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/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
115 #include <linux/user_namespace.h>
116 #include <linux/static_key.h>
117 #include <linux/memcontrol.h>
118 #include <linux/prefetch.h>
120 #include <asm/uaccess.h>
122 #include <linux/netdevice.h>
123 #include <net/protocol.h>
124 #include <linux/skbuff.h>
125 #include <net/net_namespace.h>
126 #include <net/request_sock.h>
127 #include <net/sock.h>
128 #include <linux/net_tstamp.h>
129 #include <net/xfrm.h>
130 #include <linux/ipsec.h>
131 #include <net/cls_cgroup.h>
132 #include <net/netprio_cgroup.h>
134 #include <linux/filter.h>
136 #include <trace/events/sock.h>
142 #include <net/busy_poll.h>
144 static DEFINE_MUTEX(proto_list_mutex);
145 static LIST_HEAD(proto_list);
147 #ifdef CONFIG_MEMCG_KMEM
148 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
153 mutex_lock(&proto_list_mutex);
154 list_for_each_entry(proto, &proto_list, node) {
155 if (proto->init_cgroup) {
156 ret = proto->init_cgroup(memcg, ss);
162 mutex_unlock(&proto_list_mutex);
165 list_for_each_entry_continue_reverse(proto, &proto_list, node)
166 if (proto->destroy_cgroup)
167 proto->destroy_cgroup(memcg);
168 mutex_unlock(&proto_list_mutex);
172 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
176 mutex_lock(&proto_list_mutex);
177 list_for_each_entry_reverse(proto, &proto_list, node)
178 if (proto->destroy_cgroup)
179 proto->destroy_cgroup(memcg);
180 mutex_unlock(&proto_list_mutex);
185 * Each address family might have different locking rules, so we have
186 * one slock key per address family:
188 static struct lock_class_key af_family_keys[AF_MAX];
189 static struct lock_class_key af_family_slock_keys[AF_MAX];
191 #if defined(CONFIG_MEMCG_KMEM)
192 struct static_key memcg_socket_limit_enabled;
193 EXPORT_SYMBOL(memcg_socket_limit_enabled);
197 * Make lock validator output more readable. (we pre-construct these
198 * strings build-time, so that runtime initialization of socket
201 static const char *const af_family_key_strings[AF_MAX+1] = {
202 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
203 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
204 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
205 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
206 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
207 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
208 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
209 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
210 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
211 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
212 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
213 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
214 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
215 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
217 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
218 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
219 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
220 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
221 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
222 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
223 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
224 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
225 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
226 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
227 "slock-27" , "slock-28" , "slock-AF_CAN" ,
228 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
229 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
230 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
231 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
233 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
234 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
235 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
236 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
237 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
238 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
239 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
240 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
241 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
242 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
243 "clock-27" , "clock-28" , "clock-AF_CAN" ,
244 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
245 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
246 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
247 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
251 * sk_callback_lock locking rules are per-address-family,
252 * so split the lock classes by using a per-AF key:
254 static struct lock_class_key af_callback_keys[AF_MAX];
256 /* Take into consideration the size of the struct sk_buff overhead in the
257 * determination of these values, since that is non-constant across
258 * platforms. This makes socket queueing behavior and performance
259 * not depend upon such differences.
261 #define _SK_MEM_PACKETS 256
262 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
263 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
264 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
266 /* Run time adjustable parameters. */
267 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
268 EXPORT_SYMBOL(sysctl_wmem_max);
269 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
270 EXPORT_SYMBOL(sysctl_rmem_max);
271 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
272 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
274 /* Maximal space eaten by iovec or ancillary data plus some space */
275 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
276 EXPORT_SYMBOL(sysctl_optmem_max);
278 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
279 EXPORT_SYMBOL_GPL(memalloc_socks);
282 * sk_set_memalloc - sets %SOCK_MEMALLOC
283 * @sk: socket to set it on
285 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
286 * It's the responsibility of the admin to adjust min_free_kbytes
287 * to meet the requirements
289 void sk_set_memalloc(struct sock *sk)
291 sock_set_flag(sk, SOCK_MEMALLOC);
292 sk->sk_allocation |= __GFP_MEMALLOC;
293 static_key_slow_inc(&memalloc_socks);
295 EXPORT_SYMBOL_GPL(sk_set_memalloc);
297 void sk_clear_memalloc(struct sock *sk)
299 sock_reset_flag(sk, SOCK_MEMALLOC);
300 sk->sk_allocation &= ~__GFP_MEMALLOC;
301 static_key_slow_dec(&memalloc_socks);
304 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
305 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
306 * it has rmem allocations there is a risk that the user of the
307 * socket cannot make forward progress due to exceeding the rmem
308 * limits. By rights, sk_clear_memalloc() should only be called
309 * on sockets being torn down but warn and reset the accounting if
310 * that assumption breaks.
312 if (WARN_ON(sk->sk_forward_alloc))
315 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
317 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
320 unsigned long pflags = current->flags;
322 /* these should have been dropped before queueing */
323 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
325 current->flags |= PF_MEMALLOC;
326 ret = sk->sk_backlog_rcv(sk, skb);
327 tsk_restore_flags(current, pflags, PF_MEMALLOC);
331 EXPORT_SYMBOL(__sk_backlog_rcv);
333 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
337 if (optlen < sizeof(tv))
339 if (copy_from_user(&tv, optval, sizeof(tv)))
341 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
345 static int warned __read_mostly;
348 if (warned < 10 && net_ratelimit()) {
350 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
351 __func__, current->comm, task_pid_nr(current));
355 *timeo_p = MAX_SCHEDULE_TIMEOUT;
356 if (tv.tv_sec == 0 && tv.tv_usec == 0)
358 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
359 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
363 static void sock_warn_obsolete_bsdism(const char *name)
366 static char warncomm[TASK_COMM_LEN];
367 if (strcmp(warncomm, current->comm) && warned < 5) {
368 strcpy(warncomm, current->comm);
369 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
375 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
377 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
379 if (sk->sk_flags & flags) {
380 sk->sk_flags &= ~flags;
381 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
382 net_disable_timestamp();
387 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
392 struct sk_buff_head *list = &sk->sk_receive_queue;
394 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
395 atomic_inc(&sk->sk_drops);
396 trace_sock_rcvqueue_full(sk, skb);
400 err = sk_filter(sk, skb);
404 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
405 atomic_inc(&sk->sk_drops);
410 skb_set_owner_r(skb, sk);
412 /* Cache the SKB length before we tack it onto the receive
413 * queue. Once it is added it no longer belongs to us and
414 * may be freed by other threads of control pulling packets
419 /* we escape from rcu protected region, make sure we dont leak
424 spin_lock_irqsave(&list->lock, flags);
425 skb->dropcount = atomic_read(&sk->sk_drops);
426 __skb_queue_tail(list, skb);
427 spin_unlock_irqrestore(&list->lock, flags);
429 if (!sock_flag(sk, SOCK_DEAD))
430 sk->sk_data_ready(sk, skb_len);
433 EXPORT_SYMBOL(sock_queue_rcv_skb);
435 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
437 int rc = NET_RX_SUCCESS;
439 if (sk_filter(sk, skb))
440 goto discard_and_relse;
444 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
445 atomic_inc(&sk->sk_drops);
446 goto discard_and_relse;
449 bh_lock_sock_nested(sk);
452 if (!sock_owned_by_user(sk)) {
454 * trylock + unlock semantics:
456 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
458 rc = sk_backlog_rcv(sk, skb);
460 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
461 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
463 atomic_inc(&sk->sk_drops);
464 goto discard_and_relse;
475 EXPORT_SYMBOL(sk_receive_skb);
477 void sk_reset_txq(struct sock *sk)
479 sk_tx_queue_clear(sk);
481 EXPORT_SYMBOL(sk_reset_txq);
483 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
485 struct dst_entry *dst = __sk_dst_get(sk);
487 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
488 sk_tx_queue_clear(sk);
489 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
496 EXPORT_SYMBOL(__sk_dst_check);
498 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
500 struct dst_entry *dst = sk_dst_get(sk);
502 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
510 EXPORT_SYMBOL(sk_dst_check);
512 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
515 int ret = -ENOPROTOOPT;
516 #ifdef CONFIG_NETDEVICES
517 struct net *net = sock_net(sk);
518 char devname[IFNAMSIZ];
523 if (!ns_capable(net->user_ns, CAP_NET_RAW))
530 /* Bind this socket to a particular device like "eth0",
531 * as specified in the passed interface name. If the
532 * name is "" or the option length is zero the socket
535 if (optlen > IFNAMSIZ - 1)
536 optlen = IFNAMSIZ - 1;
537 memset(devname, 0, sizeof(devname));
540 if (copy_from_user(devname, optval, optlen))
544 if (devname[0] != '\0') {
545 struct net_device *dev;
548 dev = dev_get_by_name_rcu(net, devname);
550 index = dev->ifindex;
558 sk->sk_bound_dev_if = index;
570 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
571 int __user *optlen, int len)
573 int ret = -ENOPROTOOPT;
574 #ifdef CONFIG_NETDEVICES
575 struct net *net = sock_net(sk);
576 char devname[IFNAMSIZ];
578 if (sk->sk_bound_dev_if == 0) {
587 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
591 len = strlen(devname) + 1;
594 if (copy_to_user(optval, devname, len))
599 if (put_user(len, optlen))
610 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
613 sock_set_flag(sk, bit);
615 sock_reset_flag(sk, bit);
619 * This is meant for all protocols to use and covers goings on
620 * at the socket level. Everything here is generic.
623 int sock_setsockopt(struct socket *sock, int level, int optname,
624 char __user *optval, unsigned int optlen)
626 struct sock *sk = sock->sk;
633 * Options without arguments
636 if (optname == SO_BINDTODEVICE)
637 return sock_setbindtodevice(sk, optval, optlen);
639 if (optlen < sizeof(int))
642 if (get_user(val, (int __user *)optval))
645 valbool = val ? 1 : 0;
651 if (val && !capable(CAP_NET_ADMIN))
654 sock_valbool_flag(sk, SOCK_DBG, valbool);
657 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
660 sk->sk_reuseport = valbool;
669 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
672 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
675 /* Don't error on this BSD doesn't and if you think
676 * about it this is right. Otherwise apps have to
677 * play 'guess the biggest size' games. RCVBUF/SNDBUF
678 * are treated in BSD as hints
680 val = min_t(u32, val, sysctl_wmem_max);
682 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
683 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
684 /* Wake up sending tasks if we upped the value. */
685 sk->sk_write_space(sk);
689 if (!capable(CAP_NET_ADMIN)) {
696 /* Don't error on this BSD doesn't and if you think
697 * about it this is right. Otherwise apps have to
698 * play 'guess the biggest size' games. RCVBUF/SNDBUF
699 * are treated in BSD as hints
701 val = min_t(u32, val, sysctl_rmem_max);
703 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
705 * We double it on the way in to account for
706 * "struct sk_buff" etc. overhead. Applications
707 * assume that the SO_RCVBUF setting they make will
708 * allow that much actual data to be received on that
711 * Applications are unaware that "struct sk_buff" and
712 * other overheads allocate from the receive buffer
713 * during socket buffer allocation.
715 * And after considering the possible alternatives,
716 * returning the value we actually used in getsockopt
717 * is the most desirable behavior.
719 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
723 if (!capable(CAP_NET_ADMIN)) {
731 if (sk->sk_protocol == IPPROTO_TCP &&
732 sk->sk_type == SOCK_STREAM)
733 tcp_set_keepalive(sk, valbool);
735 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
739 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
743 sk->sk_no_check = valbool;
747 if ((val >= 0 && val <= 6) ||
748 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
749 sk->sk_priority = val;
755 if (optlen < sizeof(ling)) {
756 ret = -EINVAL; /* 1003.1g */
759 if (copy_from_user(&ling, optval, sizeof(ling))) {
764 sock_reset_flag(sk, SOCK_LINGER);
766 #if (BITS_PER_LONG == 32)
767 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
768 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
771 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
772 sock_set_flag(sk, SOCK_LINGER);
777 sock_warn_obsolete_bsdism("setsockopt");
782 set_bit(SOCK_PASSCRED, &sock->flags);
784 clear_bit(SOCK_PASSCRED, &sock->flags);
790 if (optname == SO_TIMESTAMP)
791 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
793 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
794 sock_set_flag(sk, SOCK_RCVTSTAMP);
795 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
797 sock_reset_flag(sk, SOCK_RCVTSTAMP);
798 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
802 case SO_TIMESTAMPING:
803 if (val & ~SOF_TIMESTAMPING_MASK) {
807 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
808 val & SOF_TIMESTAMPING_TX_HARDWARE);
809 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
810 val & SOF_TIMESTAMPING_TX_SOFTWARE);
811 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
812 val & SOF_TIMESTAMPING_RX_HARDWARE);
813 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
814 sock_enable_timestamp(sk,
815 SOCK_TIMESTAMPING_RX_SOFTWARE);
817 sock_disable_timestamp(sk,
818 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
819 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
820 val & SOF_TIMESTAMPING_SOFTWARE);
821 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
822 val & SOF_TIMESTAMPING_SYS_HARDWARE);
823 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
824 val & SOF_TIMESTAMPING_RAW_HARDWARE);
830 sk->sk_rcvlowat = val ? : 1;
834 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
838 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
841 case SO_ATTACH_FILTER:
843 if (optlen == sizeof(struct sock_fprog)) {
844 struct sock_fprog fprog;
847 if (copy_from_user(&fprog, optval, sizeof(fprog)))
850 ret = sk_attach_filter(&fprog, sk);
854 case SO_DETACH_FILTER:
855 ret = sk_detach_filter(sk);
859 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
862 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
867 set_bit(SOCK_PASSSEC, &sock->flags);
869 clear_bit(SOCK_PASSSEC, &sock->flags);
872 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
878 /* We implement the SO_SNDLOWAT etc to
879 not be settable (1003.1g 5.3) */
881 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
885 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
889 if (sock->ops->set_peek_off)
890 sock->ops->set_peek_off(sk, val);
896 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
899 case SO_SELECT_ERR_QUEUE:
900 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
903 #ifdef CONFIG_NET_RX_BUSY_POLL
905 /* allow unprivileged users to decrease the value */
906 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
912 sk->sk_ll_usec = val;
923 EXPORT_SYMBOL(sock_setsockopt);
926 void cred_to_ucred(struct pid *pid, const struct cred *cred,
929 ucred->pid = pid_vnr(pid);
930 ucred->uid = ucred->gid = -1;
932 struct user_namespace *current_ns = current_user_ns();
934 ucred->uid = from_kuid_munged(current_ns, cred->euid);
935 ucred->gid = from_kgid_munged(current_ns, cred->egid);
938 EXPORT_SYMBOL_GPL(cred_to_ucred);
940 int sock_getsockopt(struct socket *sock, int level, int optname,
941 char __user *optval, int __user *optlen)
943 struct sock *sk = sock->sk;
951 int lv = sizeof(int);
954 if (get_user(len, optlen))
959 memset(&v, 0, sizeof(v));
963 v.val = sock_flag(sk, SOCK_DBG);
967 v.val = sock_flag(sk, SOCK_LOCALROUTE);
971 v.val = sock_flag(sk, SOCK_BROADCAST);
975 v.val = sk->sk_sndbuf;
979 v.val = sk->sk_rcvbuf;
983 v.val = sk->sk_reuse;
987 v.val = sk->sk_reuseport;
991 v.val = sock_flag(sk, SOCK_KEEPOPEN);
999 v.val = sk->sk_protocol;
1003 v.val = sk->sk_family;
1007 v.val = -sock_error(sk);
1009 v.val = xchg(&sk->sk_err_soft, 0);
1013 v.val = sock_flag(sk, SOCK_URGINLINE);
1017 v.val = sk->sk_no_check;
1021 v.val = sk->sk_priority;
1025 lv = sizeof(v.ling);
1026 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1027 v.ling.l_linger = sk->sk_lingertime / HZ;
1031 sock_warn_obsolete_bsdism("getsockopt");
1035 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1036 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1039 case SO_TIMESTAMPNS:
1040 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1043 case SO_TIMESTAMPING:
1045 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
1046 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
1047 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
1048 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
1049 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
1050 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
1051 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1052 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
1053 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
1054 v.val |= SOF_TIMESTAMPING_SOFTWARE;
1055 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
1056 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
1057 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
1058 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
1062 lv = sizeof(struct timeval);
1063 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1067 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1068 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1073 lv = sizeof(struct timeval);
1074 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1078 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1079 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1084 v.val = sk->sk_rcvlowat;
1092 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1097 struct ucred peercred;
1098 if (len > sizeof(peercred))
1099 len = sizeof(peercred);
1100 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1101 if (copy_to_user(optval, &peercred, len))
1110 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1114 if (copy_to_user(optval, address, len))
1119 /* Dubious BSD thing... Probably nobody even uses it, but
1120 * the UNIX standard wants it for whatever reason... -DaveM
1123 v.val = sk->sk_state == TCP_LISTEN;
1127 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1131 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1134 v.val = sk->sk_mark;
1138 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1141 case SO_WIFI_STATUS:
1142 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1146 if (!sock->ops->set_peek_off)
1149 v.val = sk->sk_peek_off;
1152 v.val = sock_flag(sk, SOCK_NOFCS);
1155 case SO_BINDTODEVICE:
1156 return sock_getbindtodevice(sk, optval, optlen, len);
1159 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1165 case SO_LOCK_FILTER:
1166 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1169 case SO_SELECT_ERR_QUEUE:
1170 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1173 #ifdef CONFIG_NET_RX_BUSY_POLL
1175 v.val = sk->sk_ll_usec;
1180 return -ENOPROTOOPT;
1185 if (copy_to_user(optval, &v, len))
1188 if (put_user(len, optlen))
1194 * Initialize an sk_lock.
1196 * (We also register the sk_lock with the lock validator.)
1198 static inline void sock_lock_init(struct sock *sk)
1200 sock_lock_init_class_and_name(sk,
1201 af_family_slock_key_strings[sk->sk_family],
1202 af_family_slock_keys + sk->sk_family,
1203 af_family_key_strings[sk->sk_family],
1204 af_family_keys + sk->sk_family);
1208 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1209 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1210 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1212 static void sock_copy(struct sock *nsk, const struct sock *osk)
1214 #ifdef CONFIG_SECURITY_NETWORK
1215 void *sptr = nsk->sk_security;
1217 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1219 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1220 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1222 #ifdef CONFIG_SECURITY_NETWORK
1223 nsk->sk_security = sptr;
1224 security_sk_clone(osk, nsk);
1228 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1230 unsigned long nulls1, nulls2;
1232 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1233 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1234 if (nulls1 > nulls2)
1235 swap(nulls1, nulls2);
1238 memset((char *)sk, 0, nulls1);
1239 memset((char *)sk + nulls1 + sizeof(void *), 0,
1240 nulls2 - nulls1 - sizeof(void *));
1241 memset((char *)sk + nulls2 + sizeof(void *), 0,
1242 size - nulls2 - sizeof(void *));
1244 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1246 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1250 struct kmem_cache *slab;
1254 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1257 if (priority & __GFP_ZERO) {
1259 prot->clear_sk(sk, prot->obj_size);
1261 sk_prot_clear_nulls(sk, prot->obj_size);
1264 sk = kmalloc(prot->obj_size, priority);
1267 kmemcheck_annotate_bitfield(sk, flags);
1269 if (security_sk_alloc(sk, family, priority))
1272 if (!try_module_get(prot->owner))
1274 sk_tx_queue_clear(sk);
1280 security_sk_free(sk);
1283 kmem_cache_free(slab, sk);
1289 static void sk_prot_free(struct proto *prot, struct sock *sk)
1291 struct kmem_cache *slab;
1292 struct module *owner;
1294 owner = prot->owner;
1297 security_sk_free(sk);
1299 kmem_cache_free(slab, sk);
1305 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1306 void sock_update_classid(struct sock *sk)
1310 classid = task_cls_classid(current);
1311 if (classid != sk->sk_classid)
1312 sk->sk_classid = classid;
1314 EXPORT_SYMBOL(sock_update_classid);
1317 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1318 void sock_update_netprioidx(struct sock *sk)
1323 sk->sk_cgrp_prioidx = task_netprioidx(current);
1325 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1329 * sk_alloc - All socket objects are allocated here
1330 * @net: the applicable net namespace
1331 * @family: protocol family
1332 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1333 * @prot: struct proto associated with this new sock instance
1335 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1340 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1342 sk->sk_family = family;
1344 * See comment in struct sock definition to understand
1345 * why we need sk_prot_creator -acme
1347 sk->sk_prot = sk->sk_prot_creator = prot;
1349 sock_net_set(sk, get_net(net));
1350 atomic_set(&sk->sk_wmem_alloc, 1);
1352 sock_update_classid(sk);
1353 sock_update_netprioidx(sk);
1358 EXPORT_SYMBOL(sk_alloc);
1360 static void __sk_free(struct sock *sk)
1362 struct sk_filter *filter;
1364 if (sk->sk_destruct)
1365 sk->sk_destruct(sk);
1367 filter = rcu_dereference_check(sk->sk_filter,
1368 atomic_read(&sk->sk_wmem_alloc) == 0);
1370 sk_filter_uncharge(sk, filter);
1371 RCU_INIT_POINTER(sk->sk_filter, NULL);
1374 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1376 if (atomic_read(&sk->sk_omem_alloc))
1377 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1378 __func__, atomic_read(&sk->sk_omem_alloc));
1380 if (sk->sk_peer_cred)
1381 put_cred(sk->sk_peer_cred);
1382 put_pid(sk->sk_peer_pid);
1383 put_net(sock_net(sk));
1384 sk_prot_free(sk->sk_prot_creator, sk);
1387 void sk_free(struct sock *sk)
1390 * We subtract one from sk_wmem_alloc and can know if
1391 * some packets are still in some tx queue.
1392 * If not null, sock_wfree() will call __sk_free(sk) later
1394 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1397 EXPORT_SYMBOL(sk_free);
1400 * Last sock_put should drop reference to sk->sk_net. It has already
1401 * been dropped in sk_change_net. Taking reference to stopping namespace
1403 * Take reference to a socket to remove it from hash _alive_ and after that
1404 * destroy it in the context of init_net.
1406 void sk_release_kernel(struct sock *sk)
1408 if (sk == NULL || sk->sk_socket == NULL)
1412 sock_release(sk->sk_socket);
1413 release_net(sock_net(sk));
1414 sock_net_set(sk, get_net(&init_net));
1417 EXPORT_SYMBOL(sk_release_kernel);
1419 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1421 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1422 sock_update_memcg(newsk);
1426 * sk_clone_lock - clone a socket, and lock its clone
1427 * @sk: the socket to clone
1428 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1430 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1432 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1436 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1437 if (newsk != NULL) {
1438 struct sk_filter *filter;
1440 sock_copy(newsk, sk);
1443 get_net(sock_net(newsk));
1444 sk_node_init(&newsk->sk_node);
1445 sock_lock_init(newsk);
1446 bh_lock_sock(newsk);
1447 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1448 newsk->sk_backlog.len = 0;
1450 atomic_set(&newsk->sk_rmem_alloc, 0);
1452 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1454 atomic_set(&newsk->sk_wmem_alloc, 1);
1455 atomic_set(&newsk->sk_omem_alloc, 0);
1456 skb_queue_head_init(&newsk->sk_receive_queue);
1457 skb_queue_head_init(&newsk->sk_write_queue);
1458 #ifdef CONFIG_NET_DMA
1459 skb_queue_head_init(&newsk->sk_async_wait_queue);
1462 spin_lock_init(&newsk->sk_dst_lock);
1463 rwlock_init(&newsk->sk_callback_lock);
1464 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1465 af_callback_keys + newsk->sk_family,
1466 af_family_clock_key_strings[newsk->sk_family]);
1468 newsk->sk_dst_cache = NULL;
1469 newsk->sk_wmem_queued = 0;
1470 newsk->sk_forward_alloc = 0;
1471 newsk->sk_send_head = NULL;
1472 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1474 sock_reset_flag(newsk, SOCK_DONE);
1475 skb_queue_head_init(&newsk->sk_error_queue);
1477 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1479 sk_filter_charge(newsk, filter);
1481 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1482 /* It is still raw copy of parent, so invalidate
1483 * destructor and make plain sk_free() */
1484 newsk->sk_destruct = NULL;
1485 bh_unlock_sock(newsk);
1492 newsk->sk_priority = 0;
1494 * Before updating sk_refcnt, we must commit prior changes to memory
1495 * (Documentation/RCU/rculist_nulls.txt for details)
1498 atomic_set(&newsk->sk_refcnt, 2);
1501 * Increment the counter in the same struct proto as the master
1502 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1503 * is the same as sk->sk_prot->socks, as this field was copied
1506 * This _changes_ the previous behaviour, where
1507 * tcp_create_openreq_child always was incrementing the
1508 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1509 * to be taken into account in all callers. -acme
1511 sk_refcnt_debug_inc(newsk);
1512 sk_set_socket(newsk, NULL);
1513 newsk->sk_wq = NULL;
1515 sk_update_clone(sk, newsk);
1517 if (newsk->sk_prot->sockets_allocated)
1518 sk_sockets_allocated_inc(newsk);
1520 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1521 net_enable_timestamp();
1526 EXPORT_SYMBOL_GPL(sk_clone_lock);
1528 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1530 __sk_dst_set(sk, dst);
1531 sk->sk_route_caps = dst->dev->features;
1532 if (sk->sk_route_caps & NETIF_F_GSO)
1533 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1534 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1535 if (sk_can_gso(sk)) {
1536 if (dst->header_len) {
1537 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1539 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1540 sk->sk_gso_max_size = dst->dev->gso_max_size;
1541 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1545 EXPORT_SYMBOL_GPL(sk_setup_caps);
1548 * Simple resource managers for sockets.
1553 * Write buffer destructor automatically called from kfree_skb.
1555 void sock_wfree(struct sk_buff *skb)
1557 struct sock *sk = skb->sk;
1558 unsigned int len = skb->truesize;
1560 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1562 * Keep a reference on sk_wmem_alloc, this will be released
1563 * after sk_write_space() call
1565 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1566 sk->sk_write_space(sk);
1570 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1571 * could not do because of in-flight packets
1573 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1576 EXPORT_SYMBOL(sock_wfree);
1579 * Read buffer destructor automatically called from kfree_skb.
1581 void sock_rfree(struct sk_buff *skb)
1583 struct sock *sk = skb->sk;
1584 unsigned int len = skb->truesize;
1586 atomic_sub(len, &sk->sk_rmem_alloc);
1587 sk_mem_uncharge(sk, len);
1589 EXPORT_SYMBOL(sock_rfree);
1591 void sock_edemux(struct sk_buff *skb)
1593 struct sock *sk = skb->sk;
1596 if (sk->sk_state == TCP_TIME_WAIT)
1597 inet_twsk_put(inet_twsk(sk));
1602 EXPORT_SYMBOL(sock_edemux);
1604 kuid_t sock_i_uid(struct sock *sk)
1608 read_lock_bh(&sk->sk_callback_lock);
1609 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1610 read_unlock_bh(&sk->sk_callback_lock);
1613 EXPORT_SYMBOL(sock_i_uid);
1615 unsigned long sock_i_ino(struct sock *sk)
1619 read_lock_bh(&sk->sk_callback_lock);
1620 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1621 read_unlock_bh(&sk->sk_callback_lock);
1624 EXPORT_SYMBOL(sock_i_ino);
1627 * Allocate a skb from the socket's send buffer.
1629 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1632 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1633 struct sk_buff *skb = alloc_skb(size, priority);
1635 skb_set_owner_w(skb, sk);
1641 EXPORT_SYMBOL(sock_wmalloc);
1644 * Allocate a skb from the socket's receive buffer.
1646 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1649 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1650 struct sk_buff *skb = alloc_skb(size, priority);
1652 skb_set_owner_r(skb, sk);
1660 * Allocate a memory block from the socket's option memory buffer.
1662 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1664 if ((unsigned int)size <= sysctl_optmem_max &&
1665 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1667 /* First do the add, to avoid the race if kmalloc
1670 atomic_add(size, &sk->sk_omem_alloc);
1671 mem = kmalloc(size, priority);
1674 atomic_sub(size, &sk->sk_omem_alloc);
1678 EXPORT_SYMBOL(sock_kmalloc);
1681 * Free an option memory block.
1683 void sock_kfree_s(struct sock *sk, void *mem, int size)
1686 atomic_sub(size, &sk->sk_omem_alloc);
1688 EXPORT_SYMBOL(sock_kfree_s);
1690 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1691 I think, these locks should be removed for datagram sockets.
1693 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1697 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1701 if (signal_pending(current))
1703 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1704 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1705 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1707 if (sk->sk_shutdown & SEND_SHUTDOWN)
1711 timeo = schedule_timeout(timeo);
1713 finish_wait(sk_sleep(sk), &wait);
1719 * Generic send/receive buffer handlers
1722 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1723 unsigned long data_len, int noblock,
1726 struct sk_buff *skb;
1730 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1733 if (npages > MAX_SKB_FRAGS)
1736 gfp_mask = sk->sk_allocation;
1737 if (gfp_mask & __GFP_WAIT)
1738 gfp_mask |= __GFP_REPEAT;
1740 timeo = sock_sndtimeo(sk, noblock);
1742 err = sock_error(sk);
1747 if (sk->sk_shutdown & SEND_SHUTDOWN)
1750 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1751 skb = alloc_skb(header_len, gfp_mask);
1755 /* No pages, we're done... */
1759 skb->truesize += data_len;
1760 skb_shinfo(skb)->nr_frags = npages;
1761 for (i = 0; i < npages; i++) {
1764 page = alloc_pages(sk->sk_allocation, 0);
1767 skb_shinfo(skb)->nr_frags = i;
1772 __skb_fill_page_desc(skb, i,
1774 (data_len >= PAGE_SIZE ?
1777 data_len -= PAGE_SIZE;
1780 /* Full success... */
1786 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1787 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1791 if (signal_pending(current))
1793 timeo = sock_wait_for_wmem(sk, timeo);
1796 skb_set_owner_w(skb, sk);
1800 err = sock_intr_errno(timeo);
1805 EXPORT_SYMBOL(sock_alloc_send_pskb);
1807 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1808 int noblock, int *errcode)
1810 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1812 EXPORT_SYMBOL(sock_alloc_send_skb);
1814 /* On 32bit arches, an skb frag is limited to 2^15 */
1815 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1817 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1822 if (atomic_read(&pfrag->page->_count) == 1) {
1826 if (pfrag->offset < pfrag->size)
1828 put_page(pfrag->page);
1831 /* We restrict high order allocations to users that can afford to wait */
1832 order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1835 gfp_t gfp = sk->sk_allocation;
1838 gfp |= __GFP_COMP | __GFP_NOWARN;
1839 pfrag->page = alloc_pages(gfp, order);
1840 if (likely(pfrag->page)) {
1842 pfrag->size = PAGE_SIZE << order;
1845 } while (--order >= 0);
1847 sk_enter_memory_pressure(sk);
1848 sk_stream_moderate_sndbuf(sk);
1851 EXPORT_SYMBOL(sk_page_frag_refill);
1853 static void __lock_sock(struct sock *sk)
1854 __releases(&sk->sk_lock.slock)
1855 __acquires(&sk->sk_lock.slock)
1860 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1861 TASK_UNINTERRUPTIBLE);
1862 spin_unlock_bh(&sk->sk_lock.slock);
1864 spin_lock_bh(&sk->sk_lock.slock);
1865 if (!sock_owned_by_user(sk))
1868 finish_wait(&sk->sk_lock.wq, &wait);
1871 static void __release_sock(struct sock *sk)
1872 __releases(&sk->sk_lock.slock)
1873 __acquires(&sk->sk_lock.slock)
1875 struct sk_buff *skb = sk->sk_backlog.head;
1878 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1882 struct sk_buff *next = skb->next;
1885 WARN_ON_ONCE(skb_dst_is_noref(skb));
1887 sk_backlog_rcv(sk, skb);
1890 * We are in process context here with softirqs
1891 * disabled, use cond_resched_softirq() to preempt.
1892 * This is safe to do because we've taken the backlog
1895 cond_resched_softirq();
1898 } while (skb != NULL);
1901 } while ((skb = sk->sk_backlog.head) != NULL);
1904 * Doing the zeroing here guarantee we can not loop forever
1905 * while a wild producer attempts to flood us.
1907 sk->sk_backlog.len = 0;
1911 * sk_wait_data - wait for data to arrive at sk_receive_queue
1912 * @sk: sock to wait on
1913 * @timeo: for how long
1915 * Now socket state including sk->sk_err is changed only under lock,
1916 * hence we may omit checks after joining wait queue.
1917 * We check receive queue before schedule() only as optimization;
1918 * it is very likely that release_sock() added new data.
1920 int sk_wait_data(struct sock *sk, long *timeo)
1925 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1926 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1927 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1928 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1929 finish_wait(sk_sleep(sk), &wait);
1932 EXPORT_SYMBOL(sk_wait_data);
1935 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1937 * @size: memory size to allocate
1938 * @kind: allocation type
1940 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1941 * rmem allocation. This function assumes that protocols which have
1942 * memory_pressure use sk_wmem_queued as write buffer accounting.
1944 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1946 struct proto *prot = sk->sk_prot;
1947 int amt = sk_mem_pages(size);
1949 int parent_status = UNDER_LIMIT;
1951 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1953 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1956 if (parent_status == UNDER_LIMIT &&
1957 allocated <= sk_prot_mem_limits(sk, 0)) {
1958 sk_leave_memory_pressure(sk);
1962 /* Under pressure. (we or our parents) */
1963 if ((parent_status > SOFT_LIMIT) ||
1964 allocated > sk_prot_mem_limits(sk, 1))
1965 sk_enter_memory_pressure(sk);
1967 /* Over hard limit (we or our parents) */
1968 if ((parent_status == OVER_LIMIT) ||
1969 (allocated > sk_prot_mem_limits(sk, 2)))
1970 goto suppress_allocation;
1972 /* guarantee minimum buffer size under pressure */
1973 if (kind == SK_MEM_RECV) {
1974 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1977 } else { /* SK_MEM_SEND */
1978 if (sk->sk_type == SOCK_STREAM) {
1979 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1981 } else if (atomic_read(&sk->sk_wmem_alloc) <
1982 prot->sysctl_wmem[0])
1986 if (sk_has_memory_pressure(sk)) {
1989 if (!sk_under_memory_pressure(sk))
1991 alloc = sk_sockets_allocated_read_positive(sk);
1992 if (sk_prot_mem_limits(sk, 2) > alloc *
1993 sk_mem_pages(sk->sk_wmem_queued +
1994 atomic_read(&sk->sk_rmem_alloc) +
1995 sk->sk_forward_alloc))
1999 suppress_allocation:
2001 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2002 sk_stream_moderate_sndbuf(sk);
2004 /* Fail only if socket is _under_ its sndbuf.
2005 * In this case we cannot block, so that we have to fail.
2007 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2011 trace_sock_exceed_buf_limit(sk, prot, allocated);
2013 /* Alas. Undo changes. */
2014 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2016 sk_memory_allocated_sub(sk, amt);
2020 EXPORT_SYMBOL(__sk_mem_schedule);
2023 * __sk_reclaim - reclaim memory_allocated
2026 void __sk_mem_reclaim(struct sock *sk)
2028 sk_memory_allocated_sub(sk,
2029 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2030 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2032 if (sk_under_memory_pressure(sk) &&
2033 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2034 sk_leave_memory_pressure(sk);
2036 EXPORT_SYMBOL(__sk_mem_reclaim);
2040 * Set of default routines for initialising struct proto_ops when
2041 * the protocol does not support a particular function. In certain
2042 * cases where it makes no sense for a protocol to have a "do nothing"
2043 * function, some default processing is provided.
2046 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2050 EXPORT_SYMBOL(sock_no_bind);
2052 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2057 EXPORT_SYMBOL(sock_no_connect);
2059 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2063 EXPORT_SYMBOL(sock_no_socketpair);
2065 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2069 EXPORT_SYMBOL(sock_no_accept);
2071 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2076 EXPORT_SYMBOL(sock_no_getname);
2078 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2082 EXPORT_SYMBOL(sock_no_poll);
2084 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2088 EXPORT_SYMBOL(sock_no_ioctl);
2090 int sock_no_listen(struct socket *sock, int backlog)
2094 EXPORT_SYMBOL(sock_no_listen);
2096 int sock_no_shutdown(struct socket *sock, int how)
2100 EXPORT_SYMBOL(sock_no_shutdown);
2102 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2103 char __user *optval, unsigned int optlen)
2107 EXPORT_SYMBOL(sock_no_setsockopt);
2109 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2110 char __user *optval, int __user *optlen)
2114 EXPORT_SYMBOL(sock_no_getsockopt);
2116 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2121 EXPORT_SYMBOL(sock_no_sendmsg);
2123 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2124 size_t len, int flags)
2128 EXPORT_SYMBOL(sock_no_recvmsg);
2130 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2132 /* Mirror missing mmap method error code */
2135 EXPORT_SYMBOL(sock_no_mmap);
2137 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2140 struct msghdr msg = {.msg_flags = flags};
2142 char *kaddr = kmap(page);
2143 iov.iov_base = kaddr + offset;
2145 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2149 EXPORT_SYMBOL(sock_no_sendpage);
2152 * Default Socket Callbacks
2155 static void sock_def_wakeup(struct sock *sk)
2157 struct socket_wq *wq;
2160 wq = rcu_dereference(sk->sk_wq);
2161 if (wq_has_sleeper(wq))
2162 wake_up_interruptible_all(&wq->wait);
2166 static void sock_def_error_report(struct sock *sk)
2168 struct socket_wq *wq;
2171 wq = rcu_dereference(sk->sk_wq);
2172 if (wq_has_sleeper(wq))
2173 wake_up_interruptible_poll(&wq->wait, POLLERR);
2174 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2178 static void sock_def_readable(struct sock *sk, int len)
2180 struct socket_wq *wq;
2183 wq = rcu_dereference(sk->sk_wq);
2184 if (wq_has_sleeper(wq))
2185 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2186 POLLRDNORM | POLLRDBAND);
2187 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2191 static void sock_def_write_space(struct sock *sk)
2193 struct socket_wq *wq;
2197 /* Do not wake up a writer until he can make "significant"
2200 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2201 wq = rcu_dereference(sk->sk_wq);
2202 if (wq_has_sleeper(wq))
2203 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2204 POLLWRNORM | POLLWRBAND);
2206 /* Should agree with poll, otherwise some programs break */
2207 if (sock_writeable(sk))
2208 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2214 static void sock_def_destruct(struct sock *sk)
2216 kfree(sk->sk_protinfo);
2219 void sk_send_sigurg(struct sock *sk)
2221 if (sk->sk_socket && sk->sk_socket->file)
2222 if (send_sigurg(&sk->sk_socket->file->f_owner))
2223 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2225 EXPORT_SYMBOL(sk_send_sigurg);
2227 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2228 unsigned long expires)
2230 if (!mod_timer(timer, expires))
2233 EXPORT_SYMBOL(sk_reset_timer);
2235 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2237 if (del_timer(timer))
2240 EXPORT_SYMBOL(sk_stop_timer);
2242 void sock_init_data(struct socket *sock, struct sock *sk)
2244 skb_queue_head_init(&sk->sk_receive_queue);
2245 skb_queue_head_init(&sk->sk_write_queue);
2246 skb_queue_head_init(&sk->sk_error_queue);
2247 #ifdef CONFIG_NET_DMA
2248 skb_queue_head_init(&sk->sk_async_wait_queue);
2251 sk->sk_send_head = NULL;
2253 init_timer(&sk->sk_timer);
2255 sk->sk_allocation = GFP_KERNEL;
2256 sk->sk_rcvbuf = sysctl_rmem_default;
2257 sk->sk_sndbuf = sysctl_wmem_default;
2258 sk->sk_state = TCP_CLOSE;
2259 sk_set_socket(sk, sock);
2261 sock_set_flag(sk, SOCK_ZAPPED);
2264 sk->sk_type = sock->type;
2265 sk->sk_wq = sock->wq;
2270 spin_lock_init(&sk->sk_dst_lock);
2271 rwlock_init(&sk->sk_callback_lock);
2272 lockdep_set_class_and_name(&sk->sk_callback_lock,
2273 af_callback_keys + sk->sk_family,
2274 af_family_clock_key_strings[sk->sk_family]);
2276 sk->sk_state_change = sock_def_wakeup;
2277 sk->sk_data_ready = sock_def_readable;
2278 sk->sk_write_space = sock_def_write_space;
2279 sk->sk_error_report = sock_def_error_report;
2280 sk->sk_destruct = sock_def_destruct;
2282 sk->sk_frag.page = NULL;
2283 sk->sk_frag.offset = 0;
2284 sk->sk_peek_off = -1;
2286 sk->sk_peer_pid = NULL;
2287 sk->sk_peer_cred = NULL;
2288 sk->sk_write_pending = 0;
2289 sk->sk_rcvlowat = 1;
2290 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2291 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2293 sk->sk_stamp = ktime_set(-1L, 0);
2295 #ifdef CONFIG_NET_RX_BUSY_POLL
2297 sk->sk_ll_usec = sysctl_net_busy_read;
2301 * Before updating sk_refcnt, we must commit prior changes to memory
2302 * (Documentation/RCU/rculist_nulls.txt for details)
2305 atomic_set(&sk->sk_refcnt, 1);
2306 atomic_set(&sk->sk_drops, 0);
2308 EXPORT_SYMBOL(sock_init_data);
2310 void lock_sock_nested(struct sock *sk, int subclass)
2313 spin_lock_bh(&sk->sk_lock.slock);
2314 if (sk->sk_lock.owned)
2316 sk->sk_lock.owned = 1;
2317 spin_unlock(&sk->sk_lock.slock);
2319 * The sk_lock has mutex_lock() semantics here:
2321 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2324 EXPORT_SYMBOL(lock_sock_nested);
2326 void release_sock(struct sock *sk)
2329 * The sk_lock has mutex_unlock() semantics:
2331 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2333 spin_lock_bh(&sk->sk_lock.slock);
2334 if (sk->sk_backlog.tail)
2337 if (sk->sk_prot->release_cb)
2338 sk->sk_prot->release_cb(sk);
2340 sk->sk_lock.owned = 0;
2341 if (waitqueue_active(&sk->sk_lock.wq))
2342 wake_up(&sk->sk_lock.wq);
2343 spin_unlock_bh(&sk->sk_lock.slock);
2345 EXPORT_SYMBOL(release_sock);
2348 * lock_sock_fast - fast version of lock_sock
2351 * This version should be used for very small section, where process wont block
2352 * return false if fast path is taken
2353 * sk_lock.slock locked, owned = 0, BH disabled
2354 * return true if slow path is taken
2355 * sk_lock.slock unlocked, owned = 1, BH enabled
2357 bool lock_sock_fast(struct sock *sk)
2360 spin_lock_bh(&sk->sk_lock.slock);
2362 if (!sk->sk_lock.owned)
2364 * Note : We must disable BH
2369 sk->sk_lock.owned = 1;
2370 spin_unlock(&sk->sk_lock.slock);
2372 * The sk_lock has mutex_lock() semantics here:
2374 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2378 EXPORT_SYMBOL(lock_sock_fast);
2380 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2383 if (!sock_flag(sk, SOCK_TIMESTAMP))
2384 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2385 tv = ktime_to_timeval(sk->sk_stamp);
2386 if (tv.tv_sec == -1)
2388 if (tv.tv_sec == 0) {
2389 sk->sk_stamp = ktime_get_real();
2390 tv = ktime_to_timeval(sk->sk_stamp);
2392 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2394 EXPORT_SYMBOL(sock_get_timestamp);
2396 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2399 if (!sock_flag(sk, SOCK_TIMESTAMP))
2400 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2401 ts = ktime_to_timespec(sk->sk_stamp);
2402 if (ts.tv_sec == -1)
2404 if (ts.tv_sec == 0) {
2405 sk->sk_stamp = ktime_get_real();
2406 ts = ktime_to_timespec(sk->sk_stamp);
2408 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2410 EXPORT_SYMBOL(sock_get_timestampns);
2412 void sock_enable_timestamp(struct sock *sk, int flag)
2414 if (!sock_flag(sk, flag)) {
2415 unsigned long previous_flags = sk->sk_flags;
2417 sock_set_flag(sk, flag);
2419 * we just set one of the two flags which require net
2420 * time stamping, but time stamping might have been on
2421 * already because of the other one
2423 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2424 net_enable_timestamp();
2429 * Get a socket option on an socket.
2431 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2432 * asynchronous errors should be reported by getsockopt. We assume
2433 * this means if you specify SO_ERROR (otherwise whats the point of it).
2435 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2436 char __user *optval, int __user *optlen)
2438 struct sock *sk = sock->sk;
2440 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2442 EXPORT_SYMBOL(sock_common_getsockopt);
2444 #ifdef CONFIG_COMPAT
2445 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2446 char __user *optval, int __user *optlen)
2448 struct sock *sk = sock->sk;
2450 if (sk->sk_prot->compat_getsockopt != NULL)
2451 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2453 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2455 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2458 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2459 struct msghdr *msg, size_t size, int flags)
2461 struct sock *sk = sock->sk;
2465 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2466 flags & ~MSG_DONTWAIT, &addr_len);
2468 msg->msg_namelen = addr_len;
2471 EXPORT_SYMBOL(sock_common_recvmsg);
2474 * Set socket options on an inet socket.
2476 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2477 char __user *optval, unsigned int optlen)
2479 struct sock *sk = sock->sk;
2481 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2483 EXPORT_SYMBOL(sock_common_setsockopt);
2485 #ifdef CONFIG_COMPAT
2486 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2487 char __user *optval, unsigned int optlen)
2489 struct sock *sk = sock->sk;
2491 if (sk->sk_prot->compat_setsockopt != NULL)
2492 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2494 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2496 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2499 void sk_common_release(struct sock *sk)
2501 if (sk->sk_prot->destroy)
2502 sk->sk_prot->destroy(sk);
2505 * Observation: when sock_common_release is called, processes have
2506 * no access to socket. But net still has.
2507 * Step one, detach it from networking:
2509 * A. Remove from hash tables.
2512 sk->sk_prot->unhash(sk);
2515 * In this point socket cannot receive new packets, but it is possible
2516 * that some packets are in flight because some CPU runs receiver and
2517 * did hash table lookup before we unhashed socket. They will achieve
2518 * receive queue and will be purged by socket destructor.
2520 * Also we still have packets pending on receive queue and probably,
2521 * our own packets waiting in device queues. sock_destroy will drain
2522 * receive queue, but transmitted packets will delay socket destruction
2523 * until the last reference will be released.
2528 xfrm_sk_free_policy(sk);
2530 sk_refcnt_debug_release(sk);
2532 if (sk->sk_frag.page) {
2533 put_page(sk->sk_frag.page);
2534 sk->sk_frag.page = NULL;
2539 EXPORT_SYMBOL(sk_common_release);
2541 #ifdef CONFIG_PROC_FS
2542 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2544 int val[PROTO_INUSE_NR];
2547 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2549 #ifdef CONFIG_NET_NS
2550 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2552 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2554 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2556 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2558 int cpu, idx = prot->inuse_idx;
2561 for_each_possible_cpu(cpu)
2562 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2564 return res >= 0 ? res : 0;
2566 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2568 static int __net_init sock_inuse_init_net(struct net *net)
2570 net->core.inuse = alloc_percpu(struct prot_inuse);
2571 return net->core.inuse ? 0 : -ENOMEM;
2574 static void __net_exit sock_inuse_exit_net(struct net *net)
2576 free_percpu(net->core.inuse);
2579 static struct pernet_operations net_inuse_ops = {
2580 .init = sock_inuse_init_net,
2581 .exit = sock_inuse_exit_net,
2584 static __init int net_inuse_init(void)
2586 if (register_pernet_subsys(&net_inuse_ops))
2587 panic("Cannot initialize net inuse counters");
2592 core_initcall(net_inuse_init);
2594 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2596 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2598 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2600 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2602 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2604 int cpu, idx = prot->inuse_idx;
2607 for_each_possible_cpu(cpu)
2608 res += per_cpu(prot_inuse, cpu).val[idx];
2610 return res >= 0 ? res : 0;
2612 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2615 static void assign_proto_idx(struct proto *prot)
2617 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2619 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2620 pr_err("PROTO_INUSE_NR exhausted\n");
2624 set_bit(prot->inuse_idx, proto_inuse_idx);
2627 static void release_proto_idx(struct proto *prot)
2629 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2630 clear_bit(prot->inuse_idx, proto_inuse_idx);
2633 static inline void assign_proto_idx(struct proto *prot)
2637 static inline void release_proto_idx(struct proto *prot)
2642 int proto_register(struct proto *prot, int alloc_slab)
2645 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2646 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2649 if (prot->slab == NULL) {
2650 pr_crit("%s: Can't create sock SLAB cache!\n",
2655 if (prot->rsk_prot != NULL) {
2656 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2657 if (prot->rsk_prot->slab_name == NULL)
2658 goto out_free_sock_slab;
2660 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2661 prot->rsk_prot->obj_size, 0,
2662 SLAB_HWCACHE_ALIGN, NULL);
2664 if (prot->rsk_prot->slab == NULL) {
2665 pr_crit("%s: Can't create request sock SLAB cache!\n",
2667 goto out_free_request_sock_slab_name;
2671 if (prot->twsk_prot != NULL) {
2672 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2674 if (prot->twsk_prot->twsk_slab_name == NULL)
2675 goto out_free_request_sock_slab;
2677 prot->twsk_prot->twsk_slab =
2678 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2679 prot->twsk_prot->twsk_obj_size,
2681 SLAB_HWCACHE_ALIGN |
2684 if (prot->twsk_prot->twsk_slab == NULL)
2685 goto out_free_timewait_sock_slab_name;
2689 mutex_lock(&proto_list_mutex);
2690 list_add(&prot->node, &proto_list);
2691 assign_proto_idx(prot);
2692 mutex_unlock(&proto_list_mutex);
2695 out_free_timewait_sock_slab_name:
2696 kfree(prot->twsk_prot->twsk_slab_name);
2697 out_free_request_sock_slab:
2698 if (prot->rsk_prot && prot->rsk_prot->slab) {
2699 kmem_cache_destroy(prot->rsk_prot->slab);
2700 prot->rsk_prot->slab = NULL;
2702 out_free_request_sock_slab_name:
2704 kfree(prot->rsk_prot->slab_name);
2706 kmem_cache_destroy(prot->slab);
2711 EXPORT_SYMBOL(proto_register);
2713 void proto_unregister(struct proto *prot)
2715 mutex_lock(&proto_list_mutex);
2716 release_proto_idx(prot);
2717 list_del(&prot->node);
2718 mutex_unlock(&proto_list_mutex);
2720 if (prot->slab != NULL) {
2721 kmem_cache_destroy(prot->slab);
2725 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2726 kmem_cache_destroy(prot->rsk_prot->slab);
2727 kfree(prot->rsk_prot->slab_name);
2728 prot->rsk_prot->slab = NULL;
2731 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2732 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2733 kfree(prot->twsk_prot->twsk_slab_name);
2734 prot->twsk_prot->twsk_slab = NULL;
2737 EXPORT_SYMBOL(proto_unregister);
2739 #ifdef CONFIG_PROC_FS
2740 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2741 __acquires(proto_list_mutex)
2743 mutex_lock(&proto_list_mutex);
2744 return seq_list_start_head(&proto_list, *pos);
2747 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2749 return seq_list_next(v, &proto_list, pos);
2752 static void proto_seq_stop(struct seq_file *seq, void *v)
2753 __releases(proto_list_mutex)
2755 mutex_unlock(&proto_list_mutex);
2758 static char proto_method_implemented(const void *method)
2760 return method == NULL ? 'n' : 'y';
2762 static long sock_prot_memory_allocated(struct proto *proto)
2764 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2767 static char *sock_prot_memory_pressure(struct proto *proto)
2769 return proto->memory_pressure != NULL ?
2770 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2773 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2776 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2777 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2780 sock_prot_inuse_get(seq_file_net(seq), proto),
2781 sock_prot_memory_allocated(proto),
2782 sock_prot_memory_pressure(proto),
2784 proto->slab == NULL ? "no" : "yes",
2785 module_name(proto->owner),
2786 proto_method_implemented(proto->close),
2787 proto_method_implemented(proto->connect),
2788 proto_method_implemented(proto->disconnect),
2789 proto_method_implemented(proto->accept),
2790 proto_method_implemented(proto->ioctl),
2791 proto_method_implemented(proto->init),
2792 proto_method_implemented(proto->destroy),
2793 proto_method_implemented(proto->shutdown),
2794 proto_method_implemented(proto->setsockopt),
2795 proto_method_implemented(proto->getsockopt),
2796 proto_method_implemented(proto->sendmsg),
2797 proto_method_implemented(proto->recvmsg),
2798 proto_method_implemented(proto->sendpage),
2799 proto_method_implemented(proto->bind),
2800 proto_method_implemented(proto->backlog_rcv),
2801 proto_method_implemented(proto->hash),
2802 proto_method_implemented(proto->unhash),
2803 proto_method_implemented(proto->get_port),
2804 proto_method_implemented(proto->enter_memory_pressure));
2807 static int proto_seq_show(struct seq_file *seq, void *v)
2809 if (v == &proto_list)
2810 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2819 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2821 proto_seq_printf(seq, list_entry(v, struct proto, node));
2825 static const struct seq_operations proto_seq_ops = {
2826 .start = proto_seq_start,
2827 .next = proto_seq_next,
2828 .stop = proto_seq_stop,
2829 .show = proto_seq_show,
2832 static int proto_seq_open(struct inode *inode, struct file *file)
2834 return seq_open_net(inode, file, &proto_seq_ops,
2835 sizeof(struct seq_net_private));
2838 static const struct file_operations proto_seq_fops = {
2839 .owner = THIS_MODULE,
2840 .open = proto_seq_open,
2842 .llseek = seq_lseek,
2843 .release = seq_release_net,
2846 static __net_init int proto_init_net(struct net *net)
2848 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2854 static __net_exit void proto_exit_net(struct net *net)
2856 remove_proc_entry("protocols", net->proc_net);
2860 static __net_initdata struct pernet_operations proto_net_ops = {
2861 .init = proto_init_net,
2862 .exit = proto_exit_net,
2865 static int __init proto_init(void)
2867 return register_pernet_subsys(&proto_net_ops);
2870 subsys_initcall(proto_init);
2872 #endif /* PROC_FS */