2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <asm/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
139 #include <trace/events/sock.h>
145 #include <net/busy_poll.h>
147 static DEFINE_MUTEX(proto_list_mutex);
148 static LIST_HEAD(proto_list);
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
160 bool sk_ns_capable(const struct sock *sk,
161 struct user_namespace *user_ns, int cap)
163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
164 ns_capable(user_ns, cap);
166 EXPORT_SYMBOL(sk_ns_capable);
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
177 bool sk_capable(const struct sock *sk, int cap)
179 return sk_ns_capable(sk, &init_user_ns, cap);
181 EXPORT_SYMBOL(sk_capable);
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
192 bool sk_net_capable(const struct sock *sk, int cap)
194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
196 EXPORT_SYMBOL(sk_net_capable);
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family:
202 static struct lock_class_key af_family_keys[AF_MAX];
203 static struct lock_class_key af_family_slock_keys[AF_MAX];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
210 static const char *const af_family_key_strings[AF_MAX+1] = {
211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
226 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
227 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
228 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
229 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
230 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
231 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
232 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
233 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
234 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
235 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
236 "slock-27" , "slock-28" , "slock-AF_CAN" ,
237 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
238 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
239 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
240 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
242 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
243 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
244 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
245 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
246 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
247 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
248 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
249 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
250 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
251 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
252 "clock-27" , "clock-28" , "clock-AF_CAN" ,
253 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
254 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
255 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
256 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
260 * sk_callback_lock locking rules are per-address-family,
261 * so split the lock classes by using a per-AF key:
263 static struct lock_class_key af_callback_keys[AF_MAX];
265 /* Take into consideration the size of the struct sk_buff overhead in the
266 * determination of these values, since that is non-constant across
267 * platforms. This makes socket queueing behavior and performance
268 * not depend upon such differences.
270 #define _SK_MEM_PACKETS 256
271 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
272 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
273 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
275 /* Run time adjustable parameters. */
276 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
277 EXPORT_SYMBOL(sysctl_wmem_max);
278 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
279 EXPORT_SYMBOL(sysctl_rmem_max);
280 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
281 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
283 /* Maximal space eaten by iovec or ancillary data plus some space */
284 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
285 EXPORT_SYMBOL(sysctl_optmem_max);
287 int sysctl_tstamp_allow_data __read_mostly = 1;
289 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
290 EXPORT_SYMBOL_GPL(memalloc_socks);
293 * sk_set_memalloc - sets %SOCK_MEMALLOC
294 * @sk: socket to set it on
296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
297 * It's the responsibility of the admin to adjust min_free_kbytes
298 * to meet the requirements
300 void sk_set_memalloc(struct sock *sk)
302 sock_set_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation |= __GFP_MEMALLOC;
304 static_key_slow_inc(&memalloc_socks);
306 EXPORT_SYMBOL_GPL(sk_set_memalloc);
308 void sk_clear_memalloc(struct sock *sk)
310 sock_reset_flag(sk, SOCK_MEMALLOC);
311 sk->sk_allocation &= ~__GFP_MEMALLOC;
312 static_key_slow_dec(&memalloc_socks);
315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
316 * progress of swapping. SOCK_MEMALLOC may be cleared while
317 * it has rmem allocations due to the last swapfile being deactivated
318 * but there is a risk that the socket is unusable due to exceeding
319 * the rmem limits. Reclaim the reserves and obey rmem limits again.
323 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
325 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
328 unsigned long pflags = current->flags;
330 /* these should have been dropped before queueing */
331 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
333 current->flags |= PF_MEMALLOC;
334 ret = sk->sk_backlog_rcv(sk, skb);
335 tsk_restore_flags(current, pflags, PF_MEMALLOC);
339 EXPORT_SYMBOL(__sk_backlog_rcv);
341 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
345 if (optlen < sizeof(tv))
347 if (copy_from_user(&tv, optval, sizeof(tv)))
349 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
353 static int warned __read_mostly;
356 if (warned < 10 && net_ratelimit()) {
358 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
359 __func__, current->comm, task_pid_nr(current));
363 *timeo_p = MAX_SCHEDULE_TIMEOUT;
364 if (tv.tv_sec == 0 && tv.tv_usec == 0)
366 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
367 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
371 static void sock_warn_obsolete_bsdism(const char *name)
374 static char warncomm[TASK_COMM_LEN];
375 if (strcmp(warncomm, current->comm) && warned < 5) {
376 strcpy(warncomm, current->comm);
377 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
383 static bool sock_needs_netstamp(const struct sock *sk)
385 switch (sk->sk_family) {
394 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
396 if (sk->sk_flags & flags) {
397 sk->sk_flags &= ~flags;
398 if (sock_needs_netstamp(sk) &&
399 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
400 net_disable_timestamp();
405 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
409 struct sk_buff_head *list = &sk->sk_receive_queue;
411 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
412 atomic_inc(&sk->sk_drops);
413 trace_sock_rcvqueue_full(sk, skb);
417 err = sk_filter(sk, skb);
421 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
422 atomic_inc(&sk->sk_drops);
427 skb_set_owner_r(skb, sk);
429 /* we escape from rcu protected region, make sure we dont leak
434 spin_lock_irqsave(&list->lock, flags);
435 sock_skb_set_dropcount(sk, skb);
436 __skb_queue_tail(list, skb);
437 spin_unlock_irqrestore(&list->lock, flags);
439 if (!sock_flag(sk, SOCK_DEAD))
440 sk->sk_data_ready(sk);
443 EXPORT_SYMBOL(sock_queue_rcv_skb);
445 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
447 int rc = NET_RX_SUCCESS;
449 if (sk_filter(sk, skb))
450 goto discard_and_relse;
454 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
455 atomic_inc(&sk->sk_drops);
456 goto discard_and_relse;
459 bh_lock_sock_nested(sk);
462 if (!sock_owned_by_user(sk)) {
464 * trylock + unlock semantics:
466 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
468 rc = sk_backlog_rcv(sk, skb);
470 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
471 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
473 atomic_inc(&sk->sk_drops);
474 goto discard_and_relse;
485 EXPORT_SYMBOL(sk_receive_skb);
487 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
489 struct dst_entry *dst = __sk_dst_get(sk);
491 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
492 sk_tx_queue_clear(sk);
493 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
500 EXPORT_SYMBOL(__sk_dst_check);
502 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
504 struct dst_entry *dst = sk_dst_get(sk);
506 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
514 EXPORT_SYMBOL(sk_dst_check);
516 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
519 int ret = -ENOPROTOOPT;
520 #ifdef CONFIG_NETDEVICES
521 struct net *net = sock_net(sk);
522 char devname[IFNAMSIZ];
527 if (!ns_capable(net->user_ns, CAP_NET_RAW))
534 /* Bind this socket to a particular device like "eth0",
535 * as specified in the passed interface name. If the
536 * name is "" or the option length is zero the socket
539 if (optlen > IFNAMSIZ - 1)
540 optlen = IFNAMSIZ - 1;
541 memset(devname, 0, sizeof(devname));
544 if (copy_from_user(devname, optval, optlen))
548 if (devname[0] != '\0') {
549 struct net_device *dev;
552 dev = dev_get_by_name_rcu(net, devname);
554 index = dev->ifindex;
562 sk->sk_bound_dev_if = index;
574 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
575 int __user *optlen, int len)
577 int ret = -ENOPROTOOPT;
578 #ifdef CONFIG_NETDEVICES
579 struct net *net = sock_net(sk);
580 char devname[IFNAMSIZ];
582 if (sk->sk_bound_dev_if == 0) {
591 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
595 len = strlen(devname) + 1;
598 if (copy_to_user(optval, devname, len))
603 if (put_user(len, optlen))
614 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
617 sock_set_flag(sk, bit);
619 sock_reset_flag(sk, bit);
622 bool sk_mc_loop(struct sock *sk)
624 if (dev_recursion_level())
628 switch (sk->sk_family) {
630 return inet_sk(sk)->mc_loop;
631 #if IS_ENABLED(CONFIG_IPV6)
633 return inet6_sk(sk)->mc_loop;
639 EXPORT_SYMBOL(sk_mc_loop);
642 * This is meant for all protocols to use and covers goings on
643 * at the socket level. Everything here is generic.
646 int sock_setsockopt(struct socket *sock, int level, int optname,
647 char __user *optval, unsigned int optlen)
649 struct sock *sk = sock->sk;
656 * Options without arguments
659 if (optname == SO_BINDTODEVICE)
660 return sock_setbindtodevice(sk, optval, optlen);
662 if (optlen < sizeof(int))
665 if (get_user(val, (int __user *)optval))
668 valbool = val ? 1 : 0;
674 if (val && !capable(CAP_NET_ADMIN))
677 sock_valbool_flag(sk, SOCK_DBG, valbool);
680 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
683 sk->sk_reuseport = valbool;
692 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
695 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
698 /* Don't error on this BSD doesn't and if you think
699 * about it this is right. Otherwise apps have to
700 * play 'guess the biggest size' games. RCVBUF/SNDBUF
701 * are treated in BSD as hints
703 val = min_t(u32, val, sysctl_wmem_max);
705 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
706 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
707 /* Wake up sending tasks if we upped the value. */
708 sk->sk_write_space(sk);
712 if (!capable(CAP_NET_ADMIN)) {
719 /* Don't error on this BSD doesn't and if you think
720 * about it this is right. Otherwise apps have to
721 * play 'guess the biggest size' games. RCVBUF/SNDBUF
722 * are treated in BSD as hints
724 val = min_t(u32, val, sysctl_rmem_max);
726 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
728 * We double it on the way in to account for
729 * "struct sk_buff" etc. overhead. Applications
730 * assume that the SO_RCVBUF setting they make will
731 * allow that much actual data to be received on that
734 * Applications are unaware that "struct sk_buff" and
735 * other overheads allocate from the receive buffer
736 * during socket buffer allocation.
738 * And after considering the possible alternatives,
739 * returning the value we actually used in getsockopt
740 * is the most desirable behavior.
742 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
746 if (!capable(CAP_NET_ADMIN)) {
754 if (sk->sk_protocol == IPPROTO_TCP &&
755 sk->sk_type == SOCK_STREAM)
756 tcp_set_keepalive(sk, valbool);
758 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
762 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
766 sk->sk_no_check_tx = valbool;
770 if ((val >= 0 && val <= 6) ||
771 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
772 sk->sk_priority = val;
778 if (optlen < sizeof(ling)) {
779 ret = -EINVAL; /* 1003.1g */
782 if (copy_from_user(&ling, optval, sizeof(ling))) {
787 sock_reset_flag(sk, SOCK_LINGER);
789 #if (BITS_PER_LONG == 32)
790 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
791 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
794 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
795 sock_set_flag(sk, SOCK_LINGER);
800 sock_warn_obsolete_bsdism("setsockopt");
805 set_bit(SOCK_PASSCRED, &sock->flags);
807 clear_bit(SOCK_PASSCRED, &sock->flags);
813 if (optname == SO_TIMESTAMP)
814 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
816 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
817 sock_set_flag(sk, SOCK_RCVTSTAMP);
818 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
820 sock_reset_flag(sk, SOCK_RCVTSTAMP);
821 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
825 case SO_TIMESTAMPING:
826 if (val & ~SOF_TIMESTAMPING_MASK) {
831 if (val & SOF_TIMESTAMPING_OPT_ID &&
832 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
833 if (sk->sk_protocol == IPPROTO_TCP &&
834 sk->sk_type == SOCK_STREAM) {
835 if (sk->sk_state != TCP_ESTABLISHED) {
839 sk->sk_tskey = tcp_sk(sk)->snd_una;
844 sk->sk_tsflags = val;
845 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
846 sock_enable_timestamp(sk,
847 SOCK_TIMESTAMPING_RX_SOFTWARE);
849 sock_disable_timestamp(sk,
850 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
856 sk->sk_rcvlowat = val ? : 1;
860 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
864 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
867 case SO_ATTACH_FILTER:
869 if (optlen == sizeof(struct sock_fprog)) {
870 struct sock_fprog fprog;
873 if (copy_from_user(&fprog, optval, sizeof(fprog)))
876 ret = sk_attach_filter(&fprog, sk);
882 if (optlen == sizeof(u32)) {
886 if (copy_from_user(&ufd, optval, sizeof(ufd)))
889 ret = sk_attach_bpf(ufd, sk);
893 case SO_ATTACH_REUSEPORT_CBPF:
895 if (optlen == sizeof(struct sock_fprog)) {
896 struct sock_fprog fprog;
899 if (copy_from_user(&fprog, optval, sizeof(fprog)))
902 ret = sk_reuseport_attach_filter(&fprog, sk);
906 case SO_ATTACH_REUSEPORT_EBPF:
908 if (optlen == sizeof(u32)) {
912 if (copy_from_user(&ufd, optval, sizeof(ufd)))
915 ret = sk_reuseport_attach_bpf(ufd, sk);
919 case SO_DETACH_FILTER:
920 ret = sk_detach_filter(sk);
924 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
927 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
932 set_bit(SOCK_PASSSEC, &sock->flags);
934 clear_bit(SOCK_PASSSEC, &sock->flags);
937 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
944 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
948 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
952 if (sock->ops->set_peek_off)
953 ret = sock->ops->set_peek_off(sk, val);
959 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
962 case SO_SELECT_ERR_QUEUE:
963 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
966 #ifdef CONFIG_NET_RX_BUSY_POLL
968 /* allow unprivileged users to decrease the value */
969 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
975 sk->sk_ll_usec = val;
980 case SO_MAX_PACING_RATE:
981 sk->sk_max_pacing_rate = val;
982 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
983 sk->sk_max_pacing_rate);
986 case SO_INCOMING_CPU:
987 sk->sk_incoming_cpu = val;
992 dst_negative_advice(sk);
1001 EXPORT_SYMBOL(sock_setsockopt);
1004 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1005 struct ucred *ucred)
1007 ucred->pid = pid_vnr(pid);
1008 ucred->uid = ucred->gid = -1;
1010 struct user_namespace *current_ns = current_user_ns();
1012 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1013 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1017 int sock_getsockopt(struct socket *sock, int level, int optname,
1018 char __user *optval, int __user *optlen)
1020 struct sock *sk = sock->sk;
1028 int lv = sizeof(int);
1031 if (get_user(len, optlen))
1036 memset(&v, 0, sizeof(v));
1040 v.val = sock_flag(sk, SOCK_DBG);
1044 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1048 v.val = sock_flag(sk, SOCK_BROADCAST);
1052 v.val = sk->sk_sndbuf;
1056 v.val = sk->sk_rcvbuf;
1060 v.val = sk->sk_reuse;
1064 v.val = sk->sk_reuseport;
1068 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1072 v.val = sk->sk_type;
1076 v.val = sk->sk_protocol;
1080 v.val = sk->sk_family;
1084 v.val = -sock_error(sk);
1086 v.val = xchg(&sk->sk_err_soft, 0);
1090 v.val = sock_flag(sk, SOCK_URGINLINE);
1094 v.val = sk->sk_no_check_tx;
1098 v.val = sk->sk_priority;
1102 lv = sizeof(v.ling);
1103 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1104 v.ling.l_linger = sk->sk_lingertime / HZ;
1108 sock_warn_obsolete_bsdism("getsockopt");
1112 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1113 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1116 case SO_TIMESTAMPNS:
1117 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1120 case SO_TIMESTAMPING:
1121 v.val = sk->sk_tsflags;
1125 lv = sizeof(struct timeval);
1126 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1130 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1131 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1136 lv = sizeof(struct timeval);
1137 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1141 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1142 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1147 v.val = sk->sk_rcvlowat;
1155 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1160 struct ucred peercred;
1161 if (len > sizeof(peercred))
1162 len = sizeof(peercred);
1163 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1164 if (copy_to_user(optval, &peercred, len))
1173 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1177 if (copy_to_user(optval, address, len))
1182 /* Dubious BSD thing... Probably nobody even uses it, but
1183 * the UNIX standard wants it for whatever reason... -DaveM
1186 v.val = sk->sk_state == TCP_LISTEN;
1190 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1194 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1197 v.val = sk->sk_mark;
1201 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1204 case SO_WIFI_STATUS:
1205 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1209 if (!sock->ops->set_peek_off)
1212 v.val = sk->sk_peek_off;
1215 v.val = sock_flag(sk, SOCK_NOFCS);
1218 case SO_BINDTODEVICE:
1219 return sock_getbindtodevice(sk, optval, optlen, len);
1222 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1228 case SO_LOCK_FILTER:
1229 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1232 case SO_BPF_EXTENSIONS:
1233 v.val = bpf_tell_extensions();
1236 case SO_SELECT_ERR_QUEUE:
1237 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1240 #ifdef CONFIG_NET_RX_BUSY_POLL
1242 v.val = sk->sk_ll_usec;
1246 case SO_MAX_PACING_RATE:
1247 v.val = sk->sk_max_pacing_rate;
1250 case SO_INCOMING_CPU:
1251 v.val = sk->sk_incoming_cpu;
1255 /* We implement the SO_SNDLOWAT etc to not be settable
1258 return -ENOPROTOOPT;
1263 if (copy_to_user(optval, &v, len))
1266 if (put_user(len, optlen))
1272 * Initialize an sk_lock.
1274 * (We also register the sk_lock with the lock validator.)
1276 static inline void sock_lock_init(struct sock *sk)
1278 sock_lock_init_class_and_name(sk,
1279 af_family_slock_key_strings[sk->sk_family],
1280 af_family_slock_keys + sk->sk_family,
1281 af_family_key_strings[sk->sk_family],
1282 af_family_keys + sk->sk_family);
1286 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1287 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1288 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1290 static void sock_copy(struct sock *nsk, const struct sock *osk)
1292 #ifdef CONFIG_SECURITY_NETWORK
1293 void *sptr = nsk->sk_security;
1295 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1297 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1298 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1300 #ifdef CONFIG_SECURITY_NETWORK
1301 nsk->sk_security = sptr;
1302 security_sk_clone(osk, nsk);
1306 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1308 unsigned long nulls1, nulls2;
1310 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1311 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1312 if (nulls1 > nulls2)
1313 swap(nulls1, nulls2);
1316 memset((char *)sk, 0, nulls1);
1317 memset((char *)sk + nulls1 + sizeof(void *), 0,
1318 nulls2 - nulls1 - sizeof(void *));
1319 memset((char *)sk + nulls2 + sizeof(void *), 0,
1320 size - nulls2 - sizeof(void *));
1322 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1324 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1328 struct kmem_cache *slab;
1332 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1335 if (priority & __GFP_ZERO) {
1337 prot->clear_sk(sk, prot->obj_size);
1339 sk_prot_clear_nulls(sk, prot->obj_size);
1342 sk = kmalloc(prot->obj_size, priority);
1345 kmemcheck_annotate_bitfield(sk, flags);
1347 if (security_sk_alloc(sk, family, priority))
1350 if (!try_module_get(prot->owner))
1352 sk_tx_queue_clear(sk);
1353 cgroup_sk_alloc(&sk->sk_cgrp_data);
1359 security_sk_free(sk);
1362 kmem_cache_free(slab, sk);
1368 static void sk_prot_free(struct proto *prot, struct sock *sk)
1370 struct kmem_cache *slab;
1371 struct module *owner;
1373 owner = prot->owner;
1376 cgroup_sk_free(&sk->sk_cgrp_data);
1377 security_sk_free(sk);
1379 kmem_cache_free(slab, sk);
1386 * sk_alloc - All socket objects are allocated here
1387 * @net: the applicable net namespace
1388 * @family: protocol family
1389 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1390 * @prot: struct proto associated with this new sock instance
1391 * @kern: is this to be a kernel socket?
1393 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1394 struct proto *prot, int kern)
1398 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1400 sk->sk_family = family;
1402 * See comment in struct sock definition to understand
1403 * why we need sk_prot_creator -acme
1405 sk->sk_prot = sk->sk_prot_creator = prot;
1407 sk->sk_net_refcnt = kern ? 0 : 1;
1408 if (likely(sk->sk_net_refcnt))
1410 sock_net_set(sk, net);
1411 atomic_set(&sk->sk_wmem_alloc, 1);
1413 sock_update_classid(&sk->sk_cgrp_data);
1414 sock_update_netprioidx(&sk->sk_cgrp_data);
1419 EXPORT_SYMBOL(sk_alloc);
1421 void sk_destruct(struct sock *sk)
1423 struct sk_filter *filter;
1425 if (sk->sk_destruct)
1426 sk->sk_destruct(sk);
1428 filter = rcu_dereference_check(sk->sk_filter,
1429 atomic_read(&sk->sk_wmem_alloc) == 0);
1431 sk_filter_uncharge(sk, filter);
1432 RCU_INIT_POINTER(sk->sk_filter, NULL);
1434 if (rcu_access_pointer(sk->sk_reuseport_cb))
1435 reuseport_detach_sock(sk);
1437 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1439 if (atomic_read(&sk->sk_omem_alloc))
1440 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1441 __func__, atomic_read(&sk->sk_omem_alloc));
1443 if (sk->sk_peer_cred)
1444 put_cred(sk->sk_peer_cred);
1445 put_pid(sk->sk_peer_pid);
1446 if (likely(sk->sk_net_refcnt))
1447 put_net(sock_net(sk));
1448 sk_prot_free(sk->sk_prot_creator, sk);
1451 static void __sk_free(struct sock *sk)
1453 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1454 sock_diag_broadcast_destroy(sk);
1459 void sk_free(struct sock *sk)
1462 * We subtract one from sk_wmem_alloc and can know if
1463 * some packets are still in some tx queue.
1464 * If not null, sock_wfree() will call __sk_free(sk) later
1466 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1469 EXPORT_SYMBOL(sk_free);
1472 * sk_clone_lock - clone a socket, and lock its clone
1473 * @sk: the socket to clone
1474 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1476 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1478 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1481 bool is_charged = true;
1483 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1484 if (newsk != NULL) {
1485 struct sk_filter *filter;
1487 sock_copy(newsk, sk);
1490 if (likely(newsk->sk_net_refcnt))
1491 get_net(sock_net(newsk));
1492 sk_node_init(&newsk->sk_node);
1493 sock_lock_init(newsk);
1494 bh_lock_sock(newsk);
1495 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1496 newsk->sk_backlog.len = 0;
1498 atomic_set(&newsk->sk_rmem_alloc, 0);
1500 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1502 atomic_set(&newsk->sk_wmem_alloc, 1);
1503 atomic_set(&newsk->sk_omem_alloc, 0);
1504 skb_queue_head_init(&newsk->sk_receive_queue);
1505 skb_queue_head_init(&newsk->sk_write_queue);
1507 rwlock_init(&newsk->sk_callback_lock);
1508 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1509 af_callback_keys + newsk->sk_family,
1510 af_family_clock_key_strings[newsk->sk_family]);
1512 newsk->sk_dst_cache = NULL;
1513 newsk->sk_wmem_queued = 0;
1514 newsk->sk_forward_alloc = 0;
1515 newsk->sk_send_head = NULL;
1516 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1518 sock_reset_flag(newsk, SOCK_DONE);
1519 skb_queue_head_init(&newsk->sk_error_queue);
1521 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1523 /* though it's an empty new sock, the charging may fail
1524 * if sysctl_optmem_max was changed between creation of
1525 * original socket and cloning
1527 is_charged = sk_filter_charge(newsk, filter);
1529 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1530 /* It is still raw copy of parent, so invalidate
1531 * destructor and make plain sk_free() */
1532 newsk->sk_destruct = NULL;
1533 bh_unlock_sock(newsk);
1538 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1541 newsk->sk_priority = 0;
1542 newsk->sk_incoming_cpu = raw_smp_processor_id();
1543 atomic64_set(&newsk->sk_cookie, 0);
1545 * Before updating sk_refcnt, we must commit prior changes to memory
1546 * (Documentation/RCU/rculist_nulls.txt for details)
1549 atomic_set(&newsk->sk_refcnt, 2);
1552 * Increment the counter in the same struct proto as the master
1553 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1554 * is the same as sk->sk_prot->socks, as this field was copied
1557 * This _changes_ the previous behaviour, where
1558 * tcp_create_openreq_child always was incrementing the
1559 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1560 * to be taken into account in all callers. -acme
1562 sk_refcnt_debug_inc(newsk);
1563 sk_set_socket(newsk, NULL);
1564 newsk->sk_wq = NULL;
1566 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1567 sock_update_memcg(newsk);
1569 if (newsk->sk_prot->sockets_allocated)
1570 sk_sockets_allocated_inc(newsk);
1572 if (sock_needs_netstamp(sk) &&
1573 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1574 net_enable_timestamp();
1579 EXPORT_SYMBOL_GPL(sk_clone_lock);
1581 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1585 sk_dst_set(sk, dst);
1586 sk->sk_route_caps = dst->dev->features;
1587 if (sk->sk_route_caps & NETIF_F_GSO)
1588 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1589 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1590 if (sk_can_gso(sk)) {
1591 if (dst->header_len) {
1592 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1594 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1595 sk->sk_gso_max_size = dst->dev->gso_max_size;
1596 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1599 sk->sk_gso_max_segs = max_segs;
1601 EXPORT_SYMBOL_GPL(sk_setup_caps);
1604 * Simple resource managers for sockets.
1609 * Write buffer destructor automatically called from kfree_skb.
1611 void sock_wfree(struct sk_buff *skb)
1613 struct sock *sk = skb->sk;
1614 unsigned int len = skb->truesize;
1616 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1618 * Keep a reference on sk_wmem_alloc, this will be released
1619 * after sk_write_space() call
1621 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1622 sk->sk_write_space(sk);
1626 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1627 * could not do because of in-flight packets
1629 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1632 EXPORT_SYMBOL(sock_wfree);
1634 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1639 if (unlikely(!sk_fullsock(sk))) {
1640 skb->destructor = sock_edemux;
1645 skb->destructor = sock_wfree;
1646 skb_set_hash_from_sk(skb, sk);
1648 * We used to take a refcount on sk, but following operation
1649 * is enough to guarantee sk_free() wont free this sock until
1650 * all in-flight packets are completed
1652 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1654 EXPORT_SYMBOL(skb_set_owner_w);
1656 void skb_orphan_partial(struct sk_buff *skb)
1658 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1659 * so we do not completely orphan skb, but transfert all
1660 * accounted bytes but one, to avoid unexpected reorders.
1662 if (skb->destructor == sock_wfree
1664 || skb->destructor == tcp_wfree
1667 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1673 EXPORT_SYMBOL(skb_orphan_partial);
1676 * Read buffer destructor automatically called from kfree_skb.
1678 void sock_rfree(struct sk_buff *skb)
1680 struct sock *sk = skb->sk;
1681 unsigned int len = skb->truesize;
1683 atomic_sub(len, &sk->sk_rmem_alloc);
1684 sk_mem_uncharge(sk, len);
1686 EXPORT_SYMBOL(sock_rfree);
1689 * Buffer destructor for skbs that are not used directly in read or write
1690 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1692 void sock_efree(struct sk_buff *skb)
1696 EXPORT_SYMBOL(sock_efree);
1698 kuid_t sock_i_uid(struct sock *sk)
1702 read_lock_bh(&sk->sk_callback_lock);
1703 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1704 read_unlock_bh(&sk->sk_callback_lock);
1707 EXPORT_SYMBOL(sock_i_uid);
1709 unsigned long sock_i_ino(struct sock *sk)
1713 read_lock_bh(&sk->sk_callback_lock);
1714 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1715 read_unlock_bh(&sk->sk_callback_lock);
1718 EXPORT_SYMBOL(sock_i_ino);
1721 * Allocate a skb from the socket's send buffer.
1723 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1726 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1727 struct sk_buff *skb = alloc_skb(size, priority);
1729 skb_set_owner_w(skb, sk);
1735 EXPORT_SYMBOL(sock_wmalloc);
1738 * Allocate a memory block from the socket's option memory buffer.
1740 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1742 if ((unsigned int)size <= sysctl_optmem_max &&
1743 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1745 /* First do the add, to avoid the race if kmalloc
1748 atomic_add(size, &sk->sk_omem_alloc);
1749 mem = kmalloc(size, priority);
1752 atomic_sub(size, &sk->sk_omem_alloc);
1756 EXPORT_SYMBOL(sock_kmalloc);
1758 /* Free an option memory block. Note, we actually want the inline
1759 * here as this allows gcc to detect the nullify and fold away the
1760 * condition entirely.
1762 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1765 if (WARN_ON_ONCE(!mem))
1771 atomic_sub(size, &sk->sk_omem_alloc);
1774 void sock_kfree_s(struct sock *sk, void *mem, int size)
1776 __sock_kfree_s(sk, mem, size, false);
1778 EXPORT_SYMBOL(sock_kfree_s);
1780 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1782 __sock_kfree_s(sk, mem, size, true);
1784 EXPORT_SYMBOL(sock_kzfree_s);
1786 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1787 I think, these locks should be removed for datagram sockets.
1789 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1793 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1797 if (signal_pending(current))
1799 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1800 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1801 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1803 if (sk->sk_shutdown & SEND_SHUTDOWN)
1807 timeo = schedule_timeout(timeo);
1809 finish_wait(sk_sleep(sk), &wait);
1815 * Generic send/receive buffer handlers
1818 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1819 unsigned long data_len, int noblock,
1820 int *errcode, int max_page_order)
1822 struct sk_buff *skb;
1826 timeo = sock_sndtimeo(sk, noblock);
1828 err = sock_error(sk);
1833 if (sk->sk_shutdown & SEND_SHUTDOWN)
1836 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1839 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1840 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1844 if (signal_pending(current))
1846 timeo = sock_wait_for_wmem(sk, timeo);
1848 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1849 errcode, sk->sk_allocation);
1851 skb_set_owner_w(skb, sk);
1855 err = sock_intr_errno(timeo);
1860 EXPORT_SYMBOL(sock_alloc_send_pskb);
1862 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1863 int noblock, int *errcode)
1865 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1867 EXPORT_SYMBOL(sock_alloc_send_skb);
1869 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1870 struct sockcm_cookie *sockc)
1872 struct cmsghdr *cmsg;
1874 for_each_cmsghdr(cmsg, msg) {
1875 if (!CMSG_OK(msg, cmsg))
1877 if (cmsg->cmsg_level != SOL_SOCKET)
1879 switch (cmsg->cmsg_type) {
1881 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1883 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1885 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1893 EXPORT_SYMBOL(sock_cmsg_send);
1895 /* On 32bit arches, an skb frag is limited to 2^15 */
1896 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1899 * skb_page_frag_refill - check that a page_frag contains enough room
1900 * @sz: minimum size of the fragment we want to get
1901 * @pfrag: pointer to page_frag
1902 * @gfp: priority for memory allocation
1904 * Note: While this allocator tries to use high order pages, there is
1905 * no guarantee that allocations succeed. Therefore, @sz MUST be
1906 * less or equal than PAGE_SIZE.
1908 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1911 if (page_ref_count(pfrag->page) == 1) {
1915 if (pfrag->offset + sz <= pfrag->size)
1917 put_page(pfrag->page);
1921 if (SKB_FRAG_PAGE_ORDER) {
1922 /* Avoid direct reclaim but allow kswapd to wake */
1923 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1924 __GFP_COMP | __GFP_NOWARN |
1926 SKB_FRAG_PAGE_ORDER);
1927 if (likely(pfrag->page)) {
1928 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1932 pfrag->page = alloc_page(gfp);
1933 if (likely(pfrag->page)) {
1934 pfrag->size = PAGE_SIZE;
1939 EXPORT_SYMBOL(skb_page_frag_refill);
1941 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1943 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1946 sk_enter_memory_pressure(sk);
1947 sk_stream_moderate_sndbuf(sk);
1950 EXPORT_SYMBOL(sk_page_frag_refill);
1952 static void __lock_sock(struct sock *sk)
1953 __releases(&sk->sk_lock.slock)
1954 __acquires(&sk->sk_lock.slock)
1959 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1960 TASK_UNINTERRUPTIBLE);
1961 spin_unlock_bh(&sk->sk_lock.slock);
1963 spin_lock_bh(&sk->sk_lock.slock);
1964 if (!sock_owned_by_user(sk))
1967 finish_wait(&sk->sk_lock.wq, &wait);
1970 static void __release_sock(struct sock *sk)
1971 __releases(&sk->sk_lock.slock)
1972 __acquires(&sk->sk_lock.slock)
1974 struct sk_buff *skb = sk->sk_backlog.head;
1977 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1981 struct sk_buff *next = skb->next;
1984 WARN_ON_ONCE(skb_dst_is_noref(skb));
1986 sk_backlog_rcv(sk, skb);
1989 * We are in process context here with softirqs
1990 * disabled, use cond_resched_softirq() to preempt.
1991 * This is safe to do because we've taken the backlog
1994 cond_resched_softirq();
1997 } while (skb != NULL);
2000 } while ((skb = sk->sk_backlog.head) != NULL);
2003 * Doing the zeroing here guarantee we can not loop forever
2004 * while a wild producer attempts to flood us.
2006 sk->sk_backlog.len = 0;
2010 * sk_wait_data - wait for data to arrive at sk_receive_queue
2011 * @sk: sock to wait on
2012 * @timeo: for how long
2013 * @skb: last skb seen on sk_receive_queue
2015 * Now socket state including sk->sk_err is changed only under lock,
2016 * hence we may omit checks after joining wait queue.
2017 * We check receive queue before schedule() only as optimization;
2018 * it is very likely that release_sock() added new data.
2020 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2025 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2026 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2027 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2028 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2029 finish_wait(sk_sleep(sk), &wait);
2032 EXPORT_SYMBOL(sk_wait_data);
2035 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2037 * @size: memory size to allocate
2038 * @kind: allocation type
2040 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2041 * rmem allocation. This function assumes that protocols which have
2042 * memory_pressure use sk_wmem_queued as write buffer accounting.
2044 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2046 struct proto *prot = sk->sk_prot;
2047 int amt = sk_mem_pages(size);
2050 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2052 allocated = sk_memory_allocated_add(sk, amt);
2054 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2055 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2056 goto suppress_allocation;
2059 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2060 sk_leave_memory_pressure(sk);
2064 /* Under pressure. */
2065 if (allocated > sk_prot_mem_limits(sk, 1))
2066 sk_enter_memory_pressure(sk);
2068 /* Over hard limit. */
2069 if (allocated > sk_prot_mem_limits(sk, 2))
2070 goto suppress_allocation;
2072 /* guarantee minimum buffer size under pressure */
2073 if (kind == SK_MEM_RECV) {
2074 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2077 } else { /* SK_MEM_SEND */
2078 if (sk->sk_type == SOCK_STREAM) {
2079 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2081 } else if (atomic_read(&sk->sk_wmem_alloc) <
2082 prot->sysctl_wmem[0])
2086 if (sk_has_memory_pressure(sk)) {
2089 if (!sk_under_memory_pressure(sk))
2091 alloc = sk_sockets_allocated_read_positive(sk);
2092 if (sk_prot_mem_limits(sk, 2) > alloc *
2093 sk_mem_pages(sk->sk_wmem_queued +
2094 atomic_read(&sk->sk_rmem_alloc) +
2095 sk->sk_forward_alloc))
2099 suppress_allocation:
2101 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2102 sk_stream_moderate_sndbuf(sk);
2104 /* Fail only if socket is _under_ its sndbuf.
2105 * In this case we cannot block, so that we have to fail.
2107 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2111 trace_sock_exceed_buf_limit(sk, prot, allocated);
2113 /* Alas. Undo changes. */
2114 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2116 sk_memory_allocated_sub(sk, amt);
2118 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2119 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2123 EXPORT_SYMBOL(__sk_mem_schedule);
2126 * __sk_mem_reclaim - reclaim memory_allocated
2128 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2130 void __sk_mem_reclaim(struct sock *sk, int amount)
2132 amount >>= SK_MEM_QUANTUM_SHIFT;
2133 sk_memory_allocated_sub(sk, amount);
2134 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2136 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2137 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2139 if (sk_under_memory_pressure(sk) &&
2140 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2141 sk_leave_memory_pressure(sk);
2143 EXPORT_SYMBOL(__sk_mem_reclaim);
2147 * Set of default routines for initialising struct proto_ops when
2148 * the protocol does not support a particular function. In certain
2149 * cases where it makes no sense for a protocol to have a "do nothing"
2150 * function, some default processing is provided.
2153 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2157 EXPORT_SYMBOL(sock_no_bind);
2159 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2164 EXPORT_SYMBOL(sock_no_connect);
2166 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2170 EXPORT_SYMBOL(sock_no_socketpair);
2172 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2176 EXPORT_SYMBOL(sock_no_accept);
2178 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2183 EXPORT_SYMBOL(sock_no_getname);
2185 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2189 EXPORT_SYMBOL(sock_no_poll);
2191 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2195 EXPORT_SYMBOL(sock_no_ioctl);
2197 int sock_no_listen(struct socket *sock, int backlog)
2201 EXPORT_SYMBOL(sock_no_listen);
2203 int sock_no_shutdown(struct socket *sock, int how)
2207 EXPORT_SYMBOL(sock_no_shutdown);
2209 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2210 char __user *optval, unsigned int optlen)
2214 EXPORT_SYMBOL(sock_no_setsockopt);
2216 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2217 char __user *optval, int __user *optlen)
2221 EXPORT_SYMBOL(sock_no_getsockopt);
2223 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2227 EXPORT_SYMBOL(sock_no_sendmsg);
2229 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2234 EXPORT_SYMBOL(sock_no_recvmsg);
2236 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2238 /* Mirror missing mmap method error code */
2241 EXPORT_SYMBOL(sock_no_mmap);
2243 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2246 struct msghdr msg = {.msg_flags = flags};
2248 char *kaddr = kmap(page);
2249 iov.iov_base = kaddr + offset;
2251 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2255 EXPORT_SYMBOL(sock_no_sendpage);
2258 * Default Socket Callbacks
2261 static void sock_def_wakeup(struct sock *sk)
2263 struct socket_wq *wq;
2266 wq = rcu_dereference(sk->sk_wq);
2267 if (skwq_has_sleeper(wq))
2268 wake_up_interruptible_all(&wq->wait);
2272 static void sock_def_error_report(struct sock *sk)
2274 struct socket_wq *wq;
2277 wq = rcu_dereference(sk->sk_wq);
2278 if (skwq_has_sleeper(wq))
2279 wake_up_interruptible_poll(&wq->wait, POLLERR);
2280 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2284 static void sock_def_readable(struct sock *sk)
2286 struct socket_wq *wq;
2289 wq = rcu_dereference(sk->sk_wq);
2290 if (skwq_has_sleeper(wq))
2291 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2292 POLLRDNORM | POLLRDBAND);
2293 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2297 static void sock_def_write_space(struct sock *sk)
2299 struct socket_wq *wq;
2303 /* Do not wake up a writer until he can make "significant"
2306 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2307 wq = rcu_dereference(sk->sk_wq);
2308 if (skwq_has_sleeper(wq))
2309 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2310 POLLWRNORM | POLLWRBAND);
2312 /* Should agree with poll, otherwise some programs break */
2313 if (sock_writeable(sk))
2314 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2320 static void sock_def_destruct(struct sock *sk)
2324 void sk_send_sigurg(struct sock *sk)
2326 if (sk->sk_socket && sk->sk_socket->file)
2327 if (send_sigurg(&sk->sk_socket->file->f_owner))
2328 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2330 EXPORT_SYMBOL(sk_send_sigurg);
2332 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2333 unsigned long expires)
2335 if (!mod_timer(timer, expires))
2338 EXPORT_SYMBOL(sk_reset_timer);
2340 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2342 if (del_timer(timer))
2345 EXPORT_SYMBOL(sk_stop_timer);
2347 void sock_init_data(struct socket *sock, struct sock *sk)
2349 skb_queue_head_init(&sk->sk_receive_queue);
2350 skb_queue_head_init(&sk->sk_write_queue);
2351 skb_queue_head_init(&sk->sk_error_queue);
2353 sk->sk_send_head = NULL;
2355 init_timer(&sk->sk_timer);
2357 sk->sk_allocation = GFP_KERNEL;
2358 sk->sk_rcvbuf = sysctl_rmem_default;
2359 sk->sk_sndbuf = sysctl_wmem_default;
2360 sk->sk_state = TCP_CLOSE;
2361 sk_set_socket(sk, sock);
2363 sock_set_flag(sk, SOCK_ZAPPED);
2366 sk->sk_type = sock->type;
2367 sk->sk_wq = sock->wq;
2372 rwlock_init(&sk->sk_callback_lock);
2373 lockdep_set_class_and_name(&sk->sk_callback_lock,
2374 af_callback_keys + sk->sk_family,
2375 af_family_clock_key_strings[sk->sk_family]);
2377 sk->sk_state_change = sock_def_wakeup;
2378 sk->sk_data_ready = sock_def_readable;
2379 sk->sk_write_space = sock_def_write_space;
2380 sk->sk_error_report = sock_def_error_report;
2381 sk->sk_destruct = sock_def_destruct;
2383 sk->sk_frag.page = NULL;
2384 sk->sk_frag.offset = 0;
2385 sk->sk_peek_off = -1;
2387 sk->sk_peer_pid = NULL;
2388 sk->sk_peer_cred = NULL;
2389 sk->sk_write_pending = 0;
2390 sk->sk_rcvlowat = 1;
2391 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2392 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2394 sk->sk_stamp = ktime_set(-1L, 0);
2396 #ifdef CONFIG_NET_RX_BUSY_POLL
2398 sk->sk_ll_usec = sysctl_net_busy_read;
2401 sk->sk_max_pacing_rate = ~0U;
2402 sk->sk_pacing_rate = ~0U;
2403 sk->sk_incoming_cpu = -1;
2405 * Before updating sk_refcnt, we must commit prior changes to memory
2406 * (Documentation/RCU/rculist_nulls.txt for details)
2409 atomic_set(&sk->sk_refcnt, 1);
2410 atomic_set(&sk->sk_drops, 0);
2412 EXPORT_SYMBOL(sock_init_data);
2414 void lock_sock_nested(struct sock *sk, int subclass)
2417 spin_lock_bh(&sk->sk_lock.slock);
2418 if (sk->sk_lock.owned)
2420 sk->sk_lock.owned = 1;
2421 spin_unlock(&sk->sk_lock.slock);
2423 * The sk_lock has mutex_lock() semantics here:
2425 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2428 EXPORT_SYMBOL(lock_sock_nested);
2430 void release_sock(struct sock *sk)
2433 * The sk_lock has mutex_unlock() semantics:
2435 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2437 spin_lock_bh(&sk->sk_lock.slock);
2438 if (sk->sk_backlog.tail)
2441 /* Warning : release_cb() might need to release sk ownership,
2442 * ie call sock_release_ownership(sk) before us.
2444 if (sk->sk_prot->release_cb)
2445 sk->sk_prot->release_cb(sk);
2447 sock_release_ownership(sk);
2448 if (waitqueue_active(&sk->sk_lock.wq))
2449 wake_up(&sk->sk_lock.wq);
2450 spin_unlock_bh(&sk->sk_lock.slock);
2452 EXPORT_SYMBOL(release_sock);
2455 * lock_sock_fast - fast version of lock_sock
2458 * This version should be used for very small section, where process wont block
2459 * return false if fast path is taken
2460 * sk_lock.slock locked, owned = 0, BH disabled
2461 * return true if slow path is taken
2462 * sk_lock.slock unlocked, owned = 1, BH enabled
2464 bool lock_sock_fast(struct sock *sk)
2467 spin_lock_bh(&sk->sk_lock.slock);
2469 if (!sk->sk_lock.owned)
2471 * Note : We must disable BH
2476 sk->sk_lock.owned = 1;
2477 spin_unlock(&sk->sk_lock.slock);
2479 * The sk_lock has mutex_lock() semantics here:
2481 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2485 EXPORT_SYMBOL(lock_sock_fast);
2487 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2490 if (!sock_flag(sk, SOCK_TIMESTAMP))
2491 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2492 tv = ktime_to_timeval(sk->sk_stamp);
2493 if (tv.tv_sec == -1)
2495 if (tv.tv_sec == 0) {
2496 sk->sk_stamp = ktime_get_real();
2497 tv = ktime_to_timeval(sk->sk_stamp);
2499 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2501 EXPORT_SYMBOL(sock_get_timestamp);
2503 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2506 if (!sock_flag(sk, SOCK_TIMESTAMP))
2507 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2508 ts = ktime_to_timespec(sk->sk_stamp);
2509 if (ts.tv_sec == -1)
2511 if (ts.tv_sec == 0) {
2512 sk->sk_stamp = ktime_get_real();
2513 ts = ktime_to_timespec(sk->sk_stamp);
2515 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2517 EXPORT_SYMBOL(sock_get_timestampns);
2519 void sock_enable_timestamp(struct sock *sk, int flag)
2521 if (!sock_flag(sk, flag)) {
2522 unsigned long previous_flags = sk->sk_flags;
2524 sock_set_flag(sk, flag);
2526 * we just set one of the two flags which require net
2527 * time stamping, but time stamping might have been on
2528 * already because of the other one
2530 if (sock_needs_netstamp(sk) &&
2531 !(previous_flags & SK_FLAGS_TIMESTAMP))
2532 net_enable_timestamp();
2536 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2537 int level, int type)
2539 struct sock_exterr_skb *serr;
2540 struct sk_buff *skb;
2544 skb = sock_dequeue_err_skb(sk);
2550 msg->msg_flags |= MSG_TRUNC;
2553 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2557 sock_recv_timestamp(msg, sk, skb);
2559 serr = SKB_EXT_ERR(skb);
2560 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2562 msg->msg_flags |= MSG_ERRQUEUE;
2570 EXPORT_SYMBOL(sock_recv_errqueue);
2573 * Get a socket option on an socket.
2575 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2576 * asynchronous errors should be reported by getsockopt. We assume
2577 * this means if you specify SO_ERROR (otherwise whats the point of it).
2579 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2580 char __user *optval, int __user *optlen)
2582 struct sock *sk = sock->sk;
2584 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2586 EXPORT_SYMBOL(sock_common_getsockopt);
2588 #ifdef CONFIG_COMPAT
2589 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2590 char __user *optval, int __user *optlen)
2592 struct sock *sk = sock->sk;
2594 if (sk->sk_prot->compat_getsockopt != NULL)
2595 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2597 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2599 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2602 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2605 struct sock *sk = sock->sk;
2609 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2610 flags & ~MSG_DONTWAIT, &addr_len);
2612 msg->msg_namelen = addr_len;
2615 EXPORT_SYMBOL(sock_common_recvmsg);
2618 * Set socket options on an inet socket.
2620 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2621 char __user *optval, unsigned int optlen)
2623 struct sock *sk = sock->sk;
2625 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2627 EXPORT_SYMBOL(sock_common_setsockopt);
2629 #ifdef CONFIG_COMPAT
2630 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2631 char __user *optval, unsigned int optlen)
2633 struct sock *sk = sock->sk;
2635 if (sk->sk_prot->compat_setsockopt != NULL)
2636 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2638 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2640 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2643 void sk_common_release(struct sock *sk)
2645 if (sk->sk_prot->destroy)
2646 sk->sk_prot->destroy(sk);
2649 * Observation: when sock_common_release is called, processes have
2650 * no access to socket. But net still has.
2651 * Step one, detach it from networking:
2653 * A. Remove from hash tables.
2656 sk->sk_prot->unhash(sk);
2659 * In this point socket cannot receive new packets, but it is possible
2660 * that some packets are in flight because some CPU runs receiver and
2661 * did hash table lookup before we unhashed socket. They will achieve
2662 * receive queue and will be purged by socket destructor.
2664 * Also we still have packets pending on receive queue and probably,
2665 * our own packets waiting in device queues. sock_destroy will drain
2666 * receive queue, but transmitted packets will delay socket destruction
2667 * until the last reference will be released.
2672 xfrm_sk_free_policy(sk);
2674 sk_refcnt_debug_release(sk);
2676 if (sk->sk_frag.page) {
2677 put_page(sk->sk_frag.page);
2678 sk->sk_frag.page = NULL;
2683 EXPORT_SYMBOL(sk_common_release);
2685 #ifdef CONFIG_PROC_FS
2686 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2688 int val[PROTO_INUSE_NR];
2691 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2693 #ifdef CONFIG_NET_NS
2694 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2696 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2698 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2700 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2702 int cpu, idx = prot->inuse_idx;
2705 for_each_possible_cpu(cpu)
2706 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2708 return res >= 0 ? res : 0;
2710 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2712 static int __net_init sock_inuse_init_net(struct net *net)
2714 net->core.inuse = alloc_percpu(struct prot_inuse);
2715 return net->core.inuse ? 0 : -ENOMEM;
2718 static void __net_exit sock_inuse_exit_net(struct net *net)
2720 free_percpu(net->core.inuse);
2723 static struct pernet_operations net_inuse_ops = {
2724 .init = sock_inuse_init_net,
2725 .exit = sock_inuse_exit_net,
2728 static __init int net_inuse_init(void)
2730 if (register_pernet_subsys(&net_inuse_ops))
2731 panic("Cannot initialize net inuse counters");
2736 core_initcall(net_inuse_init);
2738 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2740 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2742 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2744 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2746 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2748 int cpu, idx = prot->inuse_idx;
2751 for_each_possible_cpu(cpu)
2752 res += per_cpu(prot_inuse, cpu).val[idx];
2754 return res >= 0 ? res : 0;
2756 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2759 static void assign_proto_idx(struct proto *prot)
2761 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2763 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2764 pr_err("PROTO_INUSE_NR exhausted\n");
2768 set_bit(prot->inuse_idx, proto_inuse_idx);
2771 static void release_proto_idx(struct proto *prot)
2773 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2774 clear_bit(prot->inuse_idx, proto_inuse_idx);
2777 static inline void assign_proto_idx(struct proto *prot)
2781 static inline void release_proto_idx(struct proto *prot)
2786 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2790 kfree(rsk_prot->slab_name);
2791 rsk_prot->slab_name = NULL;
2792 kmem_cache_destroy(rsk_prot->slab);
2793 rsk_prot->slab = NULL;
2796 static int req_prot_init(const struct proto *prot)
2798 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2803 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2805 if (!rsk_prot->slab_name)
2808 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2809 rsk_prot->obj_size, 0,
2810 prot->slab_flags, NULL);
2812 if (!rsk_prot->slab) {
2813 pr_crit("%s: Can't create request sock SLAB cache!\n",
2820 int proto_register(struct proto *prot, int alloc_slab)
2823 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2824 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2827 if (prot->slab == NULL) {
2828 pr_crit("%s: Can't create sock SLAB cache!\n",
2833 if (req_prot_init(prot))
2834 goto out_free_request_sock_slab;
2836 if (prot->twsk_prot != NULL) {
2837 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2839 if (prot->twsk_prot->twsk_slab_name == NULL)
2840 goto out_free_request_sock_slab;
2842 prot->twsk_prot->twsk_slab =
2843 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2844 prot->twsk_prot->twsk_obj_size,
2848 if (prot->twsk_prot->twsk_slab == NULL)
2849 goto out_free_timewait_sock_slab_name;
2853 mutex_lock(&proto_list_mutex);
2854 list_add(&prot->node, &proto_list);
2855 assign_proto_idx(prot);
2856 mutex_unlock(&proto_list_mutex);
2859 out_free_timewait_sock_slab_name:
2860 kfree(prot->twsk_prot->twsk_slab_name);
2861 out_free_request_sock_slab:
2862 req_prot_cleanup(prot->rsk_prot);
2864 kmem_cache_destroy(prot->slab);
2869 EXPORT_SYMBOL(proto_register);
2871 void proto_unregister(struct proto *prot)
2873 mutex_lock(&proto_list_mutex);
2874 release_proto_idx(prot);
2875 list_del(&prot->node);
2876 mutex_unlock(&proto_list_mutex);
2878 kmem_cache_destroy(prot->slab);
2881 req_prot_cleanup(prot->rsk_prot);
2883 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2884 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2885 kfree(prot->twsk_prot->twsk_slab_name);
2886 prot->twsk_prot->twsk_slab = NULL;
2889 EXPORT_SYMBOL(proto_unregister);
2891 #ifdef CONFIG_PROC_FS
2892 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2893 __acquires(proto_list_mutex)
2895 mutex_lock(&proto_list_mutex);
2896 return seq_list_start_head(&proto_list, *pos);
2899 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2901 return seq_list_next(v, &proto_list, pos);
2904 static void proto_seq_stop(struct seq_file *seq, void *v)
2905 __releases(proto_list_mutex)
2907 mutex_unlock(&proto_list_mutex);
2910 static char proto_method_implemented(const void *method)
2912 return method == NULL ? 'n' : 'y';
2914 static long sock_prot_memory_allocated(struct proto *proto)
2916 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2919 static char *sock_prot_memory_pressure(struct proto *proto)
2921 return proto->memory_pressure != NULL ?
2922 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2925 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2928 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2929 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2932 sock_prot_inuse_get(seq_file_net(seq), proto),
2933 sock_prot_memory_allocated(proto),
2934 sock_prot_memory_pressure(proto),
2936 proto->slab == NULL ? "no" : "yes",
2937 module_name(proto->owner),
2938 proto_method_implemented(proto->close),
2939 proto_method_implemented(proto->connect),
2940 proto_method_implemented(proto->disconnect),
2941 proto_method_implemented(proto->accept),
2942 proto_method_implemented(proto->ioctl),
2943 proto_method_implemented(proto->init),
2944 proto_method_implemented(proto->destroy),
2945 proto_method_implemented(proto->shutdown),
2946 proto_method_implemented(proto->setsockopt),
2947 proto_method_implemented(proto->getsockopt),
2948 proto_method_implemented(proto->sendmsg),
2949 proto_method_implemented(proto->recvmsg),
2950 proto_method_implemented(proto->sendpage),
2951 proto_method_implemented(proto->bind),
2952 proto_method_implemented(proto->backlog_rcv),
2953 proto_method_implemented(proto->hash),
2954 proto_method_implemented(proto->unhash),
2955 proto_method_implemented(proto->get_port),
2956 proto_method_implemented(proto->enter_memory_pressure));
2959 static int proto_seq_show(struct seq_file *seq, void *v)
2961 if (v == &proto_list)
2962 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2971 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2973 proto_seq_printf(seq, list_entry(v, struct proto, node));
2977 static const struct seq_operations proto_seq_ops = {
2978 .start = proto_seq_start,
2979 .next = proto_seq_next,
2980 .stop = proto_seq_stop,
2981 .show = proto_seq_show,
2984 static int proto_seq_open(struct inode *inode, struct file *file)
2986 return seq_open_net(inode, file, &proto_seq_ops,
2987 sizeof(struct seq_net_private));
2990 static const struct file_operations proto_seq_fops = {
2991 .owner = THIS_MODULE,
2992 .open = proto_seq_open,
2994 .llseek = seq_lseek,
2995 .release = seq_release_net,
2998 static __net_init int proto_init_net(struct net *net)
3000 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3006 static __net_exit void proto_exit_net(struct net *net)
3008 remove_proc_entry("protocols", net->proc_net);
3012 static __net_initdata struct pernet_operations proto_net_ops = {
3013 .init = proto_init_net,
3014 .exit = proto_exit_net,
3017 static int __init proto_init(void)
3019 return register_pernet_subsys(&proto_net_ops);
3022 subsys_initcall(proto_init);
3024 #endif /* PROC_FS */