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 <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 <asm/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>
138 #include <trace/events/sock.h>
144 #include <net/busy_poll.h>
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
150 * sk_ns_capable - General socket capability test
151 * @sk: Socket to use a capability on or through
152 * @user_ns: The user namespace of the capability to use
153 * @cap: The capability to use
155 * Test to see if the opener of the socket had when the socket was
156 * created and the current process has the capability @cap in the user
157 * namespace @user_ns.
159 bool sk_ns_capable(const struct sock *sk,
160 struct user_namespace *user_ns, int cap)
162 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
163 ns_capable(user_ns, cap);
165 EXPORT_SYMBOL(sk_ns_capable);
168 * sk_capable - Socket global capability test
169 * @sk: Socket to use a capability on or through
170 * @cap: The global capability to use
172 * Test to see if the opener of the socket had when the socket was
173 * created and the current process has the capability @cap in all user
176 bool sk_capable(const struct sock *sk, int cap)
178 return sk_ns_capable(sk, &init_user_ns, cap);
180 EXPORT_SYMBOL(sk_capable);
183 * sk_net_capable - Network namespace socket capability test
184 * @sk: Socket to use a capability on or through
185 * @cap: The capability to use
187 * Test to see if the opener of the socket had when the socket was created
188 * and the current process has the capability @cap over the network namespace
189 * the socket is a member of.
191 bool sk_net_capable(const struct sock *sk, int cap)
193 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195 EXPORT_SYMBOL(sk_net_capable);
198 #ifdef CONFIG_MEMCG_KMEM
199 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
204 mutex_lock(&proto_list_mutex);
205 list_for_each_entry(proto, &proto_list, node) {
206 if (proto->init_cgroup) {
207 ret = proto->init_cgroup(memcg, ss);
213 mutex_unlock(&proto_list_mutex);
216 list_for_each_entry_continue_reverse(proto, &proto_list, node)
217 if (proto->destroy_cgroup)
218 proto->destroy_cgroup(memcg);
219 mutex_unlock(&proto_list_mutex);
223 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
227 mutex_lock(&proto_list_mutex);
228 list_for_each_entry_reverse(proto, &proto_list, node)
229 if (proto->destroy_cgroup)
230 proto->destroy_cgroup(memcg);
231 mutex_unlock(&proto_list_mutex);
236 * Each address family might have different locking rules, so we have
237 * one slock key per address family:
239 static struct lock_class_key af_family_keys[AF_MAX];
240 static struct lock_class_key af_family_slock_keys[AF_MAX];
242 #if defined(CONFIG_MEMCG_KMEM)
243 struct static_key memcg_socket_limit_enabled;
244 EXPORT_SYMBOL(memcg_socket_limit_enabled);
248 * Make lock validator output more readable. (we pre-construct these
249 * strings build-time, so that runtime initialization of socket
252 static const char *const af_family_key_strings[AF_MAX+1] = {
253 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
254 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
255 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
256 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
257 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
258 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
259 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
260 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
261 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
262 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
263 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
264 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
265 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
266 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
268 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
269 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
270 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
271 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
272 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
273 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
274 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
275 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
276 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
277 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
278 "slock-27" , "slock-28" , "slock-AF_CAN" ,
279 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
280 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
281 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
282 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
284 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
285 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
286 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
287 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
288 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
289 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
290 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
291 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
292 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
293 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
294 "clock-27" , "clock-28" , "clock-AF_CAN" ,
295 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
296 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
297 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
298 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
302 * sk_callback_lock locking rules are per-address-family,
303 * so split the lock classes by using a per-AF key:
305 static struct lock_class_key af_callback_keys[AF_MAX];
307 /* Take into consideration the size of the struct sk_buff overhead in the
308 * determination of these values, since that is non-constant across
309 * platforms. This makes socket queueing behavior and performance
310 * not depend upon such differences.
312 #define _SK_MEM_PACKETS 256
313 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
314 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
317 /* Run time adjustable parameters. */
318 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
319 EXPORT_SYMBOL(sysctl_wmem_max);
320 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
321 EXPORT_SYMBOL(sysctl_rmem_max);
322 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
323 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
325 /* Maximal space eaten by iovec or ancillary data plus some space */
326 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
327 EXPORT_SYMBOL(sysctl_optmem_max);
329 int sysctl_tstamp_allow_data __read_mostly = 1;
331 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
332 EXPORT_SYMBOL_GPL(memalloc_socks);
335 * sk_set_memalloc - sets %SOCK_MEMALLOC
336 * @sk: socket to set it on
338 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
339 * It's the responsibility of the admin to adjust min_free_kbytes
340 * to meet the requirements
342 void sk_set_memalloc(struct sock *sk)
344 sock_set_flag(sk, SOCK_MEMALLOC);
345 sk->sk_allocation |= __GFP_MEMALLOC;
346 static_key_slow_inc(&memalloc_socks);
348 EXPORT_SYMBOL_GPL(sk_set_memalloc);
350 void sk_clear_memalloc(struct sock *sk)
352 sock_reset_flag(sk, SOCK_MEMALLOC);
353 sk->sk_allocation &= ~__GFP_MEMALLOC;
354 static_key_slow_dec(&memalloc_socks);
357 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
358 * progress of swapping. SOCK_MEMALLOC may be cleared while
359 * it has rmem allocations due to the last swapfile being deactivated
360 * but there is a risk that the socket is unusable due to exceeding
361 * the rmem limits. Reclaim the reserves and obey rmem limits again.
365 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
367 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
370 unsigned long pflags = current->flags;
372 /* these should have been dropped before queueing */
373 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
375 current->flags |= PF_MEMALLOC;
376 ret = sk->sk_backlog_rcv(sk, skb);
377 tsk_restore_flags(current, pflags, PF_MEMALLOC);
381 EXPORT_SYMBOL(__sk_backlog_rcv);
383 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
387 if (optlen < sizeof(tv))
389 if (copy_from_user(&tv, optval, sizeof(tv)))
391 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
395 static int warned __read_mostly;
398 if (warned < 10 && net_ratelimit()) {
400 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
401 __func__, current->comm, task_pid_nr(current));
405 *timeo_p = MAX_SCHEDULE_TIMEOUT;
406 if (tv.tv_sec == 0 && tv.tv_usec == 0)
408 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
409 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
413 static void sock_warn_obsolete_bsdism(const char *name)
416 static char warncomm[TASK_COMM_LEN];
417 if (strcmp(warncomm, current->comm) && warned < 5) {
418 strcpy(warncomm, current->comm);
419 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
425 static bool sock_needs_netstamp(const struct sock *sk)
427 switch (sk->sk_family) {
436 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
438 if (sk->sk_flags & flags) {
439 sk->sk_flags &= ~flags;
440 if (sock_needs_netstamp(sk) &&
441 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
442 net_disable_timestamp();
447 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
451 struct sk_buff_head *list = &sk->sk_receive_queue;
453 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
454 atomic_inc(&sk->sk_drops);
455 trace_sock_rcvqueue_full(sk, skb);
459 err = sk_filter(sk, skb);
463 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
464 atomic_inc(&sk->sk_drops);
469 skb_set_owner_r(skb, sk);
471 /* we escape from rcu protected region, make sure we dont leak
476 spin_lock_irqsave(&list->lock, flags);
477 sock_skb_set_dropcount(sk, skb);
478 __skb_queue_tail(list, skb);
479 spin_unlock_irqrestore(&list->lock, flags);
481 if (!sock_flag(sk, SOCK_DEAD))
482 sk->sk_data_ready(sk);
485 EXPORT_SYMBOL(sock_queue_rcv_skb);
487 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
489 int rc = NET_RX_SUCCESS;
491 if (sk_filter(sk, skb))
492 goto discard_and_relse;
496 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
497 atomic_inc(&sk->sk_drops);
498 goto discard_and_relse;
501 bh_lock_sock_nested(sk);
504 if (!sock_owned_by_user(sk)) {
506 * trylock + unlock semantics:
508 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
510 rc = sk_backlog_rcv(sk, skb);
512 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
513 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
515 atomic_inc(&sk->sk_drops);
516 goto discard_and_relse;
527 EXPORT_SYMBOL(sk_receive_skb);
529 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
531 struct dst_entry *dst = __sk_dst_get(sk);
533 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
534 sk_tx_queue_clear(sk);
535 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
542 EXPORT_SYMBOL(__sk_dst_check);
544 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
546 struct dst_entry *dst = sk_dst_get(sk);
548 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
556 EXPORT_SYMBOL(sk_dst_check);
558 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
561 int ret = -ENOPROTOOPT;
562 #ifdef CONFIG_NETDEVICES
563 struct net *net = sock_net(sk);
564 char devname[IFNAMSIZ];
569 if (!ns_capable(net->user_ns, CAP_NET_RAW))
576 /* Bind this socket to a particular device like "eth0",
577 * as specified in the passed interface name. If the
578 * name is "" or the option length is zero the socket
581 if (optlen > IFNAMSIZ - 1)
582 optlen = IFNAMSIZ - 1;
583 memset(devname, 0, sizeof(devname));
586 if (copy_from_user(devname, optval, optlen))
590 if (devname[0] != '\0') {
591 struct net_device *dev;
594 dev = dev_get_by_name_rcu(net, devname);
596 index = dev->ifindex;
604 sk->sk_bound_dev_if = index;
616 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
617 int __user *optlen, int len)
619 int ret = -ENOPROTOOPT;
620 #ifdef CONFIG_NETDEVICES
621 struct net *net = sock_net(sk);
622 char devname[IFNAMSIZ];
624 if (sk->sk_bound_dev_if == 0) {
633 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
637 len = strlen(devname) + 1;
640 if (copy_to_user(optval, devname, len))
645 if (put_user(len, optlen))
656 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
659 sock_set_flag(sk, bit);
661 sock_reset_flag(sk, bit);
664 bool sk_mc_loop(struct sock *sk)
666 if (dev_recursion_level())
670 switch (sk->sk_family) {
672 return inet_sk(sk)->mc_loop;
673 #if IS_ENABLED(CONFIG_IPV6)
675 return inet6_sk(sk)->mc_loop;
681 EXPORT_SYMBOL(sk_mc_loop);
684 * This is meant for all protocols to use and covers goings on
685 * at the socket level. Everything here is generic.
688 int sock_setsockopt(struct socket *sock, int level, int optname,
689 char __user *optval, unsigned int optlen)
691 struct sock *sk = sock->sk;
698 * Options without arguments
701 if (optname == SO_BINDTODEVICE)
702 return sock_setbindtodevice(sk, optval, optlen);
704 if (optlen < sizeof(int))
707 if (get_user(val, (int __user *)optval))
710 valbool = val ? 1 : 0;
716 if (val && !capable(CAP_NET_ADMIN))
719 sock_valbool_flag(sk, SOCK_DBG, valbool);
722 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
725 sk->sk_reuseport = valbool;
734 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
737 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
740 /* Don't error on this BSD doesn't and if you think
741 * about it this is right. Otherwise apps have to
742 * play 'guess the biggest size' games. RCVBUF/SNDBUF
743 * are treated in BSD as hints
745 val = min_t(u32, val, sysctl_wmem_max);
747 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
748 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
749 /* Wake up sending tasks if we upped the value. */
750 sk->sk_write_space(sk);
754 if (!capable(CAP_NET_ADMIN)) {
761 /* Don't error on this BSD doesn't and if you think
762 * about it this is right. Otherwise apps have to
763 * play 'guess the biggest size' games. RCVBUF/SNDBUF
764 * are treated in BSD as hints
766 val = min_t(u32, val, sysctl_rmem_max);
768 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
770 * We double it on the way in to account for
771 * "struct sk_buff" etc. overhead. Applications
772 * assume that the SO_RCVBUF setting they make will
773 * allow that much actual data to be received on that
776 * Applications are unaware that "struct sk_buff" and
777 * other overheads allocate from the receive buffer
778 * during socket buffer allocation.
780 * And after considering the possible alternatives,
781 * returning the value we actually used in getsockopt
782 * is the most desirable behavior.
784 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
788 if (!capable(CAP_NET_ADMIN)) {
796 if (sk->sk_protocol == IPPROTO_TCP &&
797 sk->sk_type == SOCK_STREAM)
798 tcp_set_keepalive(sk, valbool);
800 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
804 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
808 sk->sk_no_check_tx = valbool;
812 if ((val >= 0 && val <= 6) ||
813 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
814 sk->sk_priority = val;
820 if (optlen < sizeof(ling)) {
821 ret = -EINVAL; /* 1003.1g */
824 if (copy_from_user(&ling, optval, sizeof(ling))) {
829 sock_reset_flag(sk, SOCK_LINGER);
831 #if (BITS_PER_LONG == 32)
832 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
833 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
836 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
837 sock_set_flag(sk, SOCK_LINGER);
842 sock_warn_obsolete_bsdism("setsockopt");
847 set_bit(SOCK_PASSCRED, &sock->flags);
849 clear_bit(SOCK_PASSCRED, &sock->flags);
855 if (optname == SO_TIMESTAMP)
856 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
858 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
859 sock_set_flag(sk, SOCK_RCVTSTAMP);
860 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
862 sock_reset_flag(sk, SOCK_RCVTSTAMP);
863 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
867 case SO_TIMESTAMPING:
868 if (val & ~SOF_TIMESTAMPING_MASK) {
873 if (val & SOF_TIMESTAMPING_OPT_ID &&
874 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
875 if (sk->sk_protocol == IPPROTO_TCP) {
876 if (sk->sk_state != TCP_ESTABLISHED) {
880 sk->sk_tskey = tcp_sk(sk)->snd_una;
885 sk->sk_tsflags = val;
886 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
887 sock_enable_timestamp(sk,
888 SOCK_TIMESTAMPING_RX_SOFTWARE);
890 sock_disable_timestamp(sk,
891 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
897 sk->sk_rcvlowat = val ? : 1;
901 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
905 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
908 case SO_ATTACH_FILTER:
910 if (optlen == sizeof(struct sock_fprog)) {
911 struct sock_fprog fprog;
914 if (copy_from_user(&fprog, optval, sizeof(fprog)))
917 ret = sk_attach_filter(&fprog, sk);
923 if (optlen == sizeof(u32)) {
927 if (copy_from_user(&ufd, optval, sizeof(ufd)))
930 ret = sk_attach_bpf(ufd, sk);
934 case SO_DETACH_FILTER:
935 ret = sk_detach_filter(sk);
939 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
942 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
947 set_bit(SOCK_PASSSEC, &sock->flags);
949 clear_bit(SOCK_PASSSEC, &sock->flags);
952 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
959 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
963 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
967 if (sock->ops->set_peek_off)
968 ret = sock->ops->set_peek_off(sk, val);
974 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
977 case SO_SELECT_ERR_QUEUE:
978 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
981 #ifdef CONFIG_NET_RX_BUSY_POLL
983 /* allow unprivileged users to decrease the value */
984 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
990 sk->sk_ll_usec = val;
995 case SO_MAX_PACING_RATE:
996 sk->sk_max_pacing_rate = val;
997 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
998 sk->sk_max_pacing_rate);
1001 case SO_INCOMING_CPU:
1002 sk->sk_incoming_cpu = val;
1012 EXPORT_SYMBOL(sock_setsockopt);
1015 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1016 struct ucred *ucred)
1018 ucred->pid = pid_vnr(pid);
1019 ucred->uid = ucred->gid = -1;
1021 struct user_namespace *current_ns = current_user_ns();
1023 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1024 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1028 int sock_getsockopt(struct socket *sock, int level, int optname,
1029 char __user *optval, int __user *optlen)
1031 struct sock *sk = sock->sk;
1039 int lv = sizeof(int);
1042 if (get_user(len, optlen))
1047 memset(&v, 0, sizeof(v));
1051 v.val = sock_flag(sk, SOCK_DBG);
1055 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1059 v.val = sock_flag(sk, SOCK_BROADCAST);
1063 v.val = sk->sk_sndbuf;
1067 v.val = sk->sk_rcvbuf;
1071 v.val = sk->sk_reuse;
1075 v.val = sk->sk_reuseport;
1079 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1083 v.val = sk->sk_type;
1087 v.val = sk->sk_protocol;
1091 v.val = sk->sk_family;
1095 v.val = -sock_error(sk);
1097 v.val = xchg(&sk->sk_err_soft, 0);
1101 v.val = sock_flag(sk, SOCK_URGINLINE);
1105 v.val = sk->sk_no_check_tx;
1109 v.val = sk->sk_priority;
1113 lv = sizeof(v.ling);
1114 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1115 v.ling.l_linger = sk->sk_lingertime / HZ;
1119 sock_warn_obsolete_bsdism("getsockopt");
1123 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1124 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1127 case SO_TIMESTAMPNS:
1128 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1131 case SO_TIMESTAMPING:
1132 v.val = sk->sk_tsflags;
1136 lv = sizeof(struct timeval);
1137 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1141 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1142 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1147 lv = sizeof(struct timeval);
1148 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1152 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1153 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1158 v.val = sk->sk_rcvlowat;
1166 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1171 struct ucred peercred;
1172 if (len > sizeof(peercred))
1173 len = sizeof(peercred);
1174 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1175 if (copy_to_user(optval, &peercred, len))
1184 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1188 if (copy_to_user(optval, address, len))
1193 /* Dubious BSD thing... Probably nobody even uses it, but
1194 * the UNIX standard wants it for whatever reason... -DaveM
1197 v.val = sk->sk_state == TCP_LISTEN;
1201 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1205 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1208 v.val = sk->sk_mark;
1212 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1215 case SO_WIFI_STATUS:
1216 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1220 if (!sock->ops->set_peek_off)
1223 v.val = sk->sk_peek_off;
1226 v.val = sock_flag(sk, SOCK_NOFCS);
1229 case SO_BINDTODEVICE:
1230 return sock_getbindtodevice(sk, optval, optlen, len);
1233 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1239 case SO_LOCK_FILTER:
1240 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1243 case SO_BPF_EXTENSIONS:
1244 v.val = bpf_tell_extensions();
1247 case SO_SELECT_ERR_QUEUE:
1248 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1251 #ifdef CONFIG_NET_RX_BUSY_POLL
1253 v.val = sk->sk_ll_usec;
1257 case SO_MAX_PACING_RATE:
1258 v.val = sk->sk_max_pacing_rate;
1261 case SO_INCOMING_CPU:
1262 v.val = sk->sk_incoming_cpu;
1266 /* We implement the SO_SNDLOWAT etc to not be settable
1269 return -ENOPROTOOPT;
1274 if (copy_to_user(optval, &v, len))
1277 if (put_user(len, optlen))
1283 * Initialize an sk_lock.
1285 * (We also register the sk_lock with the lock validator.)
1287 static inline void sock_lock_init(struct sock *sk)
1289 sock_lock_init_class_and_name(sk,
1290 af_family_slock_key_strings[sk->sk_family],
1291 af_family_slock_keys + sk->sk_family,
1292 af_family_key_strings[sk->sk_family],
1293 af_family_keys + sk->sk_family);
1297 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1298 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1299 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1301 static void sock_copy(struct sock *nsk, const struct sock *osk)
1303 #ifdef CONFIG_SECURITY_NETWORK
1304 void *sptr = nsk->sk_security;
1306 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1308 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1309 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1311 #ifdef CONFIG_SECURITY_NETWORK
1312 nsk->sk_security = sptr;
1313 security_sk_clone(osk, nsk);
1317 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1319 unsigned long nulls1, nulls2;
1321 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1322 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1323 if (nulls1 > nulls2)
1324 swap(nulls1, nulls2);
1327 memset((char *)sk, 0, nulls1);
1328 memset((char *)sk + nulls1 + sizeof(void *), 0,
1329 nulls2 - nulls1 - sizeof(void *));
1330 memset((char *)sk + nulls2 + sizeof(void *), 0,
1331 size - nulls2 - sizeof(void *));
1333 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1335 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1339 struct kmem_cache *slab;
1343 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1346 if (priority & __GFP_ZERO) {
1348 prot->clear_sk(sk, prot->obj_size);
1350 sk_prot_clear_nulls(sk, prot->obj_size);
1353 sk = kmalloc(prot->obj_size, priority);
1356 kmemcheck_annotate_bitfield(sk, flags);
1358 if (security_sk_alloc(sk, family, priority))
1361 if (!try_module_get(prot->owner))
1363 sk_tx_queue_clear(sk);
1369 security_sk_free(sk);
1372 kmem_cache_free(slab, sk);
1378 static void sk_prot_free(struct proto *prot, struct sock *sk)
1380 struct kmem_cache *slab;
1381 struct module *owner;
1383 owner = prot->owner;
1386 security_sk_free(sk);
1388 kmem_cache_free(slab, sk);
1394 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1395 void sock_update_netprioidx(struct sock *sk)
1400 sk->sk_cgrp_prioidx = task_netprioidx(current);
1402 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1406 * sk_alloc - All socket objects are allocated here
1407 * @net: the applicable net namespace
1408 * @family: protocol family
1409 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1410 * @prot: struct proto associated with this new sock instance
1411 * @kern: is this to be a kernel socket?
1413 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1414 struct proto *prot, int kern)
1418 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1420 sk->sk_family = family;
1422 * See comment in struct sock definition to understand
1423 * why we need sk_prot_creator -acme
1425 sk->sk_prot = sk->sk_prot_creator = prot;
1427 sk->sk_net_refcnt = kern ? 0 : 1;
1428 if (likely(sk->sk_net_refcnt))
1430 sock_net_set(sk, net);
1431 atomic_set(&sk->sk_wmem_alloc, 1);
1433 sock_update_classid(sk);
1434 sock_update_netprioidx(sk);
1439 EXPORT_SYMBOL(sk_alloc);
1441 void sk_destruct(struct sock *sk)
1443 struct sk_filter *filter;
1445 if (sk->sk_destruct)
1446 sk->sk_destruct(sk);
1448 filter = rcu_dereference_check(sk->sk_filter,
1449 atomic_read(&sk->sk_wmem_alloc) == 0);
1451 sk_filter_uncharge(sk, filter);
1452 RCU_INIT_POINTER(sk->sk_filter, NULL);
1455 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1457 if (atomic_read(&sk->sk_omem_alloc))
1458 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1459 __func__, atomic_read(&sk->sk_omem_alloc));
1461 if (sk->sk_peer_cred)
1462 put_cred(sk->sk_peer_cred);
1463 put_pid(sk->sk_peer_pid);
1464 if (likely(sk->sk_net_refcnt))
1465 put_net(sock_net(sk));
1466 sk_prot_free(sk->sk_prot_creator, sk);
1469 static void __sk_free(struct sock *sk)
1471 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1472 sock_diag_broadcast_destroy(sk);
1477 void sk_free(struct sock *sk)
1480 * We subtract one from sk_wmem_alloc and can know if
1481 * some packets are still in some tx queue.
1482 * If not null, sock_wfree() will call __sk_free(sk) later
1484 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1487 EXPORT_SYMBOL(sk_free);
1489 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1491 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1492 sock_update_memcg(newsk);
1496 * sk_clone_lock - clone a socket, and lock its clone
1497 * @sk: the socket to clone
1498 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1500 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1502 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1505 bool is_charged = true;
1507 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1508 if (newsk != NULL) {
1509 struct sk_filter *filter;
1511 sock_copy(newsk, sk);
1514 if (likely(newsk->sk_net_refcnt))
1515 get_net(sock_net(newsk));
1516 sk_node_init(&newsk->sk_node);
1517 sock_lock_init(newsk);
1518 bh_lock_sock(newsk);
1519 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1520 newsk->sk_backlog.len = 0;
1522 atomic_set(&newsk->sk_rmem_alloc, 0);
1524 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1526 atomic_set(&newsk->sk_wmem_alloc, 1);
1527 atomic_set(&newsk->sk_omem_alloc, 0);
1528 skb_queue_head_init(&newsk->sk_receive_queue);
1529 skb_queue_head_init(&newsk->sk_write_queue);
1531 rwlock_init(&newsk->sk_callback_lock);
1532 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1533 af_callback_keys + newsk->sk_family,
1534 af_family_clock_key_strings[newsk->sk_family]);
1536 newsk->sk_dst_cache = NULL;
1537 newsk->sk_wmem_queued = 0;
1538 newsk->sk_forward_alloc = 0;
1539 newsk->sk_send_head = NULL;
1540 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1542 sock_reset_flag(newsk, SOCK_DONE);
1543 skb_queue_head_init(&newsk->sk_error_queue);
1545 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1547 /* though it's an empty new sock, the charging may fail
1548 * if sysctl_optmem_max was changed between creation of
1549 * original socket and cloning
1551 is_charged = sk_filter_charge(newsk, filter);
1553 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1554 /* It is still raw copy of parent, so invalidate
1555 * destructor and make plain sk_free() */
1556 newsk->sk_destruct = NULL;
1557 bh_unlock_sock(newsk);
1564 newsk->sk_priority = 0;
1565 newsk->sk_incoming_cpu = raw_smp_processor_id();
1566 atomic64_set(&newsk->sk_cookie, 0);
1568 * Before updating sk_refcnt, we must commit prior changes to memory
1569 * (Documentation/RCU/rculist_nulls.txt for details)
1572 atomic_set(&newsk->sk_refcnt, 2);
1575 * Increment the counter in the same struct proto as the master
1576 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1577 * is the same as sk->sk_prot->socks, as this field was copied
1580 * This _changes_ the previous behaviour, where
1581 * tcp_create_openreq_child always was incrementing the
1582 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1583 * to be taken into account in all callers. -acme
1585 sk_refcnt_debug_inc(newsk);
1586 sk_set_socket(newsk, NULL);
1587 newsk->sk_wq = NULL;
1589 sk_update_clone(sk, newsk);
1591 if (newsk->sk_prot->sockets_allocated)
1592 sk_sockets_allocated_inc(newsk);
1594 if (sock_needs_netstamp(sk) &&
1595 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1596 net_enable_timestamp();
1601 EXPORT_SYMBOL_GPL(sk_clone_lock);
1603 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1607 sk_dst_set(sk, dst);
1608 sk->sk_route_caps = dst->dev->features;
1609 if (sk->sk_route_caps & NETIF_F_GSO)
1610 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1611 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1612 if (sk_can_gso(sk)) {
1613 if (dst->header_len) {
1614 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1616 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1617 sk->sk_gso_max_size = dst->dev->gso_max_size;
1618 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1621 sk->sk_gso_max_segs = max_segs;
1623 EXPORT_SYMBOL_GPL(sk_setup_caps);
1626 * Simple resource managers for sockets.
1631 * Write buffer destructor automatically called from kfree_skb.
1633 void sock_wfree(struct sk_buff *skb)
1635 struct sock *sk = skb->sk;
1636 unsigned int len = skb->truesize;
1638 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1640 * Keep a reference on sk_wmem_alloc, this will be released
1641 * after sk_write_space() call
1643 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1644 sk->sk_write_space(sk);
1648 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1649 * could not do because of in-flight packets
1651 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1654 EXPORT_SYMBOL(sock_wfree);
1656 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1661 if (unlikely(!sk_fullsock(sk))) {
1662 skb->destructor = sock_edemux;
1667 skb->destructor = sock_wfree;
1668 skb_set_hash_from_sk(skb, sk);
1670 * We used to take a refcount on sk, but following operation
1671 * is enough to guarantee sk_free() wont free this sock until
1672 * all in-flight packets are completed
1674 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1676 EXPORT_SYMBOL(skb_set_owner_w);
1678 void skb_orphan_partial(struct sk_buff *skb)
1680 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1681 * so we do not completely orphan skb, but transfert all
1682 * accounted bytes but one, to avoid unexpected reorders.
1684 if (skb->destructor == sock_wfree
1686 || skb->destructor == tcp_wfree
1689 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1695 EXPORT_SYMBOL(skb_orphan_partial);
1698 * Read buffer destructor automatically called from kfree_skb.
1700 void sock_rfree(struct sk_buff *skb)
1702 struct sock *sk = skb->sk;
1703 unsigned int len = skb->truesize;
1705 atomic_sub(len, &sk->sk_rmem_alloc);
1706 sk_mem_uncharge(sk, len);
1708 EXPORT_SYMBOL(sock_rfree);
1711 * Buffer destructor for skbs that are not used directly in read or write
1712 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1714 void sock_efree(struct sk_buff *skb)
1718 EXPORT_SYMBOL(sock_efree);
1720 kuid_t sock_i_uid(struct sock *sk)
1724 read_lock_bh(&sk->sk_callback_lock);
1725 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1726 read_unlock_bh(&sk->sk_callback_lock);
1729 EXPORT_SYMBOL(sock_i_uid);
1731 unsigned long sock_i_ino(struct sock *sk)
1735 read_lock_bh(&sk->sk_callback_lock);
1736 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1737 read_unlock_bh(&sk->sk_callback_lock);
1740 EXPORT_SYMBOL(sock_i_ino);
1743 * Allocate a skb from the socket's send buffer.
1745 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1748 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1749 struct sk_buff *skb = alloc_skb(size, priority);
1751 skb_set_owner_w(skb, sk);
1757 EXPORT_SYMBOL(sock_wmalloc);
1760 * Allocate a memory block from the socket's option memory buffer.
1762 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1764 if ((unsigned int)size <= sysctl_optmem_max &&
1765 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1767 /* First do the add, to avoid the race if kmalloc
1770 atomic_add(size, &sk->sk_omem_alloc);
1771 mem = kmalloc(size, priority);
1774 atomic_sub(size, &sk->sk_omem_alloc);
1778 EXPORT_SYMBOL(sock_kmalloc);
1780 /* Free an option memory block. Note, we actually want the inline
1781 * here as this allows gcc to detect the nullify and fold away the
1782 * condition entirely.
1784 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1787 if (WARN_ON_ONCE(!mem))
1793 atomic_sub(size, &sk->sk_omem_alloc);
1796 void sock_kfree_s(struct sock *sk, void *mem, int size)
1798 __sock_kfree_s(sk, mem, size, false);
1800 EXPORT_SYMBOL(sock_kfree_s);
1802 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1804 __sock_kfree_s(sk, mem, size, true);
1806 EXPORT_SYMBOL(sock_kzfree_s);
1808 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1809 I think, these locks should be removed for datagram sockets.
1811 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1815 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1819 if (signal_pending(current))
1821 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1822 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1823 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1825 if (sk->sk_shutdown & SEND_SHUTDOWN)
1829 timeo = schedule_timeout(timeo);
1831 finish_wait(sk_sleep(sk), &wait);
1837 * Generic send/receive buffer handlers
1840 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1841 unsigned long data_len, int noblock,
1842 int *errcode, int max_page_order)
1844 struct sk_buff *skb;
1848 timeo = sock_sndtimeo(sk, noblock);
1850 err = sock_error(sk);
1855 if (sk->sk_shutdown & SEND_SHUTDOWN)
1858 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1861 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1862 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1866 if (signal_pending(current))
1868 timeo = sock_wait_for_wmem(sk, timeo);
1870 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1871 errcode, sk->sk_allocation);
1873 skb_set_owner_w(skb, sk);
1877 err = sock_intr_errno(timeo);
1882 EXPORT_SYMBOL(sock_alloc_send_pskb);
1884 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1885 int noblock, int *errcode)
1887 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1889 EXPORT_SYMBOL(sock_alloc_send_skb);
1891 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1892 struct sockcm_cookie *sockc)
1894 struct cmsghdr *cmsg;
1896 for_each_cmsghdr(cmsg, msg) {
1897 if (!CMSG_OK(msg, cmsg))
1899 if (cmsg->cmsg_level != SOL_SOCKET)
1901 switch (cmsg->cmsg_type) {
1903 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1905 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1907 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1915 EXPORT_SYMBOL(sock_cmsg_send);
1917 /* On 32bit arches, an skb frag is limited to 2^15 */
1918 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1921 * skb_page_frag_refill - check that a page_frag contains enough room
1922 * @sz: minimum size of the fragment we want to get
1923 * @pfrag: pointer to page_frag
1924 * @gfp: priority for memory allocation
1926 * Note: While this allocator tries to use high order pages, there is
1927 * no guarantee that allocations succeed. Therefore, @sz MUST be
1928 * less or equal than PAGE_SIZE.
1930 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1933 if (atomic_read(&pfrag->page->_count) == 1) {
1937 if (pfrag->offset + sz <= pfrag->size)
1939 put_page(pfrag->page);
1943 if (SKB_FRAG_PAGE_ORDER) {
1944 /* Avoid direct reclaim but allow kswapd to wake */
1945 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1946 __GFP_COMP | __GFP_NOWARN |
1948 SKB_FRAG_PAGE_ORDER);
1949 if (likely(pfrag->page)) {
1950 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1954 pfrag->page = alloc_page(gfp);
1955 if (likely(pfrag->page)) {
1956 pfrag->size = PAGE_SIZE;
1961 EXPORT_SYMBOL(skb_page_frag_refill);
1963 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1965 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1968 sk_enter_memory_pressure(sk);
1969 sk_stream_moderate_sndbuf(sk);
1972 EXPORT_SYMBOL(sk_page_frag_refill);
1974 static void __lock_sock(struct sock *sk)
1975 __releases(&sk->sk_lock.slock)
1976 __acquires(&sk->sk_lock.slock)
1981 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1982 TASK_UNINTERRUPTIBLE);
1983 spin_unlock_bh(&sk->sk_lock.slock);
1985 spin_lock_bh(&sk->sk_lock.slock);
1986 if (!sock_owned_by_user(sk))
1989 finish_wait(&sk->sk_lock.wq, &wait);
1992 static void __release_sock(struct sock *sk)
1993 __releases(&sk->sk_lock.slock)
1994 __acquires(&sk->sk_lock.slock)
1996 struct sk_buff *skb = sk->sk_backlog.head;
1999 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2003 struct sk_buff *next = skb->next;
2006 WARN_ON_ONCE(skb_dst_is_noref(skb));
2008 sk_backlog_rcv(sk, skb);
2011 * We are in process context here with softirqs
2012 * disabled, use cond_resched_softirq() to preempt.
2013 * This is safe to do because we've taken the backlog
2016 cond_resched_softirq();
2019 } while (skb != NULL);
2022 } while ((skb = sk->sk_backlog.head) != NULL);
2025 * Doing the zeroing here guarantee we can not loop forever
2026 * while a wild producer attempts to flood us.
2028 sk->sk_backlog.len = 0;
2032 * sk_wait_data - wait for data to arrive at sk_receive_queue
2033 * @sk: sock to wait on
2034 * @timeo: for how long
2035 * @skb: last skb seen on sk_receive_queue
2037 * Now socket state including sk->sk_err is changed only under lock,
2038 * hence we may omit checks after joining wait queue.
2039 * We check receive queue before schedule() only as optimization;
2040 * it is very likely that release_sock() added new data.
2042 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2047 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2048 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2049 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2050 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2051 finish_wait(sk_sleep(sk), &wait);
2054 EXPORT_SYMBOL(sk_wait_data);
2057 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2059 * @size: memory size to allocate
2060 * @kind: allocation type
2062 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2063 * rmem allocation. This function assumes that protocols which have
2064 * memory_pressure use sk_wmem_queued as write buffer accounting.
2066 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2068 struct proto *prot = sk->sk_prot;
2069 int amt = sk_mem_pages(size);
2071 int parent_status = UNDER_LIMIT;
2073 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2075 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2078 if (parent_status == UNDER_LIMIT &&
2079 allocated <= sk_prot_mem_limits(sk, 0)) {
2080 sk_leave_memory_pressure(sk);
2084 /* Under pressure. (we or our parents) */
2085 if ((parent_status > SOFT_LIMIT) ||
2086 allocated > sk_prot_mem_limits(sk, 1))
2087 sk_enter_memory_pressure(sk);
2089 /* Over hard limit (we or our parents) */
2090 if ((parent_status == OVER_LIMIT) ||
2091 (allocated > sk_prot_mem_limits(sk, 2)))
2092 goto suppress_allocation;
2094 /* guarantee minimum buffer size under pressure */
2095 if (kind == SK_MEM_RECV) {
2096 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2099 } else { /* SK_MEM_SEND */
2100 if (sk->sk_type == SOCK_STREAM) {
2101 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2103 } else if (atomic_read(&sk->sk_wmem_alloc) <
2104 prot->sysctl_wmem[0])
2108 if (sk_has_memory_pressure(sk)) {
2111 if (!sk_under_memory_pressure(sk))
2113 alloc = sk_sockets_allocated_read_positive(sk);
2114 if (sk_prot_mem_limits(sk, 2) > alloc *
2115 sk_mem_pages(sk->sk_wmem_queued +
2116 atomic_read(&sk->sk_rmem_alloc) +
2117 sk->sk_forward_alloc))
2121 suppress_allocation:
2123 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2124 sk_stream_moderate_sndbuf(sk);
2126 /* Fail only if socket is _under_ its sndbuf.
2127 * In this case we cannot block, so that we have to fail.
2129 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2133 trace_sock_exceed_buf_limit(sk, prot, allocated);
2135 /* Alas. Undo changes. */
2136 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2138 sk_memory_allocated_sub(sk, amt);
2142 EXPORT_SYMBOL(__sk_mem_schedule);
2145 * __sk_mem_reclaim - reclaim memory_allocated
2147 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2149 void __sk_mem_reclaim(struct sock *sk, int amount)
2151 amount >>= SK_MEM_QUANTUM_SHIFT;
2152 sk_memory_allocated_sub(sk, amount);
2153 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2155 if (sk_under_memory_pressure(sk) &&
2156 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2157 sk_leave_memory_pressure(sk);
2159 EXPORT_SYMBOL(__sk_mem_reclaim);
2163 * Set of default routines for initialising struct proto_ops when
2164 * the protocol does not support a particular function. In certain
2165 * cases where it makes no sense for a protocol to have a "do nothing"
2166 * function, some default processing is provided.
2169 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2173 EXPORT_SYMBOL(sock_no_bind);
2175 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2180 EXPORT_SYMBOL(sock_no_connect);
2182 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2186 EXPORT_SYMBOL(sock_no_socketpair);
2188 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2192 EXPORT_SYMBOL(sock_no_accept);
2194 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2199 EXPORT_SYMBOL(sock_no_getname);
2201 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2205 EXPORT_SYMBOL(sock_no_poll);
2207 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2211 EXPORT_SYMBOL(sock_no_ioctl);
2213 int sock_no_listen(struct socket *sock, int backlog)
2217 EXPORT_SYMBOL(sock_no_listen);
2219 int sock_no_shutdown(struct socket *sock, int how)
2223 EXPORT_SYMBOL(sock_no_shutdown);
2225 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2226 char __user *optval, unsigned int optlen)
2230 EXPORT_SYMBOL(sock_no_setsockopt);
2232 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2233 char __user *optval, int __user *optlen)
2237 EXPORT_SYMBOL(sock_no_getsockopt);
2239 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2243 EXPORT_SYMBOL(sock_no_sendmsg);
2245 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2250 EXPORT_SYMBOL(sock_no_recvmsg);
2252 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2254 /* Mirror missing mmap method error code */
2257 EXPORT_SYMBOL(sock_no_mmap);
2259 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2262 struct msghdr msg = {.msg_flags = flags};
2264 char *kaddr = kmap(page);
2265 iov.iov_base = kaddr + offset;
2267 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2271 EXPORT_SYMBOL(sock_no_sendpage);
2274 * Default Socket Callbacks
2277 static void sock_def_wakeup(struct sock *sk)
2279 struct socket_wq *wq;
2282 wq = rcu_dereference(sk->sk_wq);
2283 if (wq_has_sleeper(wq))
2284 wake_up_interruptible_all(&wq->wait);
2288 static void sock_def_error_report(struct sock *sk)
2290 struct socket_wq *wq;
2293 wq = rcu_dereference(sk->sk_wq);
2294 if (wq_has_sleeper(wq))
2295 wake_up_interruptible_poll(&wq->wait, POLLERR);
2296 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2300 static void sock_def_readable(struct sock *sk)
2302 struct socket_wq *wq;
2305 wq = rcu_dereference(sk->sk_wq);
2306 if (wq_has_sleeper(wq))
2307 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2308 POLLRDNORM | POLLRDBAND);
2309 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2313 static void sock_def_write_space(struct sock *sk)
2315 struct socket_wq *wq;
2319 /* Do not wake up a writer until he can make "significant"
2322 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2323 wq = rcu_dereference(sk->sk_wq);
2324 if (wq_has_sleeper(wq))
2325 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2326 POLLWRNORM | POLLWRBAND);
2328 /* Should agree with poll, otherwise some programs break */
2329 if (sock_writeable(sk))
2330 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2336 static void sock_def_destruct(struct sock *sk)
2340 void sk_send_sigurg(struct sock *sk)
2342 if (sk->sk_socket && sk->sk_socket->file)
2343 if (send_sigurg(&sk->sk_socket->file->f_owner))
2344 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2346 EXPORT_SYMBOL(sk_send_sigurg);
2348 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2349 unsigned long expires)
2351 if (!mod_timer(timer, expires))
2354 EXPORT_SYMBOL(sk_reset_timer);
2356 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2358 if (del_timer(timer))
2361 EXPORT_SYMBOL(sk_stop_timer);
2363 void sock_init_data(struct socket *sock, struct sock *sk)
2365 skb_queue_head_init(&sk->sk_receive_queue);
2366 skb_queue_head_init(&sk->sk_write_queue);
2367 skb_queue_head_init(&sk->sk_error_queue);
2369 sk->sk_send_head = NULL;
2371 init_timer(&sk->sk_timer);
2373 sk->sk_allocation = GFP_KERNEL;
2374 sk->sk_rcvbuf = sysctl_rmem_default;
2375 sk->sk_sndbuf = sysctl_wmem_default;
2376 sk->sk_state = TCP_CLOSE;
2377 sk_set_socket(sk, sock);
2379 sock_set_flag(sk, SOCK_ZAPPED);
2382 sk->sk_type = sock->type;
2383 sk->sk_wq = sock->wq;
2388 rwlock_init(&sk->sk_callback_lock);
2389 lockdep_set_class_and_name(&sk->sk_callback_lock,
2390 af_callback_keys + sk->sk_family,
2391 af_family_clock_key_strings[sk->sk_family]);
2393 sk->sk_state_change = sock_def_wakeup;
2394 sk->sk_data_ready = sock_def_readable;
2395 sk->sk_write_space = sock_def_write_space;
2396 sk->sk_error_report = sock_def_error_report;
2397 sk->sk_destruct = sock_def_destruct;
2399 sk->sk_frag.page = NULL;
2400 sk->sk_frag.offset = 0;
2401 sk->sk_peek_off = -1;
2403 sk->sk_peer_pid = NULL;
2404 sk->sk_peer_cred = NULL;
2405 sk->sk_write_pending = 0;
2406 sk->sk_rcvlowat = 1;
2407 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2408 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2410 sk->sk_stamp = ktime_set(-1L, 0);
2412 #ifdef CONFIG_NET_RX_BUSY_POLL
2414 sk->sk_ll_usec = sysctl_net_busy_read;
2417 sk->sk_max_pacing_rate = ~0U;
2418 sk->sk_pacing_rate = ~0U;
2419 sk->sk_incoming_cpu = -1;
2421 * Before updating sk_refcnt, we must commit prior changes to memory
2422 * (Documentation/RCU/rculist_nulls.txt for details)
2425 atomic_set(&sk->sk_refcnt, 1);
2426 atomic_set(&sk->sk_drops, 0);
2428 EXPORT_SYMBOL(sock_init_data);
2430 void lock_sock_nested(struct sock *sk, int subclass)
2433 spin_lock_bh(&sk->sk_lock.slock);
2434 if (sk->sk_lock.owned)
2436 sk->sk_lock.owned = 1;
2437 spin_unlock(&sk->sk_lock.slock);
2439 * The sk_lock has mutex_lock() semantics here:
2441 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2444 EXPORT_SYMBOL(lock_sock_nested);
2446 void release_sock(struct sock *sk)
2449 * The sk_lock has mutex_unlock() semantics:
2451 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2453 spin_lock_bh(&sk->sk_lock.slock);
2454 if (sk->sk_backlog.tail)
2457 /* Warning : release_cb() might need to release sk ownership,
2458 * ie call sock_release_ownership(sk) before us.
2460 if (sk->sk_prot->release_cb)
2461 sk->sk_prot->release_cb(sk);
2463 sock_release_ownership(sk);
2464 if (waitqueue_active(&sk->sk_lock.wq))
2465 wake_up(&sk->sk_lock.wq);
2466 spin_unlock_bh(&sk->sk_lock.slock);
2468 EXPORT_SYMBOL(release_sock);
2471 * lock_sock_fast - fast version of lock_sock
2474 * This version should be used for very small section, where process wont block
2475 * return false if fast path is taken
2476 * sk_lock.slock locked, owned = 0, BH disabled
2477 * return true if slow path is taken
2478 * sk_lock.slock unlocked, owned = 1, BH enabled
2480 bool lock_sock_fast(struct sock *sk)
2483 spin_lock_bh(&sk->sk_lock.slock);
2485 if (!sk->sk_lock.owned)
2487 * Note : We must disable BH
2492 sk->sk_lock.owned = 1;
2493 spin_unlock(&sk->sk_lock.slock);
2495 * The sk_lock has mutex_lock() semantics here:
2497 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2501 EXPORT_SYMBOL(lock_sock_fast);
2503 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2506 if (!sock_flag(sk, SOCK_TIMESTAMP))
2507 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2508 tv = ktime_to_timeval(sk->sk_stamp);
2509 if (tv.tv_sec == -1)
2511 if (tv.tv_sec == 0) {
2512 sk->sk_stamp = ktime_get_real();
2513 tv = ktime_to_timeval(sk->sk_stamp);
2515 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2517 EXPORT_SYMBOL(sock_get_timestamp);
2519 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2522 if (!sock_flag(sk, SOCK_TIMESTAMP))
2523 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2524 ts = ktime_to_timespec(sk->sk_stamp);
2525 if (ts.tv_sec == -1)
2527 if (ts.tv_sec == 0) {
2528 sk->sk_stamp = ktime_get_real();
2529 ts = ktime_to_timespec(sk->sk_stamp);
2531 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2533 EXPORT_SYMBOL(sock_get_timestampns);
2535 void sock_enable_timestamp(struct sock *sk, int flag)
2537 if (!sock_flag(sk, flag)) {
2538 unsigned long previous_flags = sk->sk_flags;
2540 sock_set_flag(sk, flag);
2542 * we just set one of the two flags which require net
2543 * time stamping, but time stamping might have been on
2544 * already because of the other one
2546 if (sock_needs_netstamp(sk) &&
2547 !(previous_flags & SK_FLAGS_TIMESTAMP))
2548 net_enable_timestamp();
2552 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2553 int level, int type)
2555 struct sock_exterr_skb *serr;
2556 struct sk_buff *skb;
2560 skb = sock_dequeue_err_skb(sk);
2566 msg->msg_flags |= MSG_TRUNC;
2569 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2573 sock_recv_timestamp(msg, sk, skb);
2575 serr = SKB_EXT_ERR(skb);
2576 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2578 msg->msg_flags |= MSG_ERRQUEUE;
2586 EXPORT_SYMBOL(sock_recv_errqueue);
2589 * Get a socket option on an socket.
2591 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2592 * asynchronous errors should be reported by getsockopt. We assume
2593 * this means if you specify SO_ERROR (otherwise whats the point of it).
2595 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2596 char __user *optval, int __user *optlen)
2598 struct sock *sk = sock->sk;
2600 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2602 EXPORT_SYMBOL(sock_common_getsockopt);
2604 #ifdef CONFIG_COMPAT
2605 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2606 char __user *optval, int __user *optlen)
2608 struct sock *sk = sock->sk;
2610 if (sk->sk_prot->compat_getsockopt != NULL)
2611 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2613 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2615 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2618 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2621 struct sock *sk = sock->sk;
2625 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2626 flags & ~MSG_DONTWAIT, &addr_len);
2628 msg->msg_namelen = addr_len;
2631 EXPORT_SYMBOL(sock_common_recvmsg);
2634 * Set socket options on an inet socket.
2636 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2637 char __user *optval, unsigned int optlen)
2639 struct sock *sk = sock->sk;
2641 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2643 EXPORT_SYMBOL(sock_common_setsockopt);
2645 #ifdef CONFIG_COMPAT
2646 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2647 char __user *optval, unsigned int optlen)
2649 struct sock *sk = sock->sk;
2651 if (sk->sk_prot->compat_setsockopt != NULL)
2652 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2654 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2656 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2659 void sk_common_release(struct sock *sk)
2661 if (sk->sk_prot->destroy)
2662 sk->sk_prot->destroy(sk);
2665 * Observation: when sock_common_release is called, processes have
2666 * no access to socket. But net still has.
2667 * Step one, detach it from networking:
2669 * A. Remove from hash tables.
2672 sk->sk_prot->unhash(sk);
2675 * In this point socket cannot receive new packets, but it is possible
2676 * that some packets are in flight because some CPU runs receiver and
2677 * did hash table lookup before we unhashed socket. They will achieve
2678 * receive queue and will be purged by socket destructor.
2680 * Also we still have packets pending on receive queue and probably,
2681 * our own packets waiting in device queues. sock_destroy will drain
2682 * receive queue, but transmitted packets will delay socket destruction
2683 * until the last reference will be released.
2688 xfrm_sk_free_policy(sk);
2690 sk_refcnt_debug_release(sk);
2692 if (sk->sk_frag.page) {
2693 put_page(sk->sk_frag.page);
2694 sk->sk_frag.page = NULL;
2699 EXPORT_SYMBOL(sk_common_release);
2701 #ifdef CONFIG_PROC_FS
2702 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2704 int val[PROTO_INUSE_NR];
2707 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2709 #ifdef CONFIG_NET_NS
2710 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2712 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2714 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2716 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2718 int cpu, idx = prot->inuse_idx;
2721 for_each_possible_cpu(cpu)
2722 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2724 return res >= 0 ? res : 0;
2726 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2728 static int __net_init sock_inuse_init_net(struct net *net)
2730 net->core.inuse = alloc_percpu(struct prot_inuse);
2731 return net->core.inuse ? 0 : -ENOMEM;
2734 static void __net_exit sock_inuse_exit_net(struct net *net)
2736 free_percpu(net->core.inuse);
2739 static struct pernet_operations net_inuse_ops = {
2740 .init = sock_inuse_init_net,
2741 .exit = sock_inuse_exit_net,
2744 static __init int net_inuse_init(void)
2746 if (register_pernet_subsys(&net_inuse_ops))
2747 panic("Cannot initialize net inuse counters");
2752 core_initcall(net_inuse_init);
2754 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2756 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2758 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2760 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2762 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2764 int cpu, idx = prot->inuse_idx;
2767 for_each_possible_cpu(cpu)
2768 res += per_cpu(prot_inuse, cpu).val[idx];
2770 return res >= 0 ? res : 0;
2772 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2775 static void assign_proto_idx(struct proto *prot)
2777 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2779 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2780 pr_err("PROTO_INUSE_NR exhausted\n");
2784 set_bit(prot->inuse_idx, proto_inuse_idx);
2787 static void release_proto_idx(struct proto *prot)
2789 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2790 clear_bit(prot->inuse_idx, proto_inuse_idx);
2793 static inline void assign_proto_idx(struct proto *prot)
2797 static inline void release_proto_idx(struct proto *prot)
2802 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2806 kfree(rsk_prot->slab_name);
2807 rsk_prot->slab_name = NULL;
2808 kmem_cache_destroy(rsk_prot->slab);
2809 rsk_prot->slab = NULL;
2812 static int req_prot_init(const struct proto *prot)
2814 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2819 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2821 if (!rsk_prot->slab_name)
2824 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2825 rsk_prot->obj_size, 0,
2826 prot->slab_flags, NULL);
2828 if (!rsk_prot->slab) {
2829 pr_crit("%s: Can't create request sock SLAB cache!\n",
2836 int proto_register(struct proto *prot, int alloc_slab)
2839 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2840 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2843 if (prot->slab == NULL) {
2844 pr_crit("%s: Can't create sock SLAB cache!\n",
2849 if (req_prot_init(prot))
2850 goto out_free_request_sock_slab;
2852 if (prot->twsk_prot != NULL) {
2853 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2855 if (prot->twsk_prot->twsk_slab_name == NULL)
2856 goto out_free_request_sock_slab;
2858 prot->twsk_prot->twsk_slab =
2859 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2860 prot->twsk_prot->twsk_obj_size,
2864 if (prot->twsk_prot->twsk_slab == NULL)
2865 goto out_free_timewait_sock_slab_name;
2869 mutex_lock(&proto_list_mutex);
2870 list_add(&prot->node, &proto_list);
2871 assign_proto_idx(prot);
2872 mutex_unlock(&proto_list_mutex);
2875 out_free_timewait_sock_slab_name:
2876 kfree(prot->twsk_prot->twsk_slab_name);
2877 out_free_request_sock_slab:
2878 req_prot_cleanup(prot->rsk_prot);
2880 kmem_cache_destroy(prot->slab);
2885 EXPORT_SYMBOL(proto_register);
2887 void proto_unregister(struct proto *prot)
2889 mutex_lock(&proto_list_mutex);
2890 release_proto_idx(prot);
2891 list_del(&prot->node);
2892 mutex_unlock(&proto_list_mutex);
2894 kmem_cache_destroy(prot->slab);
2897 req_prot_cleanup(prot->rsk_prot);
2899 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2900 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2901 kfree(prot->twsk_prot->twsk_slab_name);
2902 prot->twsk_prot->twsk_slab = NULL;
2905 EXPORT_SYMBOL(proto_unregister);
2907 #ifdef CONFIG_PROC_FS
2908 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2909 __acquires(proto_list_mutex)
2911 mutex_lock(&proto_list_mutex);
2912 return seq_list_start_head(&proto_list, *pos);
2915 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2917 return seq_list_next(v, &proto_list, pos);
2920 static void proto_seq_stop(struct seq_file *seq, void *v)
2921 __releases(proto_list_mutex)
2923 mutex_unlock(&proto_list_mutex);
2926 static char proto_method_implemented(const void *method)
2928 return method == NULL ? 'n' : 'y';
2930 static long sock_prot_memory_allocated(struct proto *proto)
2932 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2935 static char *sock_prot_memory_pressure(struct proto *proto)
2937 return proto->memory_pressure != NULL ?
2938 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2941 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2944 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2945 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2948 sock_prot_inuse_get(seq_file_net(seq), proto),
2949 sock_prot_memory_allocated(proto),
2950 sock_prot_memory_pressure(proto),
2952 proto->slab == NULL ? "no" : "yes",
2953 module_name(proto->owner),
2954 proto_method_implemented(proto->close),
2955 proto_method_implemented(proto->connect),
2956 proto_method_implemented(proto->disconnect),
2957 proto_method_implemented(proto->accept),
2958 proto_method_implemented(proto->ioctl),
2959 proto_method_implemented(proto->init),
2960 proto_method_implemented(proto->destroy),
2961 proto_method_implemented(proto->shutdown),
2962 proto_method_implemented(proto->setsockopt),
2963 proto_method_implemented(proto->getsockopt),
2964 proto_method_implemented(proto->sendmsg),
2965 proto_method_implemented(proto->recvmsg),
2966 proto_method_implemented(proto->sendpage),
2967 proto_method_implemented(proto->bind),
2968 proto_method_implemented(proto->backlog_rcv),
2969 proto_method_implemented(proto->hash),
2970 proto_method_implemented(proto->unhash),
2971 proto_method_implemented(proto->get_port),
2972 proto_method_implemented(proto->enter_memory_pressure));
2975 static int proto_seq_show(struct seq_file *seq, void *v)
2977 if (v == &proto_list)
2978 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2987 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2989 proto_seq_printf(seq, list_entry(v, struct proto, node));
2993 static const struct seq_operations proto_seq_ops = {
2994 .start = proto_seq_start,
2995 .next = proto_seq_next,
2996 .stop = proto_seq_stop,
2997 .show = proto_seq_show,
3000 static int proto_seq_open(struct inode *inode, struct file *file)
3002 return seq_open_net(inode, file, &proto_seq_ops,
3003 sizeof(struct seq_net_private));
3006 static const struct file_operations proto_seq_fops = {
3007 .owner = THIS_MODULE,
3008 .open = proto_seq_open,
3010 .llseek = seq_lseek,
3011 .release = seq_release_net,
3014 static __net_init int proto_init_net(struct net *net)
3016 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3022 static __net_exit void proto_exit_net(struct net *net)
3024 remove_proc_entry("protocols", net->proc_net);
3028 static __net_initdata struct pernet_operations proto_net_ops = {
3029 .init = proto_init_net,
3030 .exit = proto_exit_net,
3033 static int __init proto_init(void)
3035 return register_pernet_subsys(&proto_net_ops);
3038 subsys_initcall(proto_init);
3040 #endif /* PROC_FS */