2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * 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 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <asm/unaligned.h>
95 #include <linux/capability.h>
96 #include <linux/errno.h>
97 #include <linux/errqueue.h>
98 #include <linux/types.h>
99 #include <linux/socket.h>
100 #include <linux/in.h>
101 #include <linux/kernel.h>
102 #include <linux/module.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <linux/sched.h>
106 #include <linux/sched/mm.h>
107 #include <linux/timer.h>
108 #include <linux/string.h>
109 #include <linux/sockios.h>
110 #include <linux/net.h>
111 #include <linux/mm.h>
112 #include <linux/slab.h>
113 #include <linux/interrupt.h>
114 #include <linux/poll.h>
115 #include <linux/tcp.h>
116 #include <linux/init.h>
117 #include <linux/highmem.h>
118 #include <linux/user_namespace.h>
119 #include <linux/static_key.h>
120 #include <linux/memcontrol.h>
121 #include <linux/prefetch.h>
123 #include <linux/uaccess.h>
125 #include <linux/netdevice.h>
126 #include <net/protocol.h>
127 #include <linux/skbuff.h>
128 #include <net/net_namespace.h>
129 #include <net/request_sock.h>
130 #include <net/sock.h>
131 #include <linux/net_tstamp.h>
132 #include <net/xfrm.h>
133 #include <linux/ipsec.h>
134 #include <net/cls_cgroup.h>
135 #include <net/netprio_cgroup.h>
136 #include <linux/sock_diag.h>
138 #include <linux/filter.h>
139 #include <net/sock_reuseport.h>
141 #include <trace/events/sock.h>
144 #include <net/busy_poll.h>
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
149 static void sock_inuse_add(struct net *net, int val);
152 * sk_ns_capable - General socket capability test
153 * @sk: Socket to use a capability on or through
154 * @user_ns: The user namespace of the capability to use
155 * @cap: The capability to use
157 * Test to see if the opener of the socket had when the socket was
158 * created and the current process has the capability @cap in the user
159 * namespace @user_ns.
161 bool sk_ns_capable(const struct sock *sk,
162 struct user_namespace *user_ns, int cap)
164 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
165 ns_capable(user_ns, cap);
167 EXPORT_SYMBOL(sk_ns_capable);
170 * sk_capable - Socket global capability test
171 * @sk: Socket to use a capability on or through
172 * @cap: The global capability to use
174 * Test to see if the opener of the socket had when the socket was
175 * created and the current process has the capability @cap in all user
178 bool sk_capable(const struct sock *sk, int cap)
180 return sk_ns_capable(sk, &init_user_ns, cap);
182 EXPORT_SYMBOL(sk_capable);
185 * sk_net_capable - Network namespace socket capability test
186 * @sk: Socket to use a capability on or through
187 * @cap: The capability to use
189 * Test to see if the opener of the socket had when the socket was created
190 * and the current process has the capability @cap over the network namespace
191 * the socket is a member of.
193 bool sk_net_capable(const struct sock *sk, int cap)
195 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
197 EXPORT_SYMBOL(sk_net_capable);
200 * Each address family might have different locking rules, so we have
201 * one slock key per address family and separate keys for internal and
204 static struct lock_class_key af_family_keys[AF_MAX];
205 static struct lock_class_key af_family_kern_keys[AF_MAX];
206 static struct lock_class_key af_family_slock_keys[AF_MAX];
207 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
210 * Make lock validator output more readable. (we pre-construct these
211 * strings build-time, so that runtime initialization of socket
215 #define _sock_locks(x) \
216 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
217 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
218 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
219 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
220 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
221 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
222 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
223 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
224 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
225 x "27" , x "28" , x "AF_CAN" , \
226 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
227 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
228 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
229 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
230 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
233 static const char *const af_family_key_strings[AF_MAX+1] = {
234 _sock_locks("sk_lock-")
236 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
237 _sock_locks("slock-")
239 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
240 _sock_locks("clock-")
243 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
244 _sock_locks("k-sk_lock-")
246 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
247 _sock_locks("k-slock-")
249 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
250 _sock_locks("k-clock-")
252 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
253 _sock_locks("rlock-")
255 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
256 _sock_locks("wlock-")
258 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
259 _sock_locks("elock-")
263 * sk_callback_lock and sk queues locking rules are per-address-family,
264 * so split the lock classes by using a per-AF key:
266 static struct lock_class_key af_callback_keys[AF_MAX];
267 static struct lock_class_key af_rlock_keys[AF_MAX];
268 static struct lock_class_key af_wlock_keys[AF_MAX];
269 static struct lock_class_key af_elock_keys[AF_MAX];
270 static struct lock_class_key af_kern_callback_keys[AF_MAX];
272 /* Run time adjustable parameters. */
273 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
274 EXPORT_SYMBOL(sysctl_wmem_max);
275 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
276 EXPORT_SYMBOL(sysctl_rmem_max);
277 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
278 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
280 /* Maximal space eaten by iovec or ancillary data plus some space */
281 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
282 EXPORT_SYMBOL(sysctl_optmem_max);
284 int sysctl_tstamp_allow_data __read_mostly = 1;
286 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
287 EXPORT_SYMBOL_GPL(memalloc_socks_key);
290 * sk_set_memalloc - sets %SOCK_MEMALLOC
291 * @sk: socket to set it on
293 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
294 * It's the responsibility of the admin to adjust min_free_kbytes
295 * to meet the requirements
297 void sk_set_memalloc(struct sock *sk)
299 sock_set_flag(sk, SOCK_MEMALLOC);
300 sk->sk_allocation |= __GFP_MEMALLOC;
301 static_branch_inc(&memalloc_socks_key);
303 EXPORT_SYMBOL_GPL(sk_set_memalloc);
305 void sk_clear_memalloc(struct sock *sk)
307 sock_reset_flag(sk, SOCK_MEMALLOC);
308 sk->sk_allocation &= ~__GFP_MEMALLOC;
309 static_branch_dec(&memalloc_socks_key);
312 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
313 * progress of swapping. SOCK_MEMALLOC may be cleared while
314 * it has rmem allocations due to the last swapfile being deactivated
315 * but there is a risk that the socket is unusable due to exceeding
316 * the rmem limits. Reclaim the reserves and obey rmem limits again.
320 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
322 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
325 unsigned int noreclaim_flag;
327 /* these should have been dropped before queueing */
328 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
330 noreclaim_flag = memalloc_noreclaim_save();
331 ret = sk->sk_backlog_rcv(sk, skb);
332 memalloc_noreclaim_restore(noreclaim_flag);
336 EXPORT_SYMBOL(__sk_backlog_rcv);
338 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
342 if (optlen < sizeof(tv))
344 if (copy_from_user(&tv, optval, sizeof(tv)))
346 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
350 static int warned __read_mostly;
353 if (warned < 10 && net_ratelimit()) {
355 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
356 __func__, current->comm, task_pid_nr(current));
360 *timeo_p = MAX_SCHEDULE_TIMEOUT;
361 if (tv.tv_sec == 0 && tv.tv_usec == 0)
363 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
364 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
368 static void sock_warn_obsolete_bsdism(const char *name)
371 static char warncomm[TASK_COMM_LEN];
372 if (strcmp(warncomm, current->comm) && warned < 5) {
373 strcpy(warncomm, current->comm);
374 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
380 static bool sock_needs_netstamp(const struct sock *sk)
382 switch (sk->sk_family) {
391 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
393 if (sk->sk_flags & flags) {
394 sk->sk_flags &= ~flags;
395 if (sock_needs_netstamp(sk) &&
396 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
397 net_disable_timestamp();
402 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
405 struct sk_buff_head *list = &sk->sk_receive_queue;
407 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
408 atomic_inc(&sk->sk_drops);
409 trace_sock_rcvqueue_full(sk, skb);
413 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
414 atomic_inc(&sk->sk_drops);
419 skb_set_owner_r(skb, sk);
421 /* we escape from rcu protected region, make sure we dont leak
426 spin_lock_irqsave(&list->lock, flags);
427 sock_skb_set_dropcount(sk, skb);
428 __skb_queue_tail(list, skb);
429 spin_unlock_irqrestore(&list->lock, flags);
431 if (!sock_flag(sk, SOCK_DEAD))
432 sk->sk_data_ready(sk);
435 EXPORT_SYMBOL(__sock_queue_rcv_skb);
437 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
441 err = sk_filter(sk, skb);
445 return __sock_queue_rcv_skb(sk, skb);
447 EXPORT_SYMBOL(sock_queue_rcv_skb);
449 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
450 const int nested, unsigned int trim_cap, bool refcounted)
452 int rc = NET_RX_SUCCESS;
454 if (sk_filter_trim_cap(sk, skb, trim_cap))
455 goto discard_and_relse;
459 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
460 atomic_inc(&sk->sk_drops);
461 goto discard_and_relse;
464 bh_lock_sock_nested(sk);
467 if (!sock_owned_by_user(sk)) {
469 * trylock + unlock semantics:
471 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
473 rc = sk_backlog_rcv(sk, skb);
475 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
476 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
478 atomic_inc(&sk->sk_drops);
479 goto discard_and_relse;
491 EXPORT_SYMBOL(__sk_receive_skb);
493 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
495 struct dst_entry *dst = __sk_dst_get(sk);
497 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
498 sk_tx_queue_clear(sk);
499 sk->sk_dst_pending_confirm = 0;
500 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
507 EXPORT_SYMBOL(__sk_dst_check);
509 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
511 struct dst_entry *dst = sk_dst_get(sk);
513 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
521 EXPORT_SYMBOL(sk_dst_check);
523 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
526 int ret = -ENOPROTOOPT;
527 #ifdef CONFIG_NETDEVICES
528 struct net *net = sock_net(sk);
529 char devname[IFNAMSIZ];
534 if (!ns_capable(net->user_ns, CAP_NET_RAW))
541 /* Bind this socket to a particular device like "eth0",
542 * as specified in the passed interface name. If the
543 * name is "" or the option length is zero the socket
546 if (optlen > IFNAMSIZ - 1)
547 optlen = IFNAMSIZ - 1;
548 memset(devname, 0, sizeof(devname));
551 if (copy_from_user(devname, optval, optlen))
555 if (devname[0] != '\0') {
556 struct net_device *dev;
559 dev = dev_get_by_name_rcu(net, devname);
561 index = dev->ifindex;
569 sk->sk_bound_dev_if = index;
581 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
582 int __user *optlen, int len)
584 int ret = -ENOPROTOOPT;
585 #ifdef CONFIG_NETDEVICES
586 struct net *net = sock_net(sk);
587 char devname[IFNAMSIZ];
589 if (sk->sk_bound_dev_if == 0) {
598 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
602 len = strlen(devname) + 1;
605 if (copy_to_user(optval, devname, len))
610 if (put_user(len, optlen))
621 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
624 sock_set_flag(sk, bit);
626 sock_reset_flag(sk, bit);
629 bool sk_mc_loop(struct sock *sk)
631 if (dev_recursion_level())
635 switch (sk->sk_family) {
637 return inet_sk(sk)->mc_loop;
638 #if IS_ENABLED(CONFIG_IPV6)
640 return inet6_sk(sk)->mc_loop;
646 EXPORT_SYMBOL(sk_mc_loop);
649 * This is meant for all protocols to use and covers goings on
650 * at the socket level. Everything here is generic.
653 int sock_setsockopt(struct socket *sock, int level, int optname,
654 char __user *optval, unsigned int optlen)
656 struct sock_txtime sk_txtime;
657 struct sock *sk = sock->sk;
664 * Options without arguments
667 if (optname == SO_BINDTODEVICE)
668 return sock_setbindtodevice(sk, optval, optlen);
670 if (optlen < sizeof(int))
673 if (get_user(val, (int __user *)optval))
676 valbool = val ? 1 : 0;
682 if (val && !capable(CAP_NET_ADMIN))
685 sock_valbool_flag(sk, SOCK_DBG, valbool);
688 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
691 sk->sk_reuseport = valbool;
700 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
704 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
707 /* Don't error on this BSD doesn't and if you think
708 * about it this is right. Otherwise apps have to
709 * play 'guess the biggest size' games. RCVBUF/SNDBUF
710 * are treated in BSD as hints
712 val = min_t(u32, val, sysctl_wmem_max);
714 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
715 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
716 /* Wake up sending tasks if we upped the value. */
717 sk->sk_write_space(sk);
721 if (!capable(CAP_NET_ADMIN)) {
728 /* Don't error on this BSD doesn't and if you think
729 * about it this is right. Otherwise apps have to
730 * play 'guess the biggest size' games. RCVBUF/SNDBUF
731 * are treated in BSD as hints
733 val = min_t(u32, val, sysctl_rmem_max);
735 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
737 * We double it on the way in to account for
738 * "struct sk_buff" etc. overhead. Applications
739 * assume that the SO_RCVBUF setting they make will
740 * allow that much actual data to be received on that
743 * Applications are unaware that "struct sk_buff" and
744 * other overheads allocate from the receive buffer
745 * during socket buffer allocation.
747 * And after considering the possible alternatives,
748 * returning the value we actually used in getsockopt
749 * is the most desirable behavior.
751 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
755 if (!capable(CAP_NET_ADMIN)) {
762 if (sk->sk_prot->keepalive)
763 sk->sk_prot->keepalive(sk, valbool);
764 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
768 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
772 sk->sk_no_check_tx = valbool;
776 if ((val >= 0 && val <= 6) ||
777 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
778 sk->sk_priority = val;
784 if (optlen < sizeof(ling)) {
785 ret = -EINVAL; /* 1003.1g */
788 if (copy_from_user(&ling, optval, sizeof(ling))) {
793 sock_reset_flag(sk, SOCK_LINGER);
795 #if (BITS_PER_LONG == 32)
796 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
797 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
800 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
801 sock_set_flag(sk, SOCK_LINGER);
806 sock_warn_obsolete_bsdism("setsockopt");
811 set_bit(SOCK_PASSCRED, &sock->flags);
813 clear_bit(SOCK_PASSCRED, &sock->flags);
819 if (optname == SO_TIMESTAMP)
820 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
822 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
823 sock_set_flag(sk, SOCK_RCVTSTAMP);
824 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
826 sock_reset_flag(sk, SOCK_RCVTSTAMP);
827 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
831 case SO_TIMESTAMPING:
832 if (val & ~SOF_TIMESTAMPING_MASK) {
837 if (val & SOF_TIMESTAMPING_OPT_ID &&
838 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
839 if (sk->sk_protocol == IPPROTO_TCP &&
840 sk->sk_type == SOCK_STREAM) {
841 if ((1 << sk->sk_state) &
842 (TCPF_CLOSE | TCPF_LISTEN)) {
846 sk->sk_tskey = tcp_sk(sk)->snd_una;
852 if (val & SOF_TIMESTAMPING_OPT_STATS &&
853 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
858 sk->sk_tsflags = val;
859 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
860 sock_enable_timestamp(sk,
861 SOCK_TIMESTAMPING_RX_SOFTWARE);
863 sock_disable_timestamp(sk,
864 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
870 if (sock->ops->set_rcvlowat)
871 ret = sock->ops->set_rcvlowat(sk, val);
873 sk->sk_rcvlowat = val ? : 1;
877 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
881 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
884 case SO_ATTACH_FILTER:
886 if (optlen == sizeof(struct sock_fprog)) {
887 struct sock_fprog fprog;
890 if (copy_from_user(&fprog, optval, sizeof(fprog)))
893 ret = sk_attach_filter(&fprog, sk);
899 if (optlen == sizeof(u32)) {
903 if (copy_from_user(&ufd, optval, sizeof(ufd)))
906 ret = sk_attach_bpf(ufd, sk);
910 case SO_ATTACH_REUSEPORT_CBPF:
912 if (optlen == sizeof(struct sock_fprog)) {
913 struct sock_fprog fprog;
916 if (copy_from_user(&fprog, optval, sizeof(fprog)))
919 ret = sk_reuseport_attach_filter(&fprog, sk);
923 case SO_ATTACH_REUSEPORT_EBPF:
925 if (optlen == sizeof(u32)) {
929 if (copy_from_user(&ufd, optval, sizeof(ufd)))
932 ret = sk_reuseport_attach_bpf(ufd, sk);
936 case SO_DETACH_FILTER:
937 ret = sk_detach_filter(sk);
941 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
944 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
949 set_bit(SOCK_PASSSEC, &sock->flags);
951 clear_bit(SOCK_PASSSEC, &sock->flags);
954 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
961 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
965 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
969 if (sock->ops->set_peek_off)
970 ret = sock->ops->set_peek_off(sk, val);
976 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
979 case SO_SELECT_ERR_QUEUE:
980 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
983 #ifdef CONFIG_NET_RX_BUSY_POLL
985 /* allow unprivileged users to decrease the value */
986 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
992 WRITE_ONCE(sk->sk_ll_usec, val);
997 case SO_MAX_PACING_RATE:
999 cmpxchg(&sk->sk_pacing_status,
1002 sk->sk_max_pacing_rate = val;
1003 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1004 sk->sk_max_pacing_rate);
1007 case SO_INCOMING_CPU:
1008 WRITE_ONCE(sk->sk_incoming_cpu, val);
1013 dst_negative_advice(sk);
1017 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1018 if (sk->sk_protocol != IPPROTO_TCP)
1020 } else if (sk->sk_family != PF_RDS) {
1024 if (val < 0 || val > 1)
1027 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1032 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1034 } else if (optlen != sizeof(struct sock_txtime)) {
1036 } else if (copy_from_user(&sk_txtime, optval,
1037 sizeof(struct sock_txtime))) {
1039 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1042 sock_valbool_flag(sk, SOCK_TXTIME, true);
1043 sk->sk_clockid = sk_txtime.clockid;
1044 sk->sk_txtime_deadline_mode =
1045 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1046 sk->sk_txtime_report_errors =
1047 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1058 EXPORT_SYMBOL(sock_setsockopt);
1060 static const struct cred *sk_get_peer_cred(struct sock *sk)
1062 const struct cred *cred;
1064 spin_lock(&sk->sk_peer_lock);
1065 cred = get_cred(sk->sk_peer_cred);
1066 spin_unlock(&sk->sk_peer_lock);
1071 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1072 struct ucred *ucred)
1074 ucred->pid = pid_vnr(pid);
1075 ucred->uid = ucred->gid = -1;
1077 struct user_namespace *current_ns = current_user_ns();
1079 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1080 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1084 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1086 struct user_namespace *user_ns = current_user_ns();
1089 for (i = 0; i < src->ngroups; i++)
1090 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1096 int sock_getsockopt(struct socket *sock, int level, int optname,
1097 char __user *optval, int __user *optlen)
1099 struct sock *sk = sock->sk;
1106 struct sock_txtime txtime;
1109 int lv = sizeof(int);
1112 if (get_user(len, optlen))
1117 memset(&v, 0, sizeof(v));
1121 v.val = sock_flag(sk, SOCK_DBG);
1125 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1129 v.val = sock_flag(sk, SOCK_BROADCAST);
1133 v.val = sk->sk_sndbuf;
1137 v.val = sk->sk_rcvbuf;
1141 v.val = sk->sk_reuse;
1145 v.val = sk->sk_reuseport;
1149 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1153 v.val = sk->sk_type;
1157 v.val = sk->sk_protocol;
1161 v.val = sk->sk_family;
1165 v.val = -sock_error(sk);
1167 v.val = xchg(&sk->sk_err_soft, 0);
1171 v.val = sock_flag(sk, SOCK_URGINLINE);
1175 v.val = sk->sk_no_check_tx;
1179 v.val = sk->sk_priority;
1183 lv = sizeof(v.ling);
1184 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1185 v.ling.l_linger = sk->sk_lingertime / HZ;
1189 sock_warn_obsolete_bsdism("getsockopt");
1193 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1194 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1197 case SO_TIMESTAMPNS:
1198 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1201 case SO_TIMESTAMPING:
1202 v.val = sk->sk_tsflags;
1206 lv = sizeof(struct timeval);
1207 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1211 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1212 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1217 lv = sizeof(struct timeval);
1218 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1222 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1223 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1228 v.val = sk->sk_rcvlowat;
1236 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1241 struct ucred peercred;
1242 if (len > sizeof(peercred))
1243 len = sizeof(peercred);
1245 spin_lock(&sk->sk_peer_lock);
1246 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1247 spin_unlock(&sk->sk_peer_lock);
1249 if (copy_to_user(optval, &peercred, len))
1256 const struct cred *cred;
1259 cred = sk_get_peer_cred(sk);
1263 n = cred->group_info->ngroups;
1264 if (len < n * sizeof(gid_t)) {
1265 len = n * sizeof(gid_t);
1267 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1269 len = n * sizeof(gid_t);
1271 ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1282 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1287 if (copy_to_user(optval, address, len))
1292 /* Dubious BSD thing... Probably nobody even uses it, but
1293 * the UNIX standard wants it for whatever reason... -DaveM
1296 v.val = sk->sk_state == TCP_LISTEN;
1300 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1304 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1307 v.val = sk->sk_mark;
1311 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1314 case SO_WIFI_STATUS:
1315 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1319 if (!sock->ops->set_peek_off)
1322 v.val = sk->sk_peek_off;
1325 v.val = sock_flag(sk, SOCK_NOFCS);
1328 case SO_BINDTODEVICE:
1329 return sock_getbindtodevice(sk, optval, optlen, len);
1332 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1338 case SO_LOCK_FILTER:
1339 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1342 case SO_BPF_EXTENSIONS:
1343 v.val = bpf_tell_extensions();
1346 case SO_SELECT_ERR_QUEUE:
1347 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1350 #ifdef CONFIG_NET_RX_BUSY_POLL
1352 v.val = sk->sk_ll_usec;
1356 case SO_MAX_PACING_RATE:
1357 v.val = sk->sk_max_pacing_rate;
1360 case SO_INCOMING_CPU:
1361 v.val = READ_ONCE(sk->sk_incoming_cpu);
1366 u32 meminfo[SK_MEMINFO_VARS];
1368 sk_get_meminfo(sk, meminfo);
1370 len = min_t(unsigned int, len, sizeof(meminfo));
1371 if (copy_to_user(optval, &meminfo, len))
1377 #ifdef CONFIG_NET_RX_BUSY_POLL
1378 case SO_INCOMING_NAPI_ID:
1379 v.val = READ_ONCE(sk->sk_napi_id);
1381 /* aggregate non-NAPI IDs down to 0 */
1382 if (v.val < MIN_NAPI_ID)
1392 v.val64 = sock_gen_cookie(sk);
1396 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1400 lv = sizeof(v.txtime);
1401 v.txtime.clockid = sk->sk_clockid;
1402 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1403 SOF_TXTIME_DEADLINE_MODE : 0;
1404 v.txtime.flags |= sk->sk_txtime_report_errors ?
1405 SOF_TXTIME_REPORT_ERRORS : 0;
1409 /* We implement the SO_SNDLOWAT etc to not be settable
1412 return -ENOPROTOOPT;
1417 if (copy_to_user(optval, &v, len))
1420 if (put_user(len, optlen))
1426 * Initialize an sk_lock.
1428 * (We also register the sk_lock with the lock validator.)
1430 static inline void sock_lock_init(struct sock *sk)
1432 if (sk->sk_kern_sock)
1433 sock_lock_init_class_and_name(
1435 af_family_kern_slock_key_strings[sk->sk_family],
1436 af_family_kern_slock_keys + sk->sk_family,
1437 af_family_kern_key_strings[sk->sk_family],
1438 af_family_kern_keys + sk->sk_family);
1440 sock_lock_init_class_and_name(
1442 af_family_slock_key_strings[sk->sk_family],
1443 af_family_slock_keys + sk->sk_family,
1444 af_family_key_strings[sk->sk_family],
1445 af_family_keys + sk->sk_family);
1449 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1450 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1451 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1453 static void sock_copy(struct sock *nsk, const struct sock *osk)
1455 #ifdef CONFIG_SECURITY_NETWORK
1456 void *sptr = nsk->sk_security;
1458 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1460 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1461 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1463 #ifdef CONFIG_SECURITY_NETWORK
1464 nsk->sk_security = sptr;
1465 security_sk_clone(osk, nsk);
1469 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1473 struct kmem_cache *slab;
1477 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1480 if (priority & __GFP_ZERO)
1481 sk_prot_clear_nulls(sk, prot->obj_size);
1483 sk = kmalloc(prot->obj_size, priority);
1486 if (security_sk_alloc(sk, family, priority))
1489 if (!try_module_get(prot->owner))
1491 sk_tx_queue_clear(sk);
1497 security_sk_free(sk);
1500 kmem_cache_free(slab, sk);
1506 static void sk_prot_free(struct proto *prot, struct sock *sk)
1508 struct kmem_cache *slab;
1509 struct module *owner;
1511 owner = prot->owner;
1514 cgroup_sk_free(&sk->sk_cgrp_data);
1515 mem_cgroup_sk_free(sk);
1516 security_sk_free(sk);
1518 kmem_cache_free(slab, sk);
1525 * sk_alloc - All socket objects are allocated here
1526 * @net: the applicable net namespace
1527 * @family: protocol family
1528 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1529 * @prot: struct proto associated with this new sock instance
1530 * @kern: is this to be a kernel socket?
1532 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1533 struct proto *prot, int kern)
1537 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1539 sk->sk_family = family;
1541 * See comment in struct sock definition to understand
1542 * why we need sk_prot_creator -acme
1544 sk->sk_prot = sk->sk_prot_creator = prot;
1545 sk->sk_kern_sock = kern;
1547 sk->sk_net_refcnt = kern ? 0 : 1;
1548 if (likely(sk->sk_net_refcnt)) {
1550 sock_inuse_add(net, 1);
1553 sock_net_set(sk, net);
1554 refcount_set(&sk->sk_wmem_alloc, 1);
1556 mem_cgroup_sk_alloc(sk);
1557 cgroup_sk_alloc(&sk->sk_cgrp_data);
1558 sock_update_classid(&sk->sk_cgrp_data);
1559 sock_update_netprioidx(&sk->sk_cgrp_data);
1560 sk_tx_queue_clear(sk);
1565 EXPORT_SYMBOL(sk_alloc);
1567 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1568 * grace period. This is the case for UDP sockets and TCP listeners.
1570 static void __sk_destruct(struct rcu_head *head)
1572 struct sock *sk = container_of(head, struct sock, sk_rcu);
1573 struct sk_filter *filter;
1575 if (sk->sk_destruct)
1576 sk->sk_destruct(sk);
1578 filter = rcu_dereference_check(sk->sk_filter,
1579 refcount_read(&sk->sk_wmem_alloc) == 0);
1581 sk_filter_uncharge(sk, filter);
1582 RCU_INIT_POINTER(sk->sk_filter, NULL);
1585 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1587 if (atomic_read(&sk->sk_omem_alloc))
1588 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1589 __func__, atomic_read(&sk->sk_omem_alloc));
1591 if (sk->sk_frag.page) {
1592 put_page(sk->sk_frag.page);
1593 sk->sk_frag.page = NULL;
1596 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
1597 put_cred(sk->sk_peer_cred);
1598 put_pid(sk->sk_peer_pid);
1600 if (likely(sk->sk_net_refcnt))
1601 put_net(sock_net(sk));
1602 sk_prot_free(sk->sk_prot_creator, sk);
1605 void sk_destruct(struct sock *sk)
1607 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1609 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1610 reuseport_detach_sock(sk);
1611 use_call_rcu = true;
1615 call_rcu(&sk->sk_rcu, __sk_destruct);
1617 __sk_destruct(&sk->sk_rcu);
1620 static void __sk_free(struct sock *sk)
1622 if (likely(sk->sk_net_refcnt))
1623 sock_inuse_add(sock_net(sk), -1);
1625 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1626 sock_diag_broadcast_destroy(sk);
1631 void sk_free(struct sock *sk)
1634 * We subtract one from sk_wmem_alloc and can know if
1635 * some packets are still in some tx queue.
1636 * If not null, sock_wfree() will call __sk_free(sk) later
1638 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1641 EXPORT_SYMBOL(sk_free);
1643 static void sk_init_common(struct sock *sk)
1645 skb_queue_head_init(&sk->sk_receive_queue);
1646 skb_queue_head_init(&sk->sk_write_queue);
1647 skb_queue_head_init(&sk->sk_error_queue);
1649 rwlock_init(&sk->sk_callback_lock);
1650 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1651 af_rlock_keys + sk->sk_family,
1652 af_family_rlock_key_strings[sk->sk_family]);
1653 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1654 af_wlock_keys + sk->sk_family,
1655 af_family_wlock_key_strings[sk->sk_family]);
1656 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1657 af_elock_keys + sk->sk_family,
1658 af_family_elock_key_strings[sk->sk_family]);
1659 lockdep_set_class_and_name(&sk->sk_callback_lock,
1660 af_callback_keys + sk->sk_family,
1661 af_family_clock_key_strings[sk->sk_family]);
1665 * sk_clone_lock - clone a socket, and lock its clone
1666 * @sk: the socket to clone
1667 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1669 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1671 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1674 bool is_charged = true;
1676 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1677 if (newsk != NULL) {
1678 struct sk_filter *filter;
1680 sock_copy(newsk, sk);
1682 newsk->sk_prot_creator = sk->sk_prot;
1685 if (likely(newsk->sk_net_refcnt))
1686 get_net(sock_net(newsk));
1687 sk_node_init(&newsk->sk_node);
1688 sock_lock_init(newsk);
1689 bh_lock_sock(newsk);
1690 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1691 newsk->sk_backlog.len = 0;
1693 atomic_set(&newsk->sk_rmem_alloc, 0);
1695 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1697 refcount_set(&newsk->sk_wmem_alloc, 1);
1698 atomic_set(&newsk->sk_omem_alloc, 0);
1699 sk_init_common(newsk);
1701 newsk->sk_dst_cache = NULL;
1702 newsk->sk_dst_pending_confirm = 0;
1703 newsk->sk_wmem_queued = 0;
1704 newsk->sk_forward_alloc = 0;
1705 atomic_set(&newsk->sk_drops, 0);
1706 newsk->sk_send_head = NULL;
1707 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1708 atomic_set(&newsk->sk_zckey, 0);
1710 sock_reset_flag(newsk, SOCK_DONE);
1712 /* sk->sk_memcg will be populated at accept() time */
1713 newsk->sk_memcg = NULL;
1715 cgroup_sk_clone(&newsk->sk_cgrp_data);
1718 filter = rcu_dereference(sk->sk_filter);
1720 /* though it's an empty new sock, the charging may fail
1721 * if sysctl_optmem_max was changed between creation of
1722 * original socket and cloning
1724 is_charged = sk_filter_charge(newsk, filter);
1725 RCU_INIT_POINTER(newsk->sk_filter, filter);
1728 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1729 /* We need to make sure that we don't uncharge the new
1730 * socket if we couldn't charge it in the first place
1731 * as otherwise we uncharge the parent's filter.
1734 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1735 sk_free_unlock_clone(newsk);
1739 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1742 newsk->sk_err_soft = 0;
1743 newsk->sk_priority = 0;
1744 newsk->sk_incoming_cpu = raw_smp_processor_id();
1745 atomic64_set(&newsk->sk_cookie, 0);
1746 if (likely(newsk->sk_net_refcnt))
1747 sock_inuse_add(sock_net(newsk), 1);
1750 * Before updating sk_refcnt, we must commit prior changes to memory
1751 * (Documentation/RCU/rculist_nulls.txt for details)
1754 refcount_set(&newsk->sk_refcnt, 2);
1757 * Increment the counter in the same struct proto as the master
1758 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1759 * is the same as sk->sk_prot->socks, as this field was copied
1762 * This _changes_ the previous behaviour, where
1763 * tcp_create_openreq_child always was incrementing the
1764 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1765 * to be taken into account in all callers. -acme
1767 sk_refcnt_debug_inc(newsk);
1768 sk_set_socket(newsk, NULL);
1769 sk_tx_queue_clear(newsk);
1770 newsk->sk_wq = NULL;
1772 if (newsk->sk_prot->sockets_allocated)
1773 sk_sockets_allocated_inc(newsk);
1775 if (sock_needs_netstamp(sk) &&
1776 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1777 net_enable_timestamp();
1782 EXPORT_SYMBOL_GPL(sk_clone_lock);
1784 void sk_free_unlock_clone(struct sock *sk)
1786 /* It is still raw copy of parent, so invalidate
1787 * destructor and make plain sk_free() */
1788 sk->sk_destruct = NULL;
1792 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1794 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1798 sk_dst_set(sk, dst);
1799 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1800 if (sk->sk_route_caps & NETIF_F_GSO)
1801 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1802 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1803 if (sk_can_gso(sk)) {
1804 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1805 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1807 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1808 sk->sk_gso_max_size = dst->dev->gso_max_size;
1809 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1812 sk->sk_gso_max_segs = max_segs;
1814 EXPORT_SYMBOL_GPL(sk_setup_caps);
1817 * Simple resource managers for sockets.
1822 * Write buffer destructor automatically called from kfree_skb.
1824 void sock_wfree(struct sk_buff *skb)
1826 struct sock *sk = skb->sk;
1827 unsigned int len = skb->truesize;
1829 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1831 * Keep a reference on sk_wmem_alloc, this will be released
1832 * after sk_write_space() call
1834 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1835 sk->sk_write_space(sk);
1839 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1840 * could not do because of in-flight packets
1842 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1845 EXPORT_SYMBOL(sock_wfree);
1847 /* This variant of sock_wfree() is used by TCP,
1848 * since it sets SOCK_USE_WRITE_QUEUE.
1850 void __sock_wfree(struct sk_buff *skb)
1852 struct sock *sk = skb->sk;
1854 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1858 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1863 if (unlikely(!sk_fullsock(sk))) {
1864 skb->destructor = sock_edemux;
1869 skb->destructor = sock_wfree;
1870 skb_set_hash_from_sk(skb, sk);
1872 * We used to take a refcount on sk, but following operation
1873 * is enough to guarantee sk_free() wont free this sock until
1874 * all in-flight packets are completed
1876 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1878 EXPORT_SYMBOL(skb_set_owner_w);
1880 /* This helper is used by netem, as it can hold packets in its
1881 * delay queue. We want to allow the owner socket to send more
1882 * packets, as if they were already TX completed by a typical driver.
1883 * But we also want to keep skb->sk set because some packet schedulers
1884 * rely on it (sch_fq for example).
1886 void skb_orphan_partial(struct sk_buff *skb)
1888 if (skb_is_tcp_pure_ack(skb))
1891 if (skb->destructor == sock_wfree
1893 || skb->destructor == tcp_wfree
1896 struct sock *sk = skb->sk;
1898 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1899 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
1900 skb->destructor = sock_efree;
1906 EXPORT_SYMBOL(skb_orphan_partial);
1909 * Read buffer destructor automatically called from kfree_skb.
1911 void sock_rfree(struct sk_buff *skb)
1913 struct sock *sk = skb->sk;
1914 unsigned int len = skb->truesize;
1916 atomic_sub(len, &sk->sk_rmem_alloc);
1917 sk_mem_uncharge(sk, len);
1919 EXPORT_SYMBOL(sock_rfree);
1922 * Buffer destructor for skbs that are not used directly in read or write
1923 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1925 void sock_efree(struct sk_buff *skb)
1929 EXPORT_SYMBOL(sock_efree);
1931 kuid_t sock_i_uid(struct sock *sk)
1935 read_lock_bh(&sk->sk_callback_lock);
1936 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1937 read_unlock_bh(&sk->sk_callback_lock);
1940 EXPORT_SYMBOL(sock_i_uid);
1942 unsigned long sock_i_ino(struct sock *sk)
1946 read_lock_bh(&sk->sk_callback_lock);
1947 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1948 read_unlock_bh(&sk->sk_callback_lock);
1951 EXPORT_SYMBOL(sock_i_ino);
1954 * Allocate a skb from the socket's send buffer.
1956 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1959 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1960 struct sk_buff *skb = alloc_skb(size, priority);
1962 skb_set_owner_w(skb, sk);
1968 EXPORT_SYMBOL(sock_wmalloc);
1970 static void sock_ofree(struct sk_buff *skb)
1972 struct sock *sk = skb->sk;
1974 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
1977 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1980 struct sk_buff *skb;
1982 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
1983 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
1987 skb = alloc_skb(size, priority);
1991 atomic_add(skb->truesize, &sk->sk_omem_alloc);
1993 skb->destructor = sock_ofree;
1998 * Allocate a memory block from the socket's option memory buffer.
2000 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2002 if ((unsigned int)size <= sysctl_optmem_max &&
2003 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2005 /* First do the add, to avoid the race if kmalloc
2008 atomic_add(size, &sk->sk_omem_alloc);
2009 mem = kmalloc(size, priority);
2012 atomic_sub(size, &sk->sk_omem_alloc);
2016 EXPORT_SYMBOL(sock_kmalloc);
2018 /* Free an option memory block. Note, we actually want the inline
2019 * here as this allows gcc to detect the nullify and fold away the
2020 * condition entirely.
2022 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2025 if (WARN_ON_ONCE(!mem))
2031 atomic_sub(size, &sk->sk_omem_alloc);
2034 void sock_kfree_s(struct sock *sk, void *mem, int size)
2036 __sock_kfree_s(sk, mem, size, false);
2038 EXPORT_SYMBOL(sock_kfree_s);
2040 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2042 __sock_kfree_s(sk, mem, size, true);
2044 EXPORT_SYMBOL(sock_kzfree_s);
2046 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2047 I think, these locks should be removed for datagram sockets.
2049 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2053 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2057 if (signal_pending(current))
2059 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2060 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2061 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2063 if (sk->sk_shutdown & SEND_SHUTDOWN)
2067 timeo = schedule_timeout(timeo);
2069 finish_wait(sk_sleep(sk), &wait);
2075 * Generic send/receive buffer handlers
2078 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2079 unsigned long data_len, int noblock,
2080 int *errcode, int max_page_order)
2082 struct sk_buff *skb;
2086 timeo = sock_sndtimeo(sk, noblock);
2088 err = sock_error(sk);
2093 if (sk->sk_shutdown & SEND_SHUTDOWN)
2096 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2099 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2100 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2104 if (signal_pending(current))
2106 timeo = sock_wait_for_wmem(sk, timeo);
2108 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2109 errcode, sk->sk_allocation);
2111 skb_set_owner_w(skb, sk);
2115 err = sock_intr_errno(timeo);
2120 EXPORT_SYMBOL(sock_alloc_send_pskb);
2122 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2123 int noblock, int *errcode)
2125 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2127 EXPORT_SYMBOL(sock_alloc_send_skb);
2129 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2130 struct sockcm_cookie *sockc)
2134 switch (cmsg->cmsg_type) {
2136 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2138 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2140 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2142 case SO_TIMESTAMPING:
2143 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2146 tsflags = *(u32 *)CMSG_DATA(cmsg);
2147 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2150 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2151 sockc->tsflags |= tsflags;
2154 if (!sock_flag(sk, SOCK_TXTIME))
2156 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2158 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2160 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2162 case SCM_CREDENTIALS:
2169 EXPORT_SYMBOL(__sock_cmsg_send);
2171 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2172 struct sockcm_cookie *sockc)
2174 struct cmsghdr *cmsg;
2177 for_each_cmsghdr(cmsg, msg) {
2178 if (!CMSG_OK(msg, cmsg))
2180 if (cmsg->cmsg_level != SOL_SOCKET)
2182 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2188 EXPORT_SYMBOL(sock_cmsg_send);
2190 static void sk_enter_memory_pressure(struct sock *sk)
2192 if (!sk->sk_prot->enter_memory_pressure)
2195 sk->sk_prot->enter_memory_pressure(sk);
2198 static void sk_leave_memory_pressure(struct sock *sk)
2200 if (sk->sk_prot->leave_memory_pressure) {
2201 sk->sk_prot->leave_memory_pressure(sk);
2203 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2205 if (memory_pressure && READ_ONCE(*memory_pressure))
2206 WRITE_ONCE(*memory_pressure, 0);
2210 /* On 32bit arches, an skb frag is limited to 2^15 */
2211 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2214 * skb_page_frag_refill - check that a page_frag contains enough room
2215 * @sz: minimum size of the fragment we want to get
2216 * @pfrag: pointer to page_frag
2217 * @gfp: priority for memory allocation
2219 * Note: While this allocator tries to use high order pages, there is
2220 * no guarantee that allocations succeed. Therefore, @sz MUST be
2221 * less or equal than PAGE_SIZE.
2223 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2226 if (page_ref_count(pfrag->page) == 1) {
2230 if (pfrag->offset + sz <= pfrag->size)
2232 put_page(pfrag->page);
2236 if (SKB_FRAG_PAGE_ORDER) {
2237 /* Avoid direct reclaim but allow kswapd to wake */
2238 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2239 __GFP_COMP | __GFP_NOWARN |
2241 SKB_FRAG_PAGE_ORDER);
2242 if (likely(pfrag->page)) {
2243 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2247 pfrag->page = alloc_page(gfp);
2248 if (likely(pfrag->page)) {
2249 pfrag->size = PAGE_SIZE;
2254 EXPORT_SYMBOL(skb_page_frag_refill);
2256 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2258 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2261 sk_enter_memory_pressure(sk);
2262 sk_stream_moderate_sndbuf(sk);
2265 EXPORT_SYMBOL(sk_page_frag_refill);
2267 int sk_alloc_sg(struct sock *sk, int len, struct scatterlist *sg,
2268 int sg_start, int *sg_curr_index, unsigned int *sg_curr_size,
2271 int sg_curr = *sg_curr_index, use = 0, rc = 0;
2272 unsigned int size = *sg_curr_size;
2273 struct page_frag *pfrag;
2274 struct scatterlist *sge;
2277 pfrag = sk_page_frag(sk);
2280 unsigned int orig_offset;
2282 if (!sk_page_frag_refill(sk, pfrag)) {
2287 use = min_t(int, len, pfrag->size - pfrag->offset);
2289 if (!sk_wmem_schedule(sk, use)) {
2294 sk_mem_charge(sk, use);
2296 orig_offset = pfrag->offset;
2297 pfrag->offset += use;
2299 sge = sg + sg_curr - 1;
2300 if (sg_curr > first_coalesce && sg_page(sge) == pfrag->page &&
2301 sge->offset + sge->length == orig_offset) {
2306 sg_set_page(sge, pfrag->page, use, orig_offset);
2307 get_page(pfrag->page);
2310 if (sg_curr == MAX_SKB_FRAGS)
2313 if (sg_curr == sg_start) {
2322 *sg_curr_size = size;
2323 *sg_curr_index = sg_curr;
2326 EXPORT_SYMBOL(sk_alloc_sg);
2328 static void __lock_sock(struct sock *sk)
2329 __releases(&sk->sk_lock.slock)
2330 __acquires(&sk->sk_lock.slock)
2335 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2336 TASK_UNINTERRUPTIBLE);
2337 spin_unlock_bh(&sk->sk_lock.slock);
2339 spin_lock_bh(&sk->sk_lock.slock);
2340 if (!sock_owned_by_user(sk))
2343 finish_wait(&sk->sk_lock.wq, &wait);
2346 void __release_sock(struct sock *sk)
2347 __releases(&sk->sk_lock.slock)
2348 __acquires(&sk->sk_lock.slock)
2350 struct sk_buff *skb, *next;
2352 while ((skb = sk->sk_backlog.head) != NULL) {
2353 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2355 spin_unlock_bh(&sk->sk_lock.slock);
2360 WARN_ON_ONCE(skb_dst_is_noref(skb));
2362 sk_backlog_rcv(sk, skb);
2367 } while (skb != NULL);
2369 spin_lock_bh(&sk->sk_lock.slock);
2373 * Doing the zeroing here guarantee we can not loop forever
2374 * while a wild producer attempts to flood us.
2376 sk->sk_backlog.len = 0;
2379 void __sk_flush_backlog(struct sock *sk)
2381 spin_lock_bh(&sk->sk_lock.slock);
2383 spin_unlock_bh(&sk->sk_lock.slock);
2387 * sk_wait_data - wait for data to arrive at sk_receive_queue
2388 * @sk: sock to wait on
2389 * @timeo: for how long
2390 * @skb: last skb seen on sk_receive_queue
2392 * Now socket state including sk->sk_err is changed only under lock,
2393 * hence we may omit checks after joining wait queue.
2394 * We check receive queue before schedule() only as optimization;
2395 * it is very likely that release_sock() added new data.
2397 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2399 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2402 add_wait_queue(sk_sleep(sk), &wait);
2403 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2404 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2405 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2406 remove_wait_queue(sk_sleep(sk), &wait);
2409 EXPORT_SYMBOL(sk_wait_data);
2412 * __sk_mem_raise_allocated - increase memory_allocated
2414 * @size: memory size to allocate
2415 * @amt: pages to allocate
2416 * @kind: allocation type
2418 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2420 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2422 struct proto *prot = sk->sk_prot;
2423 long allocated = sk_memory_allocated_add(sk, amt);
2424 bool charged = true;
2426 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2427 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2428 goto suppress_allocation;
2431 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2432 sk_leave_memory_pressure(sk);
2436 /* Under pressure. */
2437 if (allocated > sk_prot_mem_limits(sk, 1))
2438 sk_enter_memory_pressure(sk);
2440 /* Over hard limit. */
2441 if (allocated > sk_prot_mem_limits(sk, 2))
2442 goto suppress_allocation;
2444 /* guarantee minimum buffer size under pressure */
2445 if (kind == SK_MEM_RECV) {
2446 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2449 } else { /* SK_MEM_SEND */
2450 int wmem0 = sk_get_wmem0(sk, prot);
2452 if (sk->sk_type == SOCK_STREAM) {
2453 if (sk->sk_wmem_queued < wmem0)
2455 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2460 if (sk_has_memory_pressure(sk)) {
2463 if (!sk_under_memory_pressure(sk))
2465 alloc = sk_sockets_allocated_read_positive(sk);
2466 if (sk_prot_mem_limits(sk, 2) > alloc *
2467 sk_mem_pages(sk->sk_wmem_queued +
2468 atomic_read(&sk->sk_rmem_alloc) +
2469 sk->sk_forward_alloc))
2473 suppress_allocation:
2475 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2476 sk_stream_moderate_sndbuf(sk);
2478 /* Fail only if socket is _under_ its sndbuf.
2479 * In this case we cannot block, so that we have to fail.
2481 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2485 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2486 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2488 sk_memory_allocated_sub(sk, amt);
2490 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2491 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2495 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2498 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2500 * @size: memory size to allocate
2501 * @kind: allocation type
2503 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2504 * rmem allocation. This function assumes that protocols which have
2505 * memory_pressure use sk_wmem_queued as write buffer accounting.
2507 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2509 int ret, amt = sk_mem_pages(size);
2511 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2512 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2514 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2517 EXPORT_SYMBOL(__sk_mem_schedule);
2520 * __sk_mem_reduce_allocated - reclaim memory_allocated
2522 * @amount: number of quanta
2524 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2526 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2528 sk_memory_allocated_sub(sk, amount);
2530 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2531 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2533 if (sk_under_memory_pressure(sk) &&
2534 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2535 sk_leave_memory_pressure(sk);
2537 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2540 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2542 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2544 void __sk_mem_reclaim(struct sock *sk, int amount)
2546 amount >>= SK_MEM_QUANTUM_SHIFT;
2547 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2548 __sk_mem_reduce_allocated(sk, amount);
2550 EXPORT_SYMBOL(__sk_mem_reclaim);
2552 int sk_set_peek_off(struct sock *sk, int val)
2554 sk->sk_peek_off = val;
2557 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2560 * Set of default routines for initialising struct proto_ops when
2561 * the protocol does not support a particular function. In certain
2562 * cases where it makes no sense for a protocol to have a "do nothing"
2563 * function, some default processing is provided.
2566 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2570 EXPORT_SYMBOL(sock_no_bind);
2572 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2577 EXPORT_SYMBOL(sock_no_connect);
2579 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2583 EXPORT_SYMBOL(sock_no_socketpair);
2585 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2590 EXPORT_SYMBOL(sock_no_accept);
2592 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2597 EXPORT_SYMBOL(sock_no_getname);
2599 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2603 EXPORT_SYMBOL(sock_no_ioctl);
2605 int sock_no_listen(struct socket *sock, int backlog)
2609 EXPORT_SYMBOL(sock_no_listen);
2611 int sock_no_shutdown(struct socket *sock, int how)
2615 EXPORT_SYMBOL(sock_no_shutdown);
2617 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2618 char __user *optval, unsigned int optlen)
2622 EXPORT_SYMBOL(sock_no_setsockopt);
2624 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2625 char __user *optval, int __user *optlen)
2629 EXPORT_SYMBOL(sock_no_getsockopt);
2631 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2635 EXPORT_SYMBOL(sock_no_sendmsg);
2637 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2641 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2643 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2648 EXPORT_SYMBOL(sock_no_recvmsg);
2650 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2652 /* Mirror missing mmap method error code */
2655 EXPORT_SYMBOL(sock_no_mmap);
2658 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2659 * various sock-based usage counts.
2661 void __receive_sock(struct file *file)
2663 struct socket *sock;
2667 * The resulting value of "error" is ignored here since we only
2668 * need to take action when the file is a socket and testing
2669 * "sock" for NULL is sufficient.
2671 sock = sock_from_file(file, &error);
2673 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2674 sock_update_classid(&sock->sk->sk_cgrp_data);
2678 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2681 struct msghdr msg = {.msg_flags = flags};
2683 char *kaddr = kmap(page);
2684 iov.iov_base = kaddr + offset;
2686 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2690 EXPORT_SYMBOL(sock_no_sendpage);
2692 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2693 int offset, size_t size, int flags)
2696 struct msghdr msg = {.msg_flags = flags};
2698 char *kaddr = kmap(page);
2700 iov.iov_base = kaddr + offset;
2702 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2706 EXPORT_SYMBOL(sock_no_sendpage_locked);
2709 * Default Socket Callbacks
2712 static void sock_def_wakeup(struct sock *sk)
2714 struct socket_wq *wq;
2717 wq = rcu_dereference(sk->sk_wq);
2718 if (skwq_has_sleeper(wq))
2719 wake_up_interruptible_all(&wq->wait);
2723 static void sock_def_error_report(struct sock *sk)
2725 struct socket_wq *wq;
2728 wq = rcu_dereference(sk->sk_wq);
2729 if (skwq_has_sleeper(wq))
2730 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2731 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2735 static void sock_def_readable(struct sock *sk)
2737 struct socket_wq *wq;
2740 wq = rcu_dereference(sk->sk_wq);
2741 if (skwq_has_sleeper(wq))
2742 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2743 EPOLLRDNORM | EPOLLRDBAND);
2744 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2748 static void sock_def_write_space(struct sock *sk)
2750 struct socket_wq *wq;
2754 /* Do not wake up a writer until he can make "significant"
2757 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2758 wq = rcu_dereference(sk->sk_wq);
2759 if (skwq_has_sleeper(wq))
2760 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2761 EPOLLWRNORM | EPOLLWRBAND);
2763 /* Should agree with poll, otherwise some programs break */
2764 if (sock_writeable(sk))
2765 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2771 static void sock_def_destruct(struct sock *sk)
2775 void sk_send_sigurg(struct sock *sk)
2777 if (sk->sk_socket && sk->sk_socket->file)
2778 if (send_sigurg(&sk->sk_socket->file->f_owner))
2779 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2781 EXPORT_SYMBOL(sk_send_sigurg);
2783 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2784 unsigned long expires)
2786 if (!mod_timer(timer, expires))
2789 EXPORT_SYMBOL(sk_reset_timer);
2791 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2793 if (del_timer(timer))
2796 EXPORT_SYMBOL(sk_stop_timer);
2798 void sock_init_data(struct socket *sock, struct sock *sk)
2801 sk->sk_send_head = NULL;
2803 timer_setup(&sk->sk_timer, NULL, 0);
2805 sk->sk_allocation = GFP_KERNEL;
2806 sk->sk_rcvbuf = sysctl_rmem_default;
2807 sk->sk_sndbuf = sysctl_wmem_default;
2808 sk->sk_state = TCP_CLOSE;
2809 sk_set_socket(sk, sock);
2811 sock_set_flag(sk, SOCK_ZAPPED);
2814 sk->sk_type = sock->type;
2815 sk->sk_wq = sock->wq;
2817 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2820 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2823 rwlock_init(&sk->sk_callback_lock);
2824 if (sk->sk_kern_sock)
2825 lockdep_set_class_and_name(
2826 &sk->sk_callback_lock,
2827 af_kern_callback_keys + sk->sk_family,
2828 af_family_kern_clock_key_strings[sk->sk_family]);
2830 lockdep_set_class_and_name(
2831 &sk->sk_callback_lock,
2832 af_callback_keys + sk->sk_family,
2833 af_family_clock_key_strings[sk->sk_family]);
2835 sk->sk_state_change = sock_def_wakeup;
2836 sk->sk_data_ready = sock_def_readable;
2837 sk->sk_write_space = sock_def_write_space;
2838 sk->sk_error_report = sock_def_error_report;
2839 sk->sk_destruct = sock_def_destruct;
2841 sk->sk_frag.page = NULL;
2842 sk->sk_frag.offset = 0;
2843 sk->sk_peek_off = -1;
2845 sk->sk_peer_pid = NULL;
2846 sk->sk_peer_cred = NULL;
2847 spin_lock_init(&sk->sk_peer_lock);
2849 sk->sk_write_pending = 0;
2850 sk->sk_rcvlowat = 1;
2851 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2852 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2854 sk->sk_stamp = SK_DEFAULT_STAMP;
2855 #if BITS_PER_LONG==32
2856 seqlock_init(&sk->sk_stamp_seq);
2858 atomic_set(&sk->sk_zckey, 0);
2860 #ifdef CONFIG_NET_RX_BUSY_POLL
2862 sk->sk_ll_usec = sysctl_net_busy_read;
2865 sk->sk_max_pacing_rate = ~0U;
2866 sk->sk_pacing_rate = ~0U;
2867 sk->sk_pacing_shift = 10;
2868 sk->sk_incoming_cpu = -1;
2870 sk_rx_queue_clear(sk);
2872 * Before updating sk_refcnt, we must commit prior changes to memory
2873 * (Documentation/RCU/rculist_nulls.txt for details)
2876 refcount_set(&sk->sk_refcnt, 1);
2877 atomic_set(&sk->sk_drops, 0);
2879 EXPORT_SYMBOL(sock_init_data);
2881 void lock_sock_nested(struct sock *sk, int subclass)
2884 spin_lock_bh(&sk->sk_lock.slock);
2885 if (sk->sk_lock.owned)
2887 sk->sk_lock.owned = 1;
2888 spin_unlock(&sk->sk_lock.slock);
2890 * The sk_lock has mutex_lock() semantics here:
2892 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2895 EXPORT_SYMBOL(lock_sock_nested);
2897 void release_sock(struct sock *sk)
2899 spin_lock_bh(&sk->sk_lock.slock);
2900 if (sk->sk_backlog.tail)
2903 /* Warning : release_cb() might need to release sk ownership,
2904 * ie call sock_release_ownership(sk) before us.
2906 if (sk->sk_prot->release_cb)
2907 sk->sk_prot->release_cb(sk);
2909 sock_release_ownership(sk);
2910 if (waitqueue_active(&sk->sk_lock.wq))
2911 wake_up(&sk->sk_lock.wq);
2912 spin_unlock_bh(&sk->sk_lock.slock);
2914 EXPORT_SYMBOL(release_sock);
2917 * lock_sock_fast - fast version of lock_sock
2920 * This version should be used for very small section, where process wont block
2921 * return false if fast path is taken:
2923 * sk_lock.slock locked, owned = 0, BH disabled
2925 * return true if slow path is taken:
2927 * sk_lock.slock unlocked, owned = 1, BH enabled
2929 bool lock_sock_fast(struct sock *sk)
2932 spin_lock_bh(&sk->sk_lock.slock);
2934 if (!sk->sk_lock.owned)
2936 * Note : We must disable BH
2941 sk->sk_lock.owned = 1;
2942 spin_unlock(&sk->sk_lock.slock);
2944 * The sk_lock has mutex_lock() semantics here:
2946 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2950 EXPORT_SYMBOL(lock_sock_fast);
2952 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2956 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2957 tv = ktime_to_timeval(sock_read_timestamp(sk));
2958 if (tv.tv_sec == -1)
2960 if (tv.tv_sec == 0) {
2961 ktime_t kt = ktime_get_real();
2962 sock_write_timestamp(sk, kt);
2963 tv = ktime_to_timeval(kt);
2965 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2967 EXPORT_SYMBOL(sock_get_timestamp);
2969 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2973 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2974 ts = ktime_to_timespec(sock_read_timestamp(sk));
2975 if (ts.tv_sec == -1)
2977 if (ts.tv_sec == 0) {
2978 ktime_t kt = ktime_get_real();
2979 sock_write_timestamp(sk, kt);
2980 ts = ktime_to_timespec(sk->sk_stamp);
2982 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2984 EXPORT_SYMBOL(sock_get_timestampns);
2986 void sock_enable_timestamp(struct sock *sk, int flag)
2988 if (!sock_flag(sk, flag)) {
2989 unsigned long previous_flags = sk->sk_flags;
2991 sock_set_flag(sk, flag);
2993 * we just set one of the two flags which require net
2994 * time stamping, but time stamping might have been on
2995 * already because of the other one
2997 if (sock_needs_netstamp(sk) &&
2998 !(previous_flags & SK_FLAGS_TIMESTAMP))
2999 net_enable_timestamp();
3003 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3004 int level, int type)
3006 struct sock_exterr_skb *serr;
3007 struct sk_buff *skb;
3011 skb = sock_dequeue_err_skb(sk);
3017 msg->msg_flags |= MSG_TRUNC;
3020 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3024 sock_recv_timestamp(msg, sk, skb);
3026 serr = SKB_EXT_ERR(skb);
3027 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3029 msg->msg_flags |= MSG_ERRQUEUE;
3037 EXPORT_SYMBOL(sock_recv_errqueue);
3040 * Get a socket option on an socket.
3042 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3043 * asynchronous errors should be reported by getsockopt. We assume
3044 * this means if you specify SO_ERROR (otherwise whats the point of it).
3046 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3047 char __user *optval, int __user *optlen)
3049 struct sock *sk = sock->sk;
3051 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3053 EXPORT_SYMBOL(sock_common_getsockopt);
3055 #ifdef CONFIG_COMPAT
3056 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3057 char __user *optval, int __user *optlen)
3059 struct sock *sk = sock->sk;
3061 if (sk->sk_prot->compat_getsockopt != NULL)
3062 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3064 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3066 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3069 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3072 struct sock *sk = sock->sk;
3076 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3077 flags & ~MSG_DONTWAIT, &addr_len);
3079 msg->msg_namelen = addr_len;
3082 EXPORT_SYMBOL(sock_common_recvmsg);
3085 * Set socket options on an inet socket.
3087 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3088 char __user *optval, unsigned int optlen)
3090 struct sock *sk = sock->sk;
3092 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3094 EXPORT_SYMBOL(sock_common_setsockopt);
3096 #ifdef CONFIG_COMPAT
3097 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3098 char __user *optval, unsigned int optlen)
3100 struct sock *sk = sock->sk;
3102 if (sk->sk_prot->compat_setsockopt != NULL)
3103 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3105 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3107 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3110 void sk_common_release(struct sock *sk)
3112 if (sk->sk_prot->destroy)
3113 sk->sk_prot->destroy(sk);
3116 * Observation: when sock_common_release is called, processes have
3117 * no access to socket. But net still has.
3118 * Step one, detach it from networking:
3120 * A. Remove from hash tables.
3123 sk->sk_prot->unhash(sk);
3126 * In this point socket cannot receive new packets, but it is possible
3127 * that some packets are in flight because some CPU runs receiver and
3128 * did hash table lookup before we unhashed socket. They will achieve
3129 * receive queue and will be purged by socket destructor.
3131 * Also we still have packets pending on receive queue and probably,
3132 * our own packets waiting in device queues. sock_destroy will drain
3133 * receive queue, but transmitted packets will delay socket destruction
3134 * until the last reference will be released.
3139 xfrm_sk_free_policy(sk);
3141 sk_refcnt_debug_release(sk);
3145 EXPORT_SYMBOL(sk_common_release);
3147 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3149 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3151 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3152 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3153 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3154 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3155 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3156 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3157 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3158 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3159 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3162 #ifdef CONFIG_PROC_FS
3163 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3165 int val[PROTO_INUSE_NR];
3168 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3170 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3172 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3174 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3176 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3178 int cpu, idx = prot->inuse_idx;
3181 for_each_possible_cpu(cpu)
3182 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3184 return res >= 0 ? res : 0;
3186 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3188 static void sock_inuse_add(struct net *net, int val)
3190 this_cpu_add(*net->core.sock_inuse, val);
3193 int sock_inuse_get(struct net *net)
3197 for_each_possible_cpu(cpu)
3198 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3203 EXPORT_SYMBOL_GPL(sock_inuse_get);
3205 static int __net_init sock_inuse_init_net(struct net *net)
3207 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3208 if (net->core.prot_inuse == NULL)
3211 net->core.sock_inuse = alloc_percpu(int);
3212 if (net->core.sock_inuse == NULL)
3218 free_percpu(net->core.prot_inuse);
3222 static void __net_exit sock_inuse_exit_net(struct net *net)
3224 free_percpu(net->core.prot_inuse);
3225 free_percpu(net->core.sock_inuse);
3228 static struct pernet_operations net_inuse_ops = {
3229 .init = sock_inuse_init_net,
3230 .exit = sock_inuse_exit_net,
3233 static __init int net_inuse_init(void)
3235 if (register_pernet_subsys(&net_inuse_ops))
3236 panic("Cannot initialize net inuse counters");
3241 core_initcall(net_inuse_init);
3243 static void assign_proto_idx(struct proto *prot)
3245 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3247 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3248 pr_err("PROTO_INUSE_NR exhausted\n");
3252 set_bit(prot->inuse_idx, proto_inuse_idx);
3255 static void release_proto_idx(struct proto *prot)
3257 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3258 clear_bit(prot->inuse_idx, proto_inuse_idx);
3261 static inline void assign_proto_idx(struct proto *prot)
3265 static inline void release_proto_idx(struct proto *prot)
3269 static void sock_inuse_add(struct net *net, int val)
3274 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3278 kfree(rsk_prot->slab_name);
3279 rsk_prot->slab_name = NULL;
3280 kmem_cache_destroy(rsk_prot->slab);
3281 rsk_prot->slab = NULL;
3284 static int req_prot_init(const struct proto *prot)
3286 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3291 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3293 if (!rsk_prot->slab_name)
3296 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3297 rsk_prot->obj_size, 0,
3298 SLAB_ACCOUNT | prot->slab_flags,
3301 if (!rsk_prot->slab) {
3302 pr_crit("%s: Can't create request sock SLAB cache!\n",
3309 int proto_register(struct proto *prot, int alloc_slab)
3312 prot->slab = kmem_cache_create_usercopy(prot->name,
3314 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3316 prot->useroffset, prot->usersize,
3319 if (prot->slab == NULL) {
3320 pr_crit("%s: Can't create sock SLAB cache!\n",
3325 if (req_prot_init(prot))
3326 goto out_free_request_sock_slab;
3328 if (prot->twsk_prot != NULL) {
3329 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3331 if (prot->twsk_prot->twsk_slab_name == NULL)
3332 goto out_free_request_sock_slab;
3334 prot->twsk_prot->twsk_slab =
3335 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3336 prot->twsk_prot->twsk_obj_size,
3341 if (prot->twsk_prot->twsk_slab == NULL)
3342 goto out_free_timewait_sock_slab_name;
3346 mutex_lock(&proto_list_mutex);
3347 list_add(&prot->node, &proto_list);
3348 assign_proto_idx(prot);
3349 mutex_unlock(&proto_list_mutex);
3352 out_free_timewait_sock_slab_name:
3353 kfree(prot->twsk_prot->twsk_slab_name);
3354 out_free_request_sock_slab:
3355 req_prot_cleanup(prot->rsk_prot);
3357 kmem_cache_destroy(prot->slab);
3362 EXPORT_SYMBOL(proto_register);
3364 void proto_unregister(struct proto *prot)
3366 mutex_lock(&proto_list_mutex);
3367 release_proto_idx(prot);
3368 list_del(&prot->node);
3369 mutex_unlock(&proto_list_mutex);
3371 kmem_cache_destroy(prot->slab);
3374 req_prot_cleanup(prot->rsk_prot);
3376 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3377 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3378 kfree(prot->twsk_prot->twsk_slab_name);
3379 prot->twsk_prot->twsk_slab = NULL;
3382 EXPORT_SYMBOL(proto_unregister);
3384 int sock_load_diag_module(int family, int protocol)
3387 if (!sock_is_registered(family))
3390 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3391 NETLINK_SOCK_DIAG, family);
3395 if (family == AF_INET &&
3396 protocol != IPPROTO_RAW &&
3397 !rcu_access_pointer(inet_protos[protocol]))
3401 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3402 NETLINK_SOCK_DIAG, family, protocol);
3404 EXPORT_SYMBOL(sock_load_diag_module);
3406 #ifdef CONFIG_PROC_FS
3407 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3408 __acquires(proto_list_mutex)
3410 mutex_lock(&proto_list_mutex);
3411 return seq_list_start_head(&proto_list, *pos);
3414 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3416 return seq_list_next(v, &proto_list, pos);
3419 static void proto_seq_stop(struct seq_file *seq, void *v)
3420 __releases(proto_list_mutex)
3422 mutex_unlock(&proto_list_mutex);
3425 static char proto_method_implemented(const void *method)
3427 return method == NULL ? 'n' : 'y';
3429 static long sock_prot_memory_allocated(struct proto *proto)
3431 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3434 static char *sock_prot_memory_pressure(struct proto *proto)
3436 return proto->memory_pressure != NULL ?
3437 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3440 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3443 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3444 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3447 sock_prot_inuse_get(seq_file_net(seq), proto),
3448 sock_prot_memory_allocated(proto),
3449 sock_prot_memory_pressure(proto),
3451 proto->slab == NULL ? "no" : "yes",
3452 module_name(proto->owner),
3453 proto_method_implemented(proto->close),
3454 proto_method_implemented(proto->connect),
3455 proto_method_implemented(proto->disconnect),
3456 proto_method_implemented(proto->accept),
3457 proto_method_implemented(proto->ioctl),
3458 proto_method_implemented(proto->init),
3459 proto_method_implemented(proto->destroy),
3460 proto_method_implemented(proto->shutdown),
3461 proto_method_implemented(proto->setsockopt),
3462 proto_method_implemented(proto->getsockopt),
3463 proto_method_implemented(proto->sendmsg),
3464 proto_method_implemented(proto->recvmsg),
3465 proto_method_implemented(proto->sendpage),
3466 proto_method_implemented(proto->bind),
3467 proto_method_implemented(proto->backlog_rcv),
3468 proto_method_implemented(proto->hash),
3469 proto_method_implemented(proto->unhash),
3470 proto_method_implemented(proto->get_port),
3471 proto_method_implemented(proto->enter_memory_pressure));
3474 static int proto_seq_show(struct seq_file *seq, void *v)
3476 if (v == &proto_list)
3477 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3486 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3488 proto_seq_printf(seq, list_entry(v, struct proto, node));
3492 static const struct seq_operations proto_seq_ops = {
3493 .start = proto_seq_start,
3494 .next = proto_seq_next,
3495 .stop = proto_seq_stop,
3496 .show = proto_seq_show,
3499 static __net_init int proto_init_net(struct net *net)
3501 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3502 sizeof(struct seq_net_private)))
3508 static __net_exit void proto_exit_net(struct net *net)
3510 remove_proc_entry("protocols", net->proc_net);
3514 static __net_initdata struct pernet_operations proto_net_ops = {
3515 .init = proto_init_net,
3516 .exit = proto_exit_net,
3519 static int __init proto_init(void)
3521 return register_pernet_subsys(&proto_net_ops);
3524 subsys_initcall(proto_init);
3526 #endif /* PROC_FS */
3528 #ifdef CONFIG_NET_RX_BUSY_POLL
3529 bool sk_busy_loop_end(void *p, unsigned long start_time)
3531 struct sock *sk = p;
3533 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3534 sk_busy_loop_timeout(sk, start_time);
3536 EXPORT_SYMBOL(sk_busy_loop_end);
3537 #endif /* CONFIG_NET_RX_BUSY_POLL */