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.
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 <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <linux/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
139 #include <trace/events/sock.h>
145 #include <net/busy_poll.h>
147 static DEFINE_MUTEX(proto_list_mutex);
148 static LIST_HEAD(proto_list);
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
160 bool sk_ns_capable(const struct sock *sk,
161 struct user_namespace *user_ns, int cap)
163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
164 ns_capable(user_ns, cap);
166 EXPORT_SYMBOL(sk_ns_capable);
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
177 bool sk_capable(const struct sock *sk, int cap)
179 return sk_ns_capable(sk, &init_user_ns, cap);
181 EXPORT_SYMBOL(sk_capable);
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
192 bool sk_net_capable(const struct sock *sk, int cap)
194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
196 EXPORT_SYMBOL(sk_net_capable);
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family and separate keys for internal and
203 static struct lock_class_key af_family_keys[AF_MAX];
204 static struct lock_class_key af_family_kern_keys[AF_MAX];
205 static struct lock_class_key af_family_slock_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
209 * Make lock validator output more readable. (we pre-construct these
210 * strings build-time, so that runtime initialization of socket
214 #define _sock_locks(x) \
215 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
216 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
217 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
218 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
219 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
220 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
221 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
222 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
223 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
224 x "27" , x "28" , x "AF_CAN" , \
225 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
226 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
227 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
228 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
229 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX"
231 static const char *const af_family_key_strings[AF_MAX+1] = {
232 _sock_locks("sk_lock-")
234 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
235 _sock_locks("slock-")
237 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
238 _sock_locks("clock-")
241 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
242 _sock_locks("k-sk_lock-")
244 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
245 _sock_locks("k-slock-")
247 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
248 _sock_locks("k-clock-")
252 * sk_callback_lock locking rules are per-address-family,
253 * so split the lock classes by using a per-AF key:
255 static struct lock_class_key af_callback_keys[AF_MAX];
256 static struct lock_class_key af_kern_callback_keys[AF_MAX];
258 /* Take into consideration the size of the struct sk_buff overhead in the
259 * determination of these values, since that is non-constant across
260 * platforms. This makes socket queueing behavior and performance
261 * not depend upon such differences.
263 #define _SK_MEM_PACKETS 256
264 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
265 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
266 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
268 /* Run time adjustable parameters. */
269 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
270 EXPORT_SYMBOL(sysctl_wmem_max);
271 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
272 EXPORT_SYMBOL(sysctl_rmem_max);
273 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
274 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
276 /* Maximal space eaten by iovec or ancillary data plus some space */
277 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
278 EXPORT_SYMBOL(sysctl_optmem_max);
280 int sysctl_tstamp_allow_data __read_mostly = 1;
282 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
283 EXPORT_SYMBOL_GPL(memalloc_socks);
286 * sk_set_memalloc - sets %SOCK_MEMALLOC
287 * @sk: socket to set it on
289 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
290 * It's the responsibility of the admin to adjust min_free_kbytes
291 * to meet the requirements
293 void sk_set_memalloc(struct sock *sk)
295 sock_set_flag(sk, SOCK_MEMALLOC);
296 sk->sk_allocation |= __GFP_MEMALLOC;
297 static_key_slow_inc(&memalloc_socks);
299 EXPORT_SYMBOL_GPL(sk_set_memalloc);
301 void sk_clear_memalloc(struct sock *sk)
303 sock_reset_flag(sk, SOCK_MEMALLOC);
304 sk->sk_allocation &= ~__GFP_MEMALLOC;
305 static_key_slow_dec(&memalloc_socks);
308 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
309 * progress of swapping. SOCK_MEMALLOC may be cleared while
310 * it has rmem allocations due to the last swapfile being deactivated
311 * but there is a risk that the socket is unusable due to exceeding
312 * the rmem limits. Reclaim the reserves and obey rmem limits again.
316 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
318 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
321 unsigned long pflags = current->flags;
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
326 current->flags |= PF_MEMALLOC;
327 ret = sk->sk_backlog_rcv(sk, skb);
328 tsk_restore_flags(current, pflags, PF_MEMALLOC);
332 EXPORT_SYMBOL(__sk_backlog_rcv);
334 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
338 if (optlen < sizeof(tv))
340 if (copy_from_user(&tv, optval, sizeof(tv)))
342 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
346 static int warned __read_mostly;
349 if (warned < 10 && net_ratelimit()) {
351 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
352 __func__, current->comm, task_pid_nr(current));
356 *timeo_p = MAX_SCHEDULE_TIMEOUT;
357 if (tv.tv_sec == 0 && tv.tv_usec == 0)
359 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
360 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
364 static void sock_warn_obsolete_bsdism(const char *name)
367 static char warncomm[TASK_COMM_LEN];
368 if (strcmp(warncomm, current->comm) && warned < 5) {
369 strcpy(warncomm, current->comm);
370 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
376 static bool sock_needs_netstamp(const struct sock *sk)
378 switch (sk->sk_family) {
387 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
389 if (sk->sk_flags & flags) {
390 sk->sk_flags &= ~flags;
391 if (sock_needs_netstamp(sk) &&
392 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
393 net_disable_timestamp();
398 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
401 struct sk_buff_head *list = &sk->sk_receive_queue;
403 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
404 atomic_inc(&sk->sk_drops);
405 trace_sock_rcvqueue_full(sk, skb);
409 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
410 atomic_inc(&sk->sk_drops);
415 skb_set_owner_r(skb, sk);
417 /* we escape from rcu protected region, make sure we dont leak
422 spin_lock_irqsave(&list->lock, flags);
423 sock_skb_set_dropcount(sk, skb);
424 __skb_queue_tail(list, skb);
425 spin_unlock_irqrestore(&list->lock, flags);
427 if (!sock_flag(sk, SOCK_DEAD))
428 sk->sk_data_ready(sk);
431 EXPORT_SYMBOL(__sock_queue_rcv_skb);
433 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
437 err = sk_filter(sk, skb);
441 return __sock_queue_rcv_skb(sk, skb);
443 EXPORT_SYMBOL(sock_queue_rcv_skb);
445 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
446 const int nested, unsigned int trim_cap, bool refcounted)
448 int rc = NET_RX_SUCCESS;
450 if (sk_filter_trim_cap(sk, skb, trim_cap))
451 goto discard_and_relse;
455 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
456 atomic_inc(&sk->sk_drops);
457 goto discard_and_relse;
460 bh_lock_sock_nested(sk);
463 if (!sock_owned_by_user(sk)) {
465 * trylock + unlock semantics:
467 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
469 rc = sk_backlog_rcv(sk, skb);
471 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
472 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
474 atomic_inc(&sk->sk_drops);
475 goto discard_and_relse;
487 EXPORT_SYMBOL(__sk_receive_skb);
489 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
491 struct dst_entry *dst = __sk_dst_get(sk);
493 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
494 sk_tx_queue_clear(sk);
495 sk->sk_dst_pending_confirm = 0;
496 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
503 EXPORT_SYMBOL(__sk_dst_check);
505 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
507 struct dst_entry *dst = sk_dst_get(sk);
509 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
517 EXPORT_SYMBOL(sk_dst_check);
519 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
522 int ret = -ENOPROTOOPT;
523 #ifdef CONFIG_NETDEVICES
524 struct net *net = sock_net(sk);
525 char devname[IFNAMSIZ];
530 if (!ns_capable(net->user_ns, CAP_NET_RAW))
537 /* Bind this socket to a particular device like "eth0",
538 * as specified in the passed interface name. If the
539 * name is "" or the option length is zero the socket
542 if (optlen > IFNAMSIZ - 1)
543 optlen = IFNAMSIZ - 1;
544 memset(devname, 0, sizeof(devname));
547 if (copy_from_user(devname, optval, optlen))
551 if (devname[0] != '\0') {
552 struct net_device *dev;
555 dev = dev_get_by_name_rcu(net, devname);
557 index = dev->ifindex;
565 sk->sk_bound_dev_if = index;
577 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
578 int __user *optlen, int len)
580 int ret = -ENOPROTOOPT;
581 #ifdef CONFIG_NETDEVICES
582 struct net *net = sock_net(sk);
583 char devname[IFNAMSIZ];
585 if (sk->sk_bound_dev_if == 0) {
594 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
598 len = strlen(devname) + 1;
601 if (copy_to_user(optval, devname, len))
606 if (put_user(len, optlen))
617 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
620 sock_set_flag(sk, bit);
622 sock_reset_flag(sk, bit);
625 bool sk_mc_loop(struct sock *sk)
627 if (dev_recursion_level())
631 switch (sk->sk_family) {
633 return inet_sk(sk)->mc_loop;
634 #if IS_ENABLED(CONFIG_IPV6)
636 return inet6_sk(sk)->mc_loop;
642 EXPORT_SYMBOL(sk_mc_loop);
645 * This is meant for all protocols to use and covers goings on
646 * at the socket level. Everything here is generic.
649 int sock_setsockopt(struct socket *sock, int level, int optname,
650 char __user *optval, unsigned int optlen)
652 struct sock *sk = sock->sk;
659 * Options without arguments
662 if (optname == SO_BINDTODEVICE)
663 return sock_setbindtodevice(sk, optval, optlen);
665 if (optlen < sizeof(int))
668 if (get_user(val, (int __user *)optval))
671 valbool = val ? 1 : 0;
677 if (val && !capable(CAP_NET_ADMIN))
680 sock_valbool_flag(sk, SOCK_DBG, valbool);
683 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
686 sk->sk_reuseport = valbool;
695 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
698 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
701 /* Don't error on this BSD doesn't and if you think
702 * about it this is right. Otherwise apps have to
703 * play 'guess the biggest size' games. RCVBUF/SNDBUF
704 * are treated in BSD as hints
706 val = min_t(u32, val, sysctl_wmem_max);
708 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
709 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
710 /* Wake up sending tasks if we upped the value. */
711 sk->sk_write_space(sk);
715 if (!capable(CAP_NET_ADMIN)) {
722 /* Don't error on this BSD doesn't and if you think
723 * about it this is right. Otherwise apps have to
724 * play 'guess the biggest size' games. RCVBUF/SNDBUF
725 * are treated in BSD as hints
727 val = min_t(u32, val, sysctl_rmem_max);
729 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
731 * We double it on the way in to account for
732 * "struct sk_buff" etc. overhead. Applications
733 * assume that the SO_RCVBUF setting they make will
734 * allow that much actual data to be received on that
737 * Applications are unaware that "struct sk_buff" and
738 * other overheads allocate from the receive buffer
739 * during socket buffer allocation.
741 * And after considering the possible alternatives,
742 * returning the value we actually used in getsockopt
743 * is the most desirable behavior.
745 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
749 if (!capable(CAP_NET_ADMIN)) {
756 if (sk->sk_prot->keepalive)
757 sk->sk_prot->keepalive(sk, valbool);
758 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
762 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
766 sk->sk_no_check_tx = valbool;
770 if ((val >= 0 && val <= 6) ||
771 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
772 sk->sk_priority = val;
778 if (optlen < sizeof(ling)) {
779 ret = -EINVAL; /* 1003.1g */
782 if (copy_from_user(&ling, optval, sizeof(ling))) {
787 sock_reset_flag(sk, SOCK_LINGER);
789 #if (BITS_PER_LONG == 32)
790 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
791 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
794 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
795 sock_set_flag(sk, SOCK_LINGER);
800 sock_warn_obsolete_bsdism("setsockopt");
805 set_bit(SOCK_PASSCRED, &sock->flags);
807 clear_bit(SOCK_PASSCRED, &sock->flags);
813 if (optname == SO_TIMESTAMP)
814 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
816 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
817 sock_set_flag(sk, SOCK_RCVTSTAMP);
818 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
820 sock_reset_flag(sk, SOCK_RCVTSTAMP);
821 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
825 case SO_TIMESTAMPING:
826 if (val & ~SOF_TIMESTAMPING_MASK) {
831 if (val & SOF_TIMESTAMPING_OPT_ID &&
832 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
833 if (sk->sk_protocol == IPPROTO_TCP &&
834 sk->sk_type == SOCK_STREAM) {
835 if ((1 << sk->sk_state) &
836 (TCPF_CLOSE | TCPF_LISTEN)) {
840 sk->sk_tskey = tcp_sk(sk)->snd_una;
846 if (val & SOF_TIMESTAMPING_OPT_STATS &&
847 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
852 sk->sk_tsflags = val;
853 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
854 sock_enable_timestamp(sk,
855 SOCK_TIMESTAMPING_RX_SOFTWARE);
857 sock_disable_timestamp(sk,
858 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
864 sk->sk_rcvlowat = val ? : 1;
868 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
872 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
875 case SO_ATTACH_FILTER:
877 if (optlen == sizeof(struct sock_fprog)) {
878 struct sock_fprog fprog;
881 if (copy_from_user(&fprog, optval, sizeof(fprog)))
884 ret = sk_attach_filter(&fprog, sk);
890 if (optlen == sizeof(u32)) {
894 if (copy_from_user(&ufd, optval, sizeof(ufd)))
897 ret = sk_attach_bpf(ufd, sk);
901 case SO_ATTACH_REUSEPORT_CBPF:
903 if (optlen == sizeof(struct sock_fprog)) {
904 struct sock_fprog fprog;
907 if (copy_from_user(&fprog, optval, sizeof(fprog)))
910 ret = sk_reuseport_attach_filter(&fprog, sk);
914 case SO_ATTACH_REUSEPORT_EBPF:
916 if (optlen == sizeof(u32)) {
920 if (copy_from_user(&ufd, optval, sizeof(ufd)))
923 ret = sk_reuseport_attach_bpf(ufd, sk);
927 case SO_DETACH_FILTER:
928 ret = sk_detach_filter(sk);
932 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
935 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
940 set_bit(SOCK_PASSSEC, &sock->flags);
942 clear_bit(SOCK_PASSSEC, &sock->flags);
945 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
952 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
956 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
960 if (sock->ops->set_peek_off)
961 ret = sock->ops->set_peek_off(sk, val);
967 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
970 case SO_SELECT_ERR_QUEUE:
971 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
974 #ifdef CONFIG_NET_RX_BUSY_POLL
976 /* allow unprivileged users to decrease the value */
977 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
983 sk->sk_ll_usec = val;
988 case SO_MAX_PACING_RATE:
989 sk->sk_max_pacing_rate = val;
990 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
991 sk->sk_max_pacing_rate);
994 case SO_INCOMING_CPU:
995 sk->sk_incoming_cpu = val;
1000 dst_negative_advice(sk);
1009 EXPORT_SYMBOL(sock_setsockopt);
1012 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1013 struct ucred *ucred)
1015 ucred->pid = pid_vnr(pid);
1016 ucred->uid = ucred->gid = -1;
1018 struct user_namespace *current_ns = current_user_ns();
1020 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1021 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1025 int sock_getsockopt(struct socket *sock, int level, int optname,
1026 char __user *optval, int __user *optlen)
1028 struct sock *sk = sock->sk;
1036 int lv = sizeof(int);
1039 if (get_user(len, optlen))
1044 memset(&v, 0, sizeof(v));
1048 v.val = sock_flag(sk, SOCK_DBG);
1052 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1056 v.val = sock_flag(sk, SOCK_BROADCAST);
1060 v.val = sk->sk_sndbuf;
1064 v.val = sk->sk_rcvbuf;
1068 v.val = sk->sk_reuse;
1072 v.val = sk->sk_reuseport;
1076 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1080 v.val = sk->sk_type;
1084 v.val = sk->sk_protocol;
1088 v.val = sk->sk_family;
1092 v.val = -sock_error(sk);
1094 v.val = xchg(&sk->sk_err_soft, 0);
1098 v.val = sock_flag(sk, SOCK_URGINLINE);
1102 v.val = sk->sk_no_check_tx;
1106 v.val = sk->sk_priority;
1110 lv = sizeof(v.ling);
1111 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1112 v.ling.l_linger = sk->sk_lingertime / HZ;
1116 sock_warn_obsolete_bsdism("getsockopt");
1120 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1121 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1124 case SO_TIMESTAMPNS:
1125 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1128 case SO_TIMESTAMPING:
1129 v.val = sk->sk_tsflags;
1133 lv = sizeof(struct timeval);
1134 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1138 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1139 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1144 lv = sizeof(struct timeval);
1145 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1149 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1150 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1155 v.val = sk->sk_rcvlowat;
1163 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1168 struct ucred peercred;
1169 if (len > sizeof(peercred))
1170 len = sizeof(peercred);
1171 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1172 if (copy_to_user(optval, &peercred, len))
1181 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1185 if (copy_to_user(optval, address, len))
1190 /* Dubious BSD thing... Probably nobody even uses it, but
1191 * the UNIX standard wants it for whatever reason... -DaveM
1194 v.val = sk->sk_state == TCP_LISTEN;
1198 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1202 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1205 v.val = sk->sk_mark;
1209 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1212 case SO_WIFI_STATUS:
1213 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1217 if (!sock->ops->set_peek_off)
1220 v.val = sk->sk_peek_off;
1223 v.val = sock_flag(sk, SOCK_NOFCS);
1226 case SO_BINDTODEVICE:
1227 return sock_getbindtodevice(sk, optval, optlen, len);
1230 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1236 case SO_LOCK_FILTER:
1237 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1240 case SO_BPF_EXTENSIONS:
1241 v.val = bpf_tell_extensions();
1244 case SO_SELECT_ERR_QUEUE:
1245 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1248 #ifdef CONFIG_NET_RX_BUSY_POLL
1250 v.val = sk->sk_ll_usec;
1254 case SO_MAX_PACING_RATE:
1255 v.val = sk->sk_max_pacing_rate;
1258 case SO_INCOMING_CPU:
1259 v.val = sk->sk_incoming_cpu;
1263 /* We implement the SO_SNDLOWAT etc to not be settable
1266 return -ENOPROTOOPT;
1271 if (copy_to_user(optval, &v, len))
1274 if (put_user(len, optlen))
1280 * Initialize an sk_lock.
1282 * (We also register the sk_lock with the lock validator.)
1284 static inline void sock_lock_init(struct sock *sk)
1286 if (sk->sk_kern_sock)
1287 sock_lock_init_class_and_name(
1289 af_family_kern_slock_key_strings[sk->sk_family],
1290 af_family_kern_slock_keys + sk->sk_family,
1291 af_family_kern_key_strings[sk->sk_family],
1292 af_family_kern_keys + sk->sk_family);
1294 sock_lock_init_class_and_name(
1296 af_family_slock_key_strings[sk->sk_family],
1297 af_family_slock_keys + sk->sk_family,
1298 af_family_key_strings[sk->sk_family],
1299 af_family_keys + sk->sk_family);
1303 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1304 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1305 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1307 static void sock_copy(struct sock *nsk, const struct sock *osk)
1309 #ifdef CONFIG_SECURITY_NETWORK
1310 void *sptr = nsk->sk_security;
1312 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1314 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1315 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1317 #ifdef CONFIG_SECURITY_NETWORK
1318 nsk->sk_security = sptr;
1319 security_sk_clone(osk, nsk);
1323 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1327 struct kmem_cache *slab;
1331 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1334 if (priority & __GFP_ZERO)
1335 sk_prot_clear_nulls(sk, prot->obj_size);
1337 sk = kmalloc(prot->obj_size, priority);
1340 kmemcheck_annotate_bitfield(sk, flags);
1342 if (security_sk_alloc(sk, family, priority))
1345 if (!try_module_get(prot->owner))
1347 sk_tx_queue_clear(sk);
1353 security_sk_free(sk);
1356 kmem_cache_free(slab, sk);
1362 static void sk_prot_free(struct proto *prot, struct sock *sk)
1364 struct kmem_cache *slab;
1365 struct module *owner;
1367 owner = prot->owner;
1370 cgroup_sk_free(&sk->sk_cgrp_data);
1371 mem_cgroup_sk_free(sk);
1372 security_sk_free(sk);
1374 kmem_cache_free(slab, sk);
1381 * sk_alloc - All socket objects are allocated here
1382 * @net: the applicable net namespace
1383 * @family: protocol family
1384 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1385 * @prot: struct proto associated with this new sock instance
1386 * @kern: is this to be a kernel socket?
1388 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1389 struct proto *prot, int kern)
1393 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1395 sk->sk_family = family;
1397 * See comment in struct sock definition to understand
1398 * why we need sk_prot_creator -acme
1400 sk->sk_prot = sk->sk_prot_creator = prot;
1401 sk->sk_kern_sock = kern;
1403 sk->sk_net_refcnt = kern ? 0 : 1;
1404 if (likely(sk->sk_net_refcnt))
1406 sock_net_set(sk, net);
1407 atomic_set(&sk->sk_wmem_alloc, 1);
1409 mem_cgroup_sk_alloc(sk);
1410 cgroup_sk_alloc(&sk->sk_cgrp_data);
1411 sock_update_classid(&sk->sk_cgrp_data);
1412 sock_update_netprioidx(&sk->sk_cgrp_data);
1417 EXPORT_SYMBOL(sk_alloc);
1419 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1420 * grace period. This is the case for UDP sockets and TCP listeners.
1422 static void __sk_destruct(struct rcu_head *head)
1424 struct sock *sk = container_of(head, struct sock, sk_rcu);
1425 struct sk_filter *filter;
1427 if (sk->sk_destruct)
1428 sk->sk_destruct(sk);
1430 filter = rcu_dereference_check(sk->sk_filter,
1431 atomic_read(&sk->sk_wmem_alloc) == 0);
1433 sk_filter_uncharge(sk, filter);
1434 RCU_INIT_POINTER(sk->sk_filter, NULL);
1436 if (rcu_access_pointer(sk->sk_reuseport_cb))
1437 reuseport_detach_sock(sk);
1439 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1441 if (atomic_read(&sk->sk_omem_alloc))
1442 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1443 __func__, atomic_read(&sk->sk_omem_alloc));
1445 if (sk->sk_frag.page) {
1446 put_page(sk->sk_frag.page);
1447 sk->sk_frag.page = NULL;
1450 if (sk->sk_peer_cred)
1451 put_cred(sk->sk_peer_cred);
1452 put_pid(sk->sk_peer_pid);
1453 if (likely(sk->sk_net_refcnt))
1454 put_net(sock_net(sk));
1455 sk_prot_free(sk->sk_prot_creator, sk);
1458 void sk_destruct(struct sock *sk)
1460 if (sock_flag(sk, SOCK_RCU_FREE))
1461 call_rcu(&sk->sk_rcu, __sk_destruct);
1463 __sk_destruct(&sk->sk_rcu);
1466 static void __sk_free(struct sock *sk)
1468 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1469 sock_diag_broadcast_destroy(sk);
1474 void sk_free(struct sock *sk)
1477 * We subtract one from sk_wmem_alloc and can know if
1478 * some packets are still in some tx queue.
1479 * If not null, sock_wfree() will call __sk_free(sk) later
1481 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1484 EXPORT_SYMBOL(sk_free);
1487 * sk_clone_lock - clone a socket, and lock its clone
1488 * @sk: the socket to clone
1489 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1491 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1493 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1496 bool is_charged = true;
1498 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1499 if (newsk != NULL) {
1500 struct sk_filter *filter;
1502 sock_copy(newsk, sk);
1505 if (likely(newsk->sk_net_refcnt))
1506 get_net(sock_net(newsk));
1507 sk_node_init(&newsk->sk_node);
1508 sock_lock_init(newsk);
1509 bh_lock_sock(newsk);
1510 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1511 newsk->sk_backlog.len = 0;
1513 atomic_set(&newsk->sk_rmem_alloc, 0);
1515 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1517 atomic_set(&newsk->sk_wmem_alloc, 1);
1518 atomic_set(&newsk->sk_omem_alloc, 0);
1519 skb_queue_head_init(&newsk->sk_receive_queue);
1520 skb_queue_head_init(&newsk->sk_write_queue);
1522 rwlock_init(&newsk->sk_callback_lock);
1523 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1524 af_callback_keys + newsk->sk_family,
1525 af_family_clock_key_strings[newsk->sk_family]);
1527 newsk->sk_dst_cache = NULL;
1528 newsk->sk_dst_pending_confirm = 0;
1529 newsk->sk_wmem_queued = 0;
1530 newsk->sk_forward_alloc = 0;
1531 atomic_set(&newsk->sk_drops, 0);
1532 newsk->sk_send_head = NULL;
1533 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1535 sock_reset_flag(newsk, SOCK_DONE);
1536 skb_queue_head_init(&newsk->sk_error_queue);
1538 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1540 /* though it's an empty new sock, the charging may fail
1541 * if sysctl_optmem_max was changed between creation of
1542 * original socket and cloning
1544 is_charged = sk_filter_charge(newsk, filter);
1546 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1547 /* We need to make sure that we don't uncharge the new
1548 * socket if we couldn't charge it in the first place
1549 * as otherwise we uncharge the parent's filter.
1552 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1553 sk_free_unlock_clone(newsk);
1557 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1560 newsk->sk_err_soft = 0;
1561 newsk->sk_priority = 0;
1562 newsk->sk_incoming_cpu = raw_smp_processor_id();
1563 atomic64_set(&newsk->sk_cookie, 0);
1565 mem_cgroup_sk_alloc(newsk);
1566 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1569 * Before updating sk_refcnt, we must commit prior changes to memory
1570 * (Documentation/RCU/rculist_nulls.txt for details)
1573 atomic_set(&newsk->sk_refcnt, 2);
1576 * Increment the counter in the same struct proto as the master
1577 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1578 * is the same as sk->sk_prot->socks, as this field was copied
1581 * This _changes_ the previous behaviour, where
1582 * tcp_create_openreq_child always was incrementing the
1583 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1584 * to be taken into account in all callers. -acme
1586 sk_refcnt_debug_inc(newsk);
1587 sk_set_socket(newsk, NULL);
1588 newsk->sk_wq = NULL;
1590 if (newsk->sk_prot->sockets_allocated)
1591 sk_sockets_allocated_inc(newsk);
1593 if (sock_needs_netstamp(sk) &&
1594 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1595 net_enable_timestamp();
1600 EXPORT_SYMBOL_GPL(sk_clone_lock);
1602 void sk_free_unlock_clone(struct sock *sk)
1604 /* It is still raw copy of parent, so invalidate
1605 * destructor and make plain sk_free() */
1606 sk->sk_destruct = NULL;
1610 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1612 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1616 sk_dst_set(sk, dst);
1617 sk->sk_route_caps = dst->dev->features;
1618 if (sk->sk_route_caps & NETIF_F_GSO)
1619 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1620 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1621 if (sk_can_gso(sk)) {
1622 if (dst->header_len) {
1623 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1625 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1626 sk->sk_gso_max_size = dst->dev->gso_max_size;
1627 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1630 sk->sk_gso_max_segs = max_segs;
1632 EXPORT_SYMBOL_GPL(sk_setup_caps);
1635 * Simple resource managers for sockets.
1640 * Write buffer destructor automatically called from kfree_skb.
1642 void sock_wfree(struct sk_buff *skb)
1644 struct sock *sk = skb->sk;
1645 unsigned int len = skb->truesize;
1647 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1649 * Keep a reference on sk_wmem_alloc, this will be released
1650 * after sk_write_space() call
1652 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1653 sk->sk_write_space(sk);
1657 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1658 * could not do because of in-flight packets
1660 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1663 EXPORT_SYMBOL(sock_wfree);
1665 /* This variant of sock_wfree() is used by TCP,
1666 * since it sets SOCK_USE_WRITE_QUEUE.
1668 void __sock_wfree(struct sk_buff *skb)
1670 struct sock *sk = skb->sk;
1672 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1676 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1681 if (unlikely(!sk_fullsock(sk))) {
1682 skb->destructor = sock_edemux;
1687 skb->destructor = sock_wfree;
1688 skb_set_hash_from_sk(skb, sk);
1690 * We used to take a refcount on sk, but following operation
1691 * is enough to guarantee sk_free() wont free this sock until
1692 * all in-flight packets are completed
1694 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1696 EXPORT_SYMBOL(skb_set_owner_w);
1698 /* This helper is used by netem, as it can hold packets in its
1699 * delay queue. We want to allow the owner socket to send more
1700 * packets, as if they were already TX completed by a typical driver.
1701 * But we also want to keep skb->sk set because some packet schedulers
1702 * rely on it (sch_fq for example). So we set skb->truesize to a small
1703 * amount (1) and decrease sk_wmem_alloc accordingly.
1705 void skb_orphan_partial(struct sk_buff *skb)
1707 /* If this skb is a TCP pure ACK or already went here,
1708 * we have nothing to do. 2 is already a very small truesize.
1710 if (skb->truesize <= 2)
1713 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1714 * so we do not completely orphan skb, but transfert all
1715 * accounted bytes but one, to avoid unexpected reorders.
1717 if (skb->destructor == sock_wfree
1719 || skb->destructor == tcp_wfree
1722 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1728 EXPORT_SYMBOL(skb_orphan_partial);
1731 * Read buffer destructor automatically called from kfree_skb.
1733 void sock_rfree(struct sk_buff *skb)
1735 struct sock *sk = skb->sk;
1736 unsigned int len = skb->truesize;
1738 atomic_sub(len, &sk->sk_rmem_alloc);
1739 sk_mem_uncharge(sk, len);
1741 EXPORT_SYMBOL(sock_rfree);
1744 * Buffer destructor for skbs that are not used directly in read or write
1745 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1747 void sock_efree(struct sk_buff *skb)
1751 EXPORT_SYMBOL(sock_efree);
1753 kuid_t sock_i_uid(struct sock *sk)
1757 read_lock_bh(&sk->sk_callback_lock);
1758 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1759 read_unlock_bh(&sk->sk_callback_lock);
1762 EXPORT_SYMBOL(sock_i_uid);
1764 unsigned long sock_i_ino(struct sock *sk)
1768 read_lock_bh(&sk->sk_callback_lock);
1769 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1770 read_unlock_bh(&sk->sk_callback_lock);
1773 EXPORT_SYMBOL(sock_i_ino);
1776 * Allocate a skb from the socket's send buffer.
1778 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1781 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1782 struct sk_buff *skb = alloc_skb(size, priority);
1784 skb_set_owner_w(skb, sk);
1790 EXPORT_SYMBOL(sock_wmalloc);
1793 * Allocate a memory block from the socket's option memory buffer.
1795 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1797 if ((unsigned int)size <= sysctl_optmem_max &&
1798 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1800 /* First do the add, to avoid the race if kmalloc
1803 atomic_add(size, &sk->sk_omem_alloc);
1804 mem = kmalloc(size, priority);
1807 atomic_sub(size, &sk->sk_omem_alloc);
1811 EXPORT_SYMBOL(sock_kmalloc);
1813 /* Free an option memory block. Note, we actually want the inline
1814 * here as this allows gcc to detect the nullify and fold away the
1815 * condition entirely.
1817 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1820 if (WARN_ON_ONCE(!mem))
1826 atomic_sub(size, &sk->sk_omem_alloc);
1829 void sock_kfree_s(struct sock *sk, void *mem, int size)
1831 __sock_kfree_s(sk, mem, size, false);
1833 EXPORT_SYMBOL(sock_kfree_s);
1835 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1837 __sock_kfree_s(sk, mem, size, true);
1839 EXPORT_SYMBOL(sock_kzfree_s);
1841 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1842 I think, these locks should be removed for datagram sockets.
1844 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1848 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1852 if (signal_pending(current))
1854 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1855 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1856 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1858 if (sk->sk_shutdown & SEND_SHUTDOWN)
1862 timeo = schedule_timeout(timeo);
1864 finish_wait(sk_sleep(sk), &wait);
1870 * Generic send/receive buffer handlers
1873 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1874 unsigned long data_len, int noblock,
1875 int *errcode, int max_page_order)
1877 struct sk_buff *skb;
1881 timeo = sock_sndtimeo(sk, noblock);
1883 err = sock_error(sk);
1888 if (sk->sk_shutdown & SEND_SHUTDOWN)
1891 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1894 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1895 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1899 if (signal_pending(current))
1901 timeo = sock_wait_for_wmem(sk, timeo);
1903 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1904 errcode, sk->sk_allocation);
1906 skb_set_owner_w(skb, sk);
1910 err = sock_intr_errno(timeo);
1915 EXPORT_SYMBOL(sock_alloc_send_pskb);
1917 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1918 int noblock, int *errcode)
1920 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1922 EXPORT_SYMBOL(sock_alloc_send_skb);
1924 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1925 struct sockcm_cookie *sockc)
1929 switch (cmsg->cmsg_type) {
1931 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1933 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1935 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1937 case SO_TIMESTAMPING:
1938 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1941 tsflags = *(u32 *)CMSG_DATA(cmsg);
1942 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1945 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1946 sockc->tsflags |= tsflags;
1948 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1950 case SCM_CREDENTIALS:
1957 EXPORT_SYMBOL(__sock_cmsg_send);
1959 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1960 struct sockcm_cookie *sockc)
1962 struct cmsghdr *cmsg;
1965 for_each_cmsghdr(cmsg, msg) {
1966 if (!CMSG_OK(msg, cmsg))
1968 if (cmsg->cmsg_level != SOL_SOCKET)
1970 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1976 EXPORT_SYMBOL(sock_cmsg_send);
1978 /* On 32bit arches, an skb frag is limited to 2^15 */
1979 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1982 * skb_page_frag_refill - check that a page_frag contains enough room
1983 * @sz: minimum size of the fragment we want to get
1984 * @pfrag: pointer to page_frag
1985 * @gfp: priority for memory allocation
1987 * Note: While this allocator tries to use high order pages, there is
1988 * no guarantee that allocations succeed. Therefore, @sz MUST be
1989 * less or equal than PAGE_SIZE.
1991 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1994 if (page_ref_count(pfrag->page) == 1) {
1998 if (pfrag->offset + sz <= pfrag->size)
2000 put_page(pfrag->page);
2004 if (SKB_FRAG_PAGE_ORDER) {
2005 /* Avoid direct reclaim but allow kswapd to wake */
2006 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2007 __GFP_COMP | __GFP_NOWARN |
2009 SKB_FRAG_PAGE_ORDER);
2010 if (likely(pfrag->page)) {
2011 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2015 pfrag->page = alloc_page(gfp);
2016 if (likely(pfrag->page)) {
2017 pfrag->size = PAGE_SIZE;
2022 EXPORT_SYMBOL(skb_page_frag_refill);
2024 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2026 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2029 sk_enter_memory_pressure(sk);
2030 sk_stream_moderate_sndbuf(sk);
2033 EXPORT_SYMBOL(sk_page_frag_refill);
2035 static void __lock_sock(struct sock *sk)
2036 __releases(&sk->sk_lock.slock)
2037 __acquires(&sk->sk_lock.slock)
2042 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2043 TASK_UNINTERRUPTIBLE);
2044 spin_unlock_bh(&sk->sk_lock.slock);
2046 spin_lock_bh(&sk->sk_lock.slock);
2047 if (!sock_owned_by_user(sk))
2050 finish_wait(&sk->sk_lock.wq, &wait);
2053 static void __release_sock(struct sock *sk)
2054 __releases(&sk->sk_lock.slock)
2055 __acquires(&sk->sk_lock.slock)
2057 struct sk_buff *skb, *next;
2059 while ((skb = sk->sk_backlog.head) != NULL) {
2060 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2062 spin_unlock_bh(&sk->sk_lock.slock);
2067 WARN_ON_ONCE(skb_dst_is_noref(skb));
2069 sk_backlog_rcv(sk, skb);
2074 } while (skb != NULL);
2076 spin_lock_bh(&sk->sk_lock.slock);
2080 * Doing the zeroing here guarantee we can not loop forever
2081 * while a wild producer attempts to flood us.
2083 sk->sk_backlog.len = 0;
2086 void __sk_flush_backlog(struct sock *sk)
2088 spin_lock_bh(&sk->sk_lock.slock);
2090 spin_unlock_bh(&sk->sk_lock.slock);
2094 * sk_wait_data - wait for data to arrive at sk_receive_queue
2095 * @sk: sock to wait on
2096 * @timeo: for how long
2097 * @skb: last skb seen on sk_receive_queue
2099 * Now socket state including sk->sk_err is changed only under lock,
2100 * hence we may omit checks after joining wait queue.
2101 * We check receive queue before schedule() only as optimization;
2102 * it is very likely that release_sock() added new data.
2104 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2106 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2109 add_wait_queue(sk_sleep(sk), &wait);
2110 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2111 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2112 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2113 remove_wait_queue(sk_sleep(sk), &wait);
2116 EXPORT_SYMBOL(sk_wait_data);
2119 * __sk_mem_raise_allocated - increase memory_allocated
2121 * @size: memory size to allocate
2122 * @amt: pages to allocate
2123 * @kind: allocation type
2125 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2127 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2129 struct proto *prot = sk->sk_prot;
2130 long allocated = sk_memory_allocated_add(sk, amt);
2132 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2133 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2134 goto suppress_allocation;
2137 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2138 sk_leave_memory_pressure(sk);
2142 /* Under pressure. */
2143 if (allocated > sk_prot_mem_limits(sk, 1))
2144 sk_enter_memory_pressure(sk);
2146 /* Over hard limit. */
2147 if (allocated > sk_prot_mem_limits(sk, 2))
2148 goto suppress_allocation;
2150 /* guarantee minimum buffer size under pressure */
2151 if (kind == SK_MEM_RECV) {
2152 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2155 } else { /* SK_MEM_SEND */
2156 if (sk->sk_type == SOCK_STREAM) {
2157 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2159 } else if (atomic_read(&sk->sk_wmem_alloc) <
2160 prot->sysctl_wmem[0])
2164 if (sk_has_memory_pressure(sk)) {
2167 if (!sk_under_memory_pressure(sk))
2169 alloc = sk_sockets_allocated_read_positive(sk);
2170 if (sk_prot_mem_limits(sk, 2) > alloc *
2171 sk_mem_pages(sk->sk_wmem_queued +
2172 atomic_read(&sk->sk_rmem_alloc) +
2173 sk->sk_forward_alloc))
2177 suppress_allocation:
2179 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2180 sk_stream_moderate_sndbuf(sk);
2182 /* Fail only if socket is _under_ its sndbuf.
2183 * In this case we cannot block, so that we have to fail.
2185 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2189 trace_sock_exceed_buf_limit(sk, prot, allocated);
2191 sk_memory_allocated_sub(sk, amt);
2193 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2194 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2198 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2201 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2203 * @size: memory size to allocate
2204 * @kind: allocation type
2206 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2207 * rmem allocation. This function assumes that protocols which have
2208 * memory_pressure use sk_wmem_queued as write buffer accounting.
2210 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2212 int ret, amt = sk_mem_pages(size);
2214 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2215 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2217 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2220 EXPORT_SYMBOL(__sk_mem_schedule);
2223 * __sk_mem_reduce_allocated - reclaim memory_allocated
2225 * @amount: number of quanta
2227 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2229 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2231 sk_memory_allocated_sub(sk, amount);
2233 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2234 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2236 if (sk_under_memory_pressure(sk) &&
2237 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2238 sk_leave_memory_pressure(sk);
2240 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2243 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2245 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2247 void __sk_mem_reclaim(struct sock *sk, int amount)
2249 amount >>= SK_MEM_QUANTUM_SHIFT;
2250 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2251 __sk_mem_reduce_allocated(sk, amount);
2253 EXPORT_SYMBOL(__sk_mem_reclaim);
2255 int sk_set_peek_off(struct sock *sk, int val)
2260 sk->sk_peek_off = val;
2263 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2266 * Set of default routines for initialising struct proto_ops when
2267 * the protocol does not support a particular function. In certain
2268 * cases where it makes no sense for a protocol to have a "do nothing"
2269 * function, some default processing is provided.
2272 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2276 EXPORT_SYMBOL(sock_no_bind);
2278 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2283 EXPORT_SYMBOL(sock_no_connect);
2285 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2289 EXPORT_SYMBOL(sock_no_socketpair);
2291 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2296 EXPORT_SYMBOL(sock_no_accept);
2298 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2303 EXPORT_SYMBOL(sock_no_getname);
2305 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2309 EXPORT_SYMBOL(sock_no_poll);
2311 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2315 EXPORT_SYMBOL(sock_no_ioctl);
2317 int sock_no_listen(struct socket *sock, int backlog)
2321 EXPORT_SYMBOL(sock_no_listen);
2323 int sock_no_shutdown(struct socket *sock, int how)
2327 EXPORT_SYMBOL(sock_no_shutdown);
2329 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2330 char __user *optval, unsigned int optlen)
2334 EXPORT_SYMBOL(sock_no_setsockopt);
2336 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2337 char __user *optval, int __user *optlen)
2341 EXPORT_SYMBOL(sock_no_getsockopt);
2343 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2347 EXPORT_SYMBOL(sock_no_sendmsg);
2349 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2354 EXPORT_SYMBOL(sock_no_recvmsg);
2356 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2358 /* Mirror missing mmap method error code */
2361 EXPORT_SYMBOL(sock_no_mmap);
2363 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2366 struct msghdr msg = {.msg_flags = flags};
2368 char *kaddr = kmap(page);
2369 iov.iov_base = kaddr + offset;
2371 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2375 EXPORT_SYMBOL(sock_no_sendpage);
2378 * Default Socket Callbacks
2381 static void sock_def_wakeup(struct sock *sk)
2383 struct socket_wq *wq;
2386 wq = rcu_dereference(sk->sk_wq);
2387 if (skwq_has_sleeper(wq))
2388 wake_up_interruptible_all(&wq->wait);
2392 static void sock_def_error_report(struct sock *sk)
2394 struct socket_wq *wq;
2397 wq = rcu_dereference(sk->sk_wq);
2398 if (skwq_has_sleeper(wq))
2399 wake_up_interruptible_poll(&wq->wait, POLLERR);
2400 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2404 static void sock_def_readable(struct sock *sk)
2406 struct socket_wq *wq;
2409 wq = rcu_dereference(sk->sk_wq);
2410 if (skwq_has_sleeper(wq))
2411 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2412 POLLRDNORM | POLLRDBAND);
2413 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2417 static void sock_def_write_space(struct sock *sk)
2419 struct socket_wq *wq;
2423 /* Do not wake up a writer until he can make "significant"
2426 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2427 wq = rcu_dereference(sk->sk_wq);
2428 if (skwq_has_sleeper(wq))
2429 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2430 POLLWRNORM | POLLWRBAND);
2432 /* Should agree with poll, otherwise some programs break */
2433 if (sock_writeable(sk))
2434 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2440 static void sock_def_destruct(struct sock *sk)
2444 void sk_send_sigurg(struct sock *sk)
2446 if (sk->sk_socket && sk->sk_socket->file)
2447 if (send_sigurg(&sk->sk_socket->file->f_owner))
2448 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2450 EXPORT_SYMBOL(sk_send_sigurg);
2452 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2453 unsigned long expires)
2455 if (!mod_timer(timer, expires))
2458 EXPORT_SYMBOL(sk_reset_timer);
2460 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2462 if (del_timer(timer))
2465 EXPORT_SYMBOL(sk_stop_timer);
2467 void sock_init_data(struct socket *sock, struct sock *sk)
2469 skb_queue_head_init(&sk->sk_receive_queue);
2470 skb_queue_head_init(&sk->sk_write_queue);
2471 skb_queue_head_init(&sk->sk_error_queue);
2473 sk->sk_send_head = NULL;
2475 init_timer(&sk->sk_timer);
2477 sk->sk_allocation = GFP_KERNEL;
2478 sk->sk_rcvbuf = sysctl_rmem_default;
2479 sk->sk_sndbuf = sysctl_wmem_default;
2480 sk->sk_state = TCP_CLOSE;
2481 sk_set_socket(sk, sock);
2483 sock_set_flag(sk, SOCK_ZAPPED);
2486 sk->sk_type = sock->type;
2487 sk->sk_wq = sock->wq;
2489 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2492 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2495 rwlock_init(&sk->sk_callback_lock);
2496 if (sk->sk_kern_sock)
2497 lockdep_set_class_and_name(
2498 &sk->sk_callback_lock,
2499 af_kern_callback_keys + sk->sk_family,
2500 af_family_kern_clock_key_strings[sk->sk_family]);
2502 lockdep_set_class_and_name(
2503 &sk->sk_callback_lock,
2504 af_callback_keys + sk->sk_family,
2505 af_family_clock_key_strings[sk->sk_family]);
2507 sk->sk_state_change = sock_def_wakeup;
2508 sk->sk_data_ready = sock_def_readable;
2509 sk->sk_write_space = sock_def_write_space;
2510 sk->sk_error_report = sock_def_error_report;
2511 sk->sk_destruct = sock_def_destruct;
2513 sk->sk_frag.page = NULL;
2514 sk->sk_frag.offset = 0;
2515 sk->sk_peek_off = -1;
2517 sk->sk_peer_pid = NULL;
2518 sk->sk_peer_cred = NULL;
2519 sk->sk_write_pending = 0;
2520 sk->sk_rcvlowat = 1;
2521 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2522 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2524 sk->sk_stamp = ktime_set(-1L, 0);
2526 #ifdef CONFIG_NET_RX_BUSY_POLL
2528 sk->sk_ll_usec = sysctl_net_busy_read;
2531 sk->sk_max_pacing_rate = ~0U;
2532 sk->sk_pacing_rate = ~0U;
2533 sk->sk_incoming_cpu = -1;
2535 * Before updating sk_refcnt, we must commit prior changes to memory
2536 * (Documentation/RCU/rculist_nulls.txt for details)
2539 atomic_set(&sk->sk_refcnt, 1);
2540 atomic_set(&sk->sk_drops, 0);
2542 EXPORT_SYMBOL(sock_init_data);
2544 void lock_sock_nested(struct sock *sk, int subclass)
2547 spin_lock_bh(&sk->sk_lock.slock);
2548 if (sk->sk_lock.owned)
2550 sk->sk_lock.owned = 1;
2551 spin_unlock(&sk->sk_lock.slock);
2553 * The sk_lock has mutex_lock() semantics here:
2555 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2558 EXPORT_SYMBOL(lock_sock_nested);
2560 void release_sock(struct sock *sk)
2562 spin_lock_bh(&sk->sk_lock.slock);
2563 if (sk->sk_backlog.tail)
2566 /* Warning : release_cb() might need to release sk ownership,
2567 * ie call sock_release_ownership(sk) before us.
2569 if (sk->sk_prot->release_cb)
2570 sk->sk_prot->release_cb(sk);
2572 sock_release_ownership(sk);
2573 if (waitqueue_active(&sk->sk_lock.wq))
2574 wake_up(&sk->sk_lock.wq);
2575 spin_unlock_bh(&sk->sk_lock.slock);
2577 EXPORT_SYMBOL(release_sock);
2580 * lock_sock_fast - fast version of lock_sock
2583 * This version should be used for very small section, where process wont block
2584 * return false if fast path is taken
2585 * sk_lock.slock locked, owned = 0, BH disabled
2586 * return true if slow path is taken
2587 * sk_lock.slock unlocked, owned = 1, BH enabled
2589 bool lock_sock_fast(struct sock *sk)
2592 spin_lock_bh(&sk->sk_lock.slock);
2594 if (!sk->sk_lock.owned)
2596 * Note : We must disable BH
2601 sk->sk_lock.owned = 1;
2602 spin_unlock(&sk->sk_lock.slock);
2604 * The sk_lock has mutex_lock() semantics here:
2606 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2610 EXPORT_SYMBOL(lock_sock_fast);
2612 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2615 if (!sock_flag(sk, SOCK_TIMESTAMP))
2616 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2617 tv = ktime_to_timeval(sk->sk_stamp);
2618 if (tv.tv_sec == -1)
2620 if (tv.tv_sec == 0) {
2621 sk->sk_stamp = ktime_get_real();
2622 tv = ktime_to_timeval(sk->sk_stamp);
2624 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2626 EXPORT_SYMBOL(sock_get_timestamp);
2628 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2631 if (!sock_flag(sk, SOCK_TIMESTAMP))
2632 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2633 ts = ktime_to_timespec(sk->sk_stamp);
2634 if (ts.tv_sec == -1)
2636 if (ts.tv_sec == 0) {
2637 sk->sk_stamp = ktime_get_real();
2638 ts = ktime_to_timespec(sk->sk_stamp);
2640 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2642 EXPORT_SYMBOL(sock_get_timestampns);
2644 void sock_enable_timestamp(struct sock *sk, int flag)
2646 if (!sock_flag(sk, flag)) {
2647 unsigned long previous_flags = sk->sk_flags;
2649 sock_set_flag(sk, flag);
2651 * we just set one of the two flags which require net
2652 * time stamping, but time stamping might have been on
2653 * already because of the other one
2655 if (sock_needs_netstamp(sk) &&
2656 !(previous_flags & SK_FLAGS_TIMESTAMP))
2657 net_enable_timestamp();
2661 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2662 int level, int type)
2664 struct sock_exterr_skb *serr;
2665 struct sk_buff *skb;
2669 skb = sock_dequeue_err_skb(sk);
2675 msg->msg_flags |= MSG_TRUNC;
2678 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2682 sock_recv_timestamp(msg, sk, skb);
2684 serr = SKB_EXT_ERR(skb);
2685 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2687 msg->msg_flags |= MSG_ERRQUEUE;
2695 EXPORT_SYMBOL(sock_recv_errqueue);
2698 * Get a socket option on an socket.
2700 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2701 * asynchronous errors should be reported by getsockopt. We assume
2702 * this means if you specify SO_ERROR (otherwise whats the point of it).
2704 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2705 char __user *optval, int __user *optlen)
2707 struct sock *sk = sock->sk;
2709 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2711 EXPORT_SYMBOL(sock_common_getsockopt);
2713 #ifdef CONFIG_COMPAT
2714 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2715 char __user *optval, int __user *optlen)
2717 struct sock *sk = sock->sk;
2719 if (sk->sk_prot->compat_getsockopt != NULL)
2720 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2722 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2724 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2727 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2730 struct sock *sk = sock->sk;
2734 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2735 flags & ~MSG_DONTWAIT, &addr_len);
2737 msg->msg_namelen = addr_len;
2740 EXPORT_SYMBOL(sock_common_recvmsg);
2743 * Set socket options on an inet socket.
2745 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2746 char __user *optval, unsigned int optlen)
2748 struct sock *sk = sock->sk;
2750 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2752 EXPORT_SYMBOL(sock_common_setsockopt);
2754 #ifdef CONFIG_COMPAT
2755 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2756 char __user *optval, unsigned int optlen)
2758 struct sock *sk = sock->sk;
2760 if (sk->sk_prot->compat_setsockopt != NULL)
2761 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2763 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2765 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2768 void sk_common_release(struct sock *sk)
2770 if (sk->sk_prot->destroy)
2771 sk->sk_prot->destroy(sk);
2774 * Observation: when sock_common_release is called, processes have
2775 * no access to socket. But net still has.
2776 * Step one, detach it from networking:
2778 * A. Remove from hash tables.
2781 sk->sk_prot->unhash(sk);
2784 * In this point socket cannot receive new packets, but it is possible
2785 * that some packets are in flight because some CPU runs receiver and
2786 * did hash table lookup before we unhashed socket. They will achieve
2787 * receive queue and will be purged by socket destructor.
2789 * Also we still have packets pending on receive queue and probably,
2790 * our own packets waiting in device queues. sock_destroy will drain
2791 * receive queue, but transmitted packets will delay socket destruction
2792 * until the last reference will be released.
2797 xfrm_sk_free_policy(sk);
2799 sk_refcnt_debug_release(sk);
2803 EXPORT_SYMBOL(sk_common_release);
2805 #ifdef CONFIG_PROC_FS
2806 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2808 int val[PROTO_INUSE_NR];
2811 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2813 #ifdef CONFIG_NET_NS
2814 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2816 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2818 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2820 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2822 int cpu, idx = prot->inuse_idx;
2825 for_each_possible_cpu(cpu)
2826 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2828 return res >= 0 ? res : 0;
2830 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2832 static int __net_init sock_inuse_init_net(struct net *net)
2834 net->core.inuse = alloc_percpu(struct prot_inuse);
2835 return net->core.inuse ? 0 : -ENOMEM;
2838 static void __net_exit sock_inuse_exit_net(struct net *net)
2840 free_percpu(net->core.inuse);
2843 static struct pernet_operations net_inuse_ops = {
2844 .init = sock_inuse_init_net,
2845 .exit = sock_inuse_exit_net,
2848 static __init int net_inuse_init(void)
2850 if (register_pernet_subsys(&net_inuse_ops))
2851 panic("Cannot initialize net inuse counters");
2856 core_initcall(net_inuse_init);
2858 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2860 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2862 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2864 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2866 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2868 int cpu, idx = prot->inuse_idx;
2871 for_each_possible_cpu(cpu)
2872 res += per_cpu(prot_inuse, cpu).val[idx];
2874 return res >= 0 ? res : 0;
2876 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2879 static void assign_proto_idx(struct proto *prot)
2881 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2883 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2884 pr_err("PROTO_INUSE_NR exhausted\n");
2888 set_bit(prot->inuse_idx, proto_inuse_idx);
2891 static void release_proto_idx(struct proto *prot)
2893 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2894 clear_bit(prot->inuse_idx, proto_inuse_idx);
2897 static inline void assign_proto_idx(struct proto *prot)
2901 static inline void release_proto_idx(struct proto *prot)
2906 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2910 kfree(rsk_prot->slab_name);
2911 rsk_prot->slab_name = NULL;
2912 kmem_cache_destroy(rsk_prot->slab);
2913 rsk_prot->slab = NULL;
2916 static int req_prot_init(const struct proto *prot)
2918 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2923 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2925 if (!rsk_prot->slab_name)
2928 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2929 rsk_prot->obj_size, 0,
2930 prot->slab_flags, NULL);
2932 if (!rsk_prot->slab) {
2933 pr_crit("%s: Can't create request sock SLAB cache!\n",
2940 int proto_register(struct proto *prot, int alloc_slab)
2943 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2944 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2947 if (prot->slab == NULL) {
2948 pr_crit("%s: Can't create sock SLAB cache!\n",
2953 if (req_prot_init(prot))
2954 goto out_free_request_sock_slab;
2956 if (prot->twsk_prot != NULL) {
2957 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2959 if (prot->twsk_prot->twsk_slab_name == NULL)
2960 goto out_free_request_sock_slab;
2962 prot->twsk_prot->twsk_slab =
2963 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2964 prot->twsk_prot->twsk_obj_size,
2968 if (prot->twsk_prot->twsk_slab == NULL)
2969 goto out_free_timewait_sock_slab_name;
2973 mutex_lock(&proto_list_mutex);
2974 list_add(&prot->node, &proto_list);
2975 assign_proto_idx(prot);
2976 mutex_unlock(&proto_list_mutex);
2979 out_free_timewait_sock_slab_name:
2980 kfree(prot->twsk_prot->twsk_slab_name);
2981 out_free_request_sock_slab:
2982 req_prot_cleanup(prot->rsk_prot);
2984 kmem_cache_destroy(prot->slab);
2989 EXPORT_SYMBOL(proto_register);
2991 void proto_unregister(struct proto *prot)
2993 mutex_lock(&proto_list_mutex);
2994 release_proto_idx(prot);
2995 list_del(&prot->node);
2996 mutex_unlock(&proto_list_mutex);
2998 kmem_cache_destroy(prot->slab);
3001 req_prot_cleanup(prot->rsk_prot);
3003 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3004 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3005 kfree(prot->twsk_prot->twsk_slab_name);
3006 prot->twsk_prot->twsk_slab = NULL;
3009 EXPORT_SYMBOL(proto_unregister);
3011 #ifdef CONFIG_PROC_FS
3012 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3013 __acquires(proto_list_mutex)
3015 mutex_lock(&proto_list_mutex);
3016 return seq_list_start_head(&proto_list, *pos);
3019 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3021 return seq_list_next(v, &proto_list, pos);
3024 static void proto_seq_stop(struct seq_file *seq, void *v)
3025 __releases(proto_list_mutex)
3027 mutex_unlock(&proto_list_mutex);
3030 static char proto_method_implemented(const void *method)
3032 return method == NULL ? 'n' : 'y';
3034 static long sock_prot_memory_allocated(struct proto *proto)
3036 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3039 static char *sock_prot_memory_pressure(struct proto *proto)
3041 return proto->memory_pressure != NULL ?
3042 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3045 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3048 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3049 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3052 sock_prot_inuse_get(seq_file_net(seq), proto),
3053 sock_prot_memory_allocated(proto),
3054 sock_prot_memory_pressure(proto),
3056 proto->slab == NULL ? "no" : "yes",
3057 module_name(proto->owner),
3058 proto_method_implemented(proto->close),
3059 proto_method_implemented(proto->connect),
3060 proto_method_implemented(proto->disconnect),
3061 proto_method_implemented(proto->accept),
3062 proto_method_implemented(proto->ioctl),
3063 proto_method_implemented(proto->init),
3064 proto_method_implemented(proto->destroy),
3065 proto_method_implemented(proto->shutdown),
3066 proto_method_implemented(proto->setsockopt),
3067 proto_method_implemented(proto->getsockopt),
3068 proto_method_implemented(proto->sendmsg),
3069 proto_method_implemented(proto->recvmsg),
3070 proto_method_implemented(proto->sendpage),
3071 proto_method_implemented(proto->bind),
3072 proto_method_implemented(proto->backlog_rcv),
3073 proto_method_implemented(proto->hash),
3074 proto_method_implemented(proto->unhash),
3075 proto_method_implemented(proto->get_port),
3076 proto_method_implemented(proto->enter_memory_pressure));
3079 static int proto_seq_show(struct seq_file *seq, void *v)
3081 if (v == &proto_list)
3082 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3091 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3093 proto_seq_printf(seq, list_entry(v, struct proto, node));
3097 static const struct seq_operations proto_seq_ops = {
3098 .start = proto_seq_start,
3099 .next = proto_seq_next,
3100 .stop = proto_seq_stop,
3101 .show = proto_seq_show,
3104 static int proto_seq_open(struct inode *inode, struct file *file)
3106 return seq_open_net(inode, file, &proto_seq_ops,
3107 sizeof(struct seq_net_private));
3110 static const struct file_operations proto_seq_fops = {
3111 .owner = THIS_MODULE,
3112 .open = proto_seq_open,
3114 .llseek = seq_lseek,
3115 .release = seq_release_net,
3118 static __net_init int proto_init_net(struct net *net)
3120 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3126 static __net_exit void proto_exit_net(struct net *net)
3128 remove_proc_entry("protocols", net->proc_net);
3132 static __net_initdata struct pernet_operations proto_net_ops = {
3133 .init = proto_init_net,
3134 .exit = proto_exit_net,
3137 static int __init proto_init(void)
3139 return register_pernet_subsys(&proto_net_ops);
3142 subsys_initcall(proto_init);
3144 #endif /* PROC_FS */