuse_add(sock_net(newsk), 1);
2132 sk_node_init(&newsk->sk_node);
2133 sock_lock_init(newsk);
2134 bh_lock_sock(newsk);
2135 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2136 newsk->sk_backlog.len = 0;
2138 atomic_set(&newsk->sk_rmem_alloc, 0);
2140 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2141 refcount_set(&newsk->sk_wmem_alloc, 1);
2143 atomic_set(&newsk->sk_omem_alloc, 0);
2144 sk_init_common(newsk);
2146 newsk->sk_dst_cache = NULL;
2147 newsk->sk_dst_pending_confirm = 0;
2148 newsk->sk_wmem_queued = 0;
2149 newsk->sk_forward_alloc = 0;
2150 newsk->sk_reserved_mem = 0;
2151 atomic_set(&newsk->sk_drops, 0);
2152 newsk->sk_send_head = NULL;
2153 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2154 atomic_set(&newsk->sk_zckey, 0);
2156 sock_reset_flag(newsk, SOCK_DONE);
2158 /* sk->sk_memcg will be populated at accept() time */
2159 newsk->sk_memcg = NULL;
2161 cgroup_sk_clone(&newsk->sk_cgrp_data);
2164 filter = rcu_dereference(sk->sk_filter);
2166 /* though it's an empty new sock, the charging may fail
2167 * if sysctl_optmem_max was changed between creation of
2168 * original socket and cloning
2170 is_charged = sk_filter_charge(newsk, filter);
2171 RCU_INIT_POINTER(newsk->sk_filter, filter);
2174 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2175 /* We need to make sure that we don't uncharge the new
2176 * socket if we couldn't charge it in the first place
2177 * as otherwise we uncharge the parent's filter.
2180 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2181 sk_free_unlock_clone(newsk);
2185 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2187 if (bpf_sk_storage_clone(sk, newsk)) {
2188 sk_free_unlock_clone(newsk);
2193 /* Clear sk_user_data if parent had the pointer tagged
2194 * as not suitable for copying when cloning.
2196 if (sk_user_data_is_nocopy(newsk))
2197 newsk->sk_user_data = NULL;
2200 newsk->sk_err_soft = 0;
2201 newsk->sk_priority = 0;
2202 newsk->sk_incoming_cpu = raw_smp_processor_id();
2204 /* Before updating sk_refcnt, we must commit prior changes to memory
2205 * (Documentation/RCU/rculist_nulls.rst for details)
2208 refcount_set(&newsk->sk_refcnt, 2);
2210 /* Increment the counter in the same struct proto as the master
2211 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2212 * is the same as sk->sk_prot->socks, as this field was copied
2215 * This _changes_ the previous behaviour, where
2216 * tcp_create_openreq_child always was incrementing the
2217 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2218 * to be taken into account in all callers. -acme
2220 sk_refcnt_debug_inc(newsk);
2221 sk_set_socket(newsk, NULL);
2222 sk_tx_queue_clear(newsk);
2223 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2225 if (newsk->sk_prot->sockets_allocated)
2226 sk_sockets_allocated_inc(newsk);
2228 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2229 net_enable_timestamp();
2233 EXPORT_SYMBOL_GPL(sk_clone_lock);
2235 void sk_free_unlock_clone(struct sock *sk)
2237 /* It is still raw copy of parent, so invalidate
2238 * destructor and make plain sk_free() */
2239 sk->sk_destruct = NULL;
2243 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2245 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2249 sk_dst_set(sk, dst);
2250 sk->sk_route_caps = dst->dev->features;
2252 sk->sk_route_caps |= NETIF_F_GSO;
2253 if (sk->sk_route_caps & NETIF_F_GSO)
2254 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2255 if (unlikely(sk->sk_gso_disabled))
2256 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2257 if (sk_can_gso(sk)) {
2258 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2259 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2261 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2262 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */
2263 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size);
2264 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2265 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2268 sk->sk_gso_max_segs = max_segs;
2270 EXPORT_SYMBOL_GPL(sk_setup_caps);
2273 * Simple resource managers for sockets.
2278 * Write buffer destructor automatically called from kfree_skb.
2280 void sock_wfree(struct sk_buff *skb)
2282 struct sock *sk = skb->sk;
2283 unsigned int len = skb->truesize;
2285 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2287 * Keep a reference on sk_wmem_alloc, this will be released
2288 * after sk_write_space() call
2290 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2291 sk->sk_write_space(sk);
2295 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2296 * could not do because of in-flight packets
2298 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2301 EXPORT_SYMBOL(sock_wfree);
2303 /* This variant of sock_wfree() is used by TCP,
2304 * since it sets SOCK_USE_WRITE_QUEUE.
2306 void __sock_wfree(struct sk_buff *skb)
2308 struct sock *sk = skb->sk;
2310 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2314 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2319 if (unlikely(!sk_fullsock(sk))) {
2320 skb->destructor = sock_edemux;
2325 skb->destructor = sock_wfree;
2326 skb_set_hash_from_sk(skb, sk);
2328 * We used to take a refcount on sk, but following operation
2329 * is enough to guarantee sk_free() wont free this sock until
2330 * all in-flight packets are completed
2332 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2334 EXPORT_SYMBOL(skb_set_owner_w);
2336 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2338 #ifdef CONFIG_TLS_DEVICE
2339 /* Drivers depend on in-order delivery for crypto offload,
2340 * partial orphan breaks out-of-order-OK logic.
2345 return (skb->destructor == sock_wfree ||
2346 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2349 /* This helper is used by netem, as it can hold packets in its
2350 * delay queue. We want to allow the owner socket to send more
2351 * packets, as if they were already TX completed by a typical driver.
2352 * But we also want to keep skb->sk set because some packet schedulers
2353 * rely on it (sch_fq for example).
2355 void skb_orphan_partial(struct sk_buff *skb)
2357 if (skb_is_tcp_pure_ack(skb))
2360 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2365 EXPORT_SYMBOL(skb_orphan_partial);
2368 * Read buffer destructor automatically called from kfree_skb.
2370 void sock_rfree(struct sk_buff *skb)
2372 struct sock *sk = skb->sk;
2373 unsigned int len = skb->truesize;
2375 atomic_sub(len, &sk->sk_rmem_alloc);
2376 sk_mem_uncharge(sk, len);
2378 EXPORT_SYMBOL(sock_rfree);
2381 * Buffer destructor for skbs that are not used directly in read or write
2382 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2384 void sock_efree(struct sk_buff *skb)
2388 EXPORT_SYMBOL(sock_efree);
2390 /* Buffer destructor for prefetch/receive path where reference count may
2391 * not be held, e.g. for listen sockets.
2394 void sock_pfree(struct sk_buff *skb)
2396 if (sk_is_refcounted(skb->sk))
2397 sock_gen_put(skb->sk);
2399 EXPORT_SYMBOL(sock_pfree);
2400 #endif /* CONFIG_INET */
2402 kuid_t sock_i_uid(struct sock *sk)
2406 read_lock_bh(&sk->sk_callback_lock);
2407 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2408 read_unlock_bh(&sk->sk_callback_lock);
2411 EXPORT_SYMBOL(sock_i_uid);
2413 unsigned long sock_i_ino(struct sock *sk)
2417 read_lock_bh(&sk->sk_callback_lock);
2418 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2419 read_unlock_bh(&sk->sk_callback_lock);
2422 EXPORT_SYMBOL(sock_i_ino);
2425 * Allocate a skb from the socket's send buffer.
2427 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2431 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2432 struct sk_buff *skb = alloc_skb(size, priority);
2435 skb_set_owner_w(skb, sk);
2441 EXPORT_SYMBOL(sock_wmalloc);
2443 static void sock_ofree(struct sk_buff *skb)
2445 struct sock *sk = skb->sk;
2447 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2450 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2453 struct sk_buff *skb;
2455 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2456 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2460 skb = alloc_skb(size, priority);
2464 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2466 skb->destructor = sock_ofree;
2471 * Allocate a memory block from the socket's option memory buffer.
2473 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2475 if ((unsigned int)size <= sysctl_optmem_max &&
2476 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2478 /* First do the add, to avoid the race if kmalloc
2481 atomic_add(size, &sk->sk_omem_alloc);
2482 mem = kmalloc(size, priority);
2485 atomic_sub(size, &sk->sk_omem_alloc);
2489 EXPORT_SYMBOL(sock_kmalloc);
2491 /* Free an option memory block. Note, we actually want the inline
2492 * here as this allows gcc to detect the nullify and fold away the
2493 * condition entirely.
2495 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2498 if (WARN_ON_ONCE(!mem))
2501 kfree_sensitive(mem);
2504 atomic_sub(size, &sk->sk_omem_alloc);
2507 void sock_kfree_s(struct sock *sk, void *mem, int size)
2509 __sock_kfree_s(sk, mem, size, false);
2511 EXPORT_SYMBOL(sock_kfree_s);
2513 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2515 __sock_kfree_s(sk, mem, size, true);
2517 EXPORT_SYMBOL(sock_kzfree_s);
2519 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2520 I think, these locks should be removed for datagram sockets.
2522 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2526 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2530 if (signal_pending(current))
2532 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2533 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2534 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2536 if (sk->sk_shutdown & SEND_SHUTDOWN)
2540 timeo = schedule_timeout(timeo);
2542 finish_wait(sk_sleep(sk), &wait);
2548 * Generic send/receive buffer handlers
2551 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2552 unsigned long data_len, int noblock,
2553 int *errcode, int max_page_order)
2555 struct sk_buff *skb;
2559 timeo = sock_sndtimeo(sk, noblock);
2561 err = sock_error(sk);
2566 if (sk->sk_shutdown & SEND_SHUTDOWN)
2569 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2572 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2573 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2577 if (signal_pending(current))
2579 timeo = sock_wait_for_wmem(sk, timeo);
2581 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2582 errcode, sk->sk_allocation);
2584 skb_set_owner_w(skb, sk);
2588 err = sock_intr_errno(timeo);
2593 EXPORT_SYMBOL(sock_alloc_send_pskb);
2595 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2596 int noblock, int *errcode)
2598 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2600 EXPORT_SYMBOL(sock_alloc_send_skb);
2602 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2603 struct sockcm_cookie *sockc)
2607 switch (cmsg->cmsg_type) {
2609 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2611 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2613 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2615 case SO_TIMESTAMPING_OLD:
2616 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2619 tsflags = *(u32 *)CMSG_DATA(cmsg);
2620 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2623 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2624 sockc->tsflags |= tsflags;
2627 if (!sock_flag(sk, SOCK_TXTIME))
2629 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2631 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2633 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2635 case SCM_CREDENTIALS:
2642 EXPORT_SYMBOL(__sock_cmsg_send);
2644 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2645 struct sockcm_cookie *sockc)
2647 struct cmsghdr *cmsg;
2650 for_each_cmsghdr(cmsg, msg) {
2651 if (!CMSG_OK(msg, cmsg))
2653 if (cmsg->cmsg_level != SOL_SOCKET)
2655 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2661 EXPORT_SYMBOL(sock_cmsg_send);
2663 static void sk_enter_memory_pressure(struct sock *sk)
2665 if (!sk->sk_prot->enter_memory_pressure)
2668 sk->sk_prot->enter_memory_pressure(sk);
2671 static void sk_leave_memory_pressure(struct sock *sk)
2673 if (sk->sk_prot->leave_memory_pressure) {
2674 sk->sk_prot->leave_memory_pressure(sk);
2676 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2678 if (memory_pressure && READ_ONCE(*memory_pressure))
2679 WRITE_ONCE(*memory_pressure, 0);
2683 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2686 * skb_page_frag_refill - check that a page_frag contains enough room
2687 * @sz: minimum size of the fragment we want to get
2688 * @pfrag: pointer to page_frag
2689 * @gfp: priority for memory allocation
2691 * Note: While this allocator tries to use high order pages, there is
2692 * no guarantee that allocations succeed. Therefore, @sz MUST be
2693 * less or equal than PAGE_SIZE.
2695 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2698 if (page_ref_count(pfrag->page) == 1) {
2702 if (pfrag->offset + sz <= pfrag->size)
2704 put_page(pfrag->page);
2708 if (SKB_FRAG_PAGE_ORDER &&
2709 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2710 /* Avoid direct reclaim but allow kswapd to wake */
2711 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2712 __GFP_COMP | __GFP_NOWARN |
2714 SKB_FRAG_PAGE_ORDER);
2715 if (likely(pfrag->page)) {
2716 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2720 pfrag->page = alloc_page(gfp);
2721 if (likely(pfrag->page)) {
2722 pfrag->size = PAGE_SIZE;
2727 EXPORT_SYMBOL(skb_page_frag_refill);
2729 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2731 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2734 sk_enter_memory_pressure(sk);
2735 sk_stream_moderate_sndbuf(sk);
2738 EXPORT_SYMBOL(sk_page_frag_refill);
2740 void __lock_sock(struct sock *sk)
2741 __releases(&sk->sk_lock.slock)
2742 __acquires(&sk->sk_lock.slock)
2747 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2748 TASK_UNINTERRUPTIBLE);
2749 spin_unlock_bh(&sk->sk_lock.slock);
2751 spin_lock_bh(&sk->sk_lock.slock);
2752 if (!sock_owned_by_user(sk))
2755 finish_wait(&sk->sk_lock.wq, &wait);
2758 void __release_sock(struct sock *sk)
2759 __releases(&sk->sk_lock.slock)
2760 __acquires(&sk->sk_lock.slock)
2762 struct sk_buff *skb, *next;
2764 while ((skb = sk->sk_backlog.head) != NULL) {
2765 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2767 spin_unlock_bh(&sk->sk_lock.slock);
2772 WARN_ON_ONCE(skb_dst_is_noref(skb));
2773 skb_mark_not_on_list(skb);
2774 sk_backlog_rcv(sk, skb);
2779 } while (skb != NULL);
2781 spin_lock_bh(&sk->sk_lock.slock);
2785 * Doing the zeroing here guarantee we can not loop forever
2786 * while a wild producer attempts to flood us.
2788 sk->sk_backlog.len = 0;
2791 void __sk_flush_backlog(struct sock *sk)
2793 spin_lock_bh(&sk->sk_lock.slock);
2795 spin_unlock_bh(&sk->sk_lock.slock);
2799 * sk_wait_data - wait for data to arrive at sk_receive_queue
2800 * @sk: sock to wait on
2801 * @timeo: for how long
2802 * @skb: last skb seen on sk_receive_queue
2804 * Now socket state including sk->sk_err is changed only under lock,
2805 * hence we may omit checks after joining wait queue.
2806 * We check receive queue before schedule() only as optimization;
2807 * it is very likely that release_sock() added new data.
2809 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2811 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2814 add_wait_queue(sk_sleep(sk), &wait);
2815 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2816 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2817 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2818 remove_wait_queue(sk_sleep(sk), &wait);
2821 EXPORT_SYMBOL(sk_wait_data);
2824 * __sk_mem_raise_allocated - increase memory_allocated
2826 * @size: memory size to allocate
2827 * @amt: pages to allocate
2828 * @kind: allocation type
2830 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2832 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2834 struct proto *prot = sk->sk_prot;
2835 long allocated = sk_memory_allocated_add(sk, amt);
2836 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2837 bool charged = true;
2840 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2841 gfp_memcg_charge())))
2842 goto suppress_allocation;
2845 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2846 sk_leave_memory_pressure(sk);
2850 /* Under pressure. */
2851 if (allocated > sk_prot_mem_limits(sk, 1))
2852 sk_enter_memory_pressure(sk);
2854 /* Over hard limit. */
2855 if (allocated > sk_prot_mem_limits(sk, 2))
2856 goto suppress_allocation;
2858 /* guarantee minimum buffer size under pressure */
2859 if (kind == SK_MEM_RECV) {
2860 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2863 } else { /* SK_MEM_SEND */
2864 int wmem0 = sk_get_wmem0(sk, prot);
2866 if (sk->sk_type == SOCK_STREAM) {
2867 if (sk->sk_wmem_queued < wmem0)
2869 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2874 if (sk_has_memory_pressure(sk)) {
2877 if (!sk_under_memory_pressure(sk))
2879 alloc = sk_sockets_allocated_read_positive(sk);
2880 if (sk_prot_mem_limits(sk, 2) > alloc *
2881 sk_mem_pages(sk->sk_wmem_queued +
2882 atomic_read(&sk->sk_rmem_alloc) +
2883 sk->sk_forward_alloc))
2887 suppress_allocation:
2889 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2890 sk_stream_moderate_sndbuf(sk);
2892 /* Fail only if socket is _under_ its sndbuf.
2893 * In this case we cannot block, so that we have to fail.
2895 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
2896 /* Force charge with __GFP_NOFAIL */
2897 if (memcg_charge && !charged) {
2898 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2899 gfp_memcg_charge() | __GFP_NOFAIL);
2905 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2906 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2908 sk_memory_allocated_sub(sk, amt);
2910 if (memcg_charge && charged)
2911 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2915 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2918 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2920 * @size: memory size to allocate
2921 * @kind: allocation type
2923 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2924 * rmem allocation. This function assumes that protocols which have
2925 * memory_pressure use sk_wmem_queued as write buffer accounting.
2927 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2929 int ret, amt = sk_mem_pages(size);
2931 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2932 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2934 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2937 EXPORT_SYMBOL(__sk_mem_schedule);
2940 * __sk_mem_reduce_allocated - reclaim memory_allocated
2942 * @amount: number of quanta
2944 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2946 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2948 sk_memory_allocated_sub(sk, amount);
2950 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2951 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2953 if (sk_under_memory_pressure(sk) &&
2954 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2955 sk_leave_memory_pressure(sk);
2957 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2960 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2962 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2964 void __sk_mem_reclaim(struct sock *sk, int amount)
2966 amount >>= SK_MEM_QUANTUM_SHIFT;
2967 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2968 __sk_mem_reduce_allocated(sk, amount);
2970 EXPORT_SYMBOL(__sk_mem_reclaim);
2972 int sk_set_peek_off(struct sock *sk, int val)
2974 sk->sk_peek_off = val;
2977 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2980 * Set of default routines for initialising struct proto_ops when
2981 * the protocol does not support a particular function. In certain
2982 * cases where it makes no sense for a protocol to have a "do nothing"
2983 * function, some default processing is provided.
2986 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2990 EXPORT_SYMBOL(sock_no_bind);
2992 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2997 EXPORT_SYMBOL(sock_no_connect);
2999 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3003 EXPORT_SYMBOL(sock_no_socketpair);
3005 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3010 EXPORT_SYMBOL(sock_no_accept);
3012 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3017 EXPORT_SYMBOL(sock_no_getname);
3019 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3023 EXPORT_SYMBOL(sock_no_ioctl);
3025 int sock_no_listen(struct socket *sock, int backlog)
3029 EXPORT_SYMBOL(sock_no_listen);
3031 int sock_no_shutdown(struct socket *sock, int how)
3035 EXPORT_SYMBOL(sock_no_shutdown);
3037 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3041 EXPORT_SYMBOL(sock_no_sendmsg);
3043 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3047 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3049 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3054 EXPORT_SYMBOL(sock_no_recvmsg);
3056 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3058 /* Mirror missing mmap method error code */
3061 EXPORT_SYMBOL(sock_no_mmap);
3064 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3065 * various sock-based usage counts.
3067 void __receive_sock(struct file *file)
3069 struct socket *sock;
3071 sock = sock_from_file(file);
3073 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3074 sock_update_classid(&sock->sk->sk_cgrp_data);
3078 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3081 struct msghdr msg = {.msg_flags = flags};
3083 char *kaddr = kmap(page);
3084 iov.iov_base = kaddr + offset;
3086 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3090 EXPORT_SYMBOL(sock_no_sendpage);
3092 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3093 int offset, size_t size, int flags)
3096 struct msghdr msg = {.msg_flags = flags};
3098 char *kaddr = kmap(page);
3100 iov.iov_base = kaddr + offset;
3102 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3106 EXPORT_SYMBOL(sock_no_sendpage_locked);
3109 * Default Socket Callbacks
3112 static void sock_def_wakeup(struct sock *sk)
3114 struct socket_wq *wq;
3117 wq = rcu_dereference(sk->sk_wq);
3118 if (skwq_has_sleeper(wq))
3119 wake_up_interruptible_all(&wq->wait);
3123 static void sock_def_error_report(struct sock *sk)
3125 struct socket_wq *wq;
3128 wq = rcu_dereference(sk->sk_wq);
3129 if (skwq_has_sleeper(wq))
3130 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3131 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3135 void sock_def_readable(struct sock *sk)
3137 struct socket_wq *wq;
3140 wq = rcu_dereference(sk->sk_wq);
3141 if (skwq_has_sleeper(wq))
3142 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3143 EPOLLRDNORM | EPOLLRDBAND);
3144 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3148 static void sock_def_write_space(struct sock *sk)
3150 struct socket_wq *wq;
3154 /* Do not wake up a writer until he can make "significant"
3157 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
3158 wq = rcu_dereference(sk->sk_wq);
3159 if (skwq_has_sleeper(wq))
3160 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3161 EPOLLWRNORM | EPOLLWRBAND);
3163 /* Should agree with poll, otherwise some programs break */
3164 if (sock_writeable(sk))
3165 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3171 static void sock_def_destruct(struct sock *sk)
3175 void sk_send_sigurg(struct sock *sk)
3177 if (sk->sk_socket && sk->sk_socket->file)
3178 if (send_sigurg(&sk->sk_socket->file->f_owner))
3179 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3181 EXPORT_SYMBOL(sk_send_sigurg);
3183 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3184 unsigned long expires)
3186 if (!mod_timer(timer, expires))
3189 EXPORT_SYMBOL(sk_reset_timer);
3191 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3193 if (del_timer(timer))
3196 EXPORT_SYMBOL(sk_stop_timer);
3198 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3200 if (del_timer_sync(timer))
3203 EXPORT_SYMBOL(sk_stop_timer_sync);
3205 void sock_init_data(struct socket *sock, struct sock *sk)
3208 sk->sk_send_head = NULL;
3210 timer_setup(&sk->sk_timer, NULL, 0);
3212 sk->sk_allocation = GFP_KERNEL;
3213 sk->sk_rcvbuf = sysctl_rmem_default;
3214 sk->sk_sndbuf = sysctl_wmem_default;
3215 sk->sk_state = TCP_CLOSE;
3216 sk_set_socket(sk, sock);
3218 sock_set_flag(sk, SOCK_ZAPPED);
3221 sk->sk_type = sock->type;
3222 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3224 sk->sk_uid = SOCK_INODE(sock)->i_uid;
3226 RCU_INIT_POINTER(sk->sk_wq, NULL);
3227 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3230 rwlock_init(&sk->sk_callback_lock);
3231 if (sk->sk_kern_sock)
3232 lockdep_set_class_and_name(
3233 &sk->sk_callback_lock,
3234 af_kern_callback_keys + sk->sk_family,
3235 af_family_kern_clock_key_strings[sk->sk_family]);
3237 lockdep_set_class_and_name(
3238 &sk->sk_callback_lock,
3239 af_callback_keys + sk->sk_family,
3240 af_family_clock_key_strings[sk->sk_family]);
3242 sk->sk_state_change = sock_def_wakeup;
3243 sk->sk_data_ready = sock_def_readable;
3244 sk->sk_write_space = sock_def_write_space;
3245 sk->sk_error_report = sock_def_error_report;
3246 sk->sk_destruct = sock_def_destruct;
3248 sk->sk_frag.page = NULL;
3249 sk->sk_frag.offset = 0;
3250 sk->sk_peek_off = -1;
3252 sk->sk_peer_pid = NULL;
3253 sk->sk_peer_cred = NULL;
3254 spin_lock_init(&sk->sk_peer_lock);
3256 sk->sk_write_pending = 0;
3257 sk->sk_rcvlowat = 1;
3258 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3259 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3261 sk->sk_stamp = SK_DEFAULT_STAMP;
3262 #if BITS_PER_LONG==32
3263 seqlock_init(&sk->sk_stamp_seq);
3265 atomic_set(&sk->sk_zckey, 0);
3267 #ifdef CONFIG_NET_RX_BUSY_POLL
3269 sk->sk_ll_usec = sysctl_net_busy_read;
3272 sk->sk_max_pacing_rate = ~0UL;
3273 sk->sk_pacing_rate = ~0UL;
3274 WRITE_ONCE(sk->sk_pacing_shift, 10);
3275 sk->sk_incoming_cpu = -1;
3277 sk_rx_queue_clear(sk);
3279 * Before updating sk_refcnt, we must commit prior changes to memory
3280 * (Documentation/RCU/rculist_nulls.rst for details)
3283 refcount_set(&sk->sk_refcnt, 1);
3284 atomic_set(&sk->sk_drops, 0);
3286 EXPORT_SYMBOL(sock_init_data);
3288 void lock_sock_nested(struct sock *sk, int subclass)
3290 /* The sk_lock has mutex_lock() semantics here. */
3291 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3294 spin_lock_bh(&sk->sk_lock.slock);
3295 if (sock_owned_by_user_nocheck(sk))
3297 sk->sk_lock.owned = 1;
3298 spin_unlock_bh(&sk->sk_lock.slock);
3300 EXPORT_SYMBOL(lock_sock_nested);
3302 void release_sock(struct sock *sk)
3304 spin_lock_bh(&sk->sk_lock.slock);
3305 if (sk->sk_backlog.tail)
3308 /* Warning : release_cb() might need to release sk ownership,
3309 * ie call sock_release_ownership(sk) before us.
3311 if (sk->sk_prot->release_cb)
3312 sk->sk_prot->release_cb(sk);
3314 sock_release_ownership(sk);
3315 if (waitqueue_active(&sk->sk_lock.wq))
3316 wake_up(&sk->sk_lock.wq);
3317 spin_unlock_bh(&sk->sk_lock.slock);
3319 EXPORT_SYMBOL(release_sock);
3321 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3324 spin_lock_bh(&sk->sk_lock.slock);
3326 if (!sock_owned_by_user_nocheck(sk)) {
3328 * Fast path return with bottom halves disabled and
3329 * sock::sk_lock.slock held.
3331 * The 'mutex' is not contended and holding
3332 * sock::sk_lock.slock prevents all other lockers to
3333 * proceed so the corresponding unlock_sock_fast() can
3334 * avoid the slow path of release_sock() completely and
3335 * just release slock.
3337 * From a semantical POV this is equivalent to 'acquiring'
3338 * the 'mutex', hence the corresponding lockdep
3339 * mutex_release() has to happen in the fast path of
3340 * unlock_sock_fast().
3346 sk->sk_lock.owned = 1;
3347 __acquire(&sk->sk_lock.slock);
3348 spin_unlock_bh(&sk->sk_lock.slock);
3351 EXPORT_SYMBOL(__lock_sock_fast);
3353 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3354 bool timeval, bool time32)
3356 struct sock *sk = sock->sk;
3357 struct timespec64 ts;
3359 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3360 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3361 if (ts.tv_sec == -1)
3363 if (ts.tv_sec == 0) {
3364 ktime_t kt = ktime_get_real();
3365 sock_write_timestamp(sk, kt);
3366 ts = ktime_to_timespec64(kt);
3372 #ifdef CONFIG_COMPAT_32BIT_TIME
3374 return put_old_timespec32(&ts, userstamp);
3376 #ifdef CONFIG_SPARC64
3377 /* beware of padding in sparc64 timeval */
3378 if (timeval && !in_compat_syscall()) {
3379 struct __kernel_old_timeval __user tv = {
3380 .tv_sec = ts.tv_sec,
3381 .tv_usec = ts.tv_nsec,
3383 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3388 return put_timespec64(&ts, userstamp);
3390 EXPORT_SYMBOL(sock_gettstamp);
3392 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3394 if (!sock_flag(sk, flag)) {
3395 unsigned long previous_flags = sk->sk_flags;
3397 sock_set_flag(sk, flag);
3399 * we just set one of the two flags which require net
3400 * time stamping, but time stamping might have been on
3401 * already because of the other one
3403 if (sock_needs_netstamp(sk) &&
3404 !(previous_flags & SK_FLAGS_TIMESTAMP))
3405 net_enable_timestamp();
3409 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3410 int level, int type)
3412 struct sock_exterr_skb *serr;
3413 struct sk_buff *skb;
3417 skb = sock_dequeue_err_skb(sk);
3423 msg->msg_flags |= MSG_TRUNC;
3426 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3430 sock_recv_timestamp(msg, sk, skb);
3432 serr = SKB_EXT_ERR(skb);
3433 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3435 msg->msg_flags |= MSG_ERRQUEUE;
3443 EXPORT_SYMBOL(sock_recv_errqueue);
3446 * Get a socket option on an socket.
3448 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3449 * asynchronous errors should be reported by getsockopt. We assume
3450 * this means if you specify SO_ERROR (otherwise whats the point of it).
3452 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3453 char __user *optval, int __user *optlen)
3455 struct sock *sk = sock->sk;
3457 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3459 EXPORT_SYMBOL(sock_common_getsockopt);
3461 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3464 struct sock *sk = sock->sk;
3468 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3469 flags & ~MSG_DONTWAIT, &addr_len);
3471 msg->msg_namelen = addr_len;
3474 EXPORT_SYMBOL(sock_common_recvmsg);
3477 * Set socket options on an inet socket.
3479 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3480 sockptr_t optval, unsigned int optlen)
3482 struct sock *sk = sock->sk;
3484 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3486 EXPORT_SYMBOL(sock_common_setsockopt);
3488 void sk_common_release(struct sock *sk)
3490 if (sk->sk_prot->destroy)
3491 sk->sk_prot->destroy(sk);
3494 * Observation: when sk_common_release is called, processes have
3495 * no access to socket. But net still has.
3496 * Step one, detach it from networking:
3498 * A. Remove from hash tables.
3501 sk->sk_prot->unhash(sk);
3504 * In this point socket cannot receive new packets, but it is possible
3505 * that some packets are in flight because some CPU runs receiver and
3506 * did hash table lookup before we unhashed socket. They will achieve
3507 * receive queue and will be purged by socket destructor.
3509 * Also we still have packets pending on receive queue and probably,
3510 * our own packets waiting in device queues. sock_destroy will drain
3511 * receive queue, but transmitted packets will delay socket destruction
3512 * until the last reference will be released.
3517 xfrm_sk_free_policy(sk);
3519 sk_refcnt_debug_release(sk);
3523 EXPORT_SYMBOL(sk_common_release);
3525 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3527 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3529 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3530 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3531 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3532 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3533 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3534 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3535 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3536 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3537 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3540 #ifdef CONFIG_PROC_FS
3541 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3543 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3545 int cpu, idx = prot->inuse_idx;
3548 for_each_possible_cpu(cpu)
3549 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3551 return res >= 0 ? res : 0;
3553 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3555 int sock_inuse_get(struct net *net)
3559 for_each_possible_cpu(cpu)
3560 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3565 EXPORT_SYMBOL_GPL(sock_inuse_get);
3567 static int __net_init sock_inuse_init_net(struct net *net)
3569 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3570 if (net->core.prot_inuse == NULL)
3575 static void __net_exit sock_inuse_exit_net(struct net *net)
3577 free_percpu(net->core.prot_inuse);
3580 static struct pernet_operations net_inuse_ops = {
3581 .init = sock_inuse_init_net,
3582 .exit = sock_inuse_exit_net,
3585 static __init int net_inuse_init(void)
3587 if (register_pernet_subsys(&net_inuse_ops))
3588 panic("Cannot initialize net inuse counters");
3593 core_initcall(net_inuse_init);
3595 static int assign_proto_idx(struct proto *prot)
3597 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3599 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3600 pr_err("PROTO_INUSE_NR exhausted\n");
3604 set_bit(prot->inuse_idx, proto_inuse_idx);
3608 static void release_proto_idx(struct proto *prot)
3610 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3611 clear_bit(prot->inuse_idx, proto_inuse_idx);
3614 static inline int assign_proto_idx(struct proto *prot)
3619 static inline void release_proto_idx(struct proto *prot)
3625 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3629 kfree(twsk_prot->twsk_slab_name);
3630 twsk_prot->twsk_slab_name = NULL;
3631 kmem_cache_destroy(twsk_prot->twsk_slab);
3632 twsk_prot->twsk_slab = NULL;
3635 static int tw_prot_init(const struct proto *prot)
3637 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3642 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3644 if (!twsk_prot->twsk_slab_name)
3647 twsk_prot->twsk_slab =
3648 kmem_cache_create(twsk_prot->twsk_slab_name,
3649 twsk_prot->twsk_obj_size, 0,
3650 SLAB_ACCOUNT | prot->slab_flags,
3652 if (!twsk_prot->twsk_slab) {
3653 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3661 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3665 kfree(rsk_prot->slab_name);
3666 rsk_prot->slab_name = NULL;
3667 kmem_cache_destroy(rsk_prot->slab);
3668 rsk_prot->slab = NULL;
3671 static int req_prot_init(const struct proto *prot)
3673 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3678 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3680 if (!rsk_prot->slab_name)
3683 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3684 rsk_prot->obj_size, 0,
3685 SLAB_ACCOUNT | prot->slab_flags,
3688 if (!rsk_prot->slab) {
3689 pr_crit("%s: Can't create request sock SLAB cache!\n",
3696 int proto_register(struct proto *prot, int alloc_slab)
3701 prot->slab = kmem_cache_create_usercopy(prot->name,
3703 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3705 prot->useroffset, prot->usersize,
3708 if (prot->slab == NULL) {
3709 pr_crit("%s: Can't create sock SLAB cache!\n",
3714 if (req_prot_init(prot))
3715 goto out_free_request_sock_slab;
3717 if (tw_prot_init(prot))
3718 goto out_free_timewait_sock_slab;
3721 mutex_lock(&proto_list_mutex);
3722 ret = assign_proto_idx(prot);
3724 mutex_unlock(&proto_list_mutex);
3725 goto out_free_timewait_sock_slab;
3727 list_add(&prot->node, &proto_list);
3728 mutex_unlock(&proto_list_mutex);
3731 out_free_timewait_sock_slab:
3733 tw_prot_cleanup(prot->twsk_prot);
3734 out_free_request_sock_slab:
3736 req_prot_cleanup(prot->rsk_prot);
3738 kmem_cache_destroy(prot->slab);
3744 EXPORT_SYMBOL(proto_register);
3746 void proto_unregister(struct proto *prot)
3748 mutex_lock(&proto_list_mutex);
3749 release_proto_idx(prot);
3750 list_del(&prot->node);
3751 mutex_unlock(&proto_list_mutex);
3753 kmem_cache_destroy(prot->slab);
3756 req_prot_cleanup(prot->rsk_prot);
3757 tw_prot_cleanup(prot->twsk_prot);
3759 EXPORT_SYMBOL(proto_unregister);
3761 int sock_load_diag_module(int family, int protocol)
3764 if (!sock_is_registered(family))
3767 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3768 NETLINK_SOCK_DIAG, family);
3772 if (family == AF_INET &&
3773 protocol != IPPROTO_RAW &&
3774 protocol < MAX_INET_PROTOS &&
3775 !rcu_access_pointer(inet_protos[protocol]))
3779 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3780 NETLINK_SOCK_DIAG, family, protocol);
3782 EXPORT_SYMBOL(sock_load_diag_module);
3784 #ifdef CONFIG_PROC_FS
3785 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3786 __acquires(proto_list_mutex)
3788 mutex_lock(&proto_list_mutex);
3789 return seq_list_start_head(&proto_list, *pos);
3792 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3794 return seq_list_next(v, &proto_list, pos);
3797 static void proto_seq_stop(struct seq_file *seq, void *v)
3798 __releases(proto_list_mutex)
3800 mutex_unlock(&proto_list_mutex);
3803 static char proto_method_implemented(const void *method)
3805 return method == NULL ? 'n' : 'y';
3807 static long sock_prot_memory_allocated(struct proto *proto)
3809 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3812 static const char *sock_prot_memory_pressure(struct proto *proto)
3814 return proto->memory_pressure != NULL ?
3815 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3818 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3821 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3822 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3825 sock_prot_inuse_get(seq_file_net(seq), proto),
3826 sock_prot_memory_allocated(proto),
3827 sock_prot_memory_pressure(proto),
3829 proto->slab == NULL ? "no" : "yes",
3830 module_name(proto->owner),
3831 proto_method_implemented(proto->close),
3832 proto_method_implemented(proto->connect),
3833 proto_method_implemented(proto->disconnect),
3834 proto_method_implemented(proto->accept),
3835 proto_method_implemented(proto->ioctl),
3836 proto_method_implemented(proto->init),
3837 proto_method_implemented(proto->destroy),
3838 proto_method_implemented(proto->shutdown),
3839 proto_method_implemented(proto->setsockopt),
3840 proto_method_implemented(proto->getsockopt),
3841 proto_method_implemented(proto->sendmsg),
3842 proto_method_implemented(proto->recvmsg),
3843 proto_method_implemented(proto->sendpage),
3844 proto_method_implemented(proto->bind),
3845 proto_method_implemented(proto->backlog_rcv),
3846 proto_method_implemented(proto->hash),
3847 proto_method_implemented(proto->unhash),
3848 proto_method_implemented(proto->get_port),
3849 proto_method_implemented(proto->enter_memory_pressure));
3852 static int proto_seq_show(struct seq_file *seq, void *v)
3854 if (v == &proto_list)
3855 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3864 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3866 proto_seq_printf(seq, list_entry(v, struct proto, node));
3870 static const struct seq_operations proto_seq_ops = {
3871 .start = proto_seq_start,
3872 .next = proto_seq_next,
3873 .stop = proto_seq_stop,
3874 .show = proto_seq_show,
3877 static __net_init int proto_init_net(struct net *net)
3879 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3880 sizeof(struct seq_net_private)))
3886 static __net_exit void proto_exit_net(struct net *net)
3888 remove_proc_entry("protocols", net->proc_net);
3892 static __net_initdata struct pernet_operations proto_net_ops = {
3893 .init = proto_init_net,
3894 .exit = proto_exit_net,
3897 static int __init proto_init(void)
3899 return register_pernet_subsys(&proto_net_ops);
3902 subsys_initcall(proto_init);
3904 #endif /* PROC_FS */
3906 #ifdef CONFIG_NET_RX_BUSY_POLL
3907 bool sk_busy_loop_end(void *p, unsigned long start_time)
3909 struct sock *sk = p;
3911 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3912 sk_busy_loop_timeout(sk, start_time);
3914 EXPORT_SYMBOL(sk_busy_loop_end);
3915 #endif /* CONFIG_NET_RX_BUSY_POLL */
3917 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3919 if (!sk->sk_prot->bind_add)
3921 return sk->sk_prot->bind_add(sk, addr, addr_len);
3923 EXPORT_SYMBOL(sock_bind_add);