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);
1061 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1062 struct ucred *ucred)
1064 ucred->pid = pid_vnr(pid);
1065 ucred->uid = ucred->gid = -1;
1067 struct user_namespace *current_ns = current_user_ns();
1069 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1070 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1074 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1076 struct user_namespace *user_ns = current_user_ns();
1079 for (i = 0; i < src->ngroups; i++)
1080 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1086 int sock_getsockopt(struct socket *sock, int level, int optname,
1087 char __user *optval, int __user *optlen)
1089 struct sock *sk = sock->sk;
1096 struct sock_txtime txtime;
1099 int lv = sizeof(int);
1102 if (get_user(len, optlen))
1107 memset(&v, 0, sizeof(v));
1111 v.val = sock_flag(sk, SOCK_DBG);
1115 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1119 v.val = sock_flag(sk, SOCK_BROADCAST);
1123 v.val = sk->sk_sndbuf;
1127 v.val = sk->sk_rcvbuf;
1131 v.val = sk->sk_reuse;
1135 v.val = sk->sk_reuseport;
1139 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1143 v.val = sk->sk_type;
1147 v.val = sk->sk_protocol;
1151 v.val = sk->sk_family;
1155 v.val = -sock_error(sk);
1157 v.val = xchg(&sk->sk_err_soft, 0);
1161 v.val = sock_flag(sk, SOCK_URGINLINE);
1165 v.val = sk->sk_no_check_tx;
1169 v.val = sk->sk_priority;
1173 lv = sizeof(v.ling);
1174 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1175 v.ling.l_linger = sk->sk_lingertime / HZ;
1179 sock_warn_obsolete_bsdism("getsockopt");
1183 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1184 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1187 case SO_TIMESTAMPNS:
1188 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1191 case SO_TIMESTAMPING:
1192 v.val = sk->sk_tsflags;
1196 lv = sizeof(struct timeval);
1197 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1201 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1202 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1207 lv = sizeof(struct timeval);
1208 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1212 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1213 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1218 v.val = sk->sk_rcvlowat;
1226 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1231 struct ucred peercred;
1232 if (len > sizeof(peercred))
1233 len = sizeof(peercred);
1234 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1235 if (copy_to_user(optval, &peercred, len))
1244 if (!sk->sk_peer_cred)
1247 n = sk->sk_peer_cred->group_info->ngroups;
1248 if (len < n * sizeof(gid_t)) {
1249 len = n * sizeof(gid_t);
1250 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1252 len = n * sizeof(gid_t);
1254 ret = groups_to_user((gid_t __user *)optval,
1255 sk->sk_peer_cred->group_info);
1265 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1270 if (copy_to_user(optval, address, len))
1275 /* Dubious BSD thing... Probably nobody even uses it, but
1276 * the UNIX standard wants it for whatever reason... -DaveM
1279 v.val = sk->sk_state == TCP_LISTEN;
1283 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1287 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1290 v.val = sk->sk_mark;
1294 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1297 case SO_WIFI_STATUS:
1298 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1302 if (!sock->ops->set_peek_off)
1305 v.val = sk->sk_peek_off;
1308 v.val = sock_flag(sk, SOCK_NOFCS);
1311 case SO_BINDTODEVICE:
1312 return sock_getbindtodevice(sk, optval, optlen, len);
1315 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1321 case SO_LOCK_FILTER:
1322 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1325 case SO_BPF_EXTENSIONS:
1326 v.val = bpf_tell_extensions();
1329 case SO_SELECT_ERR_QUEUE:
1330 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1333 #ifdef CONFIG_NET_RX_BUSY_POLL
1335 v.val = sk->sk_ll_usec;
1339 case SO_MAX_PACING_RATE:
1340 v.val = sk->sk_max_pacing_rate;
1343 case SO_INCOMING_CPU:
1344 v.val = READ_ONCE(sk->sk_incoming_cpu);
1349 u32 meminfo[SK_MEMINFO_VARS];
1351 sk_get_meminfo(sk, meminfo);
1353 len = min_t(unsigned int, len, sizeof(meminfo));
1354 if (copy_to_user(optval, &meminfo, len))
1360 #ifdef CONFIG_NET_RX_BUSY_POLL
1361 case SO_INCOMING_NAPI_ID:
1362 v.val = READ_ONCE(sk->sk_napi_id);
1364 /* aggregate non-NAPI IDs down to 0 */
1365 if (v.val < MIN_NAPI_ID)
1375 v.val64 = sock_gen_cookie(sk);
1379 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1383 lv = sizeof(v.txtime);
1384 v.txtime.clockid = sk->sk_clockid;
1385 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1386 SOF_TXTIME_DEADLINE_MODE : 0;
1387 v.txtime.flags |= sk->sk_txtime_report_errors ?
1388 SOF_TXTIME_REPORT_ERRORS : 0;
1392 /* We implement the SO_SNDLOWAT etc to not be settable
1395 return -ENOPROTOOPT;
1400 if (copy_to_user(optval, &v, len))
1403 if (put_user(len, optlen))
1409 * Initialize an sk_lock.
1411 * (We also register the sk_lock with the lock validator.)
1413 static inline void sock_lock_init(struct sock *sk)
1415 if (sk->sk_kern_sock)
1416 sock_lock_init_class_and_name(
1418 af_family_kern_slock_key_strings[sk->sk_family],
1419 af_family_kern_slock_keys + sk->sk_family,
1420 af_family_kern_key_strings[sk->sk_family],
1421 af_family_kern_keys + sk->sk_family);
1423 sock_lock_init_class_and_name(
1425 af_family_slock_key_strings[sk->sk_family],
1426 af_family_slock_keys + sk->sk_family,
1427 af_family_key_strings[sk->sk_family],
1428 af_family_keys + sk->sk_family);
1432 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1433 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1434 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1436 static void sock_copy(struct sock *nsk, const struct sock *osk)
1438 #ifdef CONFIG_SECURITY_NETWORK
1439 void *sptr = nsk->sk_security;
1441 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1443 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1444 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1446 #ifdef CONFIG_SECURITY_NETWORK
1447 nsk->sk_security = sptr;
1448 security_sk_clone(osk, nsk);
1452 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1456 struct kmem_cache *slab;
1460 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1463 if (priority & __GFP_ZERO)
1464 sk_prot_clear_nulls(sk, prot->obj_size);
1466 sk = kmalloc(prot->obj_size, priority);
1469 if (security_sk_alloc(sk, family, priority))
1472 if (!try_module_get(prot->owner))
1474 sk_tx_queue_clear(sk);
1480 security_sk_free(sk);
1483 kmem_cache_free(slab, sk);
1489 static void sk_prot_free(struct proto *prot, struct sock *sk)
1491 struct kmem_cache *slab;
1492 struct module *owner;
1494 owner = prot->owner;
1497 cgroup_sk_free(&sk->sk_cgrp_data);
1498 mem_cgroup_sk_free(sk);
1499 security_sk_free(sk);
1501 kmem_cache_free(slab, sk);
1508 * sk_alloc - All socket objects are allocated here
1509 * @net: the applicable net namespace
1510 * @family: protocol family
1511 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1512 * @prot: struct proto associated with this new sock instance
1513 * @kern: is this to be a kernel socket?
1515 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1516 struct proto *prot, int kern)
1520 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1522 sk->sk_family = family;
1524 * See comment in struct sock definition to understand
1525 * why we need sk_prot_creator -acme
1527 sk->sk_prot = sk->sk_prot_creator = prot;
1528 sk->sk_kern_sock = kern;
1530 sk->sk_net_refcnt = kern ? 0 : 1;
1531 if (likely(sk->sk_net_refcnt)) {
1533 sock_inuse_add(net, 1);
1536 sock_net_set(sk, net);
1537 refcount_set(&sk->sk_wmem_alloc, 1);
1539 mem_cgroup_sk_alloc(sk);
1540 cgroup_sk_alloc(&sk->sk_cgrp_data);
1541 sock_update_classid(&sk->sk_cgrp_data);
1542 sock_update_netprioidx(&sk->sk_cgrp_data);
1543 sk_tx_queue_clear(sk);
1548 EXPORT_SYMBOL(sk_alloc);
1550 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1551 * grace period. This is the case for UDP sockets and TCP listeners.
1553 static void __sk_destruct(struct rcu_head *head)
1555 struct sock *sk = container_of(head, struct sock, sk_rcu);
1556 struct sk_filter *filter;
1558 if (sk->sk_destruct)
1559 sk->sk_destruct(sk);
1561 filter = rcu_dereference_check(sk->sk_filter,
1562 refcount_read(&sk->sk_wmem_alloc) == 0);
1564 sk_filter_uncharge(sk, filter);
1565 RCU_INIT_POINTER(sk->sk_filter, NULL);
1568 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1570 if (atomic_read(&sk->sk_omem_alloc))
1571 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1572 __func__, atomic_read(&sk->sk_omem_alloc));
1574 if (sk->sk_frag.page) {
1575 put_page(sk->sk_frag.page);
1576 sk->sk_frag.page = NULL;
1579 if (sk->sk_peer_cred)
1580 put_cred(sk->sk_peer_cred);
1581 put_pid(sk->sk_peer_pid);
1582 if (likely(sk->sk_net_refcnt))
1583 put_net(sock_net(sk));
1584 sk_prot_free(sk->sk_prot_creator, sk);
1587 void sk_destruct(struct sock *sk)
1589 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1591 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1592 reuseport_detach_sock(sk);
1593 use_call_rcu = true;
1597 call_rcu(&sk->sk_rcu, __sk_destruct);
1599 __sk_destruct(&sk->sk_rcu);
1602 static void __sk_free(struct sock *sk)
1604 if (likely(sk->sk_net_refcnt))
1605 sock_inuse_add(sock_net(sk), -1);
1607 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1608 sock_diag_broadcast_destroy(sk);
1613 void sk_free(struct sock *sk)
1616 * We subtract one from sk_wmem_alloc and can know if
1617 * some packets are still in some tx queue.
1618 * If not null, sock_wfree() will call __sk_free(sk) later
1620 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1623 EXPORT_SYMBOL(sk_free);
1625 static void sk_init_common(struct sock *sk)
1627 skb_queue_head_init(&sk->sk_receive_queue);
1628 skb_queue_head_init(&sk->sk_write_queue);
1629 skb_queue_head_init(&sk->sk_error_queue);
1631 rwlock_init(&sk->sk_callback_lock);
1632 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1633 af_rlock_keys + sk->sk_family,
1634 af_family_rlock_key_strings[sk->sk_family]);
1635 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1636 af_wlock_keys + sk->sk_family,
1637 af_family_wlock_key_strings[sk->sk_family]);
1638 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1639 af_elock_keys + sk->sk_family,
1640 af_family_elock_key_strings[sk->sk_family]);
1641 lockdep_set_class_and_name(&sk->sk_callback_lock,
1642 af_callback_keys + sk->sk_family,
1643 af_family_clock_key_strings[sk->sk_family]);
1647 * sk_clone_lock - clone a socket, and lock its clone
1648 * @sk: the socket to clone
1649 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1651 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1653 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1656 bool is_charged = true;
1658 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1659 if (newsk != NULL) {
1660 struct sk_filter *filter;
1662 sock_copy(newsk, sk);
1664 newsk->sk_prot_creator = sk->sk_prot;
1667 if (likely(newsk->sk_net_refcnt))
1668 get_net(sock_net(newsk));
1669 sk_node_init(&newsk->sk_node);
1670 sock_lock_init(newsk);
1671 bh_lock_sock(newsk);
1672 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1673 newsk->sk_backlog.len = 0;
1675 atomic_set(&newsk->sk_rmem_alloc, 0);
1677 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1679 refcount_set(&newsk->sk_wmem_alloc, 1);
1680 atomic_set(&newsk->sk_omem_alloc, 0);
1681 sk_init_common(newsk);
1683 newsk->sk_dst_cache = NULL;
1684 newsk->sk_dst_pending_confirm = 0;
1685 newsk->sk_wmem_queued = 0;
1686 newsk->sk_forward_alloc = 0;
1687 atomic_set(&newsk->sk_drops, 0);
1688 newsk->sk_send_head = NULL;
1689 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1690 atomic_set(&newsk->sk_zckey, 0);
1692 sock_reset_flag(newsk, SOCK_DONE);
1694 /* sk->sk_memcg will be populated at accept() time */
1695 newsk->sk_memcg = NULL;
1697 cgroup_sk_clone(&newsk->sk_cgrp_data);
1700 filter = rcu_dereference(sk->sk_filter);
1702 /* though it's an empty new sock, the charging may fail
1703 * if sysctl_optmem_max was changed between creation of
1704 * original socket and cloning
1706 is_charged = sk_filter_charge(newsk, filter);
1707 RCU_INIT_POINTER(newsk->sk_filter, filter);
1710 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1711 /* We need to make sure that we don't uncharge the new
1712 * socket if we couldn't charge it in the first place
1713 * as otherwise we uncharge the parent's filter.
1716 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1717 sk_free_unlock_clone(newsk);
1721 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1724 newsk->sk_err_soft = 0;
1725 newsk->sk_priority = 0;
1726 newsk->sk_incoming_cpu = raw_smp_processor_id();
1727 atomic64_set(&newsk->sk_cookie, 0);
1728 if (likely(newsk->sk_net_refcnt))
1729 sock_inuse_add(sock_net(newsk), 1);
1732 * Before updating sk_refcnt, we must commit prior changes to memory
1733 * (Documentation/RCU/rculist_nulls.txt for details)
1736 refcount_set(&newsk->sk_refcnt, 2);
1739 * Increment the counter in the same struct proto as the master
1740 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1741 * is the same as sk->sk_prot->socks, as this field was copied
1744 * This _changes_ the previous behaviour, where
1745 * tcp_create_openreq_child always was incrementing the
1746 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1747 * to be taken into account in all callers. -acme
1749 sk_refcnt_debug_inc(newsk);
1750 sk_set_socket(newsk, NULL);
1751 sk_tx_queue_clear(newsk);
1752 newsk->sk_wq = NULL;
1754 if (newsk->sk_prot->sockets_allocated)
1755 sk_sockets_allocated_inc(newsk);
1757 if (sock_needs_netstamp(sk) &&
1758 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1759 net_enable_timestamp();
1764 EXPORT_SYMBOL_GPL(sk_clone_lock);
1766 void sk_free_unlock_clone(struct sock *sk)
1768 /* It is still raw copy of parent, so invalidate
1769 * destructor and make plain sk_free() */
1770 sk->sk_destruct = NULL;
1774 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1776 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1780 sk_dst_set(sk, dst);
1781 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1782 if (sk->sk_route_caps & NETIF_F_GSO)
1783 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1784 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1785 if (sk_can_gso(sk)) {
1786 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1787 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1789 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1790 sk->sk_gso_max_size = dst->dev->gso_max_size;
1791 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1794 sk->sk_gso_max_segs = max_segs;
1796 EXPORT_SYMBOL_GPL(sk_setup_caps);
1799 * Simple resource managers for sockets.
1804 * Write buffer destructor automatically called from kfree_skb.
1806 void sock_wfree(struct sk_buff *skb)
1808 struct sock *sk = skb->sk;
1809 unsigned int len = skb->truesize;
1811 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1813 * Keep a reference on sk_wmem_alloc, this will be released
1814 * after sk_write_space() call
1816 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1817 sk->sk_write_space(sk);
1821 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1822 * could not do because of in-flight packets
1824 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1827 EXPORT_SYMBOL(sock_wfree);
1829 /* This variant of sock_wfree() is used by TCP,
1830 * since it sets SOCK_USE_WRITE_QUEUE.
1832 void __sock_wfree(struct sk_buff *skb)
1834 struct sock *sk = skb->sk;
1836 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1840 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1845 if (unlikely(!sk_fullsock(sk))) {
1846 skb->destructor = sock_edemux;
1851 skb->destructor = sock_wfree;
1852 skb_set_hash_from_sk(skb, sk);
1854 * We used to take a refcount on sk, but following operation
1855 * is enough to guarantee sk_free() wont free this sock until
1856 * all in-flight packets are completed
1858 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1860 EXPORT_SYMBOL(skb_set_owner_w);
1862 /* This helper is used by netem, as it can hold packets in its
1863 * delay queue. We want to allow the owner socket to send more
1864 * packets, as if they were already TX completed by a typical driver.
1865 * But we also want to keep skb->sk set because some packet schedulers
1866 * rely on it (sch_fq for example).
1868 void skb_orphan_partial(struct sk_buff *skb)
1870 if (skb_is_tcp_pure_ack(skb))
1873 if (skb->destructor == sock_wfree
1875 || skb->destructor == tcp_wfree
1878 struct sock *sk = skb->sk;
1880 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1881 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
1882 skb->destructor = sock_efree;
1888 EXPORT_SYMBOL(skb_orphan_partial);
1891 * Read buffer destructor automatically called from kfree_skb.
1893 void sock_rfree(struct sk_buff *skb)
1895 struct sock *sk = skb->sk;
1896 unsigned int len = skb->truesize;
1898 atomic_sub(len, &sk->sk_rmem_alloc);
1899 sk_mem_uncharge(sk, len);
1901 EXPORT_SYMBOL(sock_rfree);
1904 * Buffer destructor for skbs that are not used directly in read or write
1905 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1907 void sock_efree(struct sk_buff *skb)
1911 EXPORT_SYMBOL(sock_efree);
1913 kuid_t sock_i_uid(struct sock *sk)
1917 read_lock_bh(&sk->sk_callback_lock);
1918 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1919 read_unlock_bh(&sk->sk_callback_lock);
1922 EXPORT_SYMBOL(sock_i_uid);
1924 unsigned long sock_i_ino(struct sock *sk)
1928 read_lock_bh(&sk->sk_callback_lock);
1929 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1930 read_unlock_bh(&sk->sk_callback_lock);
1933 EXPORT_SYMBOL(sock_i_ino);
1936 * Allocate a skb from the socket's send buffer.
1938 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1941 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1942 struct sk_buff *skb = alloc_skb(size, priority);
1944 skb_set_owner_w(skb, sk);
1950 EXPORT_SYMBOL(sock_wmalloc);
1952 static void sock_ofree(struct sk_buff *skb)
1954 struct sock *sk = skb->sk;
1956 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
1959 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1962 struct sk_buff *skb;
1964 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
1965 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
1969 skb = alloc_skb(size, priority);
1973 atomic_add(skb->truesize, &sk->sk_omem_alloc);
1975 skb->destructor = sock_ofree;
1980 * Allocate a memory block from the socket's option memory buffer.
1982 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1984 if ((unsigned int)size <= sysctl_optmem_max &&
1985 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1987 /* First do the add, to avoid the race if kmalloc
1990 atomic_add(size, &sk->sk_omem_alloc);
1991 mem = kmalloc(size, priority);
1994 atomic_sub(size, &sk->sk_omem_alloc);
1998 EXPORT_SYMBOL(sock_kmalloc);
2000 /* Free an option memory block. Note, we actually want the inline
2001 * here as this allows gcc to detect the nullify and fold away the
2002 * condition entirely.
2004 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2007 if (WARN_ON_ONCE(!mem))
2013 atomic_sub(size, &sk->sk_omem_alloc);
2016 void sock_kfree_s(struct sock *sk, void *mem, int size)
2018 __sock_kfree_s(sk, mem, size, false);
2020 EXPORT_SYMBOL(sock_kfree_s);
2022 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2024 __sock_kfree_s(sk, mem, size, true);
2026 EXPORT_SYMBOL(sock_kzfree_s);
2028 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2029 I think, these locks should be removed for datagram sockets.
2031 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2035 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2039 if (signal_pending(current))
2041 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2042 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2043 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2045 if (sk->sk_shutdown & SEND_SHUTDOWN)
2049 timeo = schedule_timeout(timeo);
2051 finish_wait(sk_sleep(sk), &wait);
2057 * Generic send/receive buffer handlers
2060 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2061 unsigned long data_len, int noblock,
2062 int *errcode, int max_page_order)
2064 struct sk_buff *skb;
2068 timeo = sock_sndtimeo(sk, noblock);
2070 err = sock_error(sk);
2075 if (sk->sk_shutdown & SEND_SHUTDOWN)
2078 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2081 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2082 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2086 if (signal_pending(current))
2088 timeo = sock_wait_for_wmem(sk, timeo);
2090 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2091 errcode, sk->sk_allocation);
2093 skb_set_owner_w(skb, sk);
2097 err = sock_intr_errno(timeo);
2102 EXPORT_SYMBOL(sock_alloc_send_pskb);
2104 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2105 int noblock, int *errcode)
2107 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2109 EXPORT_SYMBOL(sock_alloc_send_skb);
2111 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2112 struct sockcm_cookie *sockc)
2116 switch (cmsg->cmsg_type) {
2118 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2120 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2122 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2124 case SO_TIMESTAMPING:
2125 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2128 tsflags = *(u32 *)CMSG_DATA(cmsg);
2129 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2132 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2133 sockc->tsflags |= tsflags;
2136 if (!sock_flag(sk, SOCK_TXTIME))
2138 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2140 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2142 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2144 case SCM_CREDENTIALS:
2151 EXPORT_SYMBOL(__sock_cmsg_send);
2153 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2154 struct sockcm_cookie *sockc)
2156 struct cmsghdr *cmsg;
2159 for_each_cmsghdr(cmsg, msg) {
2160 if (!CMSG_OK(msg, cmsg))
2162 if (cmsg->cmsg_level != SOL_SOCKET)
2164 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2170 EXPORT_SYMBOL(sock_cmsg_send);
2172 static void sk_enter_memory_pressure(struct sock *sk)
2174 if (!sk->sk_prot->enter_memory_pressure)
2177 sk->sk_prot->enter_memory_pressure(sk);
2180 static void sk_leave_memory_pressure(struct sock *sk)
2182 if (sk->sk_prot->leave_memory_pressure) {
2183 sk->sk_prot->leave_memory_pressure(sk);
2185 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2187 if (memory_pressure && READ_ONCE(*memory_pressure))
2188 WRITE_ONCE(*memory_pressure, 0);
2192 /* On 32bit arches, an skb frag is limited to 2^15 */
2193 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2196 * skb_page_frag_refill - check that a page_frag contains enough room
2197 * @sz: minimum size of the fragment we want to get
2198 * @pfrag: pointer to page_frag
2199 * @gfp: priority for memory allocation
2201 * Note: While this allocator tries to use high order pages, there is
2202 * no guarantee that allocations succeed. Therefore, @sz MUST be
2203 * less or equal than PAGE_SIZE.
2205 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2208 if (page_ref_count(pfrag->page) == 1) {
2212 if (pfrag->offset + sz <= pfrag->size)
2214 put_page(pfrag->page);
2218 if (SKB_FRAG_PAGE_ORDER) {
2219 /* Avoid direct reclaim but allow kswapd to wake */
2220 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2221 __GFP_COMP | __GFP_NOWARN |
2223 SKB_FRAG_PAGE_ORDER);
2224 if (likely(pfrag->page)) {
2225 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2229 pfrag->page = alloc_page(gfp);
2230 if (likely(pfrag->page)) {
2231 pfrag->size = PAGE_SIZE;
2236 EXPORT_SYMBOL(skb_page_frag_refill);
2238 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2240 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2243 sk_enter_memory_pressure(sk);
2244 sk_stream_moderate_sndbuf(sk);
2247 EXPORT_SYMBOL(sk_page_frag_refill);
2249 int sk_alloc_sg(struct sock *sk, int len, struct scatterlist *sg,
2250 int sg_start, int *sg_curr_index, unsigned int *sg_curr_size,
2253 int sg_curr = *sg_curr_index, use = 0, rc = 0;
2254 unsigned int size = *sg_curr_size;
2255 struct page_frag *pfrag;
2256 struct scatterlist *sge;
2259 pfrag = sk_page_frag(sk);
2262 unsigned int orig_offset;
2264 if (!sk_page_frag_refill(sk, pfrag)) {
2269 use = min_t(int, len, pfrag->size - pfrag->offset);
2271 if (!sk_wmem_schedule(sk, use)) {
2276 sk_mem_charge(sk, use);
2278 orig_offset = pfrag->offset;
2279 pfrag->offset += use;
2281 sge = sg + sg_curr - 1;
2282 if (sg_curr > first_coalesce && sg_page(sge) == pfrag->page &&
2283 sge->offset + sge->length == orig_offset) {
2288 sg_set_page(sge, pfrag->page, use, orig_offset);
2289 get_page(pfrag->page);
2292 if (sg_curr == MAX_SKB_FRAGS)
2295 if (sg_curr == sg_start) {
2304 *sg_curr_size = size;
2305 *sg_curr_index = sg_curr;
2308 EXPORT_SYMBOL(sk_alloc_sg);
2310 static void __lock_sock(struct sock *sk)
2311 __releases(&sk->sk_lock.slock)
2312 __acquires(&sk->sk_lock.slock)
2317 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2318 TASK_UNINTERRUPTIBLE);
2319 spin_unlock_bh(&sk->sk_lock.slock);
2321 spin_lock_bh(&sk->sk_lock.slock);
2322 if (!sock_owned_by_user(sk))
2325 finish_wait(&sk->sk_lock.wq, &wait);
2328 void __release_sock(struct sock *sk)
2329 __releases(&sk->sk_lock.slock)
2330 __acquires(&sk->sk_lock.slock)
2332 struct sk_buff *skb, *next;
2334 while ((skb = sk->sk_backlog.head) != NULL) {
2335 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2337 spin_unlock_bh(&sk->sk_lock.slock);
2342 WARN_ON_ONCE(skb_dst_is_noref(skb));
2344 sk_backlog_rcv(sk, skb);
2349 } while (skb != NULL);
2351 spin_lock_bh(&sk->sk_lock.slock);
2355 * Doing the zeroing here guarantee we can not loop forever
2356 * while a wild producer attempts to flood us.
2358 sk->sk_backlog.len = 0;
2361 void __sk_flush_backlog(struct sock *sk)
2363 spin_lock_bh(&sk->sk_lock.slock);
2365 spin_unlock_bh(&sk->sk_lock.slock);
2369 * sk_wait_data - wait for data to arrive at sk_receive_queue
2370 * @sk: sock to wait on
2371 * @timeo: for how long
2372 * @skb: last skb seen on sk_receive_queue
2374 * Now socket state including sk->sk_err is changed only under lock,
2375 * hence we may omit checks after joining wait queue.
2376 * We check receive queue before schedule() only as optimization;
2377 * it is very likely that release_sock() added new data.
2379 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2381 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2384 add_wait_queue(sk_sleep(sk), &wait);
2385 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2386 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2387 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2388 remove_wait_queue(sk_sleep(sk), &wait);
2391 EXPORT_SYMBOL(sk_wait_data);
2394 * __sk_mem_raise_allocated - increase memory_allocated
2396 * @size: memory size to allocate
2397 * @amt: pages to allocate
2398 * @kind: allocation type
2400 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2402 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2404 struct proto *prot = sk->sk_prot;
2405 long allocated = sk_memory_allocated_add(sk, amt);
2406 bool charged = true;
2408 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2409 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2410 goto suppress_allocation;
2413 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2414 sk_leave_memory_pressure(sk);
2418 /* Under pressure. */
2419 if (allocated > sk_prot_mem_limits(sk, 1))
2420 sk_enter_memory_pressure(sk);
2422 /* Over hard limit. */
2423 if (allocated > sk_prot_mem_limits(sk, 2))
2424 goto suppress_allocation;
2426 /* guarantee minimum buffer size under pressure */
2427 if (kind == SK_MEM_RECV) {
2428 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2431 } else { /* SK_MEM_SEND */
2432 int wmem0 = sk_get_wmem0(sk, prot);
2434 if (sk->sk_type == SOCK_STREAM) {
2435 if (sk->sk_wmem_queued < wmem0)
2437 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2442 if (sk_has_memory_pressure(sk)) {
2445 if (!sk_under_memory_pressure(sk))
2447 alloc = sk_sockets_allocated_read_positive(sk);
2448 if (sk_prot_mem_limits(sk, 2) > alloc *
2449 sk_mem_pages(sk->sk_wmem_queued +
2450 atomic_read(&sk->sk_rmem_alloc) +
2451 sk->sk_forward_alloc))
2455 suppress_allocation:
2457 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2458 sk_stream_moderate_sndbuf(sk);
2460 /* Fail only if socket is _under_ its sndbuf.
2461 * In this case we cannot block, so that we have to fail.
2463 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2467 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2468 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2470 sk_memory_allocated_sub(sk, amt);
2472 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2473 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2477 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2480 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2482 * @size: memory size to allocate
2483 * @kind: allocation type
2485 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2486 * rmem allocation. This function assumes that protocols which have
2487 * memory_pressure use sk_wmem_queued as write buffer accounting.
2489 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2491 int ret, amt = sk_mem_pages(size);
2493 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2494 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2496 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2499 EXPORT_SYMBOL(__sk_mem_schedule);
2502 * __sk_mem_reduce_allocated - reclaim memory_allocated
2504 * @amount: number of quanta
2506 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2508 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2510 sk_memory_allocated_sub(sk, amount);
2512 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2513 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2515 if (sk_under_memory_pressure(sk) &&
2516 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2517 sk_leave_memory_pressure(sk);
2519 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2522 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2524 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2526 void __sk_mem_reclaim(struct sock *sk, int amount)
2528 amount >>= SK_MEM_QUANTUM_SHIFT;
2529 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2530 __sk_mem_reduce_allocated(sk, amount);
2532 EXPORT_SYMBOL(__sk_mem_reclaim);
2534 int sk_set_peek_off(struct sock *sk, int val)
2536 sk->sk_peek_off = val;
2539 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2542 * Set of default routines for initialising struct proto_ops when
2543 * the protocol does not support a particular function. In certain
2544 * cases where it makes no sense for a protocol to have a "do nothing"
2545 * function, some default processing is provided.
2548 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2552 EXPORT_SYMBOL(sock_no_bind);
2554 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2559 EXPORT_SYMBOL(sock_no_connect);
2561 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2565 EXPORT_SYMBOL(sock_no_socketpair);
2567 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2572 EXPORT_SYMBOL(sock_no_accept);
2574 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2579 EXPORT_SYMBOL(sock_no_getname);
2581 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2585 EXPORT_SYMBOL(sock_no_ioctl);
2587 int sock_no_listen(struct socket *sock, int backlog)
2591 EXPORT_SYMBOL(sock_no_listen);
2593 int sock_no_shutdown(struct socket *sock, int how)
2597 EXPORT_SYMBOL(sock_no_shutdown);
2599 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2600 char __user *optval, unsigned int optlen)
2604 EXPORT_SYMBOL(sock_no_setsockopt);
2606 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2607 char __user *optval, int __user *optlen)
2611 EXPORT_SYMBOL(sock_no_getsockopt);
2613 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2617 EXPORT_SYMBOL(sock_no_sendmsg);
2619 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2623 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2625 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2630 EXPORT_SYMBOL(sock_no_recvmsg);
2632 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2634 /* Mirror missing mmap method error code */
2637 EXPORT_SYMBOL(sock_no_mmap);
2640 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2641 * various sock-based usage counts.
2643 void __receive_sock(struct file *file)
2645 struct socket *sock;
2649 * The resulting value of "error" is ignored here since we only
2650 * need to take action when the file is a socket and testing
2651 * "sock" for NULL is sufficient.
2653 sock = sock_from_file(file, &error);
2655 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2656 sock_update_classid(&sock->sk->sk_cgrp_data);
2660 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2663 struct msghdr msg = {.msg_flags = flags};
2665 char *kaddr = kmap(page);
2666 iov.iov_base = kaddr + offset;
2668 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2672 EXPORT_SYMBOL(sock_no_sendpage);
2674 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2675 int offset, size_t size, int flags)
2678 struct msghdr msg = {.msg_flags = flags};
2680 char *kaddr = kmap(page);
2682 iov.iov_base = kaddr + offset;
2684 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2688 EXPORT_SYMBOL(sock_no_sendpage_locked);
2691 * Default Socket Callbacks
2694 static void sock_def_wakeup(struct sock *sk)
2696 struct socket_wq *wq;
2699 wq = rcu_dereference(sk->sk_wq);
2700 if (skwq_has_sleeper(wq))
2701 wake_up_interruptible_all(&wq->wait);
2705 static void sock_def_error_report(struct sock *sk)
2707 struct socket_wq *wq;
2710 wq = rcu_dereference(sk->sk_wq);
2711 if (skwq_has_sleeper(wq))
2712 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2713 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2717 static void sock_def_readable(struct sock *sk)
2719 struct socket_wq *wq;
2722 wq = rcu_dereference(sk->sk_wq);
2723 if (skwq_has_sleeper(wq))
2724 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2725 EPOLLRDNORM | EPOLLRDBAND);
2726 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2730 static void sock_def_write_space(struct sock *sk)
2732 struct socket_wq *wq;
2736 /* Do not wake up a writer until he can make "significant"
2739 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2740 wq = rcu_dereference(sk->sk_wq);
2741 if (skwq_has_sleeper(wq))
2742 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2743 EPOLLWRNORM | EPOLLWRBAND);
2745 /* Should agree with poll, otherwise some programs break */
2746 if (sock_writeable(sk))
2747 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2753 static void sock_def_destruct(struct sock *sk)
2757 void sk_send_sigurg(struct sock *sk)
2759 if (sk->sk_socket && sk->sk_socket->file)
2760 if (send_sigurg(&sk->sk_socket->file->f_owner))
2761 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2763 EXPORT_SYMBOL(sk_send_sigurg);
2765 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2766 unsigned long expires)
2768 if (!mod_timer(timer, expires))
2771 EXPORT_SYMBOL(sk_reset_timer);
2773 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2775 if (del_timer(timer))
2778 EXPORT_SYMBOL(sk_stop_timer);
2780 void sock_init_data(struct socket *sock, struct sock *sk)
2783 sk->sk_send_head = NULL;
2785 timer_setup(&sk->sk_timer, NULL, 0);
2787 sk->sk_allocation = GFP_KERNEL;
2788 sk->sk_rcvbuf = sysctl_rmem_default;
2789 sk->sk_sndbuf = sysctl_wmem_default;
2790 sk->sk_state = TCP_CLOSE;
2791 sk_set_socket(sk, sock);
2793 sock_set_flag(sk, SOCK_ZAPPED);
2796 sk->sk_type = sock->type;
2797 sk->sk_wq = sock->wq;
2799 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2802 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2805 rwlock_init(&sk->sk_callback_lock);
2806 if (sk->sk_kern_sock)
2807 lockdep_set_class_and_name(
2808 &sk->sk_callback_lock,
2809 af_kern_callback_keys + sk->sk_family,
2810 af_family_kern_clock_key_strings[sk->sk_family]);
2812 lockdep_set_class_and_name(
2813 &sk->sk_callback_lock,
2814 af_callback_keys + sk->sk_family,
2815 af_family_clock_key_strings[sk->sk_family]);
2817 sk->sk_state_change = sock_def_wakeup;
2818 sk->sk_data_ready = sock_def_readable;
2819 sk->sk_write_space = sock_def_write_space;
2820 sk->sk_error_report = sock_def_error_report;
2821 sk->sk_destruct = sock_def_destruct;
2823 sk->sk_frag.page = NULL;
2824 sk->sk_frag.offset = 0;
2825 sk->sk_peek_off = -1;
2827 sk->sk_peer_pid = NULL;
2828 sk->sk_peer_cred = NULL;
2829 sk->sk_write_pending = 0;
2830 sk->sk_rcvlowat = 1;
2831 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2832 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2834 sk->sk_stamp = SK_DEFAULT_STAMP;
2835 #if BITS_PER_LONG==32
2836 seqlock_init(&sk->sk_stamp_seq);
2838 atomic_set(&sk->sk_zckey, 0);
2840 #ifdef CONFIG_NET_RX_BUSY_POLL
2842 sk->sk_ll_usec = sysctl_net_busy_read;
2845 sk->sk_max_pacing_rate = ~0U;
2846 sk->sk_pacing_rate = ~0U;
2847 sk->sk_pacing_shift = 10;
2848 sk->sk_incoming_cpu = -1;
2850 sk_rx_queue_clear(sk);
2852 * Before updating sk_refcnt, we must commit prior changes to memory
2853 * (Documentation/RCU/rculist_nulls.txt for details)
2856 refcount_set(&sk->sk_refcnt, 1);
2857 atomic_set(&sk->sk_drops, 0);
2859 EXPORT_SYMBOL(sock_init_data);
2861 void lock_sock_nested(struct sock *sk, int subclass)
2864 spin_lock_bh(&sk->sk_lock.slock);
2865 if (sk->sk_lock.owned)
2867 sk->sk_lock.owned = 1;
2868 spin_unlock(&sk->sk_lock.slock);
2870 * The sk_lock has mutex_lock() semantics here:
2872 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2875 EXPORT_SYMBOL(lock_sock_nested);
2877 void release_sock(struct sock *sk)
2879 spin_lock_bh(&sk->sk_lock.slock);
2880 if (sk->sk_backlog.tail)
2883 /* Warning : release_cb() might need to release sk ownership,
2884 * ie call sock_release_ownership(sk) before us.
2886 if (sk->sk_prot->release_cb)
2887 sk->sk_prot->release_cb(sk);
2889 sock_release_ownership(sk);
2890 if (waitqueue_active(&sk->sk_lock.wq))
2891 wake_up(&sk->sk_lock.wq);
2892 spin_unlock_bh(&sk->sk_lock.slock);
2894 EXPORT_SYMBOL(release_sock);
2897 * lock_sock_fast - fast version of lock_sock
2900 * This version should be used for very small section, where process wont block
2901 * return false if fast path is taken:
2903 * sk_lock.slock locked, owned = 0, BH disabled
2905 * return true if slow path is taken:
2907 * sk_lock.slock unlocked, owned = 1, BH enabled
2909 bool lock_sock_fast(struct sock *sk)
2912 spin_lock_bh(&sk->sk_lock.slock);
2914 if (!sk->sk_lock.owned)
2916 * Note : We must disable BH
2921 sk->sk_lock.owned = 1;
2922 spin_unlock(&sk->sk_lock.slock);
2924 * The sk_lock has mutex_lock() semantics here:
2926 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2930 EXPORT_SYMBOL(lock_sock_fast);
2932 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2936 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2937 tv = ktime_to_timeval(sock_read_timestamp(sk));
2938 if (tv.tv_sec == -1)
2940 if (tv.tv_sec == 0) {
2941 ktime_t kt = ktime_get_real();
2942 sock_write_timestamp(sk, kt);
2943 tv = ktime_to_timeval(kt);
2945 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2947 EXPORT_SYMBOL(sock_get_timestamp);
2949 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2953 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2954 ts = ktime_to_timespec(sock_read_timestamp(sk));
2955 if (ts.tv_sec == -1)
2957 if (ts.tv_sec == 0) {
2958 ktime_t kt = ktime_get_real();
2959 sock_write_timestamp(sk, kt);
2960 ts = ktime_to_timespec(sk->sk_stamp);
2962 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2964 EXPORT_SYMBOL(sock_get_timestampns);
2966 void sock_enable_timestamp(struct sock *sk, int flag)
2968 if (!sock_flag(sk, flag)) {
2969 unsigned long previous_flags = sk->sk_flags;
2971 sock_set_flag(sk, flag);
2973 * we just set one of the two flags which require net
2974 * time stamping, but time stamping might have been on
2975 * already because of the other one
2977 if (sock_needs_netstamp(sk) &&
2978 !(previous_flags & SK_FLAGS_TIMESTAMP))
2979 net_enable_timestamp();
2983 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2984 int level, int type)
2986 struct sock_exterr_skb *serr;
2987 struct sk_buff *skb;
2991 skb = sock_dequeue_err_skb(sk);
2997 msg->msg_flags |= MSG_TRUNC;
3000 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3004 sock_recv_timestamp(msg, sk, skb);
3006 serr = SKB_EXT_ERR(skb);
3007 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3009 msg->msg_flags |= MSG_ERRQUEUE;
3017 EXPORT_SYMBOL(sock_recv_errqueue);
3020 * Get a socket option on an socket.
3022 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3023 * asynchronous errors should be reported by getsockopt. We assume
3024 * this means if you specify SO_ERROR (otherwise whats the point of it).
3026 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3027 char __user *optval, int __user *optlen)
3029 struct sock *sk = sock->sk;
3031 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3033 EXPORT_SYMBOL(sock_common_getsockopt);
3035 #ifdef CONFIG_COMPAT
3036 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3037 char __user *optval, int __user *optlen)
3039 struct sock *sk = sock->sk;
3041 if (sk->sk_prot->compat_getsockopt != NULL)
3042 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3044 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3046 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3049 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3052 struct sock *sk = sock->sk;
3056 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3057 flags & ~MSG_DONTWAIT, &addr_len);
3059 msg->msg_namelen = addr_len;
3062 EXPORT_SYMBOL(sock_common_recvmsg);
3065 * Set socket options on an inet socket.
3067 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3068 char __user *optval, unsigned int optlen)
3070 struct sock *sk = sock->sk;
3072 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3074 EXPORT_SYMBOL(sock_common_setsockopt);
3076 #ifdef CONFIG_COMPAT
3077 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3078 char __user *optval, unsigned int optlen)
3080 struct sock *sk = sock->sk;
3082 if (sk->sk_prot->compat_setsockopt != NULL)
3083 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3085 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3087 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3090 void sk_common_release(struct sock *sk)
3092 if (sk->sk_prot->destroy)
3093 sk->sk_prot->destroy(sk);
3096 * Observation: when sock_common_release is called, processes have
3097 * no access to socket. But net still has.
3098 * Step one, detach it from networking:
3100 * A. Remove from hash tables.
3103 sk->sk_prot->unhash(sk);
3106 * In this point socket cannot receive new packets, but it is possible
3107 * that some packets are in flight because some CPU runs receiver and
3108 * did hash table lookup before we unhashed socket. They will achieve
3109 * receive queue and will be purged by socket destructor.
3111 * Also we still have packets pending on receive queue and probably,
3112 * our own packets waiting in device queues. sock_destroy will drain
3113 * receive queue, but transmitted packets will delay socket destruction
3114 * until the last reference will be released.
3119 xfrm_sk_free_policy(sk);
3121 sk_refcnt_debug_release(sk);
3125 EXPORT_SYMBOL(sk_common_release);
3127 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3129 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3131 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3132 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3133 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3134 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3135 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3136 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3137 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3138 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3139 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3142 #ifdef CONFIG_PROC_FS
3143 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3145 int val[PROTO_INUSE_NR];
3148 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3150 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3152 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3154 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3156 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3158 int cpu, idx = prot->inuse_idx;
3161 for_each_possible_cpu(cpu)
3162 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3164 return res >= 0 ? res : 0;
3166 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3168 static void sock_inuse_add(struct net *net, int val)
3170 this_cpu_add(*net->core.sock_inuse, val);
3173 int sock_inuse_get(struct net *net)
3177 for_each_possible_cpu(cpu)
3178 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3183 EXPORT_SYMBOL_GPL(sock_inuse_get);
3185 static int __net_init sock_inuse_init_net(struct net *net)
3187 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3188 if (net->core.prot_inuse == NULL)
3191 net->core.sock_inuse = alloc_percpu(int);
3192 if (net->core.sock_inuse == NULL)
3198 free_percpu(net->core.prot_inuse);
3202 static void __net_exit sock_inuse_exit_net(struct net *net)
3204 free_percpu(net->core.prot_inuse);
3205 free_percpu(net->core.sock_inuse);
3208 static struct pernet_operations net_inuse_ops = {
3209 .init = sock_inuse_init_net,
3210 .exit = sock_inuse_exit_net,
3213 static __init int net_inuse_init(void)
3215 if (register_pernet_subsys(&net_inuse_ops))
3216 panic("Cannot initialize net inuse counters");
3221 core_initcall(net_inuse_init);
3223 static void assign_proto_idx(struct proto *prot)
3225 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3227 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3228 pr_err("PROTO_INUSE_NR exhausted\n");
3232 set_bit(prot->inuse_idx, proto_inuse_idx);
3235 static void release_proto_idx(struct proto *prot)
3237 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3238 clear_bit(prot->inuse_idx, proto_inuse_idx);
3241 static inline void assign_proto_idx(struct proto *prot)
3245 static inline void release_proto_idx(struct proto *prot)
3249 static void sock_inuse_add(struct net *net, int val)
3254 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3258 kfree(rsk_prot->slab_name);
3259 rsk_prot->slab_name = NULL;
3260 kmem_cache_destroy(rsk_prot->slab);
3261 rsk_prot->slab = NULL;
3264 static int req_prot_init(const struct proto *prot)
3266 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3271 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3273 if (!rsk_prot->slab_name)
3276 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3277 rsk_prot->obj_size, 0,
3278 SLAB_ACCOUNT | prot->slab_flags,
3281 if (!rsk_prot->slab) {
3282 pr_crit("%s: Can't create request sock SLAB cache!\n",
3289 int proto_register(struct proto *prot, int alloc_slab)
3292 prot->slab = kmem_cache_create_usercopy(prot->name,
3294 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3296 prot->useroffset, prot->usersize,
3299 if (prot->slab == NULL) {
3300 pr_crit("%s: Can't create sock SLAB cache!\n",
3305 if (req_prot_init(prot))
3306 goto out_free_request_sock_slab;
3308 if (prot->twsk_prot != NULL) {
3309 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3311 if (prot->twsk_prot->twsk_slab_name == NULL)
3312 goto out_free_request_sock_slab;
3314 prot->twsk_prot->twsk_slab =
3315 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3316 prot->twsk_prot->twsk_obj_size,
3321 if (prot->twsk_prot->twsk_slab == NULL)
3322 goto out_free_timewait_sock_slab_name;
3326 mutex_lock(&proto_list_mutex);
3327 list_add(&prot->node, &proto_list);
3328 assign_proto_idx(prot);
3329 mutex_unlock(&proto_list_mutex);
3332 out_free_timewait_sock_slab_name:
3333 kfree(prot->twsk_prot->twsk_slab_name);
3334 out_free_request_sock_slab:
3335 req_prot_cleanup(prot->rsk_prot);
3337 kmem_cache_destroy(prot->slab);
3342 EXPORT_SYMBOL(proto_register);
3344 void proto_unregister(struct proto *prot)
3346 mutex_lock(&proto_list_mutex);
3347 release_proto_idx(prot);
3348 list_del(&prot->node);
3349 mutex_unlock(&proto_list_mutex);
3351 kmem_cache_destroy(prot->slab);
3354 req_prot_cleanup(prot->rsk_prot);
3356 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3357 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3358 kfree(prot->twsk_prot->twsk_slab_name);
3359 prot->twsk_prot->twsk_slab = NULL;
3362 EXPORT_SYMBOL(proto_unregister);
3364 int sock_load_diag_module(int family, int protocol)
3367 if (!sock_is_registered(family))
3370 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3371 NETLINK_SOCK_DIAG, family);
3375 if (family == AF_INET &&
3376 protocol != IPPROTO_RAW &&
3377 !rcu_access_pointer(inet_protos[protocol]))
3381 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3382 NETLINK_SOCK_DIAG, family, protocol);
3384 EXPORT_SYMBOL(sock_load_diag_module);
3386 #ifdef CONFIG_PROC_FS
3387 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3388 __acquires(proto_list_mutex)
3390 mutex_lock(&proto_list_mutex);
3391 return seq_list_start_head(&proto_list, *pos);
3394 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3396 return seq_list_next(v, &proto_list, pos);
3399 static void proto_seq_stop(struct seq_file *seq, void *v)
3400 __releases(proto_list_mutex)
3402 mutex_unlock(&proto_list_mutex);
3405 static char proto_method_implemented(const void *method)
3407 return method == NULL ? 'n' : 'y';
3409 static long sock_prot_memory_allocated(struct proto *proto)
3411 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3414 static char *sock_prot_memory_pressure(struct proto *proto)
3416 return proto->memory_pressure != NULL ?
3417 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3420 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3423 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3424 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3427 sock_prot_inuse_get(seq_file_net(seq), proto),
3428 sock_prot_memory_allocated(proto),
3429 sock_prot_memory_pressure(proto),
3431 proto->slab == NULL ? "no" : "yes",
3432 module_name(proto->owner),
3433 proto_method_implemented(proto->close),
3434 proto_method_implemented(proto->connect),
3435 proto_method_implemented(proto->disconnect),
3436 proto_method_implemented(proto->accept),
3437 proto_method_implemented(proto->ioctl),
3438 proto_method_implemented(proto->init),
3439 proto_method_implemented(proto->destroy),
3440 proto_method_implemented(proto->shutdown),
3441 proto_method_implemented(proto->setsockopt),
3442 proto_method_implemented(proto->getsockopt),
3443 proto_method_implemented(proto->sendmsg),
3444 proto_method_implemented(proto->recvmsg),
3445 proto_method_implemented(proto->sendpage),
3446 proto_method_implemented(proto->bind),
3447 proto_method_implemented(proto->backlog_rcv),
3448 proto_method_implemented(proto->hash),
3449 proto_method_implemented(proto->unhash),
3450 proto_method_implemented(proto->get_port),
3451 proto_method_implemented(proto->enter_memory_pressure));
3454 static int proto_seq_show(struct seq_file *seq, void *v)
3456 if (v == &proto_list)
3457 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3466 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3468 proto_seq_printf(seq, list_entry(v, struct proto, node));
3472 static const struct seq_operations proto_seq_ops = {
3473 .start = proto_seq_start,
3474 .next = proto_seq_next,
3475 .stop = proto_seq_stop,
3476 .show = proto_seq_show,
3479 static __net_init int proto_init_net(struct net *net)
3481 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3482 sizeof(struct seq_net_private)))
3488 static __net_exit void proto_exit_net(struct net *net)
3490 remove_proc_entry("protocols", net->proc_net);
3494 static __net_initdata struct pernet_operations proto_net_ops = {
3495 .init = proto_init_net,
3496 .exit = proto_exit_net,
3499 static int __init proto_init(void)
3501 return register_pernet_subsys(&proto_net_ops);
3504 subsys_initcall(proto_init);
3506 #endif /* PROC_FS */
3508 #ifdef CONFIG_NET_RX_BUSY_POLL
3509 bool sk_busy_loop_end(void *p, unsigned long start_time)
3511 struct sock *sk = p;
3513 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3514 sk_busy_loop_timeout(sk, start_time);
3516 EXPORT_SYMBOL(sk_busy_loop_end);
3517 #endif /* CONFIG_NET_RX_BUSY_POLL */