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 static DEFINE_MUTEX(proto_list_mutex);
143 static LIST_HEAD(proto_list);
145 #ifdef CONFIG_MEMCG_KMEM
146 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
151 mutex_lock(&proto_list_mutex);
152 list_for_each_entry(proto, &proto_list, node) {
153 if (proto->init_cgroup) {
154 ret = proto->init_cgroup(memcg, ss);
160 mutex_unlock(&proto_list_mutex);
163 list_for_each_entry_continue_reverse(proto, &proto_list, node)
164 if (proto->destroy_cgroup)
165 proto->destroy_cgroup(memcg);
166 mutex_unlock(&proto_list_mutex);
170 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
174 mutex_lock(&proto_list_mutex);
175 list_for_each_entry_reverse(proto, &proto_list, node)
176 if (proto->destroy_cgroup)
177 proto->destroy_cgroup(memcg);
178 mutex_unlock(&proto_list_mutex);
183 * Each address family might have different locking rules, so we have
184 * one slock key per address family:
186 static struct lock_class_key af_family_keys[AF_MAX];
187 static struct lock_class_key af_family_slock_keys[AF_MAX];
189 struct static_key memcg_socket_limit_enabled;
190 EXPORT_SYMBOL(memcg_socket_limit_enabled);
193 * Make lock validator output more readable. (we pre-construct these
194 * strings build-time, so that runtime initialization of socket
197 static const char *const af_family_key_strings[AF_MAX+1] = {
198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
213 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
223 "slock-27" , "slock-28" , "slock-AF_CAN" ,
224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
227 "slock-AF_NFC" , "slock-AF_MAX"
229 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
239 "clock-27" , "clock-28" , "clock-AF_CAN" ,
240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
243 "clock-AF_NFC" , "clock-AF_MAX"
247 * sk_callback_lock locking rules are per-address-family,
248 * so split the lock classes by using a per-AF key:
250 static struct lock_class_key af_callback_keys[AF_MAX];
252 /* Take into consideration the size of the struct sk_buff overhead in the
253 * determination of these values, since that is non-constant across
254 * platforms. This makes socket queueing behavior and performance
255 * not depend upon such differences.
257 #define _SK_MEM_PACKETS 256
258 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
259 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
262 /* Run time adjustable parameters. */
263 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
264 EXPORT_SYMBOL(sysctl_wmem_max);
265 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
266 EXPORT_SYMBOL(sysctl_rmem_max);
267 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
268 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
270 /* Maximal space eaten by iovec or ancillary data plus some space */
271 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
272 EXPORT_SYMBOL(sysctl_optmem_max);
274 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
275 EXPORT_SYMBOL_GPL(memalloc_socks);
278 * sk_set_memalloc - sets %SOCK_MEMALLOC
279 * @sk: socket to set it on
281 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
282 * It's the responsibility of the admin to adjust min_free_kbytes
283 * to meet the requirements
285 void sk_set_memalloc(struct sock *sk)
287 sock_set_flag(sk, SOCK_MEMALLOC);
288 sk->sk_allocation |= __GFP_MEMALLOC;
289 static_key_slow_inc(&memalloc_socks);
291 EXPORT_SYMBOL_GPL(sk_set_memalloc);
293 void sk_clear_memalloc(struct sock *sk)
295 sock_reset_flag(sk, SOCK_MEMALLOC);
296 sk->sk_allocation &= ~__GFP_MEMALLOC;
297 static_key_slow_dec(&memalloc_socks);
300 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
301 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
302 * it has rmem allocations there is a risk that the user of the
303 * socket cannot make forward progress due to exceeding the rmem
304 * limits. By rights, sk_clear_memalloc() should only be called
305 * on sockets being torn down but warn and reset the accounting if
306 * that assumption breaks.
308 if (WARN_ON(sk->sk_forward_alloc))
311 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
313 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
316 unsigned long pflags = current->flags;
318 /* these should have been dropped before queueing */
319 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
321 current->flags |= PF_MEMALLOC;
322 ret = sk->sk_backlog_rcv(sk, skb);
323 tsk_restore_flags(current, pflags, PF_MEMALLOC);
327 EXPORT_SYMBOL(__sk_backlog_rcv);
329 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
333 if (optlen < sizeof(tv))
335 if (copy_from_user(&tv, optval, sizeof(tv)))
337 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
341 static int warned __read_mostly;
344 if (warned < 10 && net_ratelimit()) {
346 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
347 __func__, current->comm, task_pid_nr(current));
351 *timeo_p = MAX_SCHEDULE_TIMEOUT;
352 if (tv.tv_sec == 0 && tv.tv_usec == 0)
354 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
355 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
359 static void sock_warn_obsolete_bsdism(const char *name)
362 static char warncomm[TASK_COMM_LEN];
363 if (strcmp(warncomm, current->comm) && warned < 5) {
364 strcpy(warncomm, current->comm);
365 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
371 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
373 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
375 if (sk->sk_flags & flags) {
376 sk->sk_flags &= ~flags;
377 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
378 net_disable_timestamp();
383 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
388 struct sk_buff_head *list = &sk->sk_receive_queue;
390 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
391 atomic_inc(&sk->sk_drops);
392 trace_sock_rcvqueue_full(sk, skb);
396 err = sk_filter(sk, skb);
400 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
401 atomic_inc(&sk->sk_drops);
406 skb_set_owner_r(skb, sk);
408 /* Cache the SKB length before we tack it onto the receive
409 * queue. Once it is added it no longer belongs to us and
410 * may be freed by other threads of control pulling packets
415 /* we escape from rcu protected region, make sure we dont leak
420 spin_lock_irqsave(&list->lock, flags);
421 skb->dropcount = atomic_read(&sk->sk_drops);
422 __skb_queue_tail(list, skb);
423 spin_unlock_irqrestore(&list->lock, flags);
425 if (!sock_flag(sk, SOCK_DEAD))
426 sk->sk_data_ready(sk, skb_len);
429 EXPORT_SYMBOL(sock_queue_rcv_skb);
431 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
433 int rc = NET_RX_SUCCESS;
435 if (sk_filter(sk, skb))
436 goto discard_and_relse;
440 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
441 atomic_inc(&sk->sk_drops);
442 goto discard_and_relse;
445 bh_lock_sock_nested(sk);
448 if (!sock_owned_by_user(sk)) {
450 * trylock + unlock semantics:
452 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
454 rc = sk_backlog_rcv(sk, skb);
456 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
457 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
459 atomic_inc(&sk->sk_drops);
460 goto discard_and_relse;
471 EXPORT_SYMBOL(sk_receive_skb);
473 void sk_reset_txq(struct sock *sk)
475 sk_tx_queue_clear(sk);
477 EXPORT_SYMBOL(sk_reset_txq);
479 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
481 struct dst_entry *dst = __sk_dst_get(sk);
483 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
484 sk_tx_queue_clear(sk);
485 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
492 EXPORT_SYMBOL(__sk_dst_check);
494 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
496 struct dst_entry *dst = sk_dst_get(sk);
498 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
506 EXPORT_SYMBOL(sk_dst_check);
508 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
511 int ret = -ENOPROTOOPT;
512 #ifdef CONFIG_NETDEVICES
513 struct net *net = sock_net(sk);
514 char devname[IFNAMSIZ];
519 if (!ns_capable(net->user_ns, CAP_NET_RAW))
526 /* Bind this socket to a particular device like "eth0",
527 * as specified in the passed interface name. If the
528 * name is "" or the option length is zero the socket
531 if (optlen > IFNAMSIZ - 1)
532 optlen = IFNAMSIZ - 1;
533 memset(devname, 0, sizeof(devname));
536 if (copy_from_user(devname, optval, optlen))
540 if (devname[0] != '\0') {
541 struct net_device *dev;
544 dev = dev_get_by_name_rcu(net, devname);
546 index = dev->ifindex;
554 sk->sk_bound_dev_if = index;
566 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
567 int __user *optlen, int len)
569 int ret = -ENOPROTOOPT;
570 #ifdef CONFIG_NETDEVICES
571 struct net *net = sock_net(sk);
572 struct net_device *dev;
573 char devname[IFNAMSIZ];
576 if (sk->sk_bound_dev_if == 0) {
586 seq = read_seqcount_begin(&devnet_rename_seq);
588 dev = dev_get_by_index_rcu(net, sk->sk_bound_dev_if);
595 strcpy(devname, dev->name);
597 if (read_seqcount_retry(&devnet_rename_seq, seq))
600 len = strlen(devname) + 1;
603 if (copy_to_user(optval, devname, len))
608 if (put_user(len, optlen))
619 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
622 sock_set_flag(sk, bit);
624 sock_reset_flag(sk, bit);
628 * This is meant for all protocols to use and covers goings on
629 * at the socket level. Everything here is generic.
632 int sock_setsockopt(struct socket *sock, int level, int optname,
633 char __user *optval, unsigned int optlen)
635 struct sock *sk = sock->sk;
642 * Options without arguments
645 if (optname == SO_BINDTODEVICE)
646 return sock_setbindtodevice(sk, optval, optlen);
648 if (optlen < sizeof(int))
651 if (get_user(val, (int __user *)optval))
654 valbool = val ? 1 : 0;
660 if (val && !capable(CAP_NET_ADMIN))
663 sock_valbool_flag(sk, SOCK_DBG, valbool);
666 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
669 sk->sk_reuseport = valbool;
678 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
681 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
684 /* Don't error on this BSD doesn't and if you think
685 * about it this is right. Otherwise apps have to
686 * play 'guess the biggest size' games. RCVBUF/SNDBUF
687 * are treated in BSD as hints
689 val = min_t(u32, val, sysctl_wmem_max);
691 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
692 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
693 /* Wake up sending tasks if we upped the value. */
694 sk->sk_write_space(sk);
698 if (!capable(CAP_NET_ADMIN)) {
705 /* Don't error on this BSD doesn't and if you think
706 * about it this is right. Otherwise apps have to
707 * play 'guess the biggest size' games. RCVBUF/SNDBUF
708 * are treated in BSD as hints
710 val = min_t(u32, val, sysctl_rmem_max);
712 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
714 * We double it on the way in to account for
715 * "struct sk_buff" etc. overhead. Applications
716 * assume that the SO_RCVBUF setting they make will
717 * allow that much actual data to be received on that
720 * Applications are unaware that "struct sk_buff" and
721 * other overheads allocate from the receive buffer
722 * during socket buffer allocation.
724 * And after considering the possible alternatives,
725 * returning the value we actually used in getsockopt
726 * is the most desirable behavior.
728 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
732 if (!capable(CAP_NET_ADMIN)) {
740 if (sk->sk_protocol == IPPROTO_TCP &&
741 sk->sk_type == SOCK_STREAM)
742 tcp_set_keepalive(sk, valbool);
744 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
748 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
752 sk->sk_no_check = valbool;
756 if ((val >= 0 && val <= 6) ||
757 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
758 sk->sk_priority = val;
764 if (optlen < sizeof(ling)) {
765 ret = -EINVAL; /* 1003.1g */
768 if (copy_from_user(&ling, optval, sizeof(ling))) {
773 sock_reset_flag(sk, SOCK_LINGER);
775 #if (BITS_PER_LONG == 32)
776 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
777 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
780 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
781 sock_set_flag(sk, SOCK_LINGER);
786 sock_warn_obsolete_bsdism("setsockopt");
791 set_bit(SOCK_PASSCRED, &sock->flags);
793 clear_bit(SOCK_PASSCRED, &sock->flags);
799 if (optname == SO_TIMESTAMP)
800 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
802 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
803 sock_set_flag(sk, SOCK_RCVTSTAMP);
804 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
806 sock_reset_flag(sk, SOCK_RCVTSTAMP);
807 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
811 case SO_TIMESTAMPING:
812 if (val & ~SOF_TIMESTAMPING_MASK) {
816 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
817 val & SOF_TIMESTAMPING_TX_HARDWARE);
818 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
819 val & SOF_TIMESTAMPING_TX_SOFTWARE);
820 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
821 val & SOF_TIMESTAMPING_RX_HARDWARE);
822 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
823 sock_enable_timestamp(sk,
824 SOCK_TIMESTAMPING_RX_SOFTWARE);
826 sock_disable_timestamp(sk,
827 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
828 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
829 val & SOF_TIMESTAMPING_SOFTWARE);
830 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
831 val & SOF_TIMESTAMPING_SYS_HARDWARE);
832 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
833 val & SOF_TIMESTAMPING_RAW_HARDWARE);
839 sk->sk_rcvlowat = val ? : 1;
843 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
847 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
850 case SO_ATTACH_FILTER:
852 if (optlen == sizeof(struct sock_fprog)) {
853 struct sock_fprog fprog;
856 if (copy_from_user(&fprog, optval, sizeof(fprog)))
859 ret = sk_attach_filter(&fprog, sk);
863 case SO_DETACH_FILTER:
864 ret = sk_detach_filter(sk);
868 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
871 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
876 set_bit(SOCK_PASSSEC, &sock->flags);
878 clear_bit(SOCK_PASSSEC, &sock->flags);
881 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
887 /* We implement the SO_SNDLOWAT etc to
888 not be settable (1003.1g 5.3) */
890 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
894 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
898 if (sock->ops->set_peek_off)
899 sock->ops->set_peek_off(sk, val);
905 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
915 EXPORT_SYMBOL(sock_setsockopt);
918 void cred_to_ucred(struct pid *pid, const struct cred *cred,
921 ucred->pid = pid_vnr(pid);
922 ucred->uid = ucred->gid = -1;
924 struct user_namespace *current_ns = current_user_ns();
926 ucred->uid = from_kuid_munged(current_ns, cred->euid);
927 ucred->gid = from_kgid_munged(current_ns, cred->egid);
930 EXPORT_SYMBOL_GPL(cred_to_ucred);
932 int sock_getsockopt(struct socket *sock, int level, int optname,
933 char __user *optval, int __user *optlen)
935 struct sock *sk = sock->sk;
943 int lv = sizeof(int);
946 if (get_user(len, optlen))
951 memset(&v, 0, sizeof(v));
955 v.val = sock_flag(sk, SOCK_DBG);
959 v.val = sock_flag(sk, SOCK_LOCALROUTE);
963 v.val = sock_flag(sk, SOCK_BROADCAST);
967 v.val = sk->sk_sndbuf;
971 v.val = sk->sk_rcvbuf;
975 v.val = sk->sk_reuse;
979 v.val = sk->sk_reuseport;
983 v.val = sock_flag(sk, SOCK_KEEPOPEN);
991 v.val = sk->sk_protocol;
995 v.val = sk->sk_family;
999 v.val = -sock_error(sk);
1001 v.val = xchg(&sk->sk_err_soft, 0);
1005 v.val = sock_flag(sk, SOCK_URGINLINE);
1009 v.val = sk->sk_no_check;
1013 v.val = sk->sk_priority;
1017 lv = sizeof(v.ling);
1018 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1019 v.ling.l_linger = sk->sk_lingertime / HZ;
1023 sock_warn_obsolete_bsdism("getsockopt");
1027 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1028 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1031 case SO_TIMESTAMPNS:
1032 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1035 case SO_TIMESTAMPING:
1037 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
1038 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
1039 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
1040 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
1041 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
1042 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
1043 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1044 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
1045 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
1046 v.val |= SOF_TIMESTAMPING_SOFTWARE;
1047 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
1048 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
1049 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
1050 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
1054 lv = sizeof(struct timeval);
1055 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1059 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1060 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1065 lv = sizeof(struct timeval);
1066 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1070 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1071 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1076 v.val = sk->sk_rcvlowat;
1084 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1089 struct ucred peercred;
1090 if (len > sizeof(peercred))
1091 len = sizeof(peercred);
1092 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1093 if (copy_to_user(optval, &peercred, len))
1102 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1106 if (copy_to_user(optval, address, len))
1111 /* Dubious BSD thing... Probably nobody even uses it, but
1112 * the UNIX standard wants it for whatever reason... -DaveM
1115 v.val = sk->sk_state == TCP_LISTEN;
1119 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1123 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1126 v.val = sk->sk_mark;
1130 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1133 case SO_WIFI_STATUS:
1134 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1138 if (!sock->ops->set_peek_off)
1141 v.val = sk->sk_peek_off;
1144 v.val = sock_flag(sk, SOCK_NOFCS);
1147 case SO_BINDTODEVICE:
1148 return sock_getbindtodevice(sk, optval, optlen, len);
1151 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1157 case SO_LOCK_FILTER:
1158 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1162 return -ENOPROTOOPT;
1167 if (copy_to_user(optval, &v, len))
1170 if (put_user(len, optlen))
1176 * Initialize an sk_lock.
1178 * (We also register the sk_lock with the lock validator.)
1180 static inline void sock_lock_init(struct sock *sk)
1182 sock_lock_init_class_and_name(sk,
1183 af_family_slock_key_strings[sk->sk_family],
1184 af_family_slock_keys + sk->sk_family,
1185 af_family_key_strings[sk->sk_family],
1186 af_family_keys + sk->sk_family);
1190 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1191 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1192 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1194 static void sock_copy(struct sock *nsk, const struct sock *osk)
1196 #ifdef CONFIG_SECURITY_NETWORK
1197 void *sptr = nsk->sk_security;
1199 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1201 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1202 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1204 #ifdef CONFIG_SECURITY_NETWORK
1205 nsk->sk_security = sptr;
1206 security_sk_clone(osk, nsk);
1211 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1212 * un-modified. Special care is taken when initializing object to zero.
1214 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1216 if (offsetof(struct sock, sk_node.next) != 0)
1217 memset(sk, 0, offsetof(struct sock, sk_node.next));
1218 memset(&sk->sk_node.pprev, 0,
1219 size - offsetof(struct sock, sk_node.pprev));
1222 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1224 unsigned long nulls1, nulls2;
1226 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1227 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1228 if (nulls1 > nulls2)
1229 swap(nulls1, nulls2);
1232 memset((char *)sk, 0, nulls1);
1233 memset((char *)sk + nulls1 + sizeof(void *), 0,
1234 nulls2 - nulls1 - sizeof(void *));
1235 memset((char *)sk + nulls2 + sizeof(void *), 0,
1236 size - nulls2 - sizeof(void *));
1238 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1240 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1244 struct kmem_cache *slab;
1248 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1251 if (priority & __GFP_ZERO) {
1253 prot->clear_sk(sk, prot->obj_size);
1255 sk_prot_clear_nulls(sk, prot->obj_size);
1258 sk = kmalloc(prot->obj_size, priority);
1261 kmemcheck_annotate_bitfield(sk, flags);
1263 if (security_sk_alloc(sk, family, priority))
1266 if (!try_module_get(prot->owner))
1268 sk_tx_queue_clear(sk);
1274 security_sk_free(sk);
1277 kmem_cache_free(slab, sk);
1283 static void sk_prot_free(struct proto *prot, struct sock *sk)
1285 struct kmem_cache *slab;
1286 struct module *owner;
1288 owner = prot->owner;
1291 security_sk_free(sk);
1293 kmem_cache_free(slab, sk);
1299 #ifdef CONFIG_CGROUPS
1300 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1301 void sock_update_classid(struct sock *sk, struct task_struct *task)
1305 classid = task_cls_classid(task);
1306 if (classid != sk->sk_classid)
1307 sk->sk_classid = classid;
1309 EXPORT_SYMBOL(sock_update_classid);
1312 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1313 void sock_update_netprioidx(struct sock *sk, struct task_struct *task)
1318 sk->sk_cgrp_prioidx = task_netprioidx(task);
1320 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1325 * sk_alloc - All socket objects are allocated here
1326 * @net: the applicable net namespace
1327 * @family: protocol family
1328 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1329 * @prot: struct proto associated with this new sock instance
1331 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1336 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1338 sk->sk_family = family;
1340 * See comment in struct sock definition to understand
1341 * why we need sk_prot_creator -acme
1343 sk->sk_prot = sk->sk_prot_creator = prot;
1345 sock_net_set(sk, get_net(net));
1346 atomic_set(&sk->sk_wmem_alloc, 1);
1348 sock_update_classid(sk, current);
1349 sock_update_netprioidx(sk, current);
1354 EXPORT_SYMBOL(sk_alloc);
1356 static void __sk_free(struct sock *sk)
1358 struct sk_filter *filter;
1360 if (sk->sk_destruct)
1361 sk->sk_destruct(sk);
1363 filter = rcu_dereference_check(sk->sk_filter,
1364 atomic_read(&sk->sk_wmem_alloc) == 0);
1366 sk_filter_uncharge(sk, filter);
1367 RCU_INIT_POINTER(sk->sk_filter, NULL);
1370 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1372 if (atomic_read(&sk->sk_omem_alloc))
1373 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1374 __func__, atomic_read(&sk->sk_omem_alloc));
1376 if (sk->sk_peer_cred)
1377 put_cred(sk->sk_peer_cred);
1378 put_pid(sk->sk_peer_pid);
1379 put_net(sock_net(sk));
1380 sk_prot_free(sk->sk_prot_creator, sk);
1383 void sk_free(struct sock *sk)
1386 * We subtract one from sk_wmem_alloc and can know if
1387 * some packets are still in some tx queue.
1388 * If not null, sock_wfree() will call __sk_free(sk) later
1390 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1393 EXPORT_SYMBOL(sk_free);
1396 * Last sock_put should drop reference to sk->sk_net. It has already
1397 * been dropped in sk_change_net. Taking reference to stopping namespace
1399 * Take reference to a socket to remove it from hash _alive_ and after that
1400 * destroy it in the context of init_net.
1402 void sk_release_kernel(struct sock *sk)
1404 if (sk == NULL || sk->sk_socket == NULL)
1408 sock_release(sk->sk_socket);
1409 release_net(sock_net(sk));
1410 sock_net_set(sk, get_net(&init_net));
1413 EXPORT_SYMBOL(sk_release_kernel);
1415 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1417 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1418 sock_update_memcg(newsk);
1422 * sk_clone_lock - clone a socket, and lock its clone
1423 * @sk: the socket to clone
1424 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1426 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1428 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1432 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1433 if (newsk != NULL) {
1434 struct sk_filter *filter;
1436 sock_copy(newsk, sk);
1439 get_net(sock_net(newsk));
1440 sk_node_init(&newsk->sk_node);
1441 sock_lock_init(newsk);
1442 bh_lock_sock(newsk);
1443 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1444 newsk->sk_backlog.len = 0;
1446 atomic_set(&newsk->sk_rmem_alloc, 0);
1448 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1450 atomic_set(&newsk->sk_wmem_alloc, 1);
1451 atomic_set(&newsk->sk_omem_alloc, 0);
1452 skb_queue_head_init(&newsk->sk_receive_queue);
1453 skb_queue_head_init(&newsk->sk_write_queue);
1454 #ifdef CONFIG_NET_DMA
1455 skb_queue_head_init(&newsk->sk_async_wait_queue);
1458 spin_lock_init(&newsk->sk_dst_lock);
1459 rwlock_init(&newsk->sk_callback_lock);
1460 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1461 af_callback_keys + newsk->sk_family,
1462 af_family_clock_key_strings[newsk->sk_family]);
1464 newsk->sk_dst_cache = NULL;
1465 newsk->sk_wmem_queued = 0;
1466 newsk->sk_forward_alloc = 0;
1467 newsk->sk_send_head = NULL;
1468 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1470 sock_reset_flag(newsk, SOCK_DONE);
1471 skb_queue_head_init(&newsk->sk_error_queue);
1473 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1475 sk_filter_charge(newsk, filter);
1477 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1478 /* It is still raw copy of parent, so invalidate
1479 * destructor and make plain sk_free() */
1480 newsk->sk_destruct = NULL;
1481 bh_unlock_sock(newsk);
1488 newsk->sk_priority = 0;
1490 * Before updating sk_refcnt, we must commit prior changes to memory
1491 * (Documentation/RCU/rculist_nulls.txt for details)
1494 atomic_set(&newsk->sk_refcnt, 2);
1497 * Increment the counter in the same struct proto as the master
1498 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1499 * is the same as sk->sk_prot->socks, as this field was copied
1502 * This _changes_ the previous behaviour, where
1503 * tcp_create_openreq_child always was incrementing the
1504 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1505 * to be taken into account in all callers. -acme
1507 sk_refcnt_debug_inc(newsk);
1508 sk_set_socket(newsk, NULL);
1509 newsk->sk_wq = NULL;
1511 sk_update_clone(sk, newsk);
1513 if (newsk->sk_prot->sockets_allocated)
1514 sk_sockets_allocated_inc(newsk);
1516 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1517 net_enable_timestamp();
1522 EXPORT_SYMBOL_GPL(sk_clone_lock);
1524 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1526 __sk_dst_set(sk, dst);
1527 sk->sk_route_caps = dst->dev->features;
1528 if (sk->sk_route_caps & NETIF_F_GSO)
1529 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1530 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1531 if (sk_can_gso(sk)) {
1532 if (dst->header_len) {
1533 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1535 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1536 sk->sk_gso_max_size = dst->dev->gso_max_size;
1537 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1541 EXPORT_SYMBOL_GPL(sk_setup_caps);
1544 * Simple resource managers for sockets.
1549 * Write buffer destructor automatically called from kfree_skb.
1551 void sock_wfree(struct sk_buff *skb)
1553 struct sock *sk = skb->sk;
1554 unsigned int len = skb->truesize;
1556 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1558 * Keep a reference on sk_wmem_alloc, this will be released
1559 * after sk_write_space() call
1561 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1562 sk->sk_write_space(sk);
1566 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1567 * could not do because of in-flight packets
1569 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1572 EXPORT_SYMBOL(sock_wfree);
1575 * Read buffer destructor automatically called from kfree_skb.
1577 void sock_rfree(struct sk_buff *skb)
1579 struct sock *sk = skb->sk;
1580 unsigned int len = skb->truesize;
1582 atomic_sub(len, &sk->sk_rmem_alloc);
1583 sk_mem_uncharge(sk, len);
1585 EXPORT_SYMBOL(sock_rfree);
1587 void sock_edemux(struct sk_buff *skb)
1589 struct sock *sk = skb->sk;
1592 if (sk->sk_state == TCP_TIME_WAIT)
1593 inet_twsk_put(inet_twsk(sk));
1598 EXPORT_SYMBOL(sock_edemux);
1600 kuid_t sock_i_uid(struct sock *sk)
1604 read_lock_bh(&sk->sk_callback_lock);
1605 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1606 read_unlock_bh(&sk->sk_callback_lock);
1609 EXPORT_SYMBOL(sock_i_uid);
1611 unsigned long sock_i_ino(struct sock *sk)
1615 read_lock_bh(&sk->sk_callback_lock);
1616 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1617 read_unlock_bh(&sk->sk_callback_lock);
1620 EXPORT_SYMBOL(sock_i_ino);
1623 * Allocate a skb from the socket's send buffer.
1625 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1628 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1629 struct sk_buff *skb = alloc_skb(size, priority);
1631 skb_set_owner_w(skb, sk);
1637 EXPORT_SYMBOL(sock_wmalloc);
1640 * Allocate a skb from the socket's receive buffer.
1642 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1645 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1646 struct sk_buff *skb = alloc_skb(size, priority);
1648 skb_set_owner_r(skb, sk);
1656 * Allocate a memory block from the socket's option memory buffer.
1658 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1660 if ((unsigned int)size <= sysctl_optmem_max &&
1661 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1663 /* First do the add, to avoid the race if kmalloc
1666 atomic_add(size, &sk->sk_omem_alloc);
1667 mem = kmalloc(size, priority);
1670 atomic_sub(size, &sk->sk_omem_alloc);
1674 EXPORT_SYMBOL(sock_kmalloc);
1677 * Free an option memory block.
1679 void sock_kfree_s(struct sock *sk, void *mem, int size)
1682 atomic_sub(size, &sk->sk_omem_alloc);
1684 EXPORT_SYMBOL(sock_kfree_s);
1686 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1687 I think, these locks should be removed for datagram sockets.
1689 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1693 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1697 if (signal_pending(current))
1699 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1700 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1701 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1703 if (sk->sk_shutdown & SEND_SHUTDOWN)
1707 timeo = schedule_timeout(timeo);
1709 finish_wait(sk_sleep(sk), &wait);
1715 * Generic send/receive buffer handlers
1718 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1719 unsigned long data_len, int noblock,
1722 struct sk_buff *skb;
1726 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1729 if (npages > MAX_SKB_FRAGS)
1732 gfp_mask = sk->sk_allocation;
1733 if (gfp_mask & __GFP_WAIT)
1734 gfp_mask |= __GFP_REPEAT;
1736 timeo = sock_sndtimeo(sk, noblock);
1738 err = sock_error(sk);
1743 if (sk->sk_shutdown & SEND_SHUTDOWN)
1746 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1747 skb = alloc_skb(header_len, gfp_mask);
1751 /* No pages, we're done... */
1755 skb->truesize += data_len;
1756 skb_shinfo(skb)->nr_frags = npages;
1757 for (i = 0; i < npages; i++) {
1760 page = alloc_pages(sk->sk_allocation, 0);
1763 skb_shinfo(skb)->nr_frags = i;
1768 __skb_fill_page_desc(skb, i,
1770 (data_len >= PAGE_SIZE ?
1773 data_len -= PAGE_SIZE;
1776 /* Full success... */
1782 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1783 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1787 if (signal_pending(current))
1789 timeo = sock_wait_for_wmem(sk, timeo);
1792 skb_set_owner_w(skb, sk);
1796 err = sock_intr_errno(timeo);
1801 EXPORT_SYMBOL(sock_alloc_send_pskb);
1803 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1804 int noblock, int *errcode)
1806 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1808 EXPORT_SYMBOL(sock_alloc_send_skb);
1810 /* On 32bit arches, an skb frag is limited to 2^15 */
1811 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1813 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1818 if (atomic_read(&pfrag->page->_count) == 1) {
1822 if (pfrag->offset < pfrag->size)
1824 put_page(pfrag->page);
1827 /* We restrict high order allocations to users that can afford to wait */
1828 order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1831 gfp_t gfp = sk->sk_allocation;
1834 gfp |= __GFP_COMP | __GFP_NOWARN;
1835 pfrag->page = alloc_pages(gfp, order);
1836 if (likely(pfrag->page)) {
1838 pfrag->size = PAGE_SIZE << order;
1841 } while (--order >= 0);
1843 sk_enter_memory_pressure(sk);
1844 sk_stream_moderate_sndbuf(sk);
1847 EXPORT_SYMBOL(sk_page_frag_refill);
1849 static void __lock_sock(struct sock *sk)
1850 __releases(&sk->sk_lock.slock)
1851 __acquires(&sk->sk_lock.slock)
1856 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1857 TASK_UNINTERRUPTIBLE);
1858 spin_unlock_bh(&sk->sk_lock.slock);
1860 spin_lock_bh(&sk->sk_lock.slock);
1861 if (!sock_owned_by_user(sk))
1864 finish_wait(&sk->sk_lock.wq, &wait);
1867 static void __release_sock(struct sock *sk)
1868 __releases(&sk->sk_lock.slock)
1869 __acquires(&sk->sk_lock.slock)
1871 struct sk_buff *skb = sk->sk_backlog.head;
1874 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1878 struct sk_buff *next = skb->next;
1881 WARN_ON_ONCE(skb_dst_is_noref(skb));
1883 sk_backlog_rcv(sk, skb);
1886 * We are in process context here with softirqs
1887 * disabled, use cond_resched_softirq() to preempt.
1888 * This is safe to do because we've taken the backlog
1891 cond_resched_softirq();
1894 } while (skb != NULL);
1897 } while ((skb = sk->sk_backlog.head) != NULL);
1900 * Doing the zeroing here guarantee we can not loop forever
1901 * while a wild producer attempts to flood us.
1903 sk->sk_backlog.len = 0;
1907 * sk_wait_data - wait for data to arrive at sk_receive_queue
1908 * @sk: sock to wait on
1909 * @timeo: for how long
1911 * Now socket state including sk->sk_err is changed only under lock,
1912 * hence we may omit checks after joining wait queue.
1913 * We check receive queue before schedule() only as optimization;
1914 * it is very likely that release_sock() added new data.
1916 int sk_wait_data(struct sock *sk, long *timeo)
1921 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1922 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1923 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1924 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1925 finish_wait(sk_sleep(sk), &wait);
1928 EXPORT_SYMBOL(sk_wait_data);
1931 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1933 * @size: memory size to allocate
1934 * @kind: allocation type
1936 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1937 * rmem allocation. This function assumes that protocols which have
1938 * memory_pressure use sk_wmem_queued as write buffer accounting.
1940 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1942 struct proto *prot = sk->sk_prot;
1943 int amt = sk_mem_pages(size);
1945 int parent_status = UNDER_LIMIT;
1947 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1949 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1952 if (parent_status == UNDER_LIMIT &&
1953 allocated <= sk_prot_mem_limits(sk, 0)) {
1954 sk_leave_memory_pressure(sk);
1958 /* Under pressure. (we or our parents) */
1959 if ((parent_status > SOFT_LIMIT) ||
1960 allocated > sk_prot_mem_limits(sk, 1))
1961 sk_enter_memory_pressure(sk);
1963 /* Over hard limit (we or our parents) */
1964 if ((parent_status == OVER_LIMIT) ||
1965 (allocated > sk_prot_mem_limits(sk, 2)))
1966 goto suppress_allocation;
1968 /* guarantee minimum buffer size under pressure */
1969 if (kind == SK_MEM_RECV) {
1970 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1973 } else { /* SK_MEM_SEND */
1974 if (sk->sk_type == SOCK_STREAM) {
1975 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1977 } else if (atomic_read(&sk->sk_wmem_alloc) <
1978 prot->sysctl_wmem[0])
1982 if (sk_has_memory_pressure(sk)) {
1985 if (!sk_under_memory_pressure(sk))
1987 alloc = sk_sockets_allocated_read_positive(sk);
1988 if (sk_prot_mem_limits(sk, 2) > alloc *
1989 sk_mem_pages(sk->sk_wmem_queued +
1990 atomic_read(&sk->sk_rmem_alloc) +
1991 sk->sk_forward_alloc))
1995 suppress_allocation:
1997 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1998 sk_stream_moderate_sndbuf(sk);
2000 /* Fail only if socket is _under_ its sndbuf.
2001 * In this case we cannot block, so that we have to fail.
2003 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2007 trace_sock_exceed_buf_limit(sk, prot, allocated);
2009 /* Alas. Undo changes. */
2010 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2012 sk_memory_allocated_sub(sk, amt);
2016 EXPORT_SYMBOL(__sk_mem_schedule);
2019 * __sk_reclaim - reclaim memory_allocated
2022 void __sk_mem_reclaim(struct sock *sk)
2024 sk_memory_allocated_sub(sk,
2025 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2026 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2028 if (sk_under_memory_pressure(sk) &&
2029 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2030 sk_leave_memory_pressure(sk);
2032 EXPORT_SYMBOL(__sk_mem_reclaim);
2036 * Set of default routines for initialising struct proto_ops when
2037 * the protocol does not support a particular function. In certain
2038 * cases where it makes no sense for a protocol to have a "do nothing"
2039 * function, some default processing is provided.
2042 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2046 EXPORT_SYMBOL(sock_no_bind);
2048 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2053 EXPORT_SYMBOL(sock_no_connect);
2055 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2059 EXPORT_SYMBOL(sock_no_socketpair);
2061 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2065 EXPORT_SYMBOL(sock_no_accept);
2067 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2072 EXPORT_SYMBOL(sock_no_getname);
2074 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2078 EXPORT_SYMBOL(sock_no_poll);
2080 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2084 EXPORT_SYMBOL(sock_no_ioctl);
2086 int sock_no_listen(struct socket *sock, int backlog)
2090 EXPORT_SYMBOL(sock_no_listen);
2092 int sock_no_shutdown(struct socket *sock, int how)
2096 EXPORT_SYMBOL(sock_no_shutdown);
2098 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2099 char __user *optval, unsigned int optlen)
2103 EXPORT_SYMBOL(sock_no_setsockopt);
2105 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2106 char __user *optval, int __user *optlen)
2110 EXPORT_SYMBOL(sock_no_getsockopt);
2112 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2117 EXPORT_SYMBOL(sock_no_sendmsg);
2119 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2120 size_t len, int flags)
2124 EXPORT_SYMBOL(sock_no_recvmsg);
2126 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2128 /* Mirror missing mmap method error code */
2131 EXPORT_SYMBOL(sock_no_mmap);
2133 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2136 struct msghdr msg = {.msg_flags = flags};
2138 char *kaddr = kmap(page);
2139 iov.iov_base = kaddr + offset;
2141 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2145 EXPORT_SYMBOL(sock_no_sendpage);
2148 * Default Socket Callbacks
2151 static void sock_def_wakeup(struct sock *sk)
2153 struct socket_wq *wq;
2156 wq = rcu_dereference(sk->sk_wq);
2157 if (wq_has_sleeper(wq))
2158 wake_up_interruptible_all(&wq->wait);
2162 static void sock_def_error_report(struct sock *sk)
2164 struct socket_wq *wq;
2167 wq = rcu_dereference(sk->sk_wq);
2168 if (wq_has_sleeper(wq))
2169 wake_up_interruptible_poll(&wq->wait, POLLERR);
2170 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2174 static void sock_def_readable(struct sock *sk, int len)
2176 struct socket_wq *wq;
2179 wq = rcu_dereference(sk->sk_wq);
2180 if (wq_has_sleeper(wq))
2181 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2182 POLLRDNORM | POLLRDBAND);
2183 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2187 static void sock_def_write_space(struct sock *sk)
2189 struct socket_wq *wq;
2193 /* Do not wake up a writer until he can make "significant"
2196 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2197 wq = rcu_dereference(sk->sk_wq);
2198 if (wq_has_sleeper(wq))
2199 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2200 POLLWRNORM | POLLWRBAND);
2202 /* Should agree with poll, otherwise some programs break */
2203 if (sock_writeable(sk))
2204 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2210 static void sock_def_destruct(struct sock *sk)
2212 kfree(sk->sk_protinfo);
2215 void sk_send_sigurg(struct sock *sk)
2217 if (sk->sk_socket && sk->sk_socket->file)
2218 if (send_sigurg(&sk->sk_socket->file->f_owner))
2219 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2221 EXPORT_SYMBOL(sk_send_sigurg);
2223 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2224 unsigned long expires)
2226 if (!mod_timer(timer, expires))
2229 EXPORT_SYMBOL(sk_reset_timer);
2231 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2233 if (del_timer(timer))
2236 EXPORT_SYMBOL(sk_stop_timer);
2238 void sock_init_data(struct socket *sock, struct sock *sk)
2240 skb_queue_head_init(&sk->sk_receive_queue);
2241 skb_queue_head_init(&sk->sk_write_queue);
2242 skb_queue_head_init(&sk->sk_error_queue);
2243 #ifdef CONFIG_NET_DMA
2244 skb_queue_head_init(&sk->sk_async_wait_queue);
2247 sk->sk_send_head = NULL;
2249 init_timer(&sk->sk_timer);
2251 sk->sk_allocation = GFP_KERNEL;
2252 sk->sk_rcvbuf = sysctl_rmem_default;
2253 sk->sk_sndbuf = sysctl_wmem_default;
2254 sk->sk_state = TCP_CLOSE;
2255 sk_set_socket(sk, sock);
2257 sock_set_flag(sk, SOCK_ZAPPED);
2260 sk->sk_type = sock->type;
2261 sk->sk_wq = sock->wq;
2266 spin_lock_init(&sk->sk_dst_lock);
2267 rwlock_init(&sk->sk_callback_lock);
2268 lockdep_set_class_and_name(&sk->sk_callback_lock,
2269 af_callback_keys + sk->sk_family,
2270 af_family_clock_key_strings[sk->sk_family]);
2272 sk->sk_state_change = sock_def_wakeup;
2273 sk->sk_data_ready = sock_def_readable;
2274 sk->sk_write_space = sock_def_write_space;
2275 sk->sk_error_report = sock_def_error_report;
2276 sk->sk_destruct = sock_def_destruct;
2278 sk->sk_frag.page = NULL;
2279 sk->sk_frag.offset = 0;
2280 sk->sk_peek_off = -1;
2282 sk->sk_peer_pid = NULL;
2283 sk->sk_peer_cred = NULL;
2284 sk->sk_write_pending = 0;
2285 sk->sk_rcvlowat = 1;
2286 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2287 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2289 sk->sk_stamp = ktime_set(-1L, 0);
2292 * Before updating sk_refcnt, we must commit prior changes to memory
2293 * (Documentation/RCU/rculist_nulls.txt for details)
2296 atomic_set(&sk->sk_refcnt, 1);
2297 atomic_set(&sk->sk_drops, 0);
2299 EXPORT_SYMBOL(sock_init_data);
2301 void lock_sock_nested(struct sock *sk, int subclass)
2304 spin_lock_bh(&sk->sk_lock.slock);
2305 if (sk->sk_lock.owned)
2307 sk->sk_lock.owned = 1;
2308 spin_unlock(&sk->sk_lock.slock);
2310 * The sk_lock has mutex_lock() semantics here:
2312 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2315 EXPORT_SYMBOL(lock_sock_nested);
2317 void release_sock(struct sock *sk)
2320 * The sk_lock has mutex_unlock() semantics:
2322 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2324 spin_lock_bh(&sk->sk_lock.slock);
2325 if (sk->sk_backlog.tail)
2328 if (sk->sk_prot->release_cb)
2329 sk->sk_prot->release_cb(sk);
2331 sk->sk_lock.owned = 0;
2332 if (waitqueue_active(&sk->sk_lock.wq))
2333 wake_up(&sk->sk_lock.wq);
2334 spin_unlock_bh(&sk->sk_lock.slock);
2336 EXPORT_SYMBOL(release_sock);
2339 * lock_sock_fast - fast version of lock_sock
2342 * This version should be used for very small section, where process wont block
2343 * return false if fast path is taken
2344 * sk_lock.slock locked, owned = 0, BH disabled
2345 * return true if slow path is taken
2346 * sk_lock.slock unlocked, owned = 1, BH enabled
2348 bool lock_sock_fast(struct sock *sk)
2351 spin_lock_bh(&sk->sk_lock.slock);
2353 if (!sk->sk_lock.owned)
2355 * Note : We must disable BH
2360 sk->sk_lock.owned = 1;
2361 spin_unlock(&sk->sk_lock.slock);
2363 * The sk_lock has mutex_lock() semantics here:
2365 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2369 EXPORT_SYMBOL(lock_sock_fast);
2371 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2374 if (!sock_flag(sk, SOCK_TIMESTAMP))
2375 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2376 tv = ktime_to_timeval(sk->sk_stamp);
2377 if (tv.tv_sec == -1)
2379 if (tv.tv_sec == 0) {
2380 sk->sk_stamp = ktime_get_real();
2381 tv = ktime_to_timeval(sk->sk_stamp);
2383 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2385 EXPORT_SYMBOL(sock_get_timestamp);
2387 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2390 if (!sock_flag(sk, SOCK_TIMESTAMP))
2391 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2392 ts = ktime_to_timespec(sk->sk_stamp);
2393 if (ts.tv_sec == -1)
2395 if (ts.tv_sec == 0) {
2396 sk->sk_stamp = ktime_get_real();
2397 ts = ktime_to_timespec(sk->sk_stamp);
2399 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2401 EXPORT_SYMBOL(sock_get_timestampns);
2403 void sock_enable_timestamp(struct sock *sk, int flag)
2405 if (!sock_flag(sk, flag)) {
2406 unsigned long previous_flags = sk->sk_flags;
2408 sock_set_flag(sk, flag);
2410 * we just set one of the two flags which require net
2411 * time stamping, but time stamping might have been on
2412 * already because of the other one
2414 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2415 net_enable_timestamp();
2420 * Get a socket option on an socket.
2422 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2423 * asynchronous errors should be reported by getsockopt. We assume
2424 * this means if you specify SO_ERROR (otherwise whats the point of it).
2426 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2427 char __user *optval, int __user *optlen)
2429 struct sock *sk = sock->sk;
2431 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2433 EXPORT_SYMBOL(sock_common_getsockopt);
2435 #ifdef CONFIG_COMPAT
2436 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2437 char __user *optval, int __user *optlen)
2439 struct sock *sk = sock->sk;
2441 if (sk->sk_prot->compat_getsockopt != NULL)
2442 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2444 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2446 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2449 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2450 struct msghdr *msg, size_t size, int flags)
2452 struct sock *sk = sock->sk;
2456 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2457 flags & ~MSG_DONTWAIT, &addr_len);
2459 msg->msg_namelen = addr_len;
2462 EXPORT_SYMBOL(sock_common_recvmsg);
2465 * Set socket options on an inet socket.
2467 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2468 char __user *optval, unsigned int optlen)
2470 struct sock *sk = sock->sk;
2472 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2474 EXPORT_SYMBOL(sock_common_setsockopt);
2476 #ifdef CONFIG_COMPAT
2477 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2478 char __user *optval, unsigned int optlen)
2480 struct sock *sk = sock->sk;
2482 if (sk->sk_prot->compat_setsockopt != NULL)
2483 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2485 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2487 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2490 void sk_common_release(struct sock *sk)
2492 if (sk->sk_prot->destroy)
2493 sk->sk_prot->destroy(sk);
2496 * Observation: when sock_common_release is called, processes have
2497 * no access to socket. But net still has.
2498 * Step one, detach it from networking:
2500 * A. Remove from hash tables.
2503 sk->sk_prot->unhash(sk);
2506 * In this point socket cannot receive new packets, but it is possible
2507 * that some packets are in flight because some CPU runs receiver and
2508 * did hash table lookup before we unhashed socket. They will achieve
2509 * receive queue and will be purged by socket destructor.
2511 * Also we still have packets pending on receive queue and probably,
2512 * our own packets waiting in device queues. sock_destroy will drain
2513 * receive queue, but transmitted packets will delay socket destruction
2514 * until the last reference will be released.
2519 xfrm_sk_free_policy(sk);
2521 sk_refcnt_debug_release(sk);
2523 if (sk->sk_frag.page) {
2524 put_page(sk->sk_frag.page);
2525 sk->sk_frag.page = NULL;
2530 EXPORT_SYMBOL(sk_common_release);
2532 #ifdef CONFIG_PROC_FS
2533 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2535 int val[PROTO_INUSE_NR];
2538 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2540 #ifdef CONFIG_NET_NS
2541 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2543 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2545 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2547 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2549 int cpu, idx = prot->inuse_idx;
2552 for_each_possible_cpu(cpu)
2553 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2555 return res >= 0 ? res : 0;
2557 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2559 static int __net_init sock_inuse_init_net(struct net *net)
2561 net->core.inuse = alloc_percpu(struct prot_inuse);
2562 return net->core.inuse ? 0 : -ENOMEM;
2565 static void __net_exit sock_inuse_exit_net(struct net *net)
2567 free_percpu(net->core.inuse);
2570 static struct pernet_operations net_inuse_ops = {
2571 .init = sock_inuse_init_net,
2572 .exit = sock_inuse_exit_net,
2575 static __init int net_inuse_init(void)
2577 if (register_pernet_subsys(&net_inuse_ops))
2578 panic("Cannot initialize net inuse counters");
2583 core_initcall(net_inuse_init);
2585 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2587 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2589 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2591 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2593 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2595 int cpu, idx = prot->inuse_idx;
2598 for_each_possible_cpu(cpu)
2599 res += per_cpu(prot_inuse, cpu).val[idx];
2601 return res >= 0 ? res : 0;
2603 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2606 static void assign_proto_idx(struct proto *prot)
2608 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2610 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2611 pr_err("PROTO_INUSE_NR exhausted\n");
2615 set_bit(prot->inuse_idx, proto_inuse_idx);
2618 static void release_proto_idx(struct proto *prot)
2620 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2621 clear_bit(prot->inuse_idx, proto_inuse_idx);
2624 static inline void assign_proto_idx(struct proto *prot)
2628 static inline void release_proto_idx(struct proto *prot)
2633 int proto_register(struct proto *prot, int alloc_slab)
2636 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2637 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2640 if (prot->slab == NULL) {
2641 pr_crit("%s: Can't create sock SLAB cache!\n",
2646 if (prot->rsk_prot != NULL) {
2647 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2648 if (prot->rsk_prot->slab_name == NULL)
2649 goto out_free_sock_slab;
2651 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2652 prot->rsk_prot->obj_size, 0,
2653 SLAB_HWCACHE_ALIGN, NULL);
2655 if (prot->rsk_prot->slab == NULL) {
2656 pr_crit("%s: Can't create request sock SLAB cache!\n",
2658 goto out_free_request_sock_slab_name;
2662 if (prot->twsk_prot != NULL) {
2663 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2665 if (prot->twsk_prot->twsk_slab_name == NULL)
2666 goto out_free_request_sock_slab;
2668 prot->twsk_prot->twsk_slab =
2669 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2670 prot->twsk_prot->twsk_obj_size,
2672 SLAB_HWCACHE_ALIGN |
2675 if (prot->twsk_prot->twsk_slab == NULL)
2676 goto out_free_timewait_sock_slab_name;
2680 mutex_lock(&proto_list_mutex);
2681 list_add(&prot->node, &proto_list);
2682 assign_proto_idx(prot);
2683 mutex_unlock(&proto_list_mutex);
2686 out_free_timewait_sock_slab_name:
2687 kfree(prot->twsk_prot->twsk_slab_name);
2688 out_free_request_sock_slab:
2689 if (prot->rsk_prot && prot->rsk_prot->slab) {
2690 kmem_cache_destroy(prot->rsk_prot->slab);
2691 prot->rsk_prot->slab = NULL;
2693 out_free_request_sock_slab_name:
2695 kfree(prot->rsk_prot->slab_name);
2697 kmem_cache_destroy(prot->slab);
2702 EXPORT_SYMBOL(proto_register);
2704 void proto_unregister(struct proto *prot)
2706 mutex_lock(&proto_list_mutex);
2707 release_proto_idx(prot);
2708 list_del(&prot->node);
2709 mutex_unlock(&proto_list_mutex);
2711 if (prot->slab != NULL) {
2712 kmem_cache_destroy(prot->slab);
2716 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2717 kmem_cache_destroy(prot->rsk_prot->slab);
2718 kfree(prot->rsk_prot->slab_name);
2719 prot->rsk_prot->slab = NULL;
2722 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2723 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2724 kfree(prot->twsk_prot->twsk_slab_name);
2725 prot->twsk_prot->twsk_slab = NULL;
2728 EXPORT_SYMBOL(proto_unregister);
2730 #ifdef CONFIG_PROC_FS
2731 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2732 __acquires(proto_list_mutex)
2734 mutex_lock(&proto_list_mutex);
2735 return seq_list_start_head(&proto_list, *pos);
2738 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2740 return seq_list_next(v, &proto_list, pos);
2743 static void proto_seq_stop(struct seq_file *seq, void *v)
2744 __releases(proto_list_mutex)
2746 mutex_unlock(&proto_list_mutex);
2749 static char proto_method_implemented(const void *method)
2751 return method == NULL ? 'n' : 'y';
2753 static long sock_prot_memory_allocated(struct proto *proto)
2755 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2758 static char *sock_prot_memory_pressure(struct proto *proto)
2760 return proto->memory_pressure != NULL ?
2761 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2764 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2767 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2768 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2771 sock_prot_inuse_get(seq_file_net(seq), proto),
2772 sock_prot_memory_allocated(proto),
2773 sock_prot_memory_pressure(proto),
2775 proto->slab == NULL ? "no" : "yes",
2776 module_name(proto->owner),
2777 proto_method_implemented(proto->close),
2778 proto_method_implemented(proto->connect),
2779 proto_method_implemented(proto->disconnect),
2780 proto_method_implemented(proto->accept),
2781 proto_method_implemented(proto->ioctl),
2782 proto_method_implemented(proto->init),
2783 proto_method_implemented(proto->destroy),
2784 proto_method_implemented(proto->shutdown),
2785 proto_method_implemented(proto->setsockopt),
2786 proto_method_implemented(proto->getsockopt),
2787 proto_method_implemented(proto->sendmsg),
2788 proto_method_implemented(proto->recvmsg),
2789 proto_method_implemented(proto->sendpage),
2790 proto_method_implemented(proto->bind),
2791 proto_method_implemented(proto->backlog_rcv),
2792 proto_method_implemented(proto->hash),
2793 proto_method_implemented(proto->unhash),
2794 proto_method_implemented(proto->get_port),
2795 proto_method_implemented(proto->enter_memory_pressure));
2798 static int proto_seq_show(struct seq_file *seq, void *v)
2800 if (v == &proto_list)
2801 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2810 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2812 proto_seq_printf(seq, list_entry(v, struct proto, node));
2816 static const struct seq_operations proto_seq_ops = {
2817 .start = proto_seq_start,
2818 .next = proto_seq_next,
2819 .stop = proto_seq_stop,
2820 .show = proto_seq_show,
2823 static int proto_seq_open(struct inode *inode, struct file *file)
2825 return seq_open_net(inode, file, &proto_seq_ops,
2826 sizeof(struct seq_net_private));
2829 static const struct file_operations proto_seq_fops = {
2830 .owner = THIS_MODULE,
2831 .open = proto_seq_open,
2833 .llseek = seq_lseek,
2834 .release = seq_release_net,
2837 static __net_init int proto_init_net(struct net *net)
2839 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2845 static __net_exit void proto_exit_net(struct net *net)
2847 proc_net_remove(net, "protocols");
2851 static __net_initdata struct pernet_operations proto_net_ops = {
2852 .init = proto_init_net,
2853 .exit = proto_exit_net,
2856 static int __init proto_init(void)
2858 return register_pernet_subsys(&proto_net_ops);
2861 subsys_initcall(proto_init);
2863 #endif /* PROC_FS */