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_KCM" ,
227 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
228 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
229 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
230 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
231 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
232 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
233 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
234 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
235 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
236 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
237 "slock-27" , "slock-28" , "slock-AF_CAN" ,
238 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
239 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
240 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
241 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" ,
244 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
245 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
246 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
247 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
248 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
249 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
250 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
251 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
252 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
253 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
254 "clock-27" , "clock-28" , "clock-AF_CAN" ,
255 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
256 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
257 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
258 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" ,
263 * sk_callback_lock locking rules are per-address-family,
264 * so split the lock classes by using a per-AF key:
266 static struct lock_class_key af_callback_keys[AF_MAX];
268 /* Take into consideration the size of the struct sk_buff overhead in the
269 * determination of these values, since that is non-constant across
270 * platforms. This makes socket queueing behavior and performance
271 * not depend upon such differences.
273 #define _SK_MEM_PACKETS 256
274 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
275 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
276 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
278 /* Run time adjustable parameters. */
279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
280 EXPORT_SYMBOL(sysctl_wmem_max);
281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
282 EXPORT_SYMBOL(sysctl_rmem_max);
283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
286 /* Maximal space eaten by iovec or ancillary data plus some space */
287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
288 EXPORT_SYMBOL(sysctl_optmem_max);
290 int sysctl_tstamp_allow_data __read_mostly = 1;
292 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
293 EXPORT_SYMBOL_GPL(memalloc_socks);
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
303 void sk_set_memalloc(struct sock *sk)
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
307 static_key_slow_inc(&memalloc_socks);
309 EXPORT_SYMBOL_GPL(sk_set_memalloc);
311 void sk_clear_memalloc(struct sock *sk)
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
315 static_key_slow_dec(&memalloc_socks);
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
326 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
331 unsigned long pflags = current->flags;
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
336 current->flags |= PF_MEMALLOC;
337 ret = sk->sk_backlog_rcv(sk, skb);
338 tsk_restore_flags(current, pflags, PF_MEMALLOC);
342 EXPORT_SYMBOL(__sk_backlog_rcv);
344 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
348 if (optlen < sizeof(tv))
350 if (copy_from_user(&tv, optval, sizeof(tv)))
352 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
356 static int warned __read_mostly;
359 if (warned < 10 && net_ratelimit()) {
361 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
362 __func__, current->comm, task_pid_nr(current));
366 *timeo_p = MAX_SCHEDULE_TIMEOUT;
367 if (tv.tv_sec == 0 && tv.tv_usec == 0)
369 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
370 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
374 static void sock_warn_obsolete_bsdism(const char *name)
377 static char warncomm[TASK_COMM_LEN];
378 if (strcmp(warncomm, current->comm) && warned < 5) {
379 strcpy(warncomm, current->comm);
380 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
386 static bool sock_needs_netstamp(const struct sock *sk)
388 switch (sk->sk_family) {
397 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
399 if (sk->sk_flags & flags) {
400 sk->sk_flags &= ~flags;
401 if (sock_needs_netstamp(sk) &&
402 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
403 net_disable_timestamp();
408 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
411 struct sk_buff_head *list = &sk->sk_receive_queue;
413 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
414 atomic_inc(&sk->sk_drops);
415 trace_sock_rcvqueue_full(sk, skb);
419 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
420 atomic_inc(&sk->sk_drops);
425 skb_set_owner_r(skb, sk);
427 /* we escape from rcu protected region, make sure we dont leak
432 spin_lock_irqsave(&list->lock, flags);
433 sock_skb_set_dropcount(sk, skb);
434 __skb_queue_tail(list, skb);
435 spin_unlock_irqrestore(&list->lock, flags);
437 if (!sock_flag(sk, SOCK_DEAD))
438 sk->sk_data_ready(sk);
441 EXPORT_SYMBOL(__sock_queue_rcv_skb);
443 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
447 err = sk_filter(sk, skb);
451 return __sock_queue_rcv_skb(sk, skb);
453 EXPORT_SYMBOL(sock_queue_rcv_skb);
455 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
456 const int nested, unsigned int trim_cap, bool refcounted)
458 int rc = NET_RX_SUCCESS;
460 if (sk_filter_trim_cap(sk, skb, trim_cap))
461 goto discard_and_relse;
465 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
466 atomic_inc(&sk->sk_drops);
467 goto discard_and_relse;
470 bh_lock_sock_nested(sk);
473 if (!sock_owned_by_user(sk)) {
475 * trylock + unlock semantics:
477 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
479 rc = sk_backlog_rcv(sk, skb);
481 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
482 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
484 atomic_inc(&sk->sk_drops);
485 goto discard_and_relse;
497 EXPORT_SYMBOL(__sk_receive_skb);
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) {
504 sk_tx_queue_clear(sk);
505 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
512 EXPORT_SYMBOL(__sk_dst_check);
514 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
516 struct dst_entry *dst = sk_dst_get(sk);
518 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
526 EXPORT_SYMBOL(sk_dst_check);
528 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
531 int ret = -ENOPROTOOPT;
532 #ifdef CONFIG_NETDEVICES
533 struct net *net = sock_net(sk);
534 char devname[IFNAMSIZ];
539 if (!ns_capable(net->user_ns, CAP_NET_RAW))
546 /* Bind this socket to a particular device like "eth0",
547 * as specified in the passed interface name. If the
548 * name is "" or the option length is zero the socket
551 if (optlen > IFNAMSIZ - 1)
552 optlen = IFNAMSIZ - 1;
553 memset(devname, 0, sizeof(devname));
556 if (copy_from_user(devname, optval, optlen))
560 if (devname[0] != '\0') {
561 struct net_device *dev;
564 dev = dev_get_by_name_rcu(net, devname);
566 index = dev->ifindex;
574 sk->sk_bound_dev_if = index;
586 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
587 int __user *optlen, int len)
589 int ret = -ENOPROTOOPT;
590 #ifdef CONFIG_NETDEVICES
591 struct net *net = sock_net(sk);
592 char devname[IFNAMSIZ];
594 if (sk->sk_bound_dev_if == 0) {
603 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
607 len = strlen(devname) + 1;
610 if (copy_to_user(optval, devname, len))
615 if (put_user(len, optlen))
626 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
629 sock_set_flag(sk, bit);
631 sock_reset_flag(sk, bit);
634 bool sk_mc_loop(struct sock *sk)
636 if (dev_recursion_level())
640 switch (sk->sk_family) {
642 return inet_sk(sk)->mc_loop;
643 #if IS_ENABLED(CONFIG_IPV6)
645 return inet6_sk(sk)->mc_loop;
651 EXPORT_SYMBOL(sk_mc_loop);
654 * This is meant for all protocols to use and covers goings on
655 * at the socket level. Everything here is generic.
658 int sock_setsockopt(struct socket *sock, int level, int optname,
659 char __user *optval, unsigned int optlen)
661 struct sock *sk = sock->sk;
668 * Options without arguments
671 if (optname == SO_BINDTODEVICE)
672 return sock_setbindtodevice(sk, optval, optlen);
674 if (optlen < sizeof(int))
677 if (get_user(val, (int __user *)optval))
680 valbool = val ? 1 : 0;
686 if (val && !capable(CAP_NET_ADMIN))
689 sock_valbool_flag(sk, SOCK_DBG, valbool);
692 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
695 sk->sk_reuseport = valbool;
704 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
707 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
710 /* Don't error on this BSD doesn't and if you think
711 * about it this is right. Otherwise apps have to
712 * play 'guess the biggest size' games. RCVBUF/SNDBUF
713 * are treated in BSD as hints
715 val = min_t(u32, val, sysctl_wmem_max);
717 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
718 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
719 /* Wake up sending tasks if we upped the value. */
720 sk->sk_write_space(sk);
724 if (!capable(CAP_NET_ADMIN)) {
731 /* Don't error on this BSD doesn't and if you think
732 * about it this is right. Otherwise apps have to
733 * play 'guess the biggest size' games. RCVBUF/SNDBUF
734 * are treated in BSD as hints
736 val = min_t(u32, val, sysctl_rmem_max);
738 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
740 * We double it on the way in to account for
741 * "struct sk_buff" etc. overhead. Applications
742 * assume that the SO_RCVBUF setting they make will
743 * allow that much actual data to be received on that
746 * Applications are unaware that "struct sk_buff" and
747 * other overheads allocate from the receive buffer
748 * during socket buffer allocation.
750 * And after considering the possible alternatives,
751 * returning the value we actually used in getsockopt
752 * is the most desirable behavior.
754 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
758 if (!capable(CAP_NET_ADMIN)) {
766 if (sk->sk_protocol == IPPROTO_TCP &&
767 sk->sk_type == SOCK_STREAM)
768 tcp_set_keepalive(sk, valbool);
770 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
774 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
778 sk->sk_no_check_tx = valbool;
782 if ((val >= 0 && val <= 6) ||
783 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
784 sk->sk_priority = val;
790 if (optlen < sizeof(ling)) {
791 ret = -EINVAL; /* 1003.1g */
794 if (copy_from_user(&ling, optval, sizeof(ling))) {
799 sock_reset_flag(sk, SOCK_LINGER);
801 #if (BITS_PER_LONG == 32)
802 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
803 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
806 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
807 sock_set_flag(sk, SOCK_LINGER);
812 sock_warn_obsolete_bsdism("setsockopt");
817 set_bit(SOCK_PASSCRED, &sock->flags);
819 clear_bit(SOCK_PASSCRED, &sock->flags);
825 if (optname == SO_TIMESTAMP)
826 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
828 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
829 sock_set_flag(sk, SOCK_RCVTSTAMP);
830 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
832 sock_reset_flag(sk, SOCK_RCVTSTAMP);
833 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
837 case SO_TIMESTAMPING:
838 if (val & ~SOF_TIMESTAMPING_MASK) {
843 if (val & SOF_TIMESTAMPING_OPT_ID &&
844 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
845 if (sk->sk_protocol == IPPROTO_TCP &&
846 sk->sk_type == SOCK_STREAM) {
847 if ((1 << sk->sk_state) &
848 (TCPF_CLOSE | TCPF_LISTEN)) {
852 sk->sk_tskey = tcp_sk(sk)->snd_una;
857 sk->sk_tsflags = val;
858 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
859 sock_enable_timestamp(sk,
860 SOCK_TIMESTAMPING_RX_SOFTWARE);
862 sock_disable_timestamp(sk,
863 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
869 sk->sk_rcvlowat = val ? : 1;
873 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
877 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
880 case SO_ATTACH_FILTER:
882 if (optlen == sizeof(struct sock_fprog)) {
883 struct sock_fprog fprog;
886 if (copy_from_user(&fprog, optval, sizeof(fprog)))
889 ret = sk_attach_filter(&fprog, sk);
895 if (optlen == sizeof(u32)) {
899 if (copy_from_user(&ufd, optval, sizeof(ufd)))
902 ret = sk_attach_bpf(ufd, sk);
906 case SO_ATTACH_REUSEPORT_CBPF:
908 if (optlen == sizeof(struct sock_fprog)) {
909 struct sock_fprog fprog;
912 if (copy_from_user(&fprog, optval, sizeof(fprog)))
915 ret = sk_reuseport_attach_filter(&fprog, sk);
919 case SO_ATTACH_REUSEPORT_EBPF:
921 if (optlen == sizeof(u32)) {
925 if (copy_from_user(&ufd, optval, sizeof(ufd)))
928 ret = sk_reuseport_attach_bpf(ufd, sk);
932 case SO_DETACH_FILTER:
933 ret = sk_detach_filter(sk);
937 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
940 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
945 set_bit(SOCK_PASSSEC, &sock->flags);
947 clear_bit(SOCK_PASSSEC, &sock->flags);
950 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
957 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
961 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
965 if (sock->ops->set_peek_off)
966 ret = sock->ops->set_peek_off(sk, val);
972 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
975 case SO_SELECT_ERR_QUEUE:
976 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
979 #ifdef CONFIG_NET_RX_BUSY_POLL
981 /* allow unprivileged users to decrease the value */
982 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
988 sk->sk_ll_usec = val;
993 case SO_MAX_PACING_RATE:
994 sk->sk_max_pacing_rate = val;
995 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
996 sk->sk_max_pacing_rate);
999 case SO_INCOMING_CPU:
1000 sk->sk_incoming_cpu = val;
1005 dst_negative_advice(sk);
1014 EXPORT_SYMBOL(sock_setsockopt);
1017 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1018 struct ucred *ucred)
1020 ucred->pid = pid_vnr(pid);
1021 ucred->uid = ucred->gid = -1;
1023 struct user_namespace *current_ns = current_user_ns();
1025 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1026 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1030 int sock_getsockopt(struct socket *sock, int level, int optname,
1031 char __user *optval, int __user *optlen)
1033 struct sock *sk = sock->sk;
1041 int lv = sizeof(int);
1044 if (get_user(len, optlen))
1049 memset(&v, 0, sizeof(v));
1053 v.val = sock_flag(sk, SOCK_DBG);
1057 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1061 v.val = sock_flag(sk, SOCK_BROADCAST);
1065 v.val = sk->sk_sndbuf;
1069 v.val = sk->sk_rcvbuf;
1073 v.val = sk->sk_reuse;
1077 v.val = sk->sk_reuseport;
1081 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1085 v.val = sk->sk_type;
1089 v.val = sk->sk_protocol;
1093 v.val = sk->sk_family;
1097 v.val = -sock_error(sk);
1099 v.val = xchg(&sk->sk_err_soft, 0);
1103 v.val = sock_flag(sk, SOCK_URGINLINE);
1107 v.val = sk->sk_no_check_tx;
1111 v.val = sk->sk_priority;
1115 lv = sizeof(v.ling);
1116 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1117 v.ling.l_linger = sk->sk_lingertime / HZ;
1121 sock_warn_obsolete_bsdism("getsockopt");
1125 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1126 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1129 case SO_TIMESTAMPNS:
1130 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1133 case SO_TIMESTAMPING:
1134 v.val = sk->sk_tsflags;
1138 lv = sizeof(struct timeval);
1139 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1143 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1144 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1149 lv = sizeof(struct timeval);
1150 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1154 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1155 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1160 v.val = sk->sk_rcvlowat;
1168 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1173 struct ucred peercred;
1174 if (len > sizeof(peercred))
1175 len = sizeof(peercred);
1176 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1177 if (copy_to_user(optval, &peercred, len))
1186 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1190 if (copy_to_user(optval, address, len))
1195 /* Dubious BSD thing... Probably nobody even uses it, but
1196 * the UNIX standard wants it for whatever reason... -DaveM
1199 v.val = sk->sk_state == TCP_LISTEN;
1203 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1207 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1210 v.val = sk->sk_mark;
1214 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1217 case SO_WIFI_STATUS:
1218 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1222 if (!sock->ops->set_peek_off)
1225 v.val = sk->sk_peek_off;
1228 v.val = sock_flag(sk, SOCK_NOFCS);
1231 case SO_BINDTODEVICE:
1232 return sock_getbindtodevice(sk, optval, optlen, len);
1235 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1241 case SO_LOCK_FILTER:
1242 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1245 case SO_BPF_EXTENSIONS:
1246 v.val = bpf_tell_extensions();
1249 case SO_SELECT_ERR_QUEUE:
1250 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1253 #ifdef CONFIG_NET_RX_BUSY_POLL
1255 v.val = sk->sk_ll_usec;
1259 case SO_MAX_PACING_RATE:
1260 v.val = sk->sk_max_pacing_rate;
1263 case SO_INCOMING_CPU:
1264 v.val = sk->sk_incoming_cpu;
1268 /* We implement the SO_SNDLOWAT etc to not be settable
1271 return -ENOPROTOOPT;
1276 if (copy_to_user(optval, &v, len))
1279 if (put_user(len, optlen))
1285 * Initialize an sk_lock.
1287 * (We also register the sk_lock with the lock validator.)
1289 static inline void sock_lock_init(struct sock *sk)
1291 sock_lock_init_class_and_name(sk,
1292 af_family_slock_key_strings[sk->sk_family],
1293 af_family_slock_keys + sk->sk_family,
1294 af_family_key_strings[sk->sk_family],
1295 af_family_keys + sk->sk_family);
1299 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1300 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1301 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1303 static void sock_copy(struct sock *nsk, const struct sock *osk)
1305 #ifdef CONFIG_SECURITY_NETWORK
1306 void *sptr = nsk->sk_security;
1308 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1310 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1311 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1313 #ifdef CONFIG_SECURITY_NETWORK
1314 nsk->sk_security = sptr;
1315 security_sk_clone(osk, nsk);
1319 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1323 struct kmem_cache *slab;
1327 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1330 if (priority & __GFP_ZERO)
1331 sk_prot_clear_nulls(sk, prot->obj_size);
1333 sk = kmalloc(prot->obj_size, priority);
1336 kmemcheck_annotate_bitfield(sk, flags);
1338 if (security_sk_alloc(sk, family, priority))
1341 if (!try_module_get(prot->owner))
1343 sk_tx_queue_clear(sk);
1349 security_sk_free(sk);
1352 kmem_cache_free(slab, sk);
1358 static void sk_prot_free(struct proto *prot, struct sock *sk)
1360 struct kmem_cache *slab;
1361 struct module *owner;
1363 owner = prot->owner;
1366 cgroup_sk_free(&sk->sk_cgrp_data);
1367 mem_cgroup_sk_free(sk);
1368 security_sk_free(sk);
1370 kmem_cache_free(slab, sk);
1377 * sk_alloc - All socket objects are allocated here
1378 * @net: the applicable net namespace
1379 * @family: protocol family
1380 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1381 * @prot: struct proto associated with this new sock instance
1382 * @kern: is this to be a kernel socket?
1384 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1385 struct proto *prot, int kern)
1389 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1391 sk->sk_family = family;
1393 * See comment in struct sock definition to understand
1394 * why we need sk_prot_creator -acme
1396 sk->sk_prot = sk->sk_prot_creator = prot;
1398 sk->sk_net_refcnt = kern ? 0 : 1;
1399 if (likely(sk->sk_net_refcnt))
1401 sock_net_set(sk, net);
1402 atomic_set(&sk->sk_wmem_alloc, 1);
1404 mem_cgroup_sk_alloc(sk);
1405 cgroup_sk_alloc(&sk->sk_cgrp_data);
1406 sock_update_classid(&sk->sk_cgrp_data);
1407 sock_update_netprioidx(&sk->sk_cgrp_data);
1412 EXPORT_SYMBOL(sk_alloc);
1414 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1415 * grace period. This is the case for UDP sockets and TCP listeners.
1417 static void __sk_destruct(struct rcu_head *head)
1419 struct sock *sk = container_of(head, struct sock, sk_rcu);
1420 struct sk_filter *filter;
1422 if (sk->sk_destruct)
1423 sk->sk_destruct(sk);
1425 filter = rcu_dereference_check(sk->sk_filter,
1426 atomic_read(&sk->sk_wmem_alloc) == 0);
1428 sk_filter_uncharge(sk, filter);
1429 RCU_INIT_POINTER(sk->sk_filter, NULL);
1431 if (rcu_access_pointer(sk->sk_reuseport_cb))
1432 reuseport_detach_sock(sk);
1434 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1436 if (atomic_read(&sk->sk_omem_alloc))
1437 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1438 __func__, atomic_read(&sk->sk_omem_alloc));
1440 if (sk->sk_peer_cred)
1441 put_cred(sk->sk_peer_cred);
1442 put_pid(sk->sk_peer_pid);
1443 if (likely(sk->sk_net_refcnt))
1444 put_net(sock_net(sk));
1445 sk_prot_free(sk->sk_prot_creator, sk);
1448 void sk_destruct(struct sock *sk)
1450 if (sock_flag(sk, SOCK_RCU_FREE))
1451 call_rcu(&sk->sk_rcu, __sk_destruct);
1453 __sk_destruct(&sk->sk_rcu);
1456 static void __sk_free(struct sock *sk)
1458 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1459 sock_diag_broadcast_destroy(sk);
1464 void sk_free(struct sock *sk)
1467 * We subtract one from sk_wmem_alloc and can know if
1468 * some packets are still in some tx queue.
1469 * If not null, sock_wfree() will call __sk_free(sk) later
1471 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1474 EXPORT_SYMBOL(sk_free);
1477 * sk_clone_lock - clone a socket, and lock its clone
1478 * @sk: the socket to clone
1479 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1481 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1483 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1486 bool is_charged = true;
1488 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1489 if (newsk != NULL) {
1490 struct sk_filter *filter;
1492 sock_copy(newsk, sk);
1495 if (likely(newsk->sk_net_refcnt))
1496 get_net(sock_net(newsk));
1497 sk_node_init(&newsk->sk_node);
1498 sock_lock_init(newsk);
1499 bh_lock_sock(newsk);
1500 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1501 newsk->sk_backlog.len = 0;
1503 atomic_set(&newsk->sk_rmem_alloc, 0);
1505 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1507 atomic_set(&newsk->sk_wmem_alloc, 1);
1508 atomic_set(&newsk->sk_omem_alloc, 0);
1509 skb_queue_head_init(&newsk->sk_receive_queue);
1510 skb_queue_head_init(&newsk->sk_write_queue);
1512 rwlock_init(&newsk->sk_callback_lock);
1513 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1514 af_callback_keys + newsk->sk_family,
1515 af_family_clock_key_strings[newsk->sk_family]);
1517 newsk->sk_dst_cache = NULL;
1518 newsk->sk_wmem_queued = 0;
1519 newsk->sk_forward_alloc = 0;
1520 atomic_set(&newsk->sk_drops, 0);
1521 newsk->sk_send_head = NULL;
1522 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1524 sock_reset_flag(newsk, SOCK_DONE);
1525 skb_queue_head_init(&newsk->sk_error_queue);
1527 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1529 /* though it's an empty new sock, the charging may fail
1530 * if sysctl_optmem_max was changed between creation of
1531 * original socket and cloning
1533 is_charged = sk_filter_charge(newsk, filter);
1535 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1536 /* It is still raw copy of parent, so invalidate
1537 * destructor and make plain sk_free() */
1538 newsk->sk_destruct = NULL;
1539 bh_unlock_sock(newsk);
1544 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1547 newsk->sk_err_soft = 0;
1548 newsk->sk_priority = 0;
1549 newsk->sk_incoming_cpu = raw_smp_processor_id();
1550 atomic64_set(&newsk->sk_cookie, 0);
1552 mem_cgroup_sk_alloc(newsk);
1553 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1556 * Before updating sk_refcnt, we must commit prior changes to memory
1557 * (Documentation/RCU/rculist_nulls.txt for details)
1560 atomic_set(&newsk->sk_refcnt, 2);
1563 * Increment the counter in the same struct proto as the master
1564 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1565 * is the same as sk->sk_prot->socks, as this field was copied
1568 * This _changes_ the previous behaviour, where
1569 * tcp_create_openreq_child always was incrementing the
1570 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1571 * to be taken into account in all callers. -acme
1573 sk_refcnt_debug_inc(newsk);
1574 sk_set_socket(newsk, NULL);
1575 newsk->sk_wq = NULL;
1577 if (newsk->sk_prot->sockets_allocated)
1578 sk_sockets_allocated_inc(newsk);
1580 if (sock_needs_netstamp(sk) &&
1581 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1582 net_enable_timestamp();
1587 EXPORT_SYMBOL_GPL(sk_clone_lock);
1589 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1593 sk_dst_set(sk, dst);
1594 sk->sk_route_caps = dst->dev->features;
1595 if (sk->sk_route_caps & NETIF_F_GSO)
1596 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1597 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1598 if (sk_can_gso(sk)) {
1599 if (dst->header_len) {
1600 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1602 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1603 sk->sk_gso_max_size = dst->dev->gso_max_size;
1604 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1607 sk->sk_gso_max_segs = max_segs;
1609 EXPORT_SYMBOL_GPL(sk_setup_caps);
1612 * Simple resource managers for sockets.
1617 * Write buffer destructor automatically called from kfree_skb.
1619 void sock_wfree(struct sk_buff *skb)
1621 struct sock *sk = skb->sk;
1622 unsigned int len = skb->truesize;
1624 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1626 * Keep a reference on sk_wmem_alloc, this will be released
1627 * after sk_write_space() call
1629 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1630 sk->sk_write_space(sk);
1634 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1635 * could not do because of in-flight packets
1637 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1640 EXPORT_SYMBOL(sock_wfree);
1642 /* This variant of sock_wfree() is used by TCP,
1643 * since it sets SOCK_USE_WRITE_QUEUE.
1645 void __sock_wfree(struct sk_buff *skb)
1647 struct sock *sk = skb->sk;
1649 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1653 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1658 if (unlikely(!sk_fullsock(sk))) {
1659 skb->destructor = sock_edemux;
1664 skb->destructor = sock_wfree;
1665 skb_set_hash_from_sk(skb, sk);
1667 * We used to take a refcount on sk, but following operation
1668 * is enough to guarantee sk_free() wont free this sock until
1669 * all in-flight packets are completed
1671 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1673 EXPORT_SYMBOL(skb_set_owner_w);
1675 /* This helper is used by netem, as it can hold packets in its
1676 * delay queue. We want to allow the owner socket to send more
1677 * packets, as if they were already TX completed by a typical driver.
1678 * But we also want to keep skb->sk set because some packet schedulers
1679 * rely on it (sch_fq for example). So we set skb->truesize to a small
1680 * amount (1) and decrease sk_wmem_alloc accordingly.
1682 void skb_orphan_partial(struct sk_buff *skb)
1684 /* If this skb is a TCP pure ACK or already went here,
1685 * we have nothing to do. 2 is already a very small truesize.
1687 if (skb->truesize <= 2)
1690 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1691 * so we do not completely orphan skb, but transfert all
1692 * accounted bytes but one, to avoid unexpected reorders.
1694 if (skb->destructor == sock_wfree
1696 || skb->destructor == tcp_wfree
1699 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1705 EXPORT_SYMBOL(skb_orphan_partial);
1708 * Read buffer destructor automatically called from kfree_skb.
1710 void sock_rfree(struct sk_buff *skb)
1712 struct sock *sk = skb->sk;
1713 unsigned int len = skb->truesize;
1715 atomic_sub(len, &sk->sk_rmem_alloc);
1716 sk_mem_uncharge(sk, len);
1718 EXPORT_SYMBOL(sock_rfree);
1721 * Buffer destructor for skbs that are not used directly in read or write
1722 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1724 void sock_efree(struct sk_buff *skb)
1728 EXPORT_SYMBOL(sock_efree);
1730 kuid_t sock_i_uid(struct sock *sk)
1734 read_lock_bh(&sk->sk_callback_lock);
1735 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1736 read_unlock_bh(&sk->sk_callback_lock);
1739 EXPORT_SYMBOL(sock_i_uid);
1741 unsigned long sock_i_ino(struct sock *sk)
1745 read_lock_bh(&sk->sk_callback_lock);
1746 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1747 read_unlock_bh(&sk->sk_callback_lock);
1750 EXPORT_SYMBOL(sock_i_ino);
1753 * Allocate a skb from the socket's send buffer.
1755 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1758 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1759 struct sk_buff *skb = alloc_skb(size, priority);
1761 skb_set_owner_w(skb, sk);
1767 EXPORT_SYMBOL(sock_wmalloc);
1770 * Allocate a memory block from the socket's option memory buffer.
1772 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1774 if ((unsigned int)size <= sysctl_optmem_max &&
1775 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1777 /* First do the add, to avoid the race if kmalloc
1780 atomic_add(size, &sk->sk_omem_alloc);
1781 mem = kmalloc(size, priority);
1784 atomic_sub(size, &sk->sk_omem_alloc);
1788 EXPORT_SYMBOL(sock_kmalloc);
1790 /* Free an option memory block. Note, we actually want the inline
1791 * here as this allows gcc to detect the nullify and fold away the
1792 * condition entirely.
1794 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1797 if (WARN_ON_ONCE(!mem))
1803 atomic_sub(size, &sk->sk_omem_alloc);
1806 void sock_kfree_s(struct sock *sk, void *mem, int size)
1808 __sock_kfree_s(sk, mem, size, false);
1810 EXPORT_SYMBOL(sock_kfree_s);
1812 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1814 __sock_kfree_s(sk, mem, size, true);
1816 EXPORT_SYMBOL(sock_kzfree_s);
1818 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1819 I think, these locks should be removed for datagram sockets.
1821 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1825 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1829 if (signal_pending(current))
1831 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1832 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1833 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1835 if (sk->sk_shutdown & SEND_SHUTDOWN)
1839 timeo = schedule_timeout(timeo);
1841 finish_wait(sk_sleep(sk), &wait);
1847 * Generic send/receive buffer handlers
1850 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1851 unsigned long data_len, int noblock,
1852 int *errcode, int max_page_order)
1854 struct sk_buff *skb;
1858 timeo = sock_sndtimeo(sk, noblock);
1860 err = sock_error(sk);
1865 if (sk->sk_shutdown & SEND_SHUTDOWN)
1868 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1871 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1872 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1876 if (signal_pending(current))
1878 timeo = sock_wait_for_wmem(sk, timeo);
1880 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1881 errcode, sk->sk_allocation);
1883 skb_set_owner_w(skb, sk);
1887 err = sock_intr_errno(timeo);
1892 EXPORT_SYMBOL(sock_alloc_send_pskb);
1894 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1895 int noblock, int *errcode)
1897 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1899 EXPORT_SYMBOL(sock_alloc_send_skb);
1901 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1902 struct sockcm_cookie *sockc)
1906 switch (cmsg->cmsg_type) {
1908 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1910 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1912 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1914 case SO_TIMESTAMPING:
1915 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1918 tsflags = *(u32 *)CMSG_DATA(cmsg);
1919 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1922 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1923 sockc->tsflags |= tsflags;
1925 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1927 case SCM_CREDENTIALS:
1934 EXPORT_SYMBOL(__sock_cmsg_send);
1936 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1937 struct sockcm_cookie *sockc)
1939 struct cmsghdr *cmsg;
1942 for_each_cmsghdr(cmsg, msg) {
1943 if (!CMSG_OK(msg, cmsg))
1945 if (cmsg->cmsg_level != SOL_SOCKET)
1947 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1953 EXPORT_SYMBOL(sock_cmsg_send);
1955 /* On 32bit arches, an skb frag is limited to 2^15 */
1956 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1959 * skb_page_frag_refill - check that a page_frag contains enough room
1960 * @sz: minimum size of the fragment we want to get
1961 * @pfrag: pointer to page_frag
1962 * @gfp: priority for memory allocation
1964 * Note: While this allocator tries to use high order pages, there is
1965 * no guarantee that allocations succeed. Therefore, @sz MUST be
1966 * less or equal than PAGE_SIZE.
1968 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1971 if (page_ref_count(pfrag->page) == 1) {
1975 if (pfrag->offset + sz <= pfrag->size)
1977 put_page(pfrag->page);
1981 if (SKB_FRAG_PAGE_ORDER) {
1982 /* Avoid direct reclaim but allow kswapd to wake */
1983 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1984 __GFP_COMP | __GFP_NOWARN |
1986 SKB_FRAG_PAGE_ORDER);
1987 if (likely(pfrag->page)) {
1988 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1992 pfrag->page = alloc_page(gfp);
1993 if (likely(pfrag->page)) {
1994 pfrag->size = PAGE_SIZE;
1999 EXPORT_SYMBOL(skb_page_frag_refill);
2001 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2003 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2006 sk_enter_memory_pressure(sk);
2007 sk_stream_moderate_sndbuf(sk);
2010 EXPORT_SYMBOL(sk_page_frag_refill);
2012 static void __lock_sock(struct sock *sk)
2013 __releases(&sk->sk_lock.slock)
2014 __acquires(&sk->sk_lock.slock)
2019 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2020 TASK_UNINTERRUPTIBLE);
2021 spin_unlock_bh(&sk->sk_lock.slock);
2023 spin_lock_bh(&sk->sk_lock.slock);
2024 if (!sock_owned_by_user(sk))
2027 finish_wait(&sk->sk_lock.wq, &wait);
2030 static void __release_sock(struct sock *sk)
2031 __releases(&sk->sk_lock.slock)
2032 __acquires(&sk->sk_lock.slock)
2034 struct sk_buff *skb, *next;
2036 while ((skb = sk->sk_backlog.head) != NULL) {
2037 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2039 spin_unlock_bh(&sk->sk_lock.slock);
2044 WARN_ON_ONCE(skb_dst_is_noref(skb));
2046 sk_backlog_rcv(sk, skb);
2051 } while (skb != NULL);
2053 spin_lock_bh(&sk->sk_lock.slock);
2057 * Doing the zeroing here guarantee we can not loop forever
2058 * while a wild producer attempts to flood us.
2060 sk->sk_backlog.len = 0;
2063 void __sk_flush_backlog(struct sock *sk)
2065 spin_lock_bh(&sk->sk_lock.slock);
2067 spin_unlock_bh(&sk->sk_lock.slock);
2071 * sk_wait_data - wait for data to arrive at sk_receive_queue
2072 * @sk: sock to wait on
2073 * @timeo: for how long
2074 * @skb: last skb seen on sk_receive_queue
2076 * Now socket state including sk->sk_err is changed only under lock,
2077 * hence we may omit checks after joining wait queue.
2078 * We check receive queue before schedule() only as optimization;
2079 * it is very likely that release_sock() added new data.
2081 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2086 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2087 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2088 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2089 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2090 finish_wait(sk_sleep(sk), &wait);
2093 EXPORT_SYMBOL(sk_wait_data);
2096 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2098 * @size: memory size to allocate
2099 * @kind: allocation type
2101 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2102 * rmem allocation. This function assumes that protocols which have
2103 * memory_pressure use sk_wmem_queued as write buffer accounting.
2105 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2107 struct proto *prot = sk->sk_prot;
2108 int amt = sk_mem_pages(size);
2111 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2113 allocated = sk_memory_allocated_add(sk, amt);
2115 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2116 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2117 goto suppress_allocation;
2120 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2121 sk_leave_memory_pressure(sk);
2125 /* Under pressure. */
2126 if (allocated > sk_prot_mem_limits(sk, 1))
2127 sk_enter_memory_pressure(sk);
2129 /* Over hard limit. */
2130 if (allocated > sk_prot_mem_limits(sk, 2))
2131 goto suppress_allocation;
2133 /* guarantee minimum buffer size under pressure */
2134 if (kind == SK_MEM_RECV) {
2135 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2138 } else { /* SK_MEM_SEND */
2139 if (sk->sk_type == SOCK_STREAM) {
2140 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2142 } else if (atomic_read(&sk->sk_wmem_alloc) <
2143 prot->sysctl_wmem[0])
2147 if (sk_has_memory_pressure(sk)) {
2150 if (!sk_under_memory_pressure(sk))
2152 alloc = sk_sockets_allocated_read_positive(sk);
2153 if (sk_prot_mem_limits(sk, 2) > alloc *
2154 sk_mem_pages(sk->sk_wmem_queued +
2155 atomic_read(&sk->sk_rmem_alloc) +
2156 sk->sk_forward_alloc))
2160 suppress_allocation:
2162 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2163 sk_stream_moderate_sndbuf(sk);
2165 /* Fail only if socket is _under_ its sndbuf.
2166 * In this case we cannot block, so that we have to fail.
2168 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2172 trace_sock_exceed_buf_limit(sk, prot, allocated);
2174 /* Alas. Undo changes. */
2175 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2177 sk_memory_allocated_sub(sk, amt);
2179 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2180 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2184 EXPORT_SYMBOL(__sk_mem_schedule);
2187 * __sk_mem_reclaim - reclaim memory_allocated
2189 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2191 void __sk_mem_reclaim(struct sock *sk, int amount)
2193 amount >>= SK_MEM_QUANTUM_SHIFT;
2194 sk_memory_allocated_sub(sk, amount);
2195 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2197 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2198 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2200 if (sk_under_memory_pressure(sk) &&
2201 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2202 sk_leave_memory_pressure(sk);
2204 EXPORT_SYMBOL(__sk_mem_reclaim);
2206 int sk_set_peek_off(struct sock *sk, int val)
2211 sk->sk_peek_off = val;
2214 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2217 * Set of default routines for initialising struct proto_ops when
2218 * the protocol does not support a particular function. In certain
2219 * cases where it makes no sense for a protocol to have a "do nothing"
2220 * function, some default processing is provided.
2223 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2227 EXPORT_SYMBOL(sock_no_bind);
2229 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2234 EXPORT_SYMBOL(sock_no_connect);
2236 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2240 EXPORT_SYMBOL(sock_no_socketpair);
2242 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2246 EXPORT_SYMBOL(sock_no_accept);
2248 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2253 EXPORT_SYMBOL(sock_no_getname);
2255 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2259 EXPORT_SYMBOL(sock_no_poll);
2261 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2265 EXPORT_SYMBOL(sock_no_ioctl);
2267 int sock_no_listen(struct socket *sock, int backlog)
2271 EXPORT_SYMBOL(sock_no_listen);
2273 int sock_no_shutdown(struct socket *sock, int how)
2277 EXPORT_SYMBOL(sock_no_shutdown);
2279 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2280 char __user *optval, unsigned int optlen)
2284 EXPORT_SYMBOL(sock_no_setsockopt);
2286 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2287 char __user *optval, int __user *optlen)
2291 EXPORT_SYMBOL(sock_no_getsockopt);
2293 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2297 EXPORT_SYMBOL(sock_no_sendmsg);
2299 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2304 EXPORT_SYMBOL(sock_no_recvmsg);
2306 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2308 /* Mirror missing mmap method error code */
2311 EXPORT_SYMBOL(sock_no_mmap);
2313 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2316 struct msghdr msg = {.msg_flags = flags};
2318 char *kaddr = kmap(page);
2319 iov.iov_base = kaddr + offset;
2321 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2325 EXPORT_SYMBOL(sock_no_sendpage);
2328 * Default Socket Callbacks
2331 static void sock_def_wakeup(struct sock *sk)
2333 struct socket_wq *wq;
2336 wq = rcu_dereference(sk->sk_wq);
2337 if (skwq_has_sleeper(wq))
2338 wake_up_interruptible_all(&wq->wait);
2342 static void sock_def_error_report(struct sock *sk)
2344 struct socket_wq *wq;
2347 wq = rcu_dereference(sk->sk_wq);
2348 if (skwq_has_sleeper(wq))
2349 wake_up_interruptible_poll(&wq->wait, POLLERR);
2350 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2354 static void sock_def_readable(struct sock *sk)
2356 struct socket_wq *wq;
2359 wq = rcu_dereference(sk->sk_wq);
2360 if (skwq_has_sleeper(wq))
2361 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2362 POLLRDNORM | POLLRDBAND);
2363 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2367 static void sock_def_write_space(struct sock *sk)
2369 struct socket_wq *wq;
2373 /* Do not wake up a writer until he can make "significant"
2376 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2377 wq = rcu_dereference(sk->sk_wq);
2378 if (skwq_has_sleeper(wq))
2379 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2380 POLLWRNORM | POLLWRBAND);
2382 /* Should agree with poll, otherwise some programs break */
2383 if (sock_writeable(sk))
2384 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2390 static void sock_def_destruct(struct sock *sk)
2394 void sk_send_sigurg(struct sock *sk)
2396 if (sk->sk_socket && sk->sk_socket->file)
2397 if (send_sigurg(&sk->sk_socket->file->f_owner))
2398 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2400 EXPORT_SYMBOL(sk_send_sigurg);
2402 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2403 unsigned long expires)
2405 if (!mod_timer(timer, expires))
2408 EXPORT_SYMBOL(sk_reset_timer);
2410 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2412 if (del_timer(timer))
2415 EXPORT_SYMBOL(sk_stop_timer);
2417 void sock_init_data(struct socket *sock, struct sock *sk)
2419 skb_queue_head_init(&sk->sk_receive_queue);
2420 skb_queue_head_init(&sk->sk_write_queue);
2421 skb_queue_head_init(&sk->sk_error_queue);
2423 sk->sk_send_head = NULL;
2425 init_timer(&sk->sk_timer);
2427 sk->sk_allocation = GFP_KERNEL;
2428 sk->sk_rcvbuf = sysctl_rmem_default;
2429 sk->sk_sndbuf = sysctl_wmem_default;
2430 sk->sk_state = TCP_CLOSE;
2431 sk_set_socket(sk, sock);
2433 sock_set_flag(sk, SOCK_ZAPPED);
2436 sk->sk_type = sock->type;
2437 sk->sk_wq = sock->wq;
2442 rwlock_init(&sk->sk_callback_lock);
2443 lockdep_set_class_and_name(&sk->sk_callback_lock,
2444 af_callback_keys + sk->sk_family,
2445 af_family_clock_key_strings[sk->sk_family]);
2447 sk->sk_state_change = sock_def_wakeup;
2448 sk->sk_data_ready = sock_def_readable;
2449 sk->sk_write_space = sock_def_write_space;
2450 sk->sk_error_report = sock_def_error_report;
2451 sk->sk_destruct = sock_def_destruct;
2453 sk->sk_frag.page = NULL;
2454 sk->sk_frag.offset = 0;
2455 sk->sk_peek_off = -1;
2457 sk->sk_peer_pid = NULL;
2458 sk->sk_peer_cred = NULL;
2459 sk->sk_write_pending = 0;
2460 sk->sk_rcvlowat = 1;
2461 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2462 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2464 sk->sk_stamp = ktime_set(-1L, 0);
2466 #ifdef CONFIG_NET_RX_BUSY_POLL
2468 sk->sk_ll_usec = sysctl_net_busy_read;
2471 sk->sk_max_pacing_rate = ~0U;
2472 sk->sk_pacing_rate = ~0U;
2473 sk->sk_incoming_cpu = -1;
2475 * Before updating sk_refcnt, we must commit prior changes to memory
2476 * (Documentation/RCU/rculist_nulls.txt for details)
2479 atomic_set(&sk->sk_refcnt, 1);
2480 atomic_set(&sk->sk_drops, 0);
2482 EXPORT_SYMBOL(sock_init_data);
2484 void lock_sock_nested(struct sock *sk, int subclass)
2487 spin_lock_bh(&sk->sk_lock.slock);
2488 if (sk->sk_lock.owned)
2490 sk->sk_lock.owned = 1;
2491 spin_unlock(&sk->sk_lock.slock);
2493 * The sk_lock has mutex_lock() semantics here:
2495 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2498 EXPORT_SYMBOL(lock_sock_nested);
2500 void release_sock(struct sock *sk)
2502 spin_lock_bh(&sk->sk_lock.slock);
2503 if (sk->sk_backlog.tail)
2506 /* Warning : release_cb() might need to release sk ownership,
2507 * ie call sock_release_ownership(sk) before us.
2509 if (sk->sk_prot->release_cb)
2510 sk->sk_prot->release_cb(sk);
2512 sock_release_ownership(sk);
2513 if (waitqueue_active(&sk->sk_lock.wq))
2514 wake_up(&sk->sk_lock.wq);
2515 spin_unlock_bh(&sk->sk_lock.slock);
2517 EXPORT_SYMBOL(release_sock);
2520 * lock_sock_fast - fast version of lock_sock
2523 * This version should be used for very small section, where process wont block
2524 * return false if fast path is taken
2525 * sk_lock.slock locked, owned = 0, BH disabled
2526 * return true if slow path is taken
2527 * sk_lock.slock unlocked, owned = 1, BH enabled
2529 bool lock_sock_fast(struct sock *sk)
2532 spin_lock_bh(&sk->sk_lock.slock);
2534 if (!sk->sk_lock.owned)
2536 * Note : We must disable BH
2541 sk->sk_lock.owned = 1;
2542 spin_unlock(&sk->sk_lock.slock);
2544 * The sk_lock has mutex_lock() semantics here:
2546 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2550 EXPORT_SYMBOL(lock_sock_fast);
2552 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2555 if (!sock_flag(sk, SOCK_TIMESTAMP))
2556 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2557 tv = ktime_to_timeval(sk->sk_stamp);
2558 if (tv.tv_sec == -1)
2560 if (tv.tv_sec == 0) {
2561 sk->sk_stamp = ktime_get_real();
2562 tv = ktime_to_timeval(sk->sk_stamp);
2564 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2566 EXPORT_SYMBOL(sock_get_timestamp);
2568 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2571 if (!sock_flag(sk, SOCK_TIMESTAMP))
2572 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2573 ts = ktime_to_timespec(sk->sk_stamp);
2574 if (ts.tv_sec == -1)
2576 if (ts.tv_sec == 0) {
2577 sk->sk_stamp = ktime_get_real();
2578 ts = ktime_to_timespec(sk->sk_stamp);
2580 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2582 EXPORT_SYMBOL(sock_get_timestampns);
2584 void sock_enable_timestamp(struct sock *sk, int flag)
2586 if (!sock_flag(sk, flag)) {
2587 unsigned long previous_flags = sk->sk_flags;
2589 sock_set_flag(sk, flag);
2591 * we just set one of the two flags which require net
2592 * time stamping, but time stamping might have been on
2593 * already because of the other one
2595 if (sock_needs_netstamp(sk) &&
2596 !(previous_flags & SK_FLAGS_TIMESTAMP))
2597 net_enable_timestamp();
2601 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2602 int level, int type)
2604 struct sock_exterr_skb *serr;
2605 struct sk_buff *skb;
2609 skb = sock_dequeue_err_skb(sk);
2615 msg->msg_flags |= MSG_TRUNC;
2618 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2622 sock_recv_timestamp(msg, sk, skb);
2624 serr = SKB_EXT_ERR(skb);
2625 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2627 msg->msg_flags |= MSG_ERRQUEUE;
2635 EXPORT_SYMBOL(sock_recv_errqueue);
2638 * Get a socket option on an socket.
2640 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2641 * asynchronous errors should be reported by getsockopt. We assume
2642 * this means if you specify SO_ERROR (otherwise whats the point of it).
2644 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2645 char __user *optval, int __user *optlen)
2647 struct sock *sk = sock->sk;
2649 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2651 EXPORT_SYMBOL(sock_common_getsockopt);
2653 #ifdef CONFIG_COMPAT
2654 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2655 char __user *optval, int __user *optlen)
2657 struct sock *sk = sock->sk;
2659 if (sk->sk_prot->compat_getsockopt != NULL)
2660 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2662 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2664 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2667 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2670 struct sock *sk = sock->sk;
2674 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2675 flags & ~MSG_DONTWAIT, &addr_len);
2677 msg->msg_namelen = addr_len;
2680 EXPORT_SYMBOL(sock_common_recvmsg);
2683 * Set socket options on an inet socket.
2685 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2686 char __user *optval, unsigned int optlen)
2688 struct sock *sk = sock->sk;
2690 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2692 EXPORT_SYMBOL(sock_common_setsockopt);
2694 #ifdef CONFIG_COMPAT
2695 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2696 char __user *optval, unsigned int optlen)
2698 struct sock *sk = sock->sk;
2700 if (sk->sk_prot->compat_setsockopt != NULL)
2701 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2703 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2705 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2708 void sk_common_release(struct sock *sk)
2710 if (sk->sk_prot->destroy)
2711 sk->sk_prot->destroy(sk);
2714 * Observation: when sock_common_release is called, processes have
2715 * no access to socket. But net still has.
2716 * Step one, detach it from networking:
2718 * A. Remove from hash tables.
2721 sk->sk_prot->unhash(sk);
2724 * In this point socket cannot receive new packets, but it is possible
2725 * that some packets are in flight because some CPU runs receiver and
2726 * did hash table lookup before we unhashed socket. They will achieve
2727 * receive queue and will be purged by socket destructor.
2729 * Also we still have packets pending on receive queue and probably,
2730 * our own packets waiting in device queues. sock_destroy will drain
2731 * receive queue, but transmitted packets will delay socket destruction
2732 * until the last reference will be released.
2737 xfrm_sk_free_policy(sk);
2739 sk_refcnt_debug_release(sk);
2741 if (sk->sk_frag.page) {
2742 put_page(sk->sk_frag.page);
2743 sk->sk_frag.page = NULL;
2748 EXPORT_SYMBOL(sk_common_release);
2750 #ifdef CONFIG_PROC_FS
2751 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2753 int val[PROTO_INUSE_NR];
2756 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2758 #ifdef CONFIG_NET_NS
2759 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2761 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2763 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2765 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2767 int cpu, idx = prot->inuse_idx;
2770 for_each_possible_cpu(cpu)
2771 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2773 return res >= 0 ? res : 0;
2775 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2777 static int __net_init sock_inuse_init_net(struct net *net)
2779 net->core.inuse = alloc_percpu(struct prot_inuse);
2780 return net->core.inuse ? 0 : -ENOMEM;
2783 static void __net_exit sock_inuse_exit_net(struct net *net)
2785 free_percpu(net->core.inuse);
2788 static struct pernet_operations net_inuse_ops = {
2789 .init = sock_inuse_init_net,
2790 .exit = sock_inuse_exit_net,
2793 static __init int net_inuse_init(void)
2795 if (register_pernet_subsys(&net_inuse_ops))
2796 panic("Cannot initialize net inuse counters");
2801 core_initcall(net_inuse_init);
2803 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2805 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2807 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2809 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2811 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2813 int cpu, idx = prot->inuse_idx;
2816 for_each_possible_cpu(cpu)
2817 res += per_cpu(prot_inuse, cpu).val[idx];
2819 return res >= 0 ? res : 0;
2821 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2824 static void assign_proto_idx(struct proto *prot)
2826 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2828 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2829 pr_err("PROTO_INUSE_NR exhausted\n");
2833 set_bit(prot->inuse_idx, proto_inuse_idx);
2836 static void release_proto_idx(struct proto *prot)
2838 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2839 clear_bit(prot->inuse_idx, proto_inuse_idx);
2842 static inline void assign_proto_idx(struct proto *prot)
2846 static inline void release_proto_idx(struct proto *prot)
2851 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2855 kfree(rsk_prot->slab_name);
2856 rsk_prot->slab_name = NULL;
2857 kmem_cache_destroy(rsk_prot->slab);
2858 rsk_prot->slab = NULL;
2861 static int req_prot_init(const struct proto *prot)
2863 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2868 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2870 if (!rsk_prot->slab_name)
2873 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2874 rsk_prot->obj_size, 0,
2875 prot->slab_flags, NULL);
2877 if (!rsk_prot->slab) {
2878 pr_crit("%s: Can't create request sock SLAB cache!\n",
2885 int proto_register(struct proto *prot, int alloc_slab)
2888 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2889 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2892 if (prot->slab == NULL) {
2893 pr_crit("%s: Can't create sock SLAB cache!\n",
2898 if (req_prot_init(prot))
2899 goto out_free_request_sock_slab;
2901 if (prot->twsk_prot != NULL) {
2902 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2904 if (prot->twsk_prot->twsk_slab_name == NULL)
2905 goto out_free_request_sock_slab;
2907 prot->twsk_prot->twsk_slab =
2908 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2909 prot->twsk_prot->twsk_obj_size,
2913 if (prot->twsk_prot->twsk_slab == NULL)
2914 goto out_free_timewait_sock_slab_name;
2918 mutex_lock(&proto_list_mutex);
2919 list_add(&prot->node, &proto_list);
2920 assign_proto_idx(prot);
2921 mutex_unlock(&proto_list_mutex);
2924 out_free_timewait_sock_slab_name:
2925 kfree(prot->twsk_prot->twsk_slab_name);
2926 out_free_request_sock_slab:
2927 req_prot_cleanup(prot->rsk_prot);
2929 kmem_cache_destroy(prot->slab);
2934 EXPORT_SYMBOL(proto_register);
2936 void proto_unregister(struct proto *prot)
2938 mutex_lock(&proto_list_mutex);
2939 release_proto_idx(prot);
2940 list_del(&prot->node);
2941 mutex_unlock(&proto_list_mutex);
2943 kmem_cache_destroy(prot->slab);
2946 req_prot_cleanup(prot->rsk_prot);
2948 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2949 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2950 kfree(prot->twsk_prot->twsk_slab_name);
2951 prot->twsk_prot->twsk_slab = NULL;
2954 EXPORT_SYMBOL(proto_unregister);
2956 #ifdef CONFIG_PROC_FS
2957 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2958 __acquires(proto_list_mutex)
2960 mutex_lock(&proto_list_mutex);
2961 return seq_list_start_head(&proto_list, *pos);
2964 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2966 return seq_list_next(v, &proto_list, pos);
2969 static void proto_seq_stop(struct seq_file *seq, void *v)
2970 __releases(proto_list_mutex)
2972 mutex_unlock(&proto_list_mutex);
2975 static char proto_method_implemented(const void *method)
2977 return method == NULL ? 'n' : 'y';
2979 static long sock_prot_memory_allocated(struct proto *proto)
2981 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2984 static char *sock_prot_memory_pressure(struct proto *proto)
2986 return proto->memory_pressure != NULL ?
2987 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2990 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2993 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2994 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2997 sock_prot_inuse_get(seq_file_net(seq), proto),
2998 sock_prot_memory_allocated(proto),
2999 sock_prot_memory_pressure(proto),
3001 proto->slab == NULL ? "no" : "yes",
3002 module_name(proto->owner),
3003 proto_method_implemented(proto->close),
3004 proto_method_implemented(proto->connect),
3005 proto_method_implemented(proto->disconnect),
3006 proto_method_implemented(proto->accept),
3007 proto_method_implemented(proto->ioctl),
3008 proto_method_implemented(proto->init),
3009 proto_method_implemented(proto->destroy),
3010 proto_method_implemented(proto->shutdown),
3011 proto_method_implemented(proto->setsockopt),
3012 proto_method_implemented(proto->getsockopt),
3013 proto_method_implemented(proto->sendmsg),
3014 proto_method_implemented(proto->recvmsg),
3015 proto_method_implemented(proto->sendpage),
3016 proto_method_implemented(proto->bind),
3017 proto_method_implemented(proto->backlog_rcv),
3018 proto_method_implemented(proto->hash),
3019 proto_method_implemented(proto->unhash),
3020 proto_method_implemented(proto->get_port),
3021 proto_method_implemented(proto->enter_memory_pressure));
3024 static int proto_seq_show(struct seq_file *seq, void *v)
3026 if (v == &proto_list)
3027 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3036 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3038 proto_seq_printf(seq, list_entry(v, struct proto, node));
3042 static const struct seq_operations proto_seq_ops = {
3043 .start = proto_seq_start,
3044 .next = proto_seq_next,
3045 .stop = proto_seq_stop,
3046 .show = proto_seq_show,
3049 static int proto_seq_open(struct inode *inode, struct file *file)
3051 return seq_open_net(inode, file, &proto_seq_ops,
3052 sizeof(struct seq_net_private));
3055 static const struct file_operations proto_seq_fops = {
3056 .owner = THIS_MODULE,
3057 .open = proto_seq_open,
3059 .llseek = seq_lseek,
3060 .release = seq_release_net,
3063 static __net_init int proto_init_net(struct net *net)
3065 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3071 static __net_exit void proto_exit_net(struct net *net)
3073 remove_proc_entry("protocols", net->proc_net);
3077 static __net_initdata struct pernet_operations proto_net_ops = {
3078 .init = proto_init_net,
3079 .exit = proto_exit_net,
3082 static int __init proto_init(void)
3084 return register_pernet_subsys(&proto_net_ops);
3087 subsys_initcall(proto_init);
3089 #endif /* PROC_FS */