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 #if defined(CONFIG_MEMCG_KMEM)
190 struct static_key memcg_socket_limit_enabled;
191 EXPORT_SYMBOL(memcg_socket_limit_enabled);
195 * Make lock validator output more readable. (we pre-construct these
196 * strings build-time, so that runtime initialization of socket
199 static const char *const af_family_key_strings[AF_MAX+1] = {
200 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
201 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
202 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
203 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
204 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
205 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
206 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
207 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
208 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
209 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
210 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
211 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
212 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
213 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
215 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
216 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
217 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
218 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
219 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
220 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
221 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
222 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
223 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
224 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
225 "slock-27" , "slock-28" , "slock-AF_CAN" ,
226 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
227 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
228 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
229 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
231 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
232 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
233 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
234 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
235 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
236 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
237 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
238 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
239 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
240 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
241 "clock-27" , "clock-28" , "clock-AF_CAN" ,
242 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
243 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
244 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
245 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
249 * sk_callback_lock locking rules are per-address-family,
250 * so split the lock classes by using a per-AF key:
252 static struct lock_class_key af_callback_keys[AF_MAX];
254 /* Take into consideration the size of the struct sk_buff overhead in the
255 * determination of these values, since that is non-constant across
256 * platforms. This makes socket queueing behavior and performance
257 * not depend upon such differences.
259 #define _SK_MEM_PACKETS 256
260 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
261 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
262 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
264 /* Run time adjustable parameters. */
265 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
266 EXPORT_SYMBOL(sysctl_wmem_max);
267 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
268 EXPORT_SYMBOL(sysctl_rmem_max);
269 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
270 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
272 /* Maximal space eaten by iovec or ancillary data plus some space */
273 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
274 EXPORT_SYMBOL(sysctl_optmem_max);
276 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
277 EXPORT_SYMBOL_GPL(memalloc_socks);
280 * sk_set_memalloc - sets %SOCK_MEMALLOC
281 * @sk: socket to set it on
283 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
284 * It's the responsibility of the admin to adjust min_free_kbytes
285 * to meet the requirements
287 void sk_set_memalloc(struct sock *sk)
289 sock_set_flag(sk, SOCK_MEMALLOC);
290 sk->sk_allocation |= __GFP_MEMALLOC;
291 static_key_slow_inc(&memalloc_socks);
293 EXPORT_SYMBOL_GPL(sk_set_memalloc);
295 void sk_clear_memalloc(struct sock *sk)
297 sock_reset_flag(sk, SOCK_MEMALLOC);
298 sk->sk_allocation &= ~__GFP_MEMALLOC;
299 static_key_slow_dec(&memalloc_socks);
302 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
303 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
304 * it has rmem allocations there is a risk that the user of the
305 * socket cannot make forward progress due to exceeding the rmem
306 * limits. By rights, sk_clear_memalloc() should only be called
307 * on sockets being torn down but warn and reset the accounting if
308 * that assumption breaks.
310 if (WARN_ON(sk->sk_forward_alloc))
313 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
315 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
318 unsigned long pflags = current->flags;
320 /* these should have been dropped before queueing */
321 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
323 current->flags |= PF_MEMALLOC;
324 ret = sk->sk_backlog_rcv(sk, skb);
325 tsk_restore_flags(current, pflags, PF_MEMALLOC);
329 EXPORT_SYMBOL(__sk_backlog_rcv);
331 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
335 if (optlen < sizeof(tv))
337 if (copy_from_user(&tv, optval, sizeof(tv)))
339 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
343 static int warned __read_mostly;
346 if (warned < 10 && net_ratelimit()) {
348 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
349 __func__, current->comm, task_pid_nr(current));
353 *timeo_p = MAX_SCHEDULE_TIMEOUT;
354 if (tv.tv_sec == 0 && tv.tv_usec == 0)
356 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
357 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
361 static void sock_warn_obsolete_bsdism(const char *name)
364 static char warncomm[TASK_COMM_LEN];
365 if (strcmp(warncomm, current->comm) && warned < 5) {
366 strcpy(warncomm, current->comm);
367 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
373 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
375 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
377 if (sk->sk_flags & flags) {
378 sk->sk_flags &= ~flags;
379 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
380 net_disable_timestamp();
385 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
390 struct sk_buff_head *list = &sk->sk_receive_queue;
392 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
393 atomic_inc(&sk->sk_drops);
394 trace_sock_rcvqueue_full(sk, skb);
398 err = sk_filter(sk, skb);
402 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
403 atomic_inc(&sk->sk_drops);
408 skb_set_owner_r(skb, sk);
410 /* Cache the SKB length before we tack it onto the receive
411 * queue. Once it is added it no longer belongs to us and
412 * may be freed by other threads of control pulling packets
417 /* we escape from rcu protected region, make sure we dont leak
422 spin_lock_irqsave(&list->lock, flags);
423 skb->dropcount = atomic_read(&sk->sk_drops);
424 __skb_queue_tail(list, skb);
425 spin_unlock_irqrestore(&list->lock, flags);
427 if (!sock_flag(sk, SOCK_DEAD))
428 sk->sk_data_ready(sk, skb_len);
431 EXPORT_SYMBOL(sock_queue_rcv_skb);
433 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
435 int rc = NET_RX_SUCCESS;
437 if (sk_filter(sk, skb))
438 goto discard_and_relse;
442 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
443 atomic_inc(&sk->sk_drops);
444 goto discard_and_relse;
447 bh_lock_sock_nested(sk);
450 if (!sock_owned_by_user(sk)) {
452 * trylock + unlock semantics:
454 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
456 rc = sk_backlog_rcv(sk, skb);
458 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
459 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
461 atomic_inc(&sk->sk_drops);
462 goto discard_and_relse;
473 EXPORT_SYMBOL(sk_receive_skb);
475 void sk_reset_txq(struct sock *sk)
477 sk_tx_queue_clear(sk);
479 EXPORT_SYMBOL(sk_reset_txq);
481 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
483 struct dst_entry *dst = __sk_dst_get(sk);
485 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
486 sk_tx_queue_clear(sk);
487 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
494 EXPORT_SYMBOL(__sk_dst_check);
496 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
498 struct dst_entry *dst = sk_dst_get(sk);
500 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
508 EXPORT_SYMBOL(sk_dst_check);
510 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
513 int ret = -ENOPROTOOPT;
514 #ifdef CONFIG_NETDEVICES
515 struct net *net = sock_net(sk);
516 char devname[IFNAMSIZ];
521 if (!ns_capable(net->user_ns, CAP_NET_RAW))
528 /* Bind this socket to a particular device like "eth0",
529 * as specified in the passed interface name. If the
530 * name is "" or the option length is zero the socket
533 if (optlen > IFNAMSIZ - 1)
534 optlen = IFNAMSIZ - 1;
535 memset(devname, 0, sizeof(devname));
538 if (copy_from_user(devname, optval, optlen))
542 if (devname[0] != '\0') {
543 struct net_device *dev;
546 dev = dev_get_by_name_rcu(net, devname);
548 index = dev->ifindex;
556 sk->sk_bound_dev_if = index;
568 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
569 int __user *optlen, int len)
571 int ret = -ENOPROTOOPT;
572 #ifdef CONFIG_NETDEVICES
573 struct net *net = sock_net(sk);
574 char devname[IFNAMSIZ];
576 if (sk->sk_bound_dev_if == 0) {
585 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
589 len = strlen(devname) + 1;
592 if (copy_to_user(optval, devname, len))
597 if (put_user(len, optlen))
608 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
611 sock_set_flag(sk, bit);
613 sock_reset_flag(sk, bit);
617 * This is meant for all protocols to use and covers goings on
618 * at the socket level. Everything here is generic.
621 int sock_setsockopt(struct socket *sock, int level, int optname,
622 char __user *optval, unsigned int optlen)
624 struct sock *sk = sock->sk;
631 * Options without arguments
634 if (optname == SO_BINDTODEVICE)
635 return sock_setbindtodevice(sk, optval, optlen);
637 if (optlen < sizeof(int))
640 if (get_user(val, (int __user *)optval))
643 valbool = val ? 1 : 0;
649 if (val && !capable(CAP_NET_ADMIN))
652 sock_valbool_flag(sk, SOCK_DBG, valbool);
655 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
658 sk->sk_reuseport = valbool;
667 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
670 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
673 /* Don't error on this BSD doesn't and if you think
674 * about it this is right. Otherwise apps have to
675 * play 'guess the biggest size' games. RCVBUF/SNDBUF
676 * are treated in BSD as hints
678 val = min_t(u32, val, sysctl_wmem_max);
680 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
681 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
682 /* Wake up sending tasks if we upped the value. */
683 sk->sk_write_space(sk);
687 if (!capable(CAP_NET_ADMIN)) {
694 /* Don't error on this BSD doesn't and if you think
695 * about it this is right. Otherwise apps have to
696 * play 'guess the biggest size' games. RCVBUF/SNDBUF
697 * are treated in BSD as hints
699 val = min_t(u32, val, sysctl_rmem_max);
701 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
703 * We double it on the way in to account for
704 * "struct sk_buff" etc. overhead. Applications
705 * assume that the SO_RCVBUF setting they make will
706 * allow that much actual data to be received on that
709 * Applications are unaware that "struct sk_buff" and
710 * other overheads allocate from the receive buffer
711 * during socket buffer allocation.
713 * And after considering the possible alternatives,
714 * returning the value we actually used in getsockopt
715 * is the most desirable behavior.
717 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
721 if (!capable(CAP_NET_ADMIN)) {
729 if (sk->sk_protocol == IPPROTO_TCP &&
730 sk->sk_type == SOCK_STREAM)
731 tcp_set_keepalive(sk, valbool);
733 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
737 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
741 sk->sk_no_check = valbool;
745 if ((val >= 0 && val <= 6) ||
746 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
747 sk->sk_priority = val;
753 if (optlen < sizeof(ling)) {
754 ret = -EINVAL; /* 1003.1g */
757 if (copy_from_user(&ling, optval, sizeof(ling))) {
762 sock_reset_flag(sk, SOCK_LINGER);
764 #if (BITS_PER_LONG == 32)
765 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
766 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
769 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
770 sock_set_flag(sk, SOCK_LINGER);
775 sock_warn_obsolete_bsdism("setsockopt");
780 set_bit(SOCK_PASSCRED, &sock->flags);
782 clear_bit(SOCK_PASSCRED, &sock->flags);
788 if (optname == SO_TIMESTAMP)
789 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
791 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
792 sock_set_flag(sk, SOCK_RCVTSTAMP);
793 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
795 sock_reset_flag(sk, SOCK_RCVTSTAMP);
796 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
800 case SO_TIMESTAMPING:
801 if (val & ~SOF_TIMESTAMPING_MASK) {
805 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
806 val & SOF_TIMESTAMPING_TX_HARDWARE);
807 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
808 val & SOF_TIMESTAMPING_TX_SOFTWARE);
809 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
810 val & SOF_TIMESTAMPING_RX_HARDWARE);
811 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
812 sock_enable_timestamp(sk,
813 SOCK_TIMESTAMPING_RX_SOFTWARE);
815 sock_disable_timestamp(sk,
816 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
817 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
818 val & SOF_TIMESTAMPING_SOFTWARE);
819 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
820 val & SOF_TIMESTAMPING_SYS_HARDWARE);
821 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
822 val & SOF_TIMESTAMPING_RAW_HARDWARE);
828 sk->sk_rcvlowat = val ? : 1;
832 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
836 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
839 case SO_ATTACH_FILTER:
841 if (optlen == sizeof(struct sock_fprog)) {
842 struct sock_fprog fprog;
845 if (copy_from_user(&fprog, optval, sizeof(fprog)))
848 ret = sk_attach_filter(&fprog, sk);
852 case SO_DETACH_FILTER:
853 ret = sk_detach_filter(sk);
857 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
860 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
865 set_bit(SOCK_PASSSEC, &sock->flags);
867 clear_bit(SOCK_PASSSEC, &sock->flags);
870 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
876 /* We implement the SO_SNDLOWAT etc to
877 not be settable (1003.1g 5.3) */
879 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
883 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
887 if (sock->ops->set_peek_off)
888 ret = sock->ops->set_peek_off(sk, val);
894 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
897 case SO_SELECT_ERR_QUEUE:
898 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
908 EXPORT_SYMBOL(sock_setsockopt);
911 void cred_to_ucred(struct pid *pid, const struct cred *cred,
914 ucred->pid = pid_vnr(pid);
915 ucred->uid = ucred->gid = -1;
917 struct user_namespace *current_ns = current_user_ns();
919 ucred->uid = from_kuid_munged(current_ns, cred->euid);
920 ucred->gid = from_kgid_munged(current_ns, cred->egid);
923 EXPORT_SYMBOL_GPL(cred_to_ucred);
925 int sock_getsockopt(struct socket *sock, int level, int optname,
926 char __user *optval, int __user *optlen)
928 struct sock *sk = sock->sk;
936 int lv = sizeof(int);
939 if (get_user(len, optlen))
944 memset(&v, 0, sizeof(v));
948 v.val = sock_flag(sk, SOCK_DBG);
952 v.val = sock_flag(sk, SOCK_LOCALROUTE);
956 v.val = sock_flag(sk, SOCK_BROADCAST);
960 v.val = sk->sk_sndbuf;
964 v.val = sk->sk_rcvbuf;
968 v.val = sk->sk_reuse;
972 v.val = sk->sk_reuseport;
976 v.val = sock_flag(sk, SOCK_KEEPOPEN);
984 v.val = sk->sk_protocol;
988 v.val = sk->sk_family;
992 v.val = -sock_error(sk);
994 v.val = xchg(&sk->sk_err_soft, 0);
998 v.val = sock_flag(sk, SOCK_URGINLINE);
1002 v.val = sk->sk_no_check;
1006 v.val = sk->sk_priority;
1010 lv = sizeof(v.ling);
1011 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1012 v.ling.l_linger = sk->sk_lingertime / HZ;
1016 sock_warn_obsolete_bsdism("getsockopt");
1020 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1021 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1024 case SO_TIMESTAMPNS:
1025 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1028 case SO_TIMESTAMPING:
1030 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
1031 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
1032 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
1033 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
1034 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
1035 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
1036 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1037 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
1038 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
1039 v.val |= SOF_TIMESTAMPING_SOFTWARE;
1040 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
1041 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
1042 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
1043 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
1047 lv = sizeof(struct timeval);
1048 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1052 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1053 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1058 lv = sizeof(struct timeval);
1059 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1063 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1064 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1069 v.val = sk->sk_rcvlowat;
1077 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1082 struct ucred peercred;
1083 if (len > sizeof(peercred))
1084 len = sizeof(peercred);
1085 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1086 if (copy_to_user(optval, &peercred, len))
1095 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1099 if (copy_to_user(optval, address, len))
1104 /* Dubious BSD thing... Probably nobody even uses it, but
1105 * the UNIX standard wants it for whatever reason... -DaveM
1108 v.val = sk->sk_state == TCP_LISTEN;
1112 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1116 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1119 v.val = sk->sk_mark;
1123 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1126 case SO_WIFI_STATUS:
1127 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1131 if (!sock->ops->set_peek_off)
1134 v.val = sk->sk_peek_off;
1137 v.val = sock_flag(sk, SOCK_NOFCS);
1140 case SO_BINDTODEVICE:
1141 return sock_getbindtodevice(sk, optval, optlen, len);
1144 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1150 case SO_LOCK_FILTER:
1151 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1154 case SO_SELECT_ERR_QUEUE:
1155 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1159 return -ENOPROTOOPT;
1164 if (copy_to_user(optval, &v, len))
1167 if (put_user(len, optlen))
1173 * Initialize an sk_lock.
1175 * (We also register the sk_lock with the lock validator.)
1177 static inline void sock_lock_init(struct sock *sk)
1179 sock_lock_init_class_and_name(sk,
1180 af_family_slock_key_strings[sk->sk_family],
1181 af_family_slock_keys + sk->sk_family,
1182 af_family_key_strings[sk->sk_family],
1183 af_family_keys + sk->sk_family);
1187 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1188 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1189 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1191 static void sock_copy(struct sock *nsk, const struct sock *osk)
1193 #ifdef CONFIG_SECURITY_NETWORK
1194 void *sptr = nsk->sk_security;
1196 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1198 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1199 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1201 #ifdef CONFIG_SECURITY_NETWORK
1202 nsk->sk_security = sptr;
1203 security_sk_clone(osk, nsk);
1207 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1209 unsigned long nulls1, nulls2;
1211 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1212 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1213 if (nulls1 > nulls2)
1214 swap(nulls1, nulls2);
1217 memset((char *)sk, 0, nulls1);
1218 memset((char *)sk + nulls1 + sizeof(void *), 0,
1219 nulls2 - nulls1 - sizeof(void *));
1220 memset((char *)sk + nulls2 + sizeof(void *), 0,
1221 size - nulls2 - sizeof(void *));
1223 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1225 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1229 struct kmem_cache *slab;
1233 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1236 if (priority & __GFP_ZERO) {
1238 prot->clear_sk(sk, prot->obj_size);
1240 sk_prot_clear_nulls(sk, prot->obj_size);
1243 sk = kmalloc(prot->obj_size, priority);
1246 kmemcheck_annotate_bitfield(sk, flags);
1248 if (security_sk_alloc(sk, family, priority))
1251 if (!try_module_get(prot->owner))
1253 sk_tx_queue_clear(sk);
1259 security_sk_free(sk);
1262 kmem_cache_free(slab, sk);
1268 static void sk_prot_free(struct proto *prot, struct sock *sk)
1270 struct kmem_cache *slab;
1271 struct module *owner;
1273 owner = prot->owner;
1276 security_sk_free(sk);
1278 kmem_cache_free(slab, sk);
1284 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1285 void sock_update_classid(struct sock *sk)
1289 classid = task_cls_classid(current);
1290 if (classid != sk->sk_classid)
1291 sk->sk_classid = classid;
1293 EXPORT_SYMBOL(sock_update_classid);
1296 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1297 void sock_update_netprioidx(struct sock *sk)
1302 sk->sk_cgrp_prioidx = task_netprioidx(current);
1304 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1308 * sk_alloc - All socket objects are allocated here
1309 * @net: the applicable net namespace
1310 * @family: protocol family
1311 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1312 * @prot: struct proto associated with this new sock instance
1314 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1319 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1321 sk->sk_family = family;
1323 * See comment in struct sock definition to understand
1324 * why we need sk_prot_creator -acme
1326 sk->sk_prot = sk->sk_prot_creator = prot;
1328 sock_net_set(sk, get_net(net));
1329 atomic_set(&sk->sk_wmem_alloc, 1);
1331 sock_update_classid(sk);
1332 sock_update_netprioidx(sk);
1337 EXPORT_SYMBOL(sk_alloc);
1339 static void __sk_free(struct sock *sk)
1341 struct sk_filter *filter;
1343 if (sk->sk_destruct)
1344 sk->sk_destruct(sk);
1346 filter = rcu_dereference_check(sk->sk_filter,
1347 atomic_read(&sk->sk_wmem_alloc) == 0);
1349 sk_filter_uncharge(sk, filter);
1350 RCU_INIT_POINTER(sk->sk_filter, NULL);
1353 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1355 if (atomic_read(&sk->sk_omem_alloc))
1356 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1357 __func__, atomic_read(&sk->sk_omem_alloc));
1359 if (sk->sk_peer_cred)
1360 put_cred(sk->sk_peer_cred);
1361 put_pid(sk->sk_peer_pid);
1362 put_net(sock_net(sk));
1363 sk_prot_free(sk->sk_prot_creator, sk);
1366 void sk_free(struct sock *sk)
1369 * We subtract one from sk_wmem_alloc and can know if
1370 * some packets are still in some tx queue.
1371 * If not null, sock_wfree() will call __sk_free(sk) later
1373 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1376 EXPORT_SYMBOL(sk_free);
1379 * Last sock_put should drop reference to sk->sk_net. It has already
1380 * been dropped in sk_change_net. Taking reference to stopping namespace
1382 * Take reference to a socket to remove it from hash _alive_ and after that
1383 * destroy it in the context of init_net.
1385 void sk_release_kernel(struct sock *sk)
1387 if (sk == NULL || sk->sk_socket == NULL)
1391 sock_release(sk->sk_socket);
1392 release_net(sock_net(sk));
1393 sock_net_set(sk, get_net(&init_net));
1396 EXPORT_SYMBOL(sk_release_kernel);
1398 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1400 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1401 sock_update_memcg(newsk);
1405 * sk_clone_lock - clone a socket, and lock its clone
1406 * @sk: the socket to clone
1407 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1409 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1411 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1415 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1416 if (newsk != NULL) {
1417 struct sk_filter *filter;
1419 sock_copy(newsk, sk);
1422 get_net(sock_net(newsk));
1423 sk_node_init(&newsk->sk_node);
1424 sock_lock_init(newsk);
1425 bh_lock_sock(newsk);
1426 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1427 newsk->sk_backlog.len = 0;
1429 atomic_set(&newsk->sk_rmem_alloc, 0);
1431 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1433 atomic_set(&newsk->sk_wmem_alloc, 1);
1434 atomic_set(&newsk->sk_omem_alloc, 0);
1435 skb_queue_head_init(&newsk->sk_receive_queue);
1436 skb_queue_head_init(&newsk->sk_write_queue);
1437 #ifdef CONFIG_NET_DMA
1438 skb_queue_head_init(&newsk->sk_async_wait_queue);
1441 spin_lock_init(&newsk->sk_dst_lock);
1442 rwlock_init(&newsk->sk_callback_lock);
1443 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1444 af_callback_keys + newsk->sk_family,
1445 af_family_clock_key_strings[newsk->sk_family]);
1447 newsk->sk_dst_cache = NULL;
1448 newsk->sk_wmem_queued = 0;
1449 newsk->sk_forward_alloc = 0;
1450 newsk->sk_send_head = NULL;
1451 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1453 sock_reset_flag(newsk, SOCK_DONE);
1454 skb_queue_head_init(&newsk->sk_error_queue);
1456 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1458 sk_filter_charge(newsk, filter);
1460 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1461 /* It is still raw copy of parent, so invalidate
1462 * destructor and make plain sk_free() */
1463 newsk->sk_destruct = NULL;
1464 bh_unlock_sock(newsk);
1471 newsk->sk_priority = 0;
1473 * Before updating sk_refcnt, we must commit prior changes to memory
1474 * (Documentation/RCU/rculist_nulls.txt for details)
1477 atomic_set(&newsk->sk_refcnt, 2);
1480 * Increment the counter in the same struct proto as the master
1481 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1482 * is the same as sk->sk_prot->socks, as this field was copied
1485 * This _changes_ the previous behaviour, where
1486 * tcp_create_openreq_child always was incrementing the
1487 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1488 * to be taken into account in all callers. -acme
1490 sk_refcnt_debug_inc(newsk);
1491 sk_set_socket(newsk, NULL);
1492 newsk->sk_wq = NULL;
1494 sk_update_clone(sk, newsk);
1496 if (newsk->sk_prot->sockets_allocated)
1497 sk_sockets_allocated_inc(newsk);
1499 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1500 net_enable_timestamp();
1505 EXPORT_SYMBOL_GPL(sk_clone_lock);
1507 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1509 __sk_dst_set(sk, dst);
1510 sk->sk_route_caps = dst->dev->features;
1511 if (sk->sk_route_caps & NETIF_F_GSO)
1512 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1513 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1514 if (sk_can_gso(sk)) {
1515 if (dst->header_len) {
1516 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1518 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1519 sk->sk_gso_max_size = dst->dev->gso_max_size;
1520 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1524 EXPORT_SYMBOL_GPL(sk_setup_caps);
1527 * Simple resource managers for sockets.
1532 * Write buffer destructor automatically called from kfree_skb.
1534 void sock_wfree(struct sk_buff *skb)
1536 struct sock *sk = skb->sk;
1537 unsigned int len = skb->truesize;
1539 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1541 * Keep a reference on sk_wmem_alloc, this will be released
1542 * after sk_write_space() call
1544 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1545 sk->sk_write_space(sk);
1549 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1550 * could not do because of in-flight packets
1552 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1555 EXPORT_SYMBOL(sock_wfree);
1558 * Read buffer destructor automatically called from kfree_skb.
1560 void sock_rfree(struct sk_buff *skb)
1562 struct sock *sk = skb->sk;
1563 unsigned int len = skb->truesize;
1565 atomic_sub(len, &sk->sk_rmem_alloc);
1566 sk_mem_uncharge(sk, len);
1568 EXPORT_SYMBOL(sock_rfree);
1570 void sock_edemux(struct sk_buff *skb)
1572 struct sock *sk = skb->sk;
1575 if (sk->sk_state == TCP_TIME_WAIT)
1576 inet_twsk_put(inet_twsk(sk));
1581 EXPORT_SYMBOL(sock_edemux);
1583 kuid_t sock_i_uid(struct sock *sk)
1587 read_lock_bh(&sk->sk_callback_lock);
1588 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1589 read_unlock_bh(&sk->sk_callback_lock);
1592 EXPORT_SYMBOL(sock_i_uid);
1594 unsigned long sock_i_ino(struct sock *sk)
1598 read_lock_bh(&sk->sk_callback_lock);
1599 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1600 read_unlock_bh(&sk->sk_callback_lock);
1603 EXPORT_SYMBOL(sock_i_ino);
1606 * Allocate a skb from the socket's send buffer.
1608 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1611 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1612 struct sk_buff *skb = alloc_skb(size, priority);
1614 skb_set_owner_w(skb, sk);
1620 EXPORT_SYMBOL(sock_wmalloc);
1623 * Allocate a skb from the socket's receive buffer.
1625 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1628 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1629 struct sk_buff *skb = alloc_skb(size, priority);
1631 skb_set_owner_r(skb, sk);
1639 * Allocate a memory block from the socket's option memory buffer.
1641 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1643 if ((unsigned int)size <= sysctl_optmem_max &&
1644 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1646 /* First do the add, to avoid the race if kmalloc
1649 atomic_add(size, &sk->sk_omem_alloc);
1650 mem = kmalloc(size, priority);
1653 atomic_sub(size, &sk->sk_omem_alloc);
1657 EXPORT_SYMBOL(sock_kmalloc);
1660 * Free an option memory block.
1662 void sock_kfree_s(struct sock *sk, void *mem, int size)
1665 atomic_sub(size, &sk->sk_omem_alloc);
1667 EXPORT_SYMBOL(sock_kfree_s);
1669 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1670 I think, these locks should be removed for datagram sockets.
1672 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1676 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1680 if (signal_pending(current))
1682 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1683 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1684 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1686 if (sk->sk_shutdown & SEND_SHUTDOWN)
1690 timeo = schedule_timeout(timeo);
1692 finish_wait(sk_sleep(sk), &wait);
1698 * Generic send/receive buffer handlers
1701 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1702 unsigned long data_len, int noblock,
1705 struct sk_buff *skb;
1709 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1712 if (npages > MAX_SKB_FRAGS)
1715 gfp_mask = sk->sk_allocation;
1716 if (gfp_mask & __GFP_WAIT)
1717 gfp_mask |= __GFP_REPEAT;
1719 timeo = sock_sndtimeo(sk, noblock);
1721 err = sock_error(sk);
1726 if (sk->sk_shutdown & SEND_SHUTDOWN)
1729 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1730 skb = alloc_skb(header_len, gfp_mask);
1734 /* No pages, we're done... */
1738 skb->truesize += data_len;
1739 skb_shinfo(skb)->nr_frags = npages;
1740 for (i = 0; i < npages; i++) {
1743 page = alloc_pages(sk->sk_allocation, 0);
1746 skb_shinfo(skb)->nr_frags = i;
1751 __skb_fill_page_desc(skb, i,
1753 (data_len >= PAGE_SIZE ?
1756 data_len -= PAGE_SIZE;
1759 /* Full success... */
1765 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1766 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1770 if (signal_pending(current))
1772 timeo = sock_wait_for_wmem(sk, timeo);
1775 skb_set_owner_w(skb, sk);
1779 err = sock_intr_errno(timeo);
1784 EXPORT_SYMBOL(sock_alloc_send_pskb);
1786 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1787 int noblock, int *errcode)
1789 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1791 EXPORT_SYMBOL(sock_alloc_send_skb);
1793 /* On 32bit arches, an skb frag is limited to 2^15 */
1794 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1796 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1801 if (atomic_read(&pfrag->page->_count) == 1) {
1805 if (pfrag->offset < pfrag->size)
1807 put_page(pfrag->page);
1810 /* We restrict high order allocations to users that can afford to wait */
1811 order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1814 gfp_t gfp = sk->sk_allocation;
1817 gfp |= __GFP_COMP | __GFP_NOWARN;
1818 pfrag->page = alloc_pages(gfp, order);
1819 if (likely(pfrag->page)) {
1821 pfrag->size = PAGE_SIZE << order;
1824 } while (--order >= 0);
1826 sk_enter_memory_pressure(sk);
1827 sk_stream_moderate_sndbuf(sk);
1830 EXPORT_SYMBOL(sk_page_frag_refill);
1832 static void __lock_sock(struct sock *sk)
1833 __releases(&sk->sk_lock.slock)
1834 __acquires(&sk->sk_lock.slock)
1839 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1840 TASK_UNINTERRUPTIBLE);
1841 spin_unlock_bh(&sk->sk_lock.slock);
1843 spin_lock_bh(&sk->sk_lock.slock);
1844 if (!sock_owned_by_user(sk))
1847 finish_wait(&sk->sk_lock.wq, &wait);
1850 static void __release_sock(struct sock *sk)
1851 __releases(&sk->sk_lock.slock)
1852 __acquires(&sk->sk_lock.slock)
1854 struct sk_buff *skb = sk->sk_backlog.head;
1857 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1861 struct sk_buff *next = skb->next;
1864 WARN_ON_ONCE(skb_dst_is_noref(skb));
1866 sk_backlog_rcv(sk, skb);
1869 * We are in process context here with softirqs
1870 * disabled, use cond_resched_softirq() to preempt.
1871 * This is safe to do because we've taken the backlog
1874 cond_resched_softirq();
1877 } while (skb != NULL);
1880 } while ((skb = sk->sk_backlog.head) != NULL);
1883 * Doing the zeroing here guarantee we can not loop forever
1884 * while a wild producer attempts to flood us.
1886 sk->sk_backlog.len = 0;
1890 * sk_wait_data - wait for data to arrive at sk_receive_queue
1891 * @sk: sock to wait on
1892 * @timeo: for how long
1894 * Now socket state including sk->sk_err is changed only under lock,
1895 * hence we may omit checks after joining wait queue.
1896 * We check receive queue before schedule() only as optimization;
1897 * it is very likely that release_sock() added new data.
1899 int sk_wait_data(struct sock *sk, long *timeo)
1904 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1905 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1906 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1907 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1908 finish_wait(sk_sleep(sk), &wait);
1911 EXPORT_SYMBOL(sk_wait_data);
1914 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1916 * @size: memory size to allocate
1917 * @kind: allocation type
1919 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1920 * rmem allocation. This function assumes that protocols which have
1921 * memory_pressure use sk_wmem_queued as write buffer accounting.
1923 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1925 struct proto *prot = sk->sk_prot;
1926 int amt = sk_mem_pages(size);
1928 int parent_status = UNDER_LIMIT;
1930 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1932 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1935 if (parent_status == UNDER_LIMIT &&
1936 allocated <= sk_prot_mem_limits(sk, 0)) {
1937 sk_leave_memory_pressure(sk);
1941 /* Under pressure. (we or our parents) */
1942 if ((parent_status > SOFT_LIMIT) ||
1943 allocated > sk_prot_mem_limits(sk, 1))
1944 sk_enter_memory_pressure(sk);
1946 /* Over hard limit (we or our parents) */
1947 if ((parent_status == OVER_LIMIT) ||
1948 (allocated > sk_prot_mem_limits(sk, 2)))
1949 goto suppress_allocation;
1951 /* guarantee minimum buffer size under pressure */
1952 if (kind == SK_MEM_RECV) {
1953 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1956 } else { /* SK_MEM_SEND */
1957 if (sk->sk_type == SOCK_STREAM) {
1958 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1960 } else if (atomic_read(&sk->sk_wmem_alloc) <
1961 prot->sysctl_wmem[0])
1965 if (sk_has_memory_pressure(sk)) {
1968 if (!sk_under_memory_pressure(sk))
1970 alloc = sk_sockets_allocated_read_positive(sk);
1971 if (sk_prot_mem_limits(sk, 2) > alloc *
1972 sk_mem_pages(sk->sk_wmem_queued +
1973 atomic_read(&sk->sk_rmem_alloc) +
1974 sk->sk_forward_alloc))
1978 suppress_allocation:
1980 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1981 sk_stream_moderate_sndbuf(sk);
1983 /* Fail only if socket is _under_ its sndbuf.
1984 * In this case we cannot block, so that we have to fail.
1986 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1990 trace_sock_exceed_buf_limit(sk, prot, allocated);
1992 /* Alas. Undo changes. */
1993 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1995 sk_memory_allocated_sub(sk, amt);
1999 EXPORT_SYMBOL(__sk_mem_schedule);
2002 * __sk_reclaim - reclaim memory_allocated
2005 void __sk_mem_reclaim(struct sock *sk)
2007 sk_memory_allocated_sub(sk,
2008 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2009 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2011 if (sk_under_memory_pressure(sk) &&
2012 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2013 sk_leave_memory_pressure(sk);
2015 EXPORT_SYMBOL(__sk_mem_reclaim);
2019 * Set of default routines for initialising struct proto_ops when
2020 * the protocol does not support a particular function. In certain
2021 * cases where it makes no sense for a protocol to have a "do nothing"
2022 * function, some default processing is provided.
2025 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2029 EXPORT_SYMBOL(sock_no_bind);
2031 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2036 EXPORT_SYMBOL(sock_no_connect);
2038 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2042 EXPORT_SYMBOL(sock_no_socketpair);
2044 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2048 EXPORT_SYMBOL(sock_no_accept);
2050 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2055 EXPORT_SYMBOL(sock_no_getname);
2057 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2061 EXPORT_SYMBOL(sock_no_poll);
2063 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2067 EXPORT_SYMBOL(sock_no_ioctl);
2069 int sock_no_listen(struct socket *sock, int backlog)
2073 EXPORT_SYMBOL(sock_no_listen);
2075 int sock_no_shutdown(struct socket *sock, int how)
2079 EXPORT_SYMBOL(sock_no_shutdown);
2081 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2082 char __user *optval, unsigned int optlen)
2086 EXPORT_SYMBOL(sock_no_setsockopt);
2088 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2089 char __user *optval, int __user *optlen)
2093 EXPORT_SYMBOL(sock_no_getsockopt);
2095 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2100 EXPORT_SYMBOL(sock_no_sendmsg);
2102 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2103 size_t len, int flags)
2107 EXPORT_SYMBOL(sock_no_recvmsg);
2109 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2111 /* Mirror missing mmap method error code */
2114 EXPORT_SYMBOL(sock_no_mmap);
2116 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2119 struct msghdr msg = {.msg_flags = flags};
2121 char *kaddr = kmap(page);
2122 iov.iov_base = kaddr + offset;
2124 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2128 EXPORT_SYMBOL(sock_no_sendpage);
2131 * Default Socket Callbacks
2134 static void sock_def_wakeup(struct sock *sk)
2136 struct socket_wq *wq;
2139 wq = rcu_dereference(sk->sk_wq);
2140 if (wq_has_sleeper(wq))
2141 wake_up_interruptible_all(&wq->wait);
2145 static void sock_def_error_report(struct sock *sk)
2147 struct socket_wq *wq;
2150 wq = rcu_dereference(sk->sk_wq);
2151 if (wq_has_sleeper(wq))
2152 wake_up_interruptible_poll(&wq->wait, POLLERR);
2153 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2157 static void sock_def_readable(struct sock *sk, int len)
2159 struct socket_wq *wq;
2162 wq = rcu_dereference(sk->sk_wq);
2163 if (wq_has_sleeper(wq))
2164 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2165 POLLRDNORM | POLLRDBAND);
2166 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2170 static void sock_def_write_space(struct sock *sk)
2172 struct socket_wq *wq;
2176 /* Do not wake up a writer until he can make "significant"
2179 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2180 wq = rcu_dereference(sk->sk_wq);
2181 if (wq_has_sleeper(wq))
2182 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2183 POLLWRNORM | POLLWRBAND);
2185 /* Should agree with poll, otherwise some programs break */
2186 if (sock_writeable(sk))
2187 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2193 static void sock_def_destruct(struct sock *sk)
2195 kfree(sk->sk_protinfo);
2198 void sk_send_sigurg(struct sock *sk)
2200 if (sk->sk_socket && sk->sk_socket->file)
2201 if (send_sigurg(&sk->sk_socket->file->f_owner))
2202 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2204 EXPORT_SYMBOL(sk_send_sigurg);
2206 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2207 unsigned long expires)
2209 if (!mod_timer(timer, expires))
2212 EXPORT_SYMBOL(sk_reset_timer);
2214 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2216 if (del_timer(timer))
2219 EXPORT_SYMBOL(sk_stop_timer);
2221 void sock_init_data(struct socket *sock, struct sock *sk)
2223 skb_queue_head_init(&sk->sk_receive_queue);
2224 skb_queue_head_init(&sk->sk_write_queue);
2225 skb_queue_head_init(&sk->sk_error_queue);
2226 #ifdef CONFIG_NET_DMA
2227 skb_queue_head_init(&sk->sk_async_wait_queue);
2230 sk->sk_send_head = NULL;
2232 init_timer(&sk->sk_timer);
2234 sk->sk_allocation = GFP_KERNEL;
2235 sk->sk_rcvbuf = sysctl_rmem_default;
2236 sk->sk_sndbuf = sysctl_wmem_default;
2237 sk->sk_state = TCP_CLOSE;
2238 sk_set_socket(sk, sock);
2240 sock_set_flag(sk, SOCK_ZAPPED);
2243 sk->sk_type = sock->type;
2244 sk->sk_wq = sock->wq;
2249 spin_lock_init(&sk->sk_dst_lock);
2250 rwlock_init(&sk->sk_callback_lock);
2251 lockdep_set_class_and_name(&sk->sk_callback_lock,
2252 af_callback_keys + sk->sk_family,
2253 af_family_clock_key_strings[sk->sk_family]);
2255 sk->sk_state_change = sock_def_wakeup;
2256 sk->sk_data_ready = sock_def_readable;
2257 sk->sk_write_space = sock_def_write_space;
2258 sk->sk_error_report = sock_def_error_report;
2259 sk->sk_destruct = sock_def_destruct;
2261 sk->sk_frag.page = NULL;
2262 sk->sk_frag.offset = 0;
2263 sk->sk_peek_off = -1;
2265 sk->sk_peer_pid = NULL;
2266 sk->sk_peer_cred = NULL;
2267 sk->sk_write_pending = 0;
2268 sk->sk_rcvlowat = 1;
2269 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2270 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2272 sk->sk_stamp = ktime_set(-1L, 0);
2274 sk->sk_pacing_rate = ~0U;
2276 * Before updating sk_refcnt, we must commit prior changes to memory
2277 * (Documentation/RCU/rculist_nulls.txt for details)
2280 atomic_set(&sk->sk_refcnt, 1);
2281 atomic_set(&sk->sk_drops, 0);
2283 EXPORT_SYMBOL(sock_init_data);
2285 void lock_sock_nested(struct sock *sk, int subclass)
2288 spin_lock_bh(&sk->sk_lock.slock);
2289 if (sk->sk_lock.owned)
2291 sk->sk_lock.owned = 1;
2292 spin_unlock(&sk->sk_lock.slock);
2294 * The sk_lock has mutex_lock() semantics here:
2296 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2299 EXPORT_SYMBOL(lock_sock_nested);
2301 void release_sock(struct sock *sk)
2304 * The sk_lock has mutex_unlock() semantics:
2306 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2308 spin_lock_bh(&sk->sk_lock.slock);
2309 if (sk->sk_backlog.tail)
2312 if (sk->sk_prot->release_cb)
2313 sk->sk_prot->release_cb(sk);
2315 sk->sk_lock.owned = 0;
2316 if (waitqueue_active(&sk->sk_lock.wq))
2317 wake_up(&sk->sk_lock.wq);
2318 spin_unlock_bh(&sk->sk_lock.slock);
2320 EXPORT_SYMBOL(release_sock);
2323 * lock_sock_fast - fast version of lock_sock
2326 * This version should be used for very small section, where process wont block
2327 * return false if fast path is taken
2328 * sk_lock.slock locked, owned = 0, BH disabled
2329 * return true if slow path is taken
2330 * sk_lock.slock unlocked, owned = 1, BH enabled
2332 bool lock_sock_fast(struct sock *sk)
2335 spin_lock_bh(&sk->sk_lock.slock);
2337 if (!sk->sk_lock.owned)
2339 * Note : We must disable BH
2344 sk->sk_lock.owned = 1;
2345 spin_unlock(&sk->sk_lock.slock);
2347 * The sk_lock has mutex_lock() semantics here:
2349 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2353 EXPORT_SYMBOL(lock_sock_fast);
2355 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2358 if (!sock_flag(sk, SOCK_TIMESTAMP))
2359 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2360 tv = ktime_to_timeval(sk->sk_stamp);
2361 if (tv.tv_sec == -1)
2363 if (tv.tv_sec == 0) {
2364 sk->sk_stamp = ktime_get_real();
2365 tv = ktime_to_timeval(sk->sk_stamp);
2367 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2369 EXPORT_SYMBOL(sock_get_timestamp);
2371 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2374 if (!sock_flag(sk, SOCK_TIMESTAMP))
2375 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2376 ts = ktime_to_timespec(sk->sk_stamp);
2377 if (ts.tv_sec == -1)
2379 if (ts.tv_sec == 0) {
2380 sk->sk_stamp = ktime_get_real();
2381 ts = ktime_to_timespec(sk->sk_stamp);
2383 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2385 EXPORT_SYMBOL(sock_get_timestampns);
2387 void sock_enable_timestamp(struct sock *sk, int flag)
2389 if (!sock_flag(sk, flag)) {
2390 unsigned long previous_flags = sk->sk_flags;
2392 sock_set_flag(sk, flag);
2394 * we just set one of the two flags which require net
2395 * time stamping, but time stamping might have been on
2396 * already because of the other one
2398 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2399 net_enable_timestamp();
2404 * Get a socket option on an socket.
2406 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2407 * asynchronous errors should be reported by getsockopt. We assume
2408 * this means if you specify SO_ERROR (otherwise whats the point of it).
2410 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2411 char __user *optval, int __user *optlen)
2413 struct sock *sk = sock->sk;
2415 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2417 EXPORT_SYMBOL(sock_common_getsockopt);
2419 #ifdef CONFIG_COMPAT
2420 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2421 char __user *optval, int __user *optlen)
2423 struct sock *sk = sock->sk;
2425 if (sk->sk_prot->compat_getsockopt != NULL)
2426 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2428 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2430 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2433 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2434 struct msghdr *msg, size_t size, int flags)
2436 struct sock *sk = sock->sk;
2440 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2441 flags & ~MSG_DONTWAIT, &addr_len);
2443 msg->msg_namelen = addr_len;
2446 EXPORT_SYMBOL(sock_common_recvmsg);
2449 * Set socket options on an inet socket.
2451 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2452 char __user *optval, unsigned int optlen)
2454 struct sock *sk = sock->sk;
2456 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2458 EXPORT_SYMBOL(sock_common_setsockopt);
2460 #ifdef CONFIG_COMPAT
2461 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2462 char __user *optval, unsigned int optlen)
2464 struct sock *sk = sock->sk;
2466 if (sk->sk_prot->compat_setsockopt != NULL)
2467 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2469 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2471 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2474 void sk_common_release(struct sock *sk)
2476 if (sk->sk_prot->destroy)
2477 sk->sk_prot->destroy(sk);
2480 * Observation: when sock_common_release is called, processes have
2481 * no access to socket. But net still has.
2482 * Step one, detach it from networking:
2484 * A. Remove from hash tables.
2487 sk->sk_prot->unhash(sk);
2490 * In this point socket cannot receive new packets, but it is possible
2491 * that some packets are in flight because some CPU runs receiver and
2492 * did hash table lookup before we unhashed socket. They will achieve
2493 * receive queue and will be purged by socket destructor.
2495 * Also we still have packets pending on receive queue and probably,
2496 * our own packets waiting in device queues. sock_destroy will drain
2497 * receive queue, but transmitted packets will delay socket destruction
2498 * until the last reference will be released.
2503 xfrm_sk_free_policy(sk);
2505 sk_refcnt_debug_release(sk);
2507 if (sk->sk_frag.page) {
2508 put_page(sk->sk_frag.page);
2509 sk->sk_frag.page = NULL;
2514 EXPORT_SYMBOL(sk_common_release);
2516 #ifdef CONFIG_PROC_FS
2517 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2519 int val[PROTO_INUSE_NR];
2522 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2524 #ifdef CONFIG_NET_NS
2525 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2527 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2529 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2531 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2533 int cpu, idx = prot->inuse_idx;
2536 for_each_possible_cpu(cpu)
2537 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2539 return res >= 0 ? res : 0;
2541 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2543 static int __net_init sock_inuse_init_net(struct net *net)
2545 net->core.inuse = alloc_percpu(struct prot_inuse);
2546 return net->core.inuse ? 0 : -ENOMEM;
2549 static void __net_exit sock_inuse_exit_net(struct net *net)
2551 free_percpu(net->core.inuse);
2554 static struct pernet_operations net_inuse_ops = {
2555 .init = sock_inuse_init_net,
2556 .exit = sock_inuse_exit_net,
2559 static __init int net_inuse_init(void)
2561 if (register_pernet_subsys(&net_inuse_ops))
2562 panic("Cannot initialize net inuse counters");
2567 core_initcall(net_inuse_init);
2569 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2571 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2573 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2575 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2577 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2579 int cpu, idx = prot->inuse_idx;
2582 for_each_possible_cpu(cpu)
2583 res += per_cpu(prot_inuse, cpu).val[idx];
2585 return res >= 0 ? res : 0;
2587 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2590 static void assign_proto_idx(struct proto *prot)
2592 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2594 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2595 pr_err("PROTO_INUSE_NR exhausted\n");
2599 set_bit(prot->inuse_idx, proto_inuse_idx);
2602 static void release_proto_idx(struct proto *prot)
2604 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2605 clear_bit(prot->inuse_idx, proto_inuse_idx);
2608 static inline void assign_proto_idx(struct proto *prot)
2612 static inline void release_proto_idx(struct proto *prot)
2617 int proto_register(struct proto *prot, int alloc_slab)
2620 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2621 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2624 if (prot->slab == NULL) {
2625 pr_crit("%s: Can't create sock SLAB cache!\n",
2630 if (prot->rsk_prot != NULL) {
2631 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2632 if (prot->rsk_prot->slab_name == NULL)
2633 goto out_free_sock_slab;
2635 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2636 prot->rsk_prot->obj_size, 0,
2637 SLAB_HWCACHE_ALIGN, NULL);
2639 if (prot->rsk_prot->slab == NULL) {
2640 pr_crit("%s: Can't create request sock SLAB cache!\n",
2642 goto out_free_request_sock_slab_name;
2646 if (prot->twsk_prot != NULL) {
2647 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2649 if (prot->twsk_prot->twsk_slab_name == NULL)
2650 goto out_free_request_sock_slab;
2652 prot->twsk_prot->twsk_slab =
2653 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2654 prot->twsk_prot->twsk_obj_size,
2656 SLAB_HWCACHE_ALIGN |
2659 if (prot->twsk_prot->twsk_slab == NULL)
2660 goto out_free_timewait_sock_slab_name;
2664 mutex_lock(&proto_list_mutex);
2665 list_add(&prot->node, &proto_list);
2666 assign_proto_idx(prot);
2667 mutex_unlock(&proto_list_mutex);
2670 out_free_timewait_sock_slab_name:
2671 kfree(prot->twsk_prot->twsk_slab_name);
2672 out_free_request_sock_slab:
2673 if (prot->rsk_prot && prot->rsk_prot->slab) {
2674 kmem_cache_destroy(prot->rsk_prot->slab);
2675 prot->rsk_prot->slab = NULL;
2677 out_free_request_sock_slab_name:
2679 kfree(prot->rsk_prot->slab_name);
2681 kmem_cache_destroy(prot->slab);
2686 EXPORT_SYMBOL(proto_register);
2688 void proto_unregister(struct proto *prot)
2690 mutex_lock(&proto_list_mutex);
2691 release_proto_idx(prot);
2692 list_del(&prot->node);
2693 mutex_unlock(&proto_list_mutex);
2695 if (prot->slab != NULL) {
2696 kmem_cache_destroy(prot->slab);
2700 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2701 kmem_cache_destroy(prot->rsk_prot->slab);
2702 kfree(prot->rsk_prot->slab_name);
2703 prot->rsk_prot->slab = NULL;
2706 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2707 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2708 kfree(prot->twsk_prot->twsk_slab_name);
2709 prot->twsk_prot->twsk_slab = NULL;
2712 EXPORT_SYMBOL(proto_unregister);
2714 #ifdef CONFIG_PROC_FS
2715 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2716 __acquires(proto_list_mutex)
2718 mutex_lock(&proto_list_mutex);
2719 return seq_list_start_head(&proto_list, *pos);
2722 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2724 return seq_list_next(v, &proto_list, pos);
2727 static void proto_seq_stop(struct seq_file *seq, void *v)
2728 __releases(proto_list_mutex)
2730 mutex_unlock(&proto_list_mutex);
2733 static char proto_method_implemented(const void *method)
2735 return method == NULL ? 'n' : 'y';
2737 static long sock_prot_memory_allocated(struct proto *proto)
2739 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2742 static char *sock_prot_memory_pressure(struct proto *proto)
2744 return proto->memory_pressure != NULL ?
2745 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2748 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2751 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2752 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2755 sock_prot_inuse_get(seq_file_net(seq), proto),
2756 sock_prot_memory_allocated(proto),
2757 sock_prot_memory_pressure(proto),
2759 proto->slab == NULL ? "no" : "yes",
2760 module_name(proto->owner),
2761 proto_method_implemented(proto->close),
2762 proto_method_implemented(proto->connect),
2763 proto_method_implemented(proto->disconnect),
2764 proto_method_implemented(proto->accept),
2765 proto_method_implemented(proto->ioctl),
2766 proto_method_implemented(proto->init),
2767 proto_method_implemented(proto->destroy),
2768 proto_method_implemented(proto->shutdown),
2769 proto_method_implemented(proto->setsockopt),
2770 proto_method_implemented(proto->getsockopt),
2771 proto_method_implemented(proto->sendmsg),
2772 proto_method_implemented(proto->recvmsg),
2773 proto_method_implemented(proto->sendpage),
2774 proto_method_implemented(proto->bind),
2775 proto_method_implemented(proto->backlog_rcv),
2776 proto_method_implemented(proto->hash),
2777 proto_method_implemented(proto->unhash),
2778 proto_method_implemented(proto->get_port),
2779 proto_method_implemented(proto->enter_memory_pressure));
2782 static int proto_seq_show(struct seq_file *seq, void *v)
2784 if (v == &proto_list)
2785 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2794 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2796 proto_seq_printf(seq, list_entry(v, struct proto, node));
2800 static const struct seq_operations proto_seq_ops = {
2801 .start = proto_seq_start,
2802 .next = proto_seq_next,
2803 .stop = proto_seq_stop,
2804 .show = proto_seq_show,
2807 static int proto_seq_open(struct inode *inode, struct file *file)
2809 return seq_open_net(inode, file, &proto_seq_ops,
2810 sizeof(struct seq_net_private));
2813 static const struct file_operations proto_seq_fops = {
2814 .owner = THIS_MODULE,
2815 .open = proto_seq_open,
2817 .llseek = seq_lseek,
2818 .release = seq_release_net,
2821 static __net_init int proto_init_net(struct net *net)
2823 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2829 static __net_exit void proto_exit_net(struct net *net)
2831 remove_proc_entry("protocols", net->proc_net);
2835 static __net_initdata struct pernet_operations proto_net_ops = {
2836 .init = proto_init_net,
2837 .exit = proto_exit_net,
2840 static int __init proto_init(void)
2842 return register_pernet_subsys(&proto_net_ops);
2845 subsys_initcall(proto_init);
2847 #endif /* PROC_FS */