2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <asm/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
135 #include <linux/filter.h>
137 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
148 #ifdef CONFIG_MEMCG_KMEM
149 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
154 mutex_lock(&proto_list_mutex);
155 list_for_each_entry(proto, &proto_list, node) {
156 if (proto->init_cgroup) {
157 ret = proto->init_cgroup(memcg, ss);
163 mutex_unlock(&proto_list_mutex);
166 list_for_each_entry_continue_reverse(proto, &proto_list, node)
167 if (proto->destroy_cgroup)
168 proto->destroy_cgroup(memcg);
169 mutex_unlock(&proto_list_mutex);
173 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
177 mutex_lock(&proto_list_mutex);
178 list_for_each_entry_reverse(proto, &proto_list, node)
179 if (proto->destroy_cgroup)
180 proto->destroy_cgroup(memcg);
181 mutex_unlock(&proto_list_mutex);
186 * Each address family might have different locking rules, so we have
187 * one slock key per address family:
189 static struct lock_class_key af_family_keys[AF_MAX];
190 static struct lock_class_key af_family_slock_keys[AF_MAX];
192 #if defined(CONFIG_MEMCG_KMEM)
193 struct static_key memcg_socket_limit_enabled;
194 EXPORT_SYMBOL(memcg_socket_limit_enabled);
198 * Make lock validator output more readable. (we pre-construct these
199 * strings build-time, so that runtime initialization of socket
202 static const char *const af_family_key_strings[AF_MAX+1] = {
203 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
204 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
205 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
206 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
207 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
208 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
209 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
210 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
211 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
212 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
213 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
214 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
215 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
216 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
218 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
219 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
220 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
221 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
222 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
223 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
224 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
225 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
226 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
227 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
228 "slock-27" , "slock-28" , "slock-AF_CAN" ,
229 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
230 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
231 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
232 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
234 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
235 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
236 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
237 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
238 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
239 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
240 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
241 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
242 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
243 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
244 "clock-27" , "clock-28" , "clock-AF_CAN" ,
245 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
246 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
247 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
248 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
252 * sk_callback_lock locking rules are per-address-family,
253 * so split the lock classes by using a per-AF key:
255 static struct lock_class_key af_callback_keys[AF_MAX];
257 /* Take into consideration the size of the struct sk_buff overhead in the
258 * determination of these values, since that is non-constant across
259 * platforms. This makes socket queueing behavior and performance
260 * not depend upon such differences.
262 #define _SK_MEM_PACKETS 256
263 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
264 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
265 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
267 /* Run time adjustable parameters. */
268 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
269 EXPORT_SYMBOL(sysctl_wmem_max);
270 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
271 EXPORT_SYMBOL(sysctl_rmem_max);
272 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
273 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
275 /* Maximal space eaten by iovec or ancillary data plus some space */
276 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
277 EXPORT_SYMBOL(sysctl_optmem_max);
279 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
280 EXPORT_SYMBOL_GPL(memalloc_socks);
283 * sk_set_memalloc - sets %SOCK_MEMALLOC
284 * @sk: socket to set it on
286 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
287 * It's the responsibility of the admin to adjust min_free_kbytes
288 * to meet the requirements
290 void sk_set_memalloc(struct sock *sk)
292 sock_set_flag(sk, SOCK_MEMALLOC);
293 sk->sk_allocation |= __GFP_MEMALLOC;
294 static_key_slow_inc(&memalloc_socks);
296 EXPORT_SYMBOL_GPL(sk_set_memalloc);
298 void sk_clear_memalloc(struct sock *sk)
300 sock_reset_flag(sk, SOCK_MEMALLOC);
301 sk->sk_allocation &= ~__GFP_MEMALLOC;
302 static_key_slow_dec(&memalloc_socks);
305 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
306 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
307 * it has rmem allocations there is a risk that the user of the
308 * socket cannot make forward progress due to exceeding the rmem
309 * limits. By rights, sk_clear_memalloc() should only be called
310 * on sockets being torn down but warn and reset the accounting if
311 * that assumption breaks.
313 if (WARN_ON(sk->sk_forward_alloc))
316 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
318 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
321 unsigned long pflags = current->flags;
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
326 current->flags |= PF_MEMALLOC;
327 ret = sk->sk_backlog_rcv(sk, skb);
328 tsk_restore_flags(current, pflags, PF_MEMALLOC);
332 EXPORT_SYMBOL(__sk_backlog_rcv);
334 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
338 if (optlen < sizeof(tv))
340 if (copy_from_user(&tv, optval, sizeof(tv)))
342 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
346 static int warned __read_mostly;
349 if (warned < 10 && net_ratelimit()) {
351 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
352 __func__, current->comm, task_pid_nr(current));
356 *timeo_p = MAX_SCHEDULE_TIMEOUT;
357 if (tv.tv_sec == 0 && tv.tv_usec == 0)
359 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
360 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
364 static void sock_warn_obsolete_bsdism(const char *name)
367 static char warncomm[TASK_COMM_LEN];
368 if (strcmp(warncomm, current->comm) && warned < 5) {
369 strcpy(warncomm, current->comm);
370 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
376 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
378 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
380 if (sk->sk_flags & flags) {
381 sk->sk_flags &= ~flags;
382 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
383 net_disable_timestamp();
388 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
393 struct sk_buff_head *list = &sk->sk_receive_queue;
395 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
396 atomic_inc(&sk->sk_drops);
397 trace_sock_rcvqueue_full(sk, skb);
401 err = sk_filter(sk, skb);
405 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
406 atomic_inc(&sk->sk_drops);
411 skb_set_owner_r(skb, sk);
413 /* Cache the SKB length before we tack it onto the receive
414 * queue. Once it is added it no longer belongs to us and
415 * may be freed by other threads of control pulling packets
420 /* we escape from rcu protected region, make sure we dont leak
425 spin_lock_irqsave(&list->lock, flags);
426 skb->dropcount = atomic_read(&sk->sk_drops);
427 __skb_queue_tail(list, skb);
428 spin_unlock_irqrestore(&list->lock, flags);
430 if (!sock_flag(sk, SOCK_DEAD))
431 sk->sk_data_ready(sk, skb_len);
434 EXPORT_SYMBOL(sock_queue_rcv_skb);
436 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
438 int rc = NET_RX_SUCCESS;
440 if (sk_filter(sk, skb))
441 goto discard_and_relse;
445 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
446 atomic_inc(&sk->sk_drops);
447 goto discard_and_relse;
450 bh_lock_sock_nested(sk);
453 if (!sock_owned_by_user(sk)) {
455 * trylock + unlock semantics:
457 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
459 rc = sk_backlog_rcv(sk, skb);
461 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
462 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
464 atomic_inc(&sk->sk_drops);
465 goto discard_and_relse;
476 EXPORT_SYMBOL(sk_receive_skb);
478 void sk_reset_txq(struct sock *sk)
480 sk_tx_queue_clear(sk);
482 EXPORT_SYMBOL(sk_reset_txq);
484 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
486 struct dst_entry *dst = __sk_dst_get(sk);
488 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
489 sk_tx_queue_clear(sk);
490 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
497 EXPORT_SYMBOL(__sk_dst_check);
499 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
501 struct dst_entry *dst = sk_dst_get(sk);
503 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
511 EXPORT_SYMBOL(sk_dst_check);
513 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
516 int ret = -ENOPROTOOPT;
517 #ifdef CONFIG_NETDEVICES
518 struct net *net = sock_net(sk);
519 char devname[IFNAMSIZ];
524 if (!ns_capable(net->user_ns, CAP_NET_RAW))
531 /* Bind this socket to a particular device like "eth0",
532 * as specified in the passed interface name. If the
533 * name is "" or the option length is zero the socket
536 if (optlen > IFNAMSIZ - 1)
537 optlen = IFNAMSIZ - 1;
538 memset(devname, 0, sizeof(devname));
541 if (copy_from_user(devname, optval, optlen))
545 if (devname[0] != '\0') {
546 struct net_device *dev;
549 dev = dev_get_by_name_rcu(net, devname);
551 index = dev->ifindex;
559 sk->sk_bound_dev_if = index;
571 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
572 int __user *optlen, int len)
574 int ret = -ENOPROTOOPT;
575 #ifdef CONFIG_NETDEVICES
576 struct net *net = sock_net(sk);
577 char devname[IFNAMSIZ];
579 if (sk->sk_bound_dev_if == 0) {
588 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
592 len = strlen(devname) + 1;
595 if (copy_to_user(optval, devname, len))
600 if (put_user(len, optlen))
611 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
614 sock_set_flag(sk, bit);
616 sock_reset_flag(sk, bit);
620 * This is meant for all protocols to use and covers goings on
621 * at the socket level. Everything here is generic.
624 int sock_setsockopt(struct socket *sock, int level, int optname,
625 char __user *optval, unsigned int optlen)
627 struct sock *sk = sock->sk;
634 * Options without arguments
637 if (optname == SO_BINDTODEVICE)
638 return sock_setbindtodevice(sk, optval, optlen);
640 if (optlen < sizeof(int))
643 if (get_user(val, (int __user *)optval))
646 valbool = val ? 1 : 0;
652 if (val && !capable(CAP_NET_ADMIN))
655 sock_valbool_flag(sk, SOCK_DBG, valbool);
658 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
661 sk->sk_reuseport = valbool;
670 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
673 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
676 /* Don't error on this BSD doesn't and if you think
677 * about it this is right. Otherwise apps have to
678 * play 'guess the biggest size' games. RCVBUF/SNDBUF
679 * are treated in BSD as hints
681 val = min_t(u32, val, sysctl_wmem_max);
683 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
684 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
685 /* Wake up sending tasks if we upped the value. */
686 sk->sk_write_space(sk);
690 if (!capable(CAP_NET_ADMIN)) {
697 /* Don't error on this BSD doesn't and if you think
698 * about it this is right. Otherwise apps have to
699 * play 'guess the biggest size' games. RCVBUF/SNDBUF
700 * are treated in BSD as hints
702 val = min_t(u32, val, sysctl_rmem_max);
704 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
706 * We double it on the way in to account for
707 * "struct sk_buff" etc. overhead. Applications
708 * assume that the SO_RCVBUF setting they make will
709 * allow that much actual data to be received on that
712 * Applications are unaware that "struct sk_buff" and
713 * other overheads allocate from the receive buffer
714 * during socket buffer allocation.
716 * And after considering the possible alternatives,
717 * returning the value we actually used in getsockopt
718 * is the most desirable behavior.
720 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
724 if (!capable(CAP_NET_ADMIN)) {
732 if (sk->sk_protocol == IPPROTO_TCP &&
733 sk->sk_type == SOCK_STREAM)
734 tcp_set_keepalive(sk, valbool);
736 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
740 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
744 sk->sk_no_check = valbool;
748 if ((val >= 0 && val <= 6) ||
749 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
750 sk->sk_priority = val;
756 if (optlen < sizeof(ling)) {
757 ret = -EINVAL; /* 1003.1g */
760 if (copy_from_user(&ling, optval, sizeof(ling))) {
765 sock_reset_flag(sk, SOCK_LINGER);
767 #if (BITS_PER_LONG == 32)
768 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
769 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
772 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
773 sock_set_flag(sk, SOCK_LINGER);
778 sock_warn_obsolete_bsdism("setsockopt");
783 set_bit(SOCK_PASSCRED, &sock->flags);
785 clear_bit(SOCK_PASSCRED, &sock->flags);
791 if (optname == SO_TIMESTAMP)
792 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
794 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
795 sock_set_flag(sk, SOCK_RCVTSTAMP);
796 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
798 sock_reset_flag(sk, SOCK_RCVTSTAMP);
799 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
803 case SO_TIMESTAMPING:
804 if (val & ~SOF_TIMESTAMPING_MASK) {
808 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
809 val & SOF_TIMESTAMPING_TX_HARDWARE);
810 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
811 val & SOF_TIMESTAMPING_TX_SOFTWARE);
812 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
813 val & SOF_TIMESTAMPING_RX_HARDWARE);
814 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
815 sock_enable_timestamp(sk,
816 SOCK_TIMESTAMPING_RX_SOFTWARE);
818 sock_disable_timestamp(sk,
819 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
820 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
821 val & SOF_TIMESTAMPING_SOFTWARE);
822 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
823 val & SOF_TIMESTAMPING_SYS_HARDWARE);
824 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
825 val & SOF_TIMESTAMPING_RAW_HARDWARE);
831 sk->sk_rcvlowat = val ? : 1;
835 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
839 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
842 case SO_ATTACH_FILTER:
844 if (optlen == sizeof(struct sock_fprog)) {
845 struct sock_fprog fprog;
848 if (copy_from_user(&fprog, optval, sizeof(fprog)))
851 ret = sk_attach_filter(&fprog, sk);
855 case SO_DETACH_FILTER:
856 ret = sk_detach_filter(sk);
860 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
863 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
868 set_bit(SOCK_PASSSEC, &sock->flags);
870 clear_bit(SOCK_PASSSEC, &sock->flags);
873 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
879 /* We implement the SO_SNDLOWAT etc to
880 not be settable (1003.1g 5.3) */
882 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
886 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
890 if (sock->ops->set_peek_off)
891 sock->ops->set_peek_off(sk, val);
897 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
900 case SO_SELECT_ERR_QUEUE:
901 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
904 #ifdef CONFIG_NET_RX_BUSY_POLL
906 /* allow unprivileged users to decrease the value */
907 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
913 sk->sk_ll_usec = val;
924 EXPORT_SYMBOL(sock_setsockopt);
927 void cred_to_ucred(struct pid *pid, const struct cred *cred,
930 ucred->pid = pid_vnr(pid);
931 ucred->uid = ucred->gid = -1;
933 struct user_namespace *current_ns = current_user_ns();
935 ucred->uid = from_kuid_munged(current_ns, cred->euid);
936 ucred->gid = from_kgid_munged(current_ns, cred->egid);
939 EXPORT_SYMBOL_GPL(cred_to_ucred);
941 int sock_getsockopt(struct socket *sock, int level, int optname,
942 char __user *optval, int __user *optlen)
944 struct sock *sk = sock->sk;
952 int lv = sizeof(int);
955 if (get_user(len, optlen))
960 memset(&v, 0, sizeof(v));
964 v.val = sock_flag(sk, SOCK_DBG);
968 v.val = sock_flag(sk, SOCK_LOCALROUTE);
972 v.val = sock_flag(sk, SOCK_BROADCAST);
976 v.val = sk->sk_sndbuf;
980 v.val = sk->sk_rcvbuf;
984 v.val = sk->sk_reuse;
988 v.val = sk->sk_reuseport;
992 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1000 v.val = sk->sk_protocol;
1004 v.val = sk->sk_family;
1008 v.val = -sock_error(sk);
1010 v.val = xchg(&sk->sk_err_soft, 0);
1014 v.val = sock_flag(sk, SOCK_URGINLINE);
1018 v.val = sk->sk_no_check;
1022 v.val = sk->sk_priority;
1026 lv = sizeof(v.ling);
1027 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1028 v.ling.l_linger = sk->sk_lingertime / HZ;
1032 sock_warn_obsolete_bsdism("getsockopt");
1036 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1037 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1040 case SO_TIMESTAMPNS:
1041 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1044 case SO_TIMESTAMPING:
1046 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
1047 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
1048 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
1049 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
1050 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
1051 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
1052 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1053 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
1054 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
1055 v.val |= SOF_TIMESTAMPING_SOFTWARE;
1056 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
1057 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
1058 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
1059 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
1063 lv = sizeof(struct timeval);
1064 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1068 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1069 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1074 lv = sizeof(struct timeval);
1075 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1079 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1080 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1085 v.val = sk->sk_rcvlowat;
1093 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1098 struct ucred peercred;
1099 if (len > sizeof(peercred))
1100 len = sizeof(peercred);
1101 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1102 if (copy_to_user(optval, &peercred, len))
1111 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1115 if (copy_to_user(optval, address, len))
1120 /* Dubious BSD thing... Probably nobody even uses it, but
1121 * the UNIX standard wants it for whatever reason... -DaveM
1124 v.val = sk->sk_state == TCP_LISTEN;
1128 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1132 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1135 v.val = sk->sk_mark;
1139 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1142 case SO_WIFI_STATUS:
1143 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1147 if (!sock->ops->set_peek_off)
1150 v.val = sk->sk_peek_off;
1153 v.val = sock_flag(sk, SOCK_NOFCS);
1156 case SO_BINDTODEVICE:
1157 return sock_getbindtodevice(sk, optval, optlen, len);
1160 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1166 case SO_LOCK_FILTER:
1167 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1170 case SO_SELECT_ERR_QUEUE:
1171 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1174 #ifdef CONFIG_NET_RX_BUSY_POLL
1176 v.val = sk->sk_ll_usec;
1181 return -ENOPROTOOPT;
1186 if (copy_to_user(optval, &v, len))
1189 if (put_user(len, optlen))
1195 * Initialize an sk_lock.
1197 * (We also register the sk_lock with the lock validator.)
1199 static inline void sock_lock_init(struct sock *sk)
1201 sock_lock_init_class_and_name(sk,
1202 af_family_slock_key_strings[sk->sk_family],
1203 af_family_slock_keys + sk->sk_family,
1204 af_family_key_strings[sk->sk_family],
1205 af_family_keys + sk->sk_family);
1209 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1210 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1211 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1213 static void sock_copy(struct sock *nsk, const struct sock *osk)
1215 #ifdef CONFIG_SECURITY_NETWORK
1216 void *sptr = nsk->sk_security;
1218 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1220 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1221 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1223 #ifdef CONFIG_SECURITY_NETWORK
1224 nsk->sk_security = sptr;
1225 security_sk_clone(osk, nsk);
1229 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1231 unsigned long nulls1, nulls2;
1233 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1234 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1235 if (nulls1 > nulls2)
1236 swap(nulls1, nulls2);
1239 memset((char *)sk, 0, nulls1);
1240 memset((char *)sk + nulls1 + sizeof(void *), 0,
1241 nulls2 - nulls1 - sizeof(void *));
1242 memset((char *)sk + nulls2 + sizeof(void *), 0,
1243 size - nulls2 - sizeof(void *));
1245 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1247 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1251 struct kmem_cache *slab;
1255 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1258 if (priority & __GFP_ZERO) {
1260 prot->clear_sk(sk, prot->obj_size);
1262 sk_prot_clear_nulls(sk, prot->obj_size);
1265 sk = kmalloc(prot->obj_size, priority);
1268 kmemcheck_annotate_bitfield(sk, flags);
1270 if (security_sk_alloc(sk, family, priority))
1273 if (!try_module_get(prot->owner))
1275 sk_tx_queue_clear(sk);
1281 security_sk_free(sk);
1284 kmem_cache_free(slab, sk);
1290 static void sk_prot_free(struct proto *prot, struct sock *sk)
1292 struct kmem_cache *slab;
1293 struct module *owner;
1295 owner = prot->owner;
1298 security_sk_free(sk);
1300 kmem_cache_free(slab, sk);
1306 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1307 void sock_update_classid(struct sock *sk)
1311 classid = task_cls_classid(current);
1312 if (classid != sk->sk_classid)
1313 sk->sk_classid = classid;
1315 EXPORT_SYMBOL(sock_update_classid);
1318 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1319 void sock_update_netprioidx(struct sock *sk)
1324 sk->sk_cgrp_prioidx = task_netprioidx(current);
1326 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1330 * sk_alloc - All socket objects are allocated here
1331 * @net: the applicable net namespace
1332 * @family: protocol family
1333 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1334 * @prot: struct proto associated with this new sock instance
1336 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1341 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1343 sk->sk_family = family;
1345 * See comment in struct sock definition to understand
1346 * why we need sk_prot_creator -acme
1348 sk->sk_prot = sk->sk_prot_creator = prot;
1350 sock_net_set(sk, get_net(net));
1351 atomic_set(&sk->sk_wmem_alloc, 1);
1353 sock_update_classid(sk);
1354 sock_update_netprioidx(sk);
1359 EXPORT_SYMBOL(sk_alloc);
1361 static void __sk_free(struct sock *sk)
1363 struct sk_filter *filter;
1365 if (sk->sk_destruct)
1366 sk->sk_destruct(sk);
1368 filter = rcu_dereference_check(sk->sk_filter,
1369 atomic_read(&sk->sk_wmem_alloc) == 0);
1371 sk_filter_uncharge(sk, filter);
1372 RCU_INIT_POINTER(sk->sk_filter, NULL);
1375 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1377 if (atomic_read(&sk->sk_omem_alloc))
1378 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1379 __func__, atomic_read(&sk->sk_omem_alloc));
1381 if (sk->sk_peer_cred)
1382 put_cred(sk->sk_peer_cred);
1383 put_pid(sk->sk_peer_pid);
1384 put_net(sock_net(sk));
1385 sk_prot_free(sk->sk_prot_creator, sk);
1388 void sk_free(struct sock *sk)
1391 * We subtract one from sk_wmem_alloc and can know if
1392 * some packets are still in some tx queue.
1393 * If not null, sock_wfree() will call __sk_free(sk) later
1395 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1398 EXPORT_SYMBOL(sk_free);
1401 * Last sock_put should drop reference to sk->sk_net. It has already
1402 * been dropped in sk_change_net. Taking reference to stopping namespace
1404 * Take reference to a socket to remove it from hash _alive_ and after that
1405 * destroy it in the context of init_net.
1407 void sk_release_kernel(struct sock *sk)
1409 if (sk == NULL || sk->sk_socket == NULL)
1413 sock_release(sk->sk_socket);
1414 release_net(sock_net(sk));
1415 sock_net_set(sk, get_net(&init_net));
1418 EXPORT_SYMBOL(sk_release_kernel);
1420 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1422 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1423 sock_update_memcg(newsk);
1427 * sk_clone_lock - clone a socket, and lock its clone
1428 * @sk: the socket to clone
1429 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1431 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1433 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1437 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1438 if (newsk != NULL) {
1439 struct sk_filter *filter;
1441 sock_copy(newsk, sk);
1444 get_net(sock_net(newsk));
1445 sk_node_init(&newsk->sk_node);
1446 sock_lock_init(newsk);
1447 bh_lock_sock(newsk);
1448 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1449 newsk->sk_backlog.len = 0;
1451 atomic_set(&newsk->sk_rmem_alloc, 0);
1453 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1455 atomic_set(&newsk->sk_wmem_alloc, 1);
1456 atomic_set(&newsk->sk_omem_alloc, 0);
1457 skb_queue_head_init(&newsk->sk_receive_queue);
1458 skb_queue_head_init(&newsk->sk_write_queue);
1459 #ifdef CONFIG_NET_DMA
1460 skb_queue_head_init(&newsk->sk_async_wait_queue);
1463 spin_lock_init(&newsk->sk_dst_lock);
1464 rwlock_init(&newsk->sk_callback_lock);
1465 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1466 af_callback_keys + newsk->sk_family,
1467 af_family_clock_key_strings[newsk->sk_family]);
1469 newsk->sk_dst_cache = NULL;
1470 newsk->sk_wmem_queued = 0;
1471 newsk->sk_forward_alloc = 0;
1472 newsk->sk_send_head = NULL;
1473 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1475 sock_reset_flag(newsk, SOCK_DONE);
1476 skb_queue_head_init(&newsk->sk_error_queue);
1478 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1480 sk_filter_charge(newsk, filter);
1482 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1483 /* It is still raw copy of parent, so invalidate
1484 * destructor and make plain sk_free() */
1485 newsk->sk_destruct = NULL;
1486 bh_unlock_sock(newsk);
1493 newsk->sk_priority = 0;
1495 * Before updating sk_refcnt, we must commit prior changes to memory
1496 * (Documentation/RCU/rculist_nulls.txt for details)
1499 atomic_set(&newsk->sk_refcnt, 2);
1502 * Increment the counter in the same struct proto as the master
1503 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1504 * is the same as sk->sk_prot->socks, as this field was copied
1507 * This _changes_ the previous behaviour, where
1508 * tcp_create_openreq_child always was incrementing the
1509 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1510 * to be taken into account in all callers. -acme
1512 sk_refcnt_debug_inc(newsk);
1513 sk_set_socket(newsk, NULL);
1514 newsk->sk_wq = NULL;
1516 sk_update_clone(sk, newsk);
1518 if (newsk->sk_prot->sockets_allocated)
1519 sk_sockets_allocated_inc(newsk);
1521 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1522 net_enable_timestamp();
1527 EXPORT_SYMBOL_GPL(sk_clone_lock);
1529 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1531 __sk_dst_set(sk, dst);
1532 sk->sk_route_caps = dst->dev->features;
1533 if (sk->sk_route_caps & NETIF_F_GSO)
1534 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1535 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1536 if (sk_can_gso(sk)) {
1537 if (dst->header_len) {
1538 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1540 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1541 sk->sk_gso_max_size = dst->dev->gso_max_size;
1542 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1546 EXPORT_SYMBOL_GPL(sk_setup_caps);
1549 * Simple resource managers for sockets.
1554 * Write buffer destructor automatically called from kfree_skb.
1556 void sock_wfree(struct sk_buff *skb)
1558 struct sock *sk = skb->sk;
1559 unsigned int len = skb->truesize;
1561 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1563 * Keep a reference on sk_wmem_alloc, this will be released
1564 * after sk_write_space() call
1566 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1567 sk->sk_write_space(sk);
1571 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1572 * could not do because of in-flight packets
1574 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1577 EXPORT_SYMBOL(sock_wfree);
1579 void skb_orphan_partial(struct sk_buff *skb)
1581 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1582 * so we do not completely orphan skb, but transfert all
1583 * accounted bytes but one, to avoid unexpected reorders.
1585 if (skb->destructor == sock_wfree
1587 || skb->destructor == tcp_wfree
1590 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1596 EXPORT_SYMBOL(skb_orphan_partial);
1599 * Read buffer destructor automatically called from kfree_skb.
1601 void sock_rfree(struct sk_buff *skb)
1603 struct sock *sk = skb->sk;
1604 unsigned int len = skb->truesize;
1606 atomic_sub(len, &sk->sk_rmem_alloc);
1607 sk_mem_uncharge(sk, len);
1609 EXPORT_SYMBOL(sock_rfree);
1611 void sock_edemux(struct sk_buff *skb)
1613 struct sock *sk = skb->sk;
1616 if (sk->sk_state == TCP_TIME_WAIT)
1617 inet_twsk_put(inet_twsk(sk));
1622 EXPORT_SYMBOL(sock_edemux);
1624 kuid_t sock_i_uid(struct sock *sk)
1628 read_lock_bh(&sk->sk_callback_lock);
1629 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1630 read_unlock_bh(&sk->sk_callback_lock);
1633 EXPORT_SYMBOL(sock_i_uid);
1635 unsigned long sock_i_ino(struct sock *sk)
1639 read_lock_bh(&sk->sk_callback_lock);
1640 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1641 read_unlock_bh(&sk->sk_callback_lock);
1644 EXPORT_SYMBOL(sock_i_ino);
1647 * Allocate a skb from the socket's send buffer.
1649 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1652 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1653 struct sk_buff *skb = alloc_skb(size, priority);
1655 skb_set_owner_w(skb, sk);
1661 EXPORT_SYMBOL(sock_wmalloc);
1664 * Allocate a skb from the socket's receive buffer.
1666 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1669 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1670 struct sk_buff *skb = alloc_skb(size, priority);
1672 skb_set_owner_r(skb, sk);
1680 * Allocate a memory block from the socket's option memory buffer.
1682 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1684 if ((unsigned int)size <= sysctl_optmem_max &&
1685 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1687 /* First do the add, to avoid the race if kmalloc
1690 atomic_add(size, &sk->sk_omem_alloc);
1691 mem = kmalloc(size, priority);
1694 atomic_sub(size, &sk->sk_omem_alloc);
1698 EXPORT_SYMBOL(sock_kmalloc);
1701 * Free an option memory block.
1703 void sock_kfree_s(struct sock *sk, void *mem, int size)
1706 atomic_sub(size, &sk->sk_omem_alloc);
1708 EXPORT_SYMBOL(sock_kfree_s);
1710 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1711 I think, these locks should be removed for datagram sockets.
1713 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1717 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1721 if (signal_pending(current))
1723 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1724 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1725 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1727 if (sk->sk_shutdown & SEND_SHUTDOWN)
1731 timeo = schedule_timeout(timeo);
1733 finish_wait(sk_sleep(sk), &wait);
1739 * Generic send/receive buffer handlers
1742 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1743 unsigned long data_len, int noblock,
1744 int *errcode, int max_page_order)
1746 struct sk_buff *skb = NULL;
1747 unsigned long chunk;
1751 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1756 if (npages > MAX_SKB_FRAGS)
1759 timeo = sock_sndtimeo(sk, noblock);
1761 err = sock_error(sk);
1766 if (sk->sk_shutdown & SEND_SHUTDOWN)
1769 if (atomic_read(&sk->sk_wmem_alloc) >= sk->sk_sndbuf) {
1770 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1771 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1775 if (signal_pending(current))
1777 timeo = sock_wait_for_wmem(sk, timeo);
1782 gfp_mask = sk->sk_allocation;
1783 if (gfp_mask & __GFP_WAIT)
1784 gfp_mask |= __GFP_REPEAT;
1786 skb = alloc_skb(header_len, gfp_mask);
1790 skb->truesize += data_len;
1792 for (i = 0; npages > 0; i++) {
1793 int order = max_page_order;
1796 if (npages >= 1 << order) {
1797 page = alloc_pages(sk->sk_allocation |
1798 __GFP_COMP | __GFP_NOWARN,
1805 page = alloc_page(sk->sk_allocation);
1809 chunk = min_t(unsigned long, data_len,
1810 PAGE_SIZE << order);
1811 skb_fill_page_desc(skb, i, page, 0, chunk);
1813 npages -= 1 << order;
1817 skb_set_owner_w(skb, sk);
1821 err = sock_intr_errno(timeo);
1827 EXPORT_SYMBOL(sock_alloc_send_pskb);
1829 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1830 int noblock, int *errcode)
1832 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1834 EXPORT_SYMBOL(sock_alloc_send_skb);
1836 /* On 32bit arches, an skb frag is limited to 2^15 */
1837 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1839 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1844 if (atomic_read(&pfrag->page->_count) == 1) {
1848 if (pfrag->offset < pfrag->size)
1850 put_page(pfrag->page);
1853 /* We restrict high order allocations to users that can afford to wait */
1854 order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1857 gfp_t gfp = sk->sk_allocation;
1860 gfp |= __GFP_COMP | __GFP_NOWARN;
1861 pfrag->page = alloc_pages(gfp, order);
1862 if (likely(pfrag->page)) {
1864 pfrag->size = PAGE_SIZE << order;
1867 } while (--order >= 0);
1869 sk_enter_memory_pressure(sk);
1870 sk_stream_moderate_sndbuf(sk);
1873 EXPORT_SYMBOL(sk_page_frag_refill);
1875 static void __lock_sock(struct sock *sk)
1876 __releases(&sk->sk_lock.slock)
1877 __acquires(&sk->sk_lock.slock)
1882 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1883 TASK_UNINTERRUPTIBLE);
1884 spin_unlock_bh(&sk->sk_lock.slock);
1886 spin_lock_bh(&sk->sk_lock.slock);
1887 if (!sock_owned_by_user(sk))
1890 finish_wait(&sk->sk_lock.wq, &wait);
1893 static void __release_sock(struct sock *sk)
1894 __releases(&sk->sk_lock.slock)
1895 __acquires(&sk->sk_lock.slock)
1897 struct sk_buff *skb = sk->sk_backlog.head;
1900 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1904 struct sk_buff *next = skb->next;
1907 WARN_ON_ONCE(skb_dst_is_noref(skb));
1909 sk_backlog_rcv(sk, skb);
1912 * We are in process context here with softirqs
1913 * disabled, use cond_resched_softirq() to preempt.
1914 * This is safe to do because we've taken the backlog
1917 cond_resched_softirq();
1920 } while (skb != NULL);
1923 } while ((skb = sk->sk_backlog.head) != NULL);
1926 * Doing the zeroing here guarantee we can not loop forever
1927 * while a wild producer attempts to flood us.
1929 sk->sk_backlog.len = 0;
1933 * sk_wait_data - wait for data to arrive at sk_receive_queue
1934 * @sk: sock to wait on
1935 * @timeo: for how long
1937 * Now socket state including sk->sk_err is changed only under lock,
1938 * hence we may omit checks after joining wait queue.
1939 * We check receive queue before schedule() only as optimization;
1940 * it is very likely that release_sock() added new data.
1942 int sk_wait_data(struct sock *sk, long *timeo)
1947 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1948 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1949 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1950 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1951 finish_wait(sk_sleep(sk), &wait);
1954 EXPORT_SYMBOL(sk_wait_data);
1957 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1959 * @size: memory size to allocate
1960 * @kind: allocation type
1962 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1963 * rmem allocation. This function assumes that protocols which have
1964 * memory_pressure use sk_wmem_queued as write buffer accounting.
1966 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1968 struct proto *prot = sk->sk_prot;
1969 int amt = sk_mem_pages(size);
1971 int parent_status = UNDER_LIMIT;
1973 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1975 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1978 if (parent_status == UNDER_LIMIT &&
1979 allocated <= sk_prot_mem_limits(sk, 0)) {
1980 sk_leave_memory_pressure(sk);
1984 /* Under pressure. (we or our parents) */
1985 if ((parent_status > SOFT_LIMIT) ||
1986 allocated > sk_prot_mem_limits(sk, 1))
1987 sk_enter_memory_pressure(sk);
1989 /* Over hard limit (we or our parents) */
1990 if ((parent_status == OVER_LIMIT) ||
1991 (allocated > sk_prot_mem_limits(sk, 2)))
1992 goto suppress_allocation;
1994 /* guarantee minimum buffer size under pressure */
1995 if (kind == SK_MEM_RECV) {
1996 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1999 } else { /* SK_MEM_SEND */
2000 if (sk->sk_type == SOCK_STREAM) {
2001 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2003 } else if (atomic_read(&sk->sk_wmem_alloc) <
2004 prot->sysctl_wmem[0])
2008 if (sk_has_memory_pressure(sk)) {
2011 if (!sk_under_memory_pressure(sk))
2013 alloc = sk_sockets_allocated_read_positive(sk);
2014 if (sk_prot_mem_limits(sk, 2) > alloc *
2015 sk_mem_pages(sk->sk_wmem_queued +
2016 atomic_read(&sk->sk_rmem_alloc) +
2017 sk->sk_forward_alloc))
2021 suppress_allocation:
2023 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2024 sk_stream_moderate_sndbuf(sk);
2026 /* Fail only if socket is _under_ its sndbuf.
2027 * In this case we cannot block, so that we have to fail.
2029 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2033 trace_sock_exceed_buf_limit(sk, prot, allocated);
2035 /* Alas. Undo changes. */
2036 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2038 sk_memory_allocated_sub(sk, amt);
2042 EXPORT_SYMBOL(__sk_mem_schedule);
2045 * __sk_reclaim - reclaim memory_allocated
2048 void __sk_mem_reclaim(struct sock *sk)
2050 sk_memory_allocated_sub(sk,
2051 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2052 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2054 if (sk_under_memory_pressure(sk) &&
2055 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2056 sk_leave_memory_pressure(sk);
2058 EXPORT_SYMBOL(__sk_mem_reclaim);
2062 * Set of default routines for initialising struct proto_ops when
2063 * the protocol does not support a particular function. In certain
2064 * cases where it makes no sense for a protocol to have a "do nothing"
2065 * function, some default processing is provided.
2068 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2072 EXPORT_SYMBOL(sock_no_bind);
2074 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2079 EXPORT_SYMBOL(sock_no_connect);
2081 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2085 EXPORT_SYMBOL(sock_no_socketpair);
2087 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2091 EXPORT_SYMBOL(sock_no_accept);
2093 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2098 EXPORT_SYMBOL(sock_no_getname);
2100 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2104 EXPORT_SYMBOL(sock_no_poll);
2106 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2110 EXPORT_SYMBOL(sock_no_ioctl);
2112 int sock_no_listen(struct socket *sock, int backlog)
2116 EXPORT_SYMBOL(sock_no_listen);
2118 int sock_no_shutdown(struct socket *sock, int how)
2122 EXPORT_SYMBOL(sock_no_shutdown);
2124 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2125 char __user *optval, unsigned int optlen)
2129 EXPORT_SYMBOL(sock_no_setsockopt);
2131 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2132 char __user *optval, int __user *optlen)
2136 EXPORT_SYMBOL(sock_no_getsockopt);
2138 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2143 EXPORT_SYMBOL(sock_no_sendmsg);
2145 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2146 size_t len, int flags)
2150 EXPORT_SYMBOL(sock_no_recvmsg);
2152 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2154 /* Mirror missing mmap method error code */
2157 EXPORT_SYMBOL(sock_no_mmap);
2159 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2162 struct msghdr msg = {.msg_flags = flags};
2164 char *kaddr = kmap(page);
2165 iov.iov_base = kaddr + offset;
2167 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2171 EXPORT_SYMBOL(sock_no_sendpage);
2174 * Default Socket Callbacks
2177 static void sock_def_wakeup(struct sock *sk)
2179 struct socket_wq *wq;
2182 wq = rcu_dereference(sk->sk_wq);
2183 if (wq_has_sleeper(wq))
2184 wake_up_interruptible_all(&wq->wait);
2188 static void sock_def_error_report(struct sock *sk)
2190 struct socket_wq *wq;
2193 wq = rcu_dereference(sk->sk_wq);
2194 if (wq_has_sleeper(wq))
2195 wake_up_interruptible_poll(&wq->wait, POLLERR);
2196 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2200 static void sock_def_readable(struct sock *sk, int len)
2202 struct socket_wq *wq;
2205 wq = rcu_dereference(sk->sk_wq);
2206 if (wq_has_sleeper(wq))
2207 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2208 POLLRDNORM | POLLRDBAND);
2209 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2213 static void sock_def_write_space(struct sock *sk)
2215 struct socket_wq *wq;
2219 /* Do not wake up a writer until he can make "significant"
2222 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2223 wq = rcu_dereference(sk->sk_wq);
2224 if (wq_has_sleeper(wq))
2225 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2226 POLLWRNORM | POLLWRBAND);
2228 /* Should agree with poll, otherwise some programs break */
2229 if (sock_writeable(sk))
2230 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2236 static void sock_def_destruct(struct sock *sk)
2238 kfree(sk->sk_protinfo);
2241 void sk_send_sigurg(struct sock *sk)
2243 if (sk->sk_socket && sk->sk_socket->file)
2244 if (send_sigurg(&sk->sk_socket->file->f_owner))
2245 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2247 EXPORT_SYMBOL(sk_send_sigurg);
2249 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2250 unsigned long expires)
2252 if (!mod_timer(timer, expires))
2255 EXPORT_SYMBOL(sk_reset_timer);
2257 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2259 if (del_timer(timer))
2262 EXPORT_SYMBOL(sk_stop_timer);
2264 void sock_init_data(struct socket *sock, struct sock *sk)
2266 skb_queue_head_init(&sk->sk_receive_queue);
2267 skb_queue_head_init(&sk->sk_write_queue);
2268 skb_queue_head_init(&sk->sk_error_queue);
2269 #ifdef CONFIG_NET_DMA
2270 skb_queue_head_init(&sk->sk_async_wait_queue);
2273 sk->sk_send_head = NULL;
2275 init_timer(&sk->sk_timer);
2277 sk->sk_allocation = GFP_KERNEL;
2278 sk->sk_rcvbuf = sysctl_rmem_default;
2279 sk->sk_sndbuf = sysctl_wmem_default;
2280 sk->sk_state = TCP_CLOSE;
2281 sk_set_socket(sk, sock);
2283 sock_set_flag(sk, SOCK_ZAPPED);
2286 sk->sk_type = sock->type;
2287 sk->sk_wq = sock->wq;
2292 spin_lock_init(&sk->sk_dst_lock);
2293 rwlock_init(&sk->sk_callback_lock);
2294 lockdep_set_class_and_name(&sk->sk_callback_lock,
2295 af_callback_keys + sk->sk_family,
2296 af_family_clock_key_strings[sk->sk_family]);
2298 sk->sk_state_change = sock_def_wakeup;
2299 sk->sk_data_ready = sock_def_readable;
2300 sk->sk_write_space = sock_def_write_space;
2301 sk->sk_error_report = sock_def_error_report;
2302 sk->sk_destruct = sock_def_destruct;
2304 sk->sk_frag.page = NULL;
2305 sk->sk_frag.offset = 0;
2306 sk->sk_peek_off = -1;
2308 sk->sk_peer_pid = NULL;
2309 sk->sk_peer_cred = NULL;
2310 sk->sk_write_pending = 0;
2311 sk->sk_rcvlowat = 1;
2312 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2313 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2315 sk->sk_stamp = ktime_set(-1L, 0);
2317 #ifdef CONFIG_NET_RX_BUSY_POLL
2319 sk->sk_ll_usec = sysctl_net_busy_read;
2323 * Before updating sk_refcnt, we must commit prior changes to memory
2324 * (Documentation/RCU/rculist_nulls.txt for details)
2327 atomic_set(&sk->sk_refcnt, 1);
2328 atomic_set(&sk->sk_drops, 0);
2330 EXPORT_SYMBOL(sock_init_data);
2332 void lock_sock_nested(struct sock *sk, int subclass)
2335 spin_lock_bh(&sk->sk_lock.slock);
2336 if (sk->sk_lock.owned)
2338 sk->sk_lock.owned = 1;
2339 spin_unlock(&sk->sk_lock.slock);
2341 * The sk_lock has mutex_lock() semantics here:
2343 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2346 EXPORT_SYMBOL(lock_sock_nested);
2348 void release_sock(struct sock *sk)
2351 * The sk_lock has mutex_unlock() semantics:
2353 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2355 spin_lock_bh(&sk->sk_lock.slock);
2356 if (sk->sk_backlog.tail)
2359 if (sk->sk_prot->release_cb)
2360 sk->sk_prot->release_cb(sk);
2362 sk->sk_lock.owned = 0;
2363 if (waitqueue_active(&sk->sk_lock.wq))
2364 wake_up(&sk->sk_lock.wq);
2365 spin_unlock_bh(&sk->sk_lock.slock);
2367 EXPORT_SYMBOL(release_sock);
2370 * lock_sock_fast - fast version of lock_sock
2373 * This version should be used for very small section, where process wont block
2374 * return false if fast path is taken
2375 * sk_lock.slock locked, owned = 0, BH disabled
2376 * return true if slow path is taken
2377 * sk_lock.slock unlocked, owned = 1, BH enabled
2379 bool lock_sock_fast(struct sock *sk)
2382 spin_lock_bh(&sk->sk_lock.slock);
2384 if (!sk->sk_lock.owned)
2386 * Note : We must disable BH
2391 sk->sk_lock.owned = 1;
2392 spin_unlock(&sk->sk_lock.slock);
2394 * The sk_lock has mutex_lock() semantics here:
2396 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2400 EXPORT_SYMBOL(lock_sock_fast);
2402 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2405 if (!sock_flag(sk, SOCK_TIMESTAMP))
2406 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2407 tv = ktime_to_timeval(sk->sk_stamp);
2408 if (tv.tv_sec == -1)
2410 if (tv.tv_sec == 0) {
2411 sk->sk_stamp = ktime_get_real();
2412 tv = ktime_to_timeval(sk->sk_stamp);
2414 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2416 EXPORT_SYMBOL(sock_get_timestamp);
2418 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2421 if (!sock_flag(sk, SOCK_TIMESTAMP))
2422 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2423 ts = ktime_to_timespec(sk->sk_stamp);
2424 if (ts.tv_sec == -1)
2426 if (ts.tv_sec == 0) {
2427 sk->sk_stamp = ktime_get_real();
2428 ts = ktime_to_timespec(sk->sk_stamp);
2430 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2432 EXPORT_SYMBOL(sock_get_timestampns);
2434 void sock_enable_timestamp(struct sock *sk, int flag)
2436 if (!sock_flag(sk, flag)) {
2437 unsigned long previous_flags = sk->sk_flags;
2439 sock_set_flag(sk, flag);
2441 * we just set one of the two flags which require net
2442 * time stamping, but time stamping might have been on
2443 * already because of the other one
2445 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2446 net_enable_timestamp();
2450 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2451 int level, int type)
2453 struct sock_exterr_skb *serr;
2454 struct sk_buff *skb, *skb2;
2458 skb = skb_dequeue(&sk->sk_error_queue);
2464 msg->msg_flags |= MSG_TRUNC;
2467 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2471 sock_recv_timestamp(msg, sk, skb);
2473 serr = SKB_EXT_ERR(skb);
2474 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2476 msg->msg_flags |= MSG_ERRQUEUE;
2479 /* Reset and regenerate socket error */
2480 spin_lock_bh(&sk->sk_error_queue.lock);
2482 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2483 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2484 spin_unlock_bh(&sk->sk_error_queue.lock);
2485 sk->sk_error_report(sk);
2487 spin_unlock_bh(&sk->sk_error_queue.lock);
2494 EXPORT_SYMBOL(sock_recv_errqueue);
2497 * Get a socket option on an socket.
2499 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2500 * asynchronous errors should be reported by getsockopt. We assume
2501 * this means if you specify SO_ERROR (otherwise whats the point of it).
2503 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2504 char __user *optval, int __user *optlen)
2506 struct sock *sk = sock->sk;
2508 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2510 EXPORT_SYMBOL(sock_common_getsockopt);
2512 #ifdef CONFIG_COMPAT
2513 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2514 char __user *optval, int __user *optlen)
2516 struct sock *sk = sock->sk;
2518 if (sk->sk_prot->compat_getsockopt != NULL)
2519 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2521 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2523 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2526 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2527 struct msghdr *msg, size_t size, int flags)
2529 struct sock *sk = sock->sk;
2533 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2534 flags & ~MSG_DONTWAIT, &addr_len);
2536 msg->msg_namelen = addr_len;
2539 EXPORT_SYMBOL(sock_common_recvmsg);
2542 * Set socket options on an inet socket.
2544 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2545 char __user *optval, unsigned int optlen)
2547 struct sock *sk = sock->sk;
2549 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2551 EXPORT_SYMBOL(sock_common_setsockopt);
2553 #ifdef CONFIG_COMPAT
2554 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2555 char __user *optval, unsigned int optlen)
2557 struct sock *sk = sock->sk;
2559 if (sk->sk_prot->compat_setsockopt != NULL)
2560 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2562 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2564 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2567 void sk_common_release(struct sock *sk)
2569 if (sk->sk_prot->destroy)
2570 sk->sk_prot->destroy(sk);
2573 * Observation: when sock_common_release is called, processes have
2574 * no access to socket. But net still has.
2575 * Step one, detach it from networking:
2577 * A. Remove from hash tables.
2580 sk->sk_prot->unhash(sk);
2583 * In this point socket cannot receive new packets, but it is possible
2584 * that some packets are in flight because some CPU runs receiver and
2585 * did hash table lookup before we unhashed socket. They will achieve
2586 * receive queue and will be purged by socket destructor.
2588 * Also we still have packets pending on receive queue and probably,
2589 * our own packets waiting in device queues. sock_destroy will drain
2590 * receive queue, but transmitted packets will delay socket destruction
2591 * until the last reference will be released.
2596 xfrm_sk_free_policy(sk);
2598 sk_refcnt_debug_release(sk);
2600 if (sk->sk_frag.page) {
2601 put_page(sk->sk_frag.page);
2602 sk->sk_frag.page = NULL;
2607 EXPORT_SYMBOL(sk_common_release);
2609 #ifdef CONFIG_PROC_FS
2610 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2612 int val[PROTO_INUSE_NR];
2615 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2617 #ifdef CONFIG_NET_NS
2618 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2620 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2622 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2624 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2626 int cpu, idx = prot->inuse_idx;
2629 for_each_possible_cpu(cpu)
2630 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2632 return res >= 0 ? res : 0;
2634 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2636 static int __net_init sock_inuse_init_net(struct net *net)
2638 net->core.inuse = alloc_percpu(struct prot_inuse);
2639 return net->core.inuse ? 0 : -ENOMEM;
2642 static void __net_exit sock_inuse_exit_net(struct net *net)
2644 free_percpu(net->core.inuse);
2647 static struct pernet_operations net_inuse_ops = {
2648 .init = sock_inuse_init_net,
2649 .exit = sock_inuse_exit_net,
2652 static __init int net_inuse_init(void)
2654 if (register_pernet_subsys(&net_inuse_ops))
2655 panic("Cannot initialize net inuse counters");
2660 core_initcall(net_inuse_init);
2662 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2664 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2666 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2668 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2670 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2672 int cpu, idx = prot->inuse_idx;
2675 for_each_possible_cpu(cpu)
2676 res += per_cpu(prot_inuse, cpu).val[idx];
2678 return res >= 0 ? res : 0;
2680 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2683 static void assign_proto_idx(struct proto *prot)
2685 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2687 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2688 pr_err("PROTO_INUSE_NR exhausted\n");
2692 set_bit(prot->inuse_idx, proto_inuse_idx);
2695 static void release_proto_idx(struct proto *prot)
2697 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2698 clear_bit(prot->inuse_idx, proto_inuse_idx);
2701 static inline void assign_proto_idx(struct proto *prot)
2705 static inline void release_proto_idx(struct proto *prot)
2710 int proto_register(struct proto *prot, int alloc_slab)
2713 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2714 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2717 if (prot->slab == NULL) {
2718 pr_crit("%s: Can't create sock SLAB cache!\n",
2723 if (prot->rsk_prot != NULL) {
2724 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2725 if (prot->rsk_prot->slab_name == NULL)
2726 goto out_free_sock_slab;
2728 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2729 prot->rsk_prot->obj_size, 0,
2730 SLAB_HWCACHE_ALIGN, NULL);
2732 if (prot->rsk_prot->slab == NULL) {
2733 pr_crit("%s: Can't create request sock SLAB cache!\n",
2735 goto out_free_request_sock_slab_name;
2739 if (prot->twsk_prot != NULL) {
2740 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2742 if (prot->twsk_prot->twsk_slab_name == NULL)
2743 goto out_free_request_sock_slab;
2745 prot->twsk_prot->twsk_slab =
2746 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2747 prot->twsk_prot->twsk_obj_size,
2749 SLAB_HWCACHE_ALIGN |
2752 if (prot->twsk_prot->twsk_slab == NULL)
2753 goto out_free_timewait_sock_slab_name;
2757 mutex_lock(&proto_list_mutex);
2758 list_add(&prot->node, &proto_list);
2759 assign_proto_idx(prot);
2760 mutex_unlock(&proto_list_mutex);
2763 out_free_timewait_sock_slab_name:
2764 kfree(prot->twsk_prot->twsk_slab_name);
2765 out_free_request_sock_slab:
2766 if (prot->rsk_prot && prot->rsk_prot->slab) {
2767 kmem_cache_destroy(prot->rsk_prot->slab);
2768 prot->rsk_prot->slab = NULL;
2770 out_free_request_sock_slab_name:
2772 kfree(prot->rsk_prot->slab_name);
2774 kmem_cache_destroy(prot->slab);
2779 EXPORT_SYMBOL(proto_register);
2781 void proto_unregister(struct proto *prot)
2783 mutex_lock(&proto_list_mutex);
2784 release_proto_idx(prot);
2785 list_del(&prot->node);
2786 mutex_unlock(&proto_list_mutex);
2788 if (prot->slab != NULL) {
2789 kmem_cache_destroy(prot->slab);
2793 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2794 kmem_cache_destroy(prot->rsk_prot->slab);
2795 kfree(prot->rsk_prot->slab_name);
2796 prot->rsk_prot->slab = NULL;
2799 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2800 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2801 kfree(prot->twsk_prot->twsk_slab_name);
2802 prot->twsk_prot->twsk_slab = NULL;
2805 EXPORT_SYMBOL(proto_unregister);
2807 #ifdef CONFIG_PROC_FS
2808 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2809 __acquires(proto_list_mutex)
2811 mutex_lock(&proto_list_mutex);
2812 return seq_list_start_head(&proto_list, *pos);
2815 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2817 return seq_list_next(v, &proto_list, pos);
2820 static void proto_seq_stop(struct seq_file *seq, void *v)
2821 __releases(proto_list_mutex)
2823 mutex_unlock(&proto_list_mutex);
2826 static char proto_method_implemented(const void *method)
2828 return method == NULL ? 'n' : 'y';
2830 static long sock_prot_memory_allocated(struct proto *proto)
2832 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2835 static char *sock_prot_memory_pressure(struct proto *proto)
2837 return proto->memory_pressure != NULL ?
2838 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2841 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2844 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2845 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2848 sock_prot_inuse_get(seq_file_net(seq), proto),
2849 sock_prot_memory_allocated(proto),
2850 sock_prot_memory_pressure(proto),
2852 proto->slab == NULL ? "no" : "yes",
2853 module_name(proto->owner),
2854 proto_method_implemented(proto->close),
2855 proto_method_implemented(proto->connect),
2856 proto_method_implemented(proto->disconnect),
2857 proto_method_implemented(proto->accept),
2858 proto_method_implemented(proto->ioctl),
2859 proto_method_implemented(proto->init),
2860 proto_method_implemented(proto->destroy),
2861 proto_method_implemented(proto->shutdown),
2862 proto_method_implemented(proto->setsockopt),
2863 proto_method_implemented(proto->getsockopt),
2864 proto_method_implemented(proto->sendmsg),
2865 proto_method_implemented(proto->recvmsg),
2866 proto_method_implemented(proto->sendpage),
2867 proto_method_implemented(proto->bind),
2868 proto_method_implemented(proto->backlog_rcv),
2869 proto_method_implemented(proto->hash),
2870 proto_method_implemented(proto->unhash),
2871 proto_method_implemented(proto->get_port),
2872 proto_method_implemented(proto->enter_memory_pressure));
2875 static int proto_seq_show(struct seq_file *seq, void *v)
2877 if (v == &proto_list)
2878 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2887 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2889 proto_seq_printf(seq, list_entry(v, struct proto, node));
2893 static const struct seq_operations proto_seq_ops = {
2894 .start = proto_seq_start,
2895 .next = proto_seq_next,
2896 .stop = proto_seq_stop,
2897 .show = proto_seq_show,
2900 static int proto_seq_open(struct inode *inode, struct file *file)
2902 return seq_open_net(inode, file, &proto_seq_ops,
2903 sizeof(struct seq_net_private));
2906 static const struct file_operations proto_seq_fops = {
2907 .owner = THIS_MODULE,
2908 .open = proto_seq_open,
2910 .llseek = seq_lseek,
2911 .release = seq_release_net,
2914 static __net_init int proto_init_net(struct net *net)
2916 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2922 static __net_exit void proto_exit_net(struct net *net)
2924 remove_proc_entry("protocols", net->proc_net);
2928 static __net_initdata struct pernet_operations proto_net_ops = {
2929 .init = proto_init_net,
2930 .exit = proto_exit_net,
2933 static int __init proto_init(void)
2935 return register_pernet_subsys(&proto_net_ops);
2938 subsys_initcall(proto_init);
2940 #endif /* PROC_FS */