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 <linux/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(int, 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(int, val * 2, SOCK_MIN_RCVBUF);
758 if (!capable(CAP_NET_ADMIN)) {
765 if (sk->sk_prot->keepalive)
766 sk->sk_prot->keepalive(sk, valbool);
767 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
771 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
775 sk->sk_no_check_tx = valbool;
779 if ((val >= 0 && val <= 6) ||
780 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
781 sk->sk_priority = val;
787 if (optlen < sizeof(ling)) {
788 ret = -EINVAL; /* 1003.1g */
791 if (copy_from_user(&ling, optval, sizeof(ling))) {
796 sock_reset_flag(sk, SOCK_LINGER);
798 #if (BITS_PER_LONG == 32)
799 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
800 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
803 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
804 sock_set_flag(sk, SOCK_LINGER);
809 sock_warn_obsolete_bsdism("setsockopt");
814 set_bit(SOCK_PASSCRED, &sock->flags);
816 clear_bit(SOCK_PASSCRED, &sock->flags);
822 if (optname == SO_TIMESTAMP)
823 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
825 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
826 sock_set_flag(sk, SOCK_RCVTSTAMP);
827 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
829 sock_reset_flag(sk, SOCK_RCVTSTAMP);
830 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
834 case SO_TIMESTAMPING:
835 if (val & ~SOF_TIMESTAMPING_MASK) {
840 if (val & SOF_TIMESTAMPING_OPT_ID &&
841 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
842 if (sk->sk_protocol == IPPROTO_TCP &&
843 sk->sk_type == SOCK_STREAM) {
844 if ((1 << sk->sk_state) &
845 (TCPF_CLOSE | TCPF_LISTEN)) {
849 sk->sk_tskey = tcp_sk(sk)->snd_una;
855 if (val & SOF_TIMESTAMPING_OPT_STATS &&
856 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
861 sk->sk_tsflags = val;
862 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
863 sock_enable_timestamp(sk,
864 SOCK_TIMESTAMPING_RX_SOFTWARE);
866 sock_disable_timestamp(sk,
867 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
873 sk->sk_rcvlowat = val ? : 1;
877 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
881 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
884 case SO_ATTACH_FILTER:
886 if (optlen == sizeof(struct sock_fprog)) {
887 struct sock_fprog fprog;
890 if (copy_from_user(&fprog, optval, sizeof(fprog)))
893 ret = sk_attach_filter(&fprog, sk);
899 if (optlen == sizeof(u32)) {
903 if (copy_from_user(&ufd, optval, sizeof(ufd)))
906 ret = sk_attach_bpf(ufd, sk);
910 case SO_ATTACH_REUSEPORT_CBPF:
912 if (optlen == sizeof(struct sock_fprog)) {
913 struct sock_fprog fprog;
916 if (copy_from_user(&fprog, optval, sizeof(fprog)))
919 ret = sk_reuseport_attach_filter(&fprog, sk);
923 case SO_ATTACH_REUSEPORT_EBPF:
925 if (optlen == sizeof(u32)) {
929 if (copy_from_user(&ufd, optval, sizeof(ufd)))
932 ret = sk_reuseport_attach_bpf(ufd, sk);
936 case SO_DETACH_FILTER:
937 ret = sk_detach_filter(sk);
941 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
944 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
949 set_bit(SOCK_PASSSEC, &sock->flags);
951 clear_bit(SOCK_PASSSEC, &sock->flags);
954 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
961 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
965 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
969 if (sock->ops->set_peek_off)
970 ret = sock->ops->set_peek_off(sk, val);
976 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
979 case SO_SELECT_ERR_QUEUE:
980 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
983 #ifdef CONFIG_NET_RX_BUSY_POLL
985 /* allow unprivileged users to decrease the value */
986 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
992 sk->sk_ll_usec = val;
997 case SO_MAX_PACING_RATE:
998 sk->sk_max_pacing_rate = val;
999 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1000 sk->sk_max_pacing_rate);
1003 case SO_INCOMING_CPU:
1004 sk->sk_incoming_cpu = val;
1009 dst_negative_advice(sk);
1018 EXPORT_SYMBOL(sock_setsockopt);
1021 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1022 struct ucred *ucred)
1024 ucred->pid = pid_vnr(pid);
1025 ucred->uid = ucred->gid = -1;
1027 struct user_namespace *current_ns = current_user_ns();
1029 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1030 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1034 int sock_getsockopt(struct socket *sock, int level, int optname,
1035 char __user *optval, int __user *optlen)
1037 struct sock *sk = sock->sk;
1045 int lv = sizeof(int);
1048 if (get_user(len, optlen))
1053 memset(&v, 0, sizeof(v));
1057 v.val = sock_flag(sk, SOCK_DBG);
1061 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1065 v.val = sock_flag(sk, SOCK_BROADCAST);
1069 v.val = sk->sk_sndbuf;
1073 v.val = sk->sk_rcvbuf;
1077 v.val = sk->sk_reuse;
1081 v.val = sk->sk_reuseport;
1085 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1089 v.val = sk->sk_type;
1093 v.val = sk->sk_protocol;
1097 v.val = sk->sk_family;
1101 v.val = -sock_error(sk);
1103 v.val = xchg(&sk->sk_err_soft, 0);
1107 v.val = sock_flag(sk, SOCK_URGINLINE);
1111 v.val = sk->sk_no_check_tx;
1115 v.val = sk->sk_priority;
1119 lv = sizeof(v.ling);
1120 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1121 v.ling.l_linger = sk->sk_lingertime / HZ;
1125 sock_warn_obsolete_bsdism("getsockopt");
1129 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1130 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1133 case SO_TIMESTAMPNS:
1134 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1137 case SO_TIMESTAMPING:
1138 v.val = sk->sk_tsflags;
1142 lv = sizeof(struct timeval);
1143 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1147 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1148 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1153 lv = sizeof(struct timeval);
1154 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1158 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1159 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1164 v.val = sk->sk_rcvlowat;
1172 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1177 struct ucred peercred;
1178 if (len > sizeof(peercred))
1179 len = sizeof(peercred);
1180 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1181 if (copy_to_user(optval, &peercred, len))
1190 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1194 if (copy_to_user(optval, address, len))
1199 /* Dubious BSD thing... Probably nobody even uses it, but
1200 * the UNIX standard wants it for whatever reason... -DaveM
1203 v.val = sk->sk_state == TCP_LISTEN;
1207 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1211 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1214 v.val = sk->sk_mark;
1218 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1221 case SO_WIFI_STATUS:
1222 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1226 if (!sock->ops->set_peek_off)
1229 v.val = sk->sk_peek_off;
1232 v.val = sock_flag(sk, SOCK_NOFCS);
1235 case SO_BINDTODEVICE:
1236 return sock_getbindtodevice(sk, optval, optlen, len);
1239 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1245 case SO_LOCK_FILTER:
1246 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1249 case SO_BPF_EXTENSIONS:
1250 v.val = bpf_tell_extensions();
1253 case SO_SELECT_ERR_QUEUE:
1254 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1257 #ifdef CONFIG_NET_RX_BUSY_POLL
1259 v.val = sk->sk_ll_usec;
1263 case SO_MAX_PACING_RATE:
1264 v.val = sk->sk_max_pacing_rate;
1267 case SO_INCOMING_CPU:
1268 v.val = sk->sk_incoming_cpu;
1272 /* We implement the SO_SNDLOWAT etc to not be settable
1275 return -ENOPROTOOPT;
1280 if (copy_to_user(optval, &v, len))
1283 if (put_user(len, optlen))
1289 * Initialize an sk_lock.
1291 * (We also register the sk_lock with the lock validator.)
1293 static inline void sock_lock_init(struct sock *sk)
1295 sock_lock_init_class_and_name(sk,
1296 af_family_slock_key_strings[sk->sk_family],
1297 af_family_slock_keys + sk->sk_family,
1298 af_family_key_strings[sk->sk_family],
1299 af_family_keys + sk->sk_family);
1303 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1304 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1305 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1307 static void sock_copy(struct sock *nsk, const struct sock *osk)
1309 #ifdef CONFIG_SECURITY_NETWORK
1310 void *sptr = nsk->sk_security;
1312 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1314 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1315 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1317 #ifdef CONFIG_SECURITY_NETWORK
1318 nsk->sk_security = sptr;
1319 security_sk_clone(osk, nsk);
1323 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1327 struct kmem_cache *slab;
1331 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1334 if (priority & __GFP_ZERO)
1335 sk_prot_clear_nulls(sk, prot->obj_size);
1337 sk = kmalloc(prot->obj_size, priority);
1340 kmemcheck_annotate_bitfield(sk, flags);
1342 if (security_sk_alloc(sk, family, priority))
1345 if (!try_module_get(prot->owner))
1347 sk_tx_queue_clear(sk);
1353 security_sk_free(sk);
1356 kmem_cache_free(slab, sk);
1362 static void sk_prot_free(struct proto *prot, struct sock *sk)
1364 struct kmem_cache *slab;
1365 struct module *owner;
1367 owner = prot->owner;
1370 cgroup_sk_free(&sk->sk_cgrp_data);
1371 mem_cgroup_sk_free(sk);
1372 security_sk_free(sk);
1374 kmem_cache_free(slab, sk);
1381 * sk_alloc - All socket objects are allocated here
1382 * @net: the applicable net namespace
1383 * @family: protocol family
1384 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1385 * @prot: struct proto associated with this new sock instance
1386 * @kern: is this to be a kernel socket?
1388 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1389 struct proto *prot, int kern)
1393 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1395 sk->sk_family = family;
1397 * See comment in struct sock definition to understand
1398 * why we need sk_prot_creator -acme
1400 sk->sk_prot = sk->sk_prot_creator = prot;
1402 sk->sk_net_refcnt = kern ? 0 : 1;
1403 if (likely(sk->sk_net_refcnt))
1405 sock_net_set(sk, net);
1406 atomic_set(&sk->sk_wmem_alloc, 1);
1408 mem_cgroup_sk_alloc(sk);
1409 cgroup_sk_alloc(&sk->sk_cgrp_data);
1410 sock_update_classid(&sk->sk_cgrp_data);
1411 sock_update_netprioidx(&sk->sk_cgrp_data);
1416 EXPORT_SYMBOL(sk_alloc);
1418 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1419 * grace period. This is the case for UDP sockets and TCP listeners.
1421 static void __sk_destruct(struct rcu_head *head)
1423 struct sock *sk = container_of(head, struct sock, sk_rcu);
1424 struct sk_filter *filter;
1426 if (sk->sk_destruct)
1427 sk->sk_destruct(sk);
1429 filter = rcu_dereference_check(sk->sk_filter,
1430 atomic_read(&sk->sk_wmem_alloc) == 0);
1432 sk_filter_uncharge(sk, filter);
1433 RCU_INIT_POINTER(sk->sk_filter, NULL);
1435 if (rcu_access_pointer(sk->sk_reuseport_cb))
1436 reuseport_detach_sock(sk);
1438 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1440 if (atomic_read(&sk->sk_omem_alloc))
1441 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1442 __func__, atomic_read(&sk->sk_omem_alloc));
1444 if (sk->sk_peer_cred)
1445 put_cred(sk->sk_peer_cred);
1446 put_pid(sk->sk_peer_pid);
1447 if (likely(sk->sk_net_refcnt))
1448 put_net(sock_net(sk));
1449 sk_prot_free(sk->sk_prot_creator, sk);
1452 void sk_destruct(struct sock *sk)
1454 if (sock_flag(sk, SOCK_RCU_FREE))
1455 call_rcu(&sk->sk_rcu, __sk_destruct);
1457 __sk_destruct(&sk->sk_rcu);
1460 static void __sk_free(struct sock *sk)
1462 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1463 sock_diag_broadcast_destroy(sk);
1468 void sk_free(struct sock *sk)
1471 * We subtract one from sk_wmem_alloc and can know if
1472 * some packets are still in some tx queue.
1473 * If not null, sock_wfree() will call __sk_free(sk) later
1475 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1478 EXPORT_SYMBOL(sk_free);
1481 * sk_clone_lock - clone a socket, and lock its clone
1482 * @sk: the socket to clone
1483 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1485 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1487 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1490 bool is_charged = true;
1492 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1493 if (newsk != NULL) {
1494 struct sk_filter *filter;
1496 sock_copy(newsk, sk);
1499 if (likely(newsk->sk_net_refcnt))
1500 get_net(sock_net(newsk));
1501 sk_node_init(&newsk->sk_node);
1502 sock_lock_init(newsk);
1503 bh_lock_sock(newsk);
1504 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1505 newsk->sk_backlog.len = 0;
1507 atomic_set(&newsk->sk_rmem_alloc, 0);
1509 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1511 atomic_set(&newsk->sk_wmem_alloc, 1);
1512 atomic_set(&newsk->sk_omem_alloc, 0);
1513 skb_queue_head_init(&newsk->sk_receive_queue);
1514 skb_queue_head_init(&newsk->sk_write_queue);
1516 rwlock_init(&newsk->sk_callback_lock);
1517 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1518 af_callback_keys + newsk->sk_family,
1519 af_family_clock_key_strings[newsk->sk_family]);
1521 newsk->sk_dst_cache = NULL;
1522 newsk->sk_wmem_queued = 0;
1523 newsk->sk_forward_alloc = 0;
1524 atomic_set(&newsk->sk_drops, 0);
1525 newsk->sk_send_head = NULL;
1526 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1528 sock_reset_flag(newsk, SOCK_DONE);
1529 skb_queue_head_init(&newsk->sk_error_queue);
1531 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1533 /* though it's an empty new sock, the charging may fail
1534 * if sysctl_optmem_max was changed between creation of
1535 * original socket and cloning
1537 is_charged = sk_filter_charge(newsk, filter);
1539 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1540 /* It is still raw copy of parent, so invalidate
1541 * destructor and make plain sk_free() */
1542 newsk->sk_destruct = NULL;
1543 bh_unlock_sock(newsk);
1548 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1551 newsk->sk_err_soft = 0;
1552 newsk->sk_priority = 0;
1553 newsk->sk_incoming_cpu = raw_smp_processor_id();
1554 atomic64_set(&newsk->sk_cookie, 0);
1556 mem_cgroup_sk_alloc(newsk);
1557 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1560 * Before updating sk_refcnt, we must commit prior changes to memory
1561 * (Documentation/RCU/rculist_nulls.txt for details)
1564 atomic_set(&newsk->sk_refcnt, 2);
1567 * Increment the counter in the same struct proto as the master
1568 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1569 * is the same as sk->sk_prot->socks, as this field was copied
1572 * This _changes_ the previous behaviour, where
1573 * tcp_create_openreq_child always was incrementing the
1574 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1575 * to be taken into account in all callers. -acme
1577 sk_refcnt_debug_inc(newsk);
1578 sk_set_socket(newsk, NULL);
1579 newsk->sk_wq = NULL;
1581 if (newsk->sk_prot->sockets_allocated)
1582 sk_sockets_allocated_inc(newsk);
1584 if (sock_needs_netstamp(sk) &&
1585 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1586 net_enable_timestamp();
1591 EXPORT_SYMBOL_GPL(sk_clone_lock);
1593 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1597 sk_dst_set(sk, dst);
1598 sk->sk_route_caps = dst->dev->features;
1599 if (sk->sk_route_caps & NETIF_F_GSO)
1600 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1601 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1602 if (sk_can_gso(sk)) {
1603 if (dst->header_len) {
1604 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1606 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1607 sk->sk_gso_max_size = dst->dev->gso_max_size;
1608 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1611 sk->sk_gso_max_segs = max_segs;
1613 EXPORT_SYMBOL_GPL(sk_setup_caps);
1616 * Simple resource managers for sockets.
1621 * Write buffer destructor automatically called from kfree_skb.
1623 void sock_wfree(struct sk_buff *skb)
1625 struct sock *sk = skb->sk;
1626 unsigned int len = skb->truesize;
1628 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1630 * Keep a reference on sk_wmem_alloc, this will be released
1631 * after sk_write_space() call
1633 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1634 sk->sk_write_space(sk);
1638 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1639 * could not do because of in-flight packets
1641 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1644 EXPORT_SYMBOL(sock_wfree);
1646 /* This variant of sock_wfree() is used by TCP,
1647 * since it sets SOCK_USE_WRITE_QUEUE.
1649 void __sock_wfree(struct sk_buff *skb)
1651 struct sock *sk = skb->sk;
1653 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1657 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1662 if (unlikely(!sk_fullsock(sk))) {
1663 skb->destructor = sock_edemux;
1668 skb->destructor = sock_wfree;
1669 skb_set_hash_from_sk(skb, sk);
1671 * We used to take a refcount on sk, but following operation
1672 * is enough to guarantee sk_free() wont free this sock until
1673 * all in-flight packets are completed
1675 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1677 EXPORT_SYMBOL(skb_set_owner_w);
1679 /* This helper is used by netem, as it can hold packets in its
1680 * delay queue. We want to allow the owner socket to send more
1681 * packets, as if they were already TX completed by a typical driver.
1682 * But we also want to keep skb->sk set because some packet schedulers
1683 * rely on it (sch_fq for example). So we set skb->truesize to a small
1684 * amount (1) and decrease sk_wmem_alloc accordingly.
1686 void skb_orphan_partial(struct sk_buff *skb)
1688 /* If this skb is a TCP pure ACK or already went here,
1689 * we have nothing to do. 2 is already a very small truesize.
1691 if (skb->truesize <= 2)
1694 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1695 * so we do not completely orphan skb, but transfert all
1696 * accounted bytes but one, to avoid unexpected reorders.
1698 if (skb->destructor == sock_wfree
1700 || skb->destructor == tcp_wfree
1703 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1709 EXPORT_SYMBOL(skb_orphan_partial);
1712 * Read buffer destructor automatically called from kfree_skb.
1714 void sock_rfree(struct sk_buff *skb)
1716 struct sock *sk = skb->sk;
1717 unsigned int len = skb->truesize;
1719 atomic_sub(len, &sk->sk_rmem_alloc);
1720 sk_mem_uncharge(sk, len);
1722 EXPORT_SYMBOL(sock_rfree);
1725 * Buffer destructor for skbs that are not used directly in read or write
1726 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1728 void sock_efree(struct sk_buff *skb)
1732 EXPORT_SYMBOL(sock_efree);
1734 kuid_t sock_i_uid(struct sock *sk)
1738 read_lock_bh(&sk->sk_callback_lock);
1739 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1740 read_unlock_bh(&sk->sk_callback_lock);
1743 EXPORT_SYMBOL(sock_i_uid);
1745 unsigned long sock_i_ino(struct sock *sk)
1749 read_lock_bh(&sk->sk_callback_lock);
1750 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1751 read_unlock_bh(&sk->sk_callback_lock);
1754 EXPORT_SYMBOL(sock_i_ino);
1757 * Allocate a skb from the socket's send buffer.
1759 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1762 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1763 struct sk_buff *skb = alloc_skb(size, priority);
1765 skb_set_owner_w(skb, sk);
1771 EXPORT_SYMBOL(sock_wmalloc);
1774 * Allocate a memory block from the socket's option memory buffer.
1776 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1778 if ((unsigned int)size <= sysctl_optmem_max &&
1779 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1781 /* First do the add, to avoid the race if kmalloc
1784 atomic_add(size, &sk->sk_omem_alloc);
1785 mem = kmalloc(size, priority);
1788 atomic_sub(size, &sk->sk_omem_alloc);
1792 EXPORT_SYMBOL(sock_kmalloc);
1794 /* Free an option memory block. Note, we actually want the inline
1795 * here as this allows gcc to detect the nullify and fold away the
1796 * condition entirely.
1798 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1801 if (WARN_ON_ONCE(!mem))
1807 atomic_sub(size, &sk->sk_omem_alloc);
1810 void sock_kfree_s(struct sock *sk, void *mem, int size)
1812 __sock_kfree_s(sk, mem, size, false);
1814 EXPORT_SYMBOL(sock_kfree_s);
1816 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1818 __sock_kfree_s(sk, mem, size, true);
1820 EXPORT_SYMBOL(sock_kzfree_s);
1822 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1823 I think, these locks should be removed for datagram sockets.
1825 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1829 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1833 if (signal_pending(current))
1835 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1836 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1837 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1839 if (sk->sk_shutdown & SEND_SHUTDOWN)
1843 timeo = schedule_timeout(timeo);
1845 finish_wait(sk_sleep(sk), &wait);
1851 * Generic send/receive buffer handlers
1854 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1855 unsigned long data_len, int noblock,
1856 int *errcode, int max_page_order)
1858 struct sk_buff *skb;
1862 timeo = sock_sndtimeo(sk, noblock);
1864 err = sock_error(sk);
1869 if (sk->sk_shutdown & SEND_SHUTDOWN)
1872 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1875 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1876 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1880 if (signal_pending(current))
1882 timeo = sock_wait_for_wmem(sk, timeo);
1884 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1885 errcode, sk->sk_allocation);
1887 skb_set_owner_w(skb, sk);
1891 err = sock_intr_errno(timeo);
1896 EXPORT_SYMBOL(sock_alloc_send_pskb);
1898 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1899 int noblock, int *errcode)
1901 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1903 EXPORT_SYMBOL(sock_alloc_send_skb);
1905 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1906 struct sockcm_cookie *sockc)
1910 switch (cmsg->cmsg_type) {
1912 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1914 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1916 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1918 case SO_TIMESTAMPING:
1919 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1922 tsflags = *(u32 *)CMSG_DATA(cmsg);
1923 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1926 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1927 sockc->tsflags |= tsflags;
1929 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1931 case SCM_CREDENTIALS:
1938 EXPORT_SYMBOL(__sock_cmsg_send);
1940 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1941 struct sockcm_cookie *sockc)
1943 struct cmsghdr *cmsg;
1946 for_each_cmsghdr(cmsg, msg) {
1947 if (!CMSG_OK(msg, cmsg))
1949 if (cmsg->cmsg_level != SOL_SOCKET)
1951 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1957 EXPORT_SYMBOL(sock_cmsg_send);
1959 /* On 32bit arches, an skb frag is limited to 2^15 */
1960 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1963 * skb_page_frag_refill - check that a page_frag contains enough room
1964 * @sz: minimum size of the fragment we want to get
1965 * @pfrag: pointer to page_frag
1966 * @gfp: priority for memory allocation
1968 * Note: While this allocator tries to use high order pages, there is
1969 * no guarantee that allocations succeed. Therefore, @sz MUST be
1970 * less or equal than PAGE_SIZE.
1972 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1975 if (page_ref_count(pfrag->page) == 1) {
1979 if (pfrag->offset + sz <= pfrag->size)
1981 put_page(pfrag->page);
1985 if (SKB_FRAG_PAGE_ORDER) {
1986 /* Avoid direct reclaim but allow kswapd to wake */
1987 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1988 __GFP_COMP | __GFP_NOWARN |
1990 SKB_FRAG_PAGE_ORDER);
1991 if (likely(pfrag->page)) {
1992 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1996 pfrag->page = alloc_page(gfp);
1997 if (likely(pfrag->page)) {
1998 pfrag->size = PAGE_SIZE;
2003 EXPORT_SYMBOL(skb_page_frag_refill);
2005 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2007 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2010 sk_enter_memory_pressure(sk);
2011 sk_stream_moderate_sndbuf(sk);
2014 EXPORT_SYMBOL(sk_page_frag_refill);
2016 static void __lock_sock(struct sock *sk)
2017 __releases(&sk->sk_lock.slock)
2018 __acquires(&sk->sk_lock.slock)
2023 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2024 TASK_UNINTERRUPTIBLE);
2025 spin_unlock_bh(&sk->sk_lock.slock);
2027 spin_lock_bh(&sk->sk_lock.slock);
2028 if (!sock_owned_by_user(sk))
2031 finish_wait(&sk->sk_lock.wq, &wait);
2034 static void __release_sock(struct sock *sk)
2035 __releases(&sk->sk_lock.slock)
2036 __acquires(&sk->sk_lock.slock)
2038 struct sk_buff *skb, *next;
2040 while ((skb = sk->sk_backlog.head) != NULL) {
2041 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2043 spin_unlock_bh(&sk->sk_lock.slock);
2048 WARN_ON_ONCE(skb_dst_is_noref(skb));
2050 sk_backlog_rcv(sk, skb);
2055 } while (skb != NULL);
2057 spin_lock_bh(&sk->sk_lock.slock);
2061 * Doing the zeroing here guarantee we can not loop forever
2062 * while a wild producer attempts to flood us.
2064 sk->sk_backlog.len = 0;
2067 void __sk_flush_backlog(struct sock *sk)
2069 spin_lock_bh(&sk->sk_lock.slock);
2071 spin_unlock_bh(&sk->sk_lock.slock);
2075 * sk_wait_data - wait for data to arrive at sk_receive_queue
2076 * @sk: sock to wait on
2077 * @timeo: for how long
2078 * @skb: last skb seen on sk_receive_queue
2080 * Now socket state including sk->sk_err is changed only under lock,
2081 * hence we may omit checks after joining wait queue.
2082 * We check receive queue before schedule() only as optimization;
2083 * it is very likely that release_sock() added new data.
2085 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2087 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2090 add_wait_queue(sk_sleep(sk), &wait);
2091 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2092 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2093 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2094 remove_wait_queue(sk_sleep(sk), &wait);
2097 EXPORT_SYMBOL(sk_wait_data);
2100 * __sk_mem_raise_allocated - increase memory_allocated
2102 * @size: memory size to allocate
2103 * @amt: pages to allocate
2104 * @kind: allocation type
2106 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2108 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2110 struct proto *prot = sk->sk_prot;
2111 long allocated = sk_memory_allocated_add(sk, amt);
2113 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2114 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2115 goto suppress_allocation;
2118 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2119 sk_leave_memory_pressure(sk);
2123 /* Under pressure. */
2124 if (allocated > sk_prot_mem_limits(sk, 1))
2125 sk_enter_memory_pressure(sk);
2127 /* Over hard limit. */
2128 if (allocated > sk_prot_mem_limits(sk, 2))
2129 goto suppress_allocation;
2131 /* guarantee minimum buffer size under pressure */
2132 if (kind == SK_MEM_RECV) {
2133 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2136 } else { /* SK_MEM_SEND */
2137 if (sk->sk_type == SOCK_STREAM) {
2138 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2140 } else if (atomic_read(&sk->sk_wmem_alloc) <
2141 prot->sysctl_wmem[0])
2145 if (sk_has_memory_pressure(sk)) {
2148 if (!sk_under_memory_pressure(sk))
2150 alloc = sk_sockets_allocated_read_positive(sk);
2151 if (sk_prot_mem_limits(sk, 2) > alloc *
2152 sk_mem_pages(sk->sk_wmem_queued +
2153 atomic_read(&sk->sk_rmem_alloc) +
2154 sk->sk_forward_alloc))
2158 suppress_allocation:
2160 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2161 sk_stream_moderate_sndbuf(sk);
2163 /* Fail only if socket is _under_ its sndbuf.
2164 * In this case we cannot block, so that we have to fail.
2166 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2170 trace_sock_exceed_buf_limit(sk, prot, allocated);
2172 sk_memory_allocated_sub(sk, amt);
2174 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2175 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2179 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2182 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2184 * @size: memory size to allocate
2185 * @kind: allocation type
2187 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2188 * rmem allocation. This function assumes that protocols which have
2189 * memory_pressure use sk_wmem_queued as write buffer accounting.
2191 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2193 int ret, amt = sk_mem_pages(size);
2195 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2196 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2198 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2201 EXPORT_SYMBOL(__sk_mem_schedule);
2204 * __sk_mem_reduce_allocated - reclaim memory_allocated
2206 * @amount: number of quanta
2208 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2210 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2212 sk_memory_allocated_sub(sk, amount);
2214 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2215 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2217 if (sk_under_memory_pressure(sk) &&
2218 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2219 sk_leave_memory_pressure(sk);
2221 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2224 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2226 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2228 void __sk_mem_reclaim(struct sock *sk, int amount)
2230 amount >>= SK_MEM_QUANTUM_SHIFT;
2231 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2232 __sk_mem_reduce_allocated(sk, amount);
2234 EXPORT_SYMBOL(__sk_mem_reclaim);
2236 int sk_set_peek_off(struct sock *sk, int val)
2241 sk->sk_peek_off = val;
2244 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2247 * Set of default routines for initialising struct proto_ops when
2248 * the protocol does not support a particular function. In certain
2249 * cases where it makes no sense for a protocol to have a "do nothing"
2250 * function, some default processing is provided.
2253 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2257 EXPORT_SYMBOL(sock_no_bind);
2259 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2264 EXPORT_SYMBOL(sock_no_connect);
2266 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2270 EXPORT_SYMBOL(sock_no_socketpair);
2272 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2276 EXPORT_SYMBOL(sock_no_accept);
2278 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2283 EXPORT_SYMBOL(sock_no_getname);
2285 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2289 EXPORT_SYMBOL(sock_no_poll);
2291 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2295 EXPORT_SYMBOL(sock_no_ioctl);
2297 int sock_no_listen(struct socket *sock, int backlog)
2301 EXPORT_SYMBOL(sock_no_listen);
2303 int sock_no_shutdown(struct socket *sock, int how)
2307 EXPORT_SYMBOL(sock_no_shutdown);
2309 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2310 char __user *optval, unsigned int optlen)
2314 EXPORT_SYMBOL(sock_no_setsockopt);
2316 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2317 char __user *optval, int __user *optlen)
2321 EXPORT_SYMBOL(sock_no_getsockopt);
2323 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2327 EXPORT_SYMBOL(sock_no_sendmsg);
2329 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2334 EXPORT_SYMBOL(sock_no_recvmsg);
2336 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2338 /* Mirror missing mmap method error code */
2341 EXPORT_SYMBOL(sock_no_mmap);
2343 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2346 struct msghdr msg = {.msg_flags = flags};
2348 char *kaddr = kmap(page);
2349 iov.iov_base = kaddr + offset;
2351 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2355 EXPORT_SYMBOL(sock_no_sendpage);
2358 * Default Socket Callbacks
2361 static void sock_def_wakeup(struct sock *sk)
2363 struct socket_wq *wq;
2366 wq = rcu_dereference(sk->sk_wq);
2367 if (skwq_has_sleeper(wq))
2368 wake_up_interruptible_all(&wq->wait);
2372 static void sock_def_error_report(struct sock *sk)
2374 struct socket_wq *wq;
2377 wq = rcu_dereference(sk->sk_wq);
2378 if (skwq_has_sleeper(wq))
2379 wake_up_interruptible_poll(&wq->wait, POLLERR);
2380 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2384 static void sock_def_readable(struct sock *sk)
2386 struct socket_wq *wq;
2389 wq = rcu_dereference(sk->sk_wq);
2390 if (skwq_has_sleeper(wq))
2391 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2392 POLLRDNORM | POLLRDBAND);
2393 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2397 static void sock_def_write_space(struct sock *sk)
2399 struct socket_wq *wq;
2403 /* Do not wake up a writer until he can make "significant"
2406 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2407 wq = rcu_dereference(sk->sk_wq);
2408 if (skwq_has_sleeper(wq))
2409 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2410 POLLWRNORM | POLLWRBAND);
2412 /* Should agree with poll, otherwise some programs break */
2413 if (sock_writeable(sk))
2414 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2420 static void sock_def_destruct(struct sock *sk)
2424 void sk_send_sigurg(struct sock *sk)
2426 if (sk->sk_socket && sk->sk_socket->file)
2427 if (send_sigurg(&sk->sk_socket->file->f_owner))
2428 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2430 EXPORT_SYMBOL(sk_send_sigurg);
2432 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2433 unsigned long expires)
2435 if (!mod_timer(timer, expires))
2438 EXPORT_SYMBOL(sk_reset_timer);
2440 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2442 if (del_timer(timer))
2445 EXPORT_SYMBOL(sk_stop_timer);
2447 void sock_init_data(struct socket *sock, struct sock *sk)
2449 skb_queue_head_init(&sk->sk_receive_queue);
2450 skb_queue_head_init(&sk->sk_write_queue);
2451 skb_queue_head_init(&sk->sk_error_queue);
2453 sk->sk_send_head = NULL;
2455 init_timer(&sk->sk_timer);
2457 sk->sk_allocation = GFP_KERNEL;
2458 sk->sk_rcvbuf = sysctl_rmem_default;
2459 sk->sk_sndbuf = sysctl_wmem_default;
2460 sk->sk_state = TCP_CLOSE;
2461 sk_set_socket(sk, sock);
2463 sock_set_flag(sk, SOCK_ZAPPED);
2466 sk->sk_type = sock->type;
2467 sk->sk_wq = sock->wq;
2469 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2472 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2475 rwlock_init(&sk->sk_callback_lock);
2476 lockdep_set_class_and_name(&sk->sk_callback_lock,
2477 af_callback_keys + sk->sk_family,
2478 af_family_clock_key_strings[sk->sk_family]);
2480 sk->sk_state_change = sock_def_wakeup;
2481 sk->sk_data_ready = sock_def_readable;
2482 sk->sk_write_space = sock_def_write_space;
2483 sk->sk_error_report = sock_def_error_report;
2484 sk->sk_destruct = sock_def_destruct;
2486 sk->sk_frag.page = NULL;
2487 sk->sk_frag.offset = 0;
2488 sk->sk_peek_off = -1;
2490 sk->sk_peer_pid = NULL;
2491 sk->sk_peer_cred = NULL;
2492 sk->sk_write_pending = 0;
2493 sk->sk_rcvlowat = 1;
2494 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2495 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2497 sk->sk_stamp = ktime_set(-1L, 0);
2499 #ifdef CONFIG_NET_RX_BUSY_POLL
2501 sk->sk_ll_usec = sysctl_net_busy_read;
2504 sk->sk_max_pacing_rate = ~0U;
2505 sk->sk_pacing_rate = ~0U;
2506 sk->sk_incoming_cpu = -1;
2508 * Before updating sk_refcnt, we must commit prior changes to memory
2509 * (Documentation/RCU/rculist_nulls.txt for details)
2512 atomic_set(&sk->sk_refcnt, 1);
2513 atomic_set(&sk->sk_drops, 0);
2515 EXPORT_SYMBOL(sock_init_data);
2517 void lock_sock_nested(struct sock *sk, int subclass)
2520 spin_lock_bh(&sk->sk_lock.slock);
2521 if (sk->sk_lock.owned)
2523 sk->sk_lock.owned = 1;
2524 spin_unlock(&sk->sk_lock.slock);
2526 * The sk_lock has mutex_lock() semantics here:
2528 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2531 EXPORT_SYMBOL(lock_sock_nested);
2533 void release_sock(struct sock *sk)
2535 spin_lock_bh(&sk->sk_lock.slock);
2536 if (sk->sk_backlog.tail)
2539 /* Warning : release_cb() might need to release sk ownership,
2540 * ie call sock_release_ownership(sk) before us.
2542 if (sk->sk_prot->release_cb)
2543 sk->sk_prot->release_cb(sk);
2545 sock_release_ownership(sk);
2546 if (waitqueue_active(&sk->sk_lock.wq))
2547 wake_up(&sk->sk_lock.wq);
2548 spin_unlock_bh(&sk->sk_lock.slock);
2550 EXPORT_SYMBOL(release_sock);
2553 * lock_sock_fast - fast version of lock_sock
2556 * This version should be used for very small section, where process wont block
2557 * return false if fast path is taken
2558 * sk_lock.slock locked, owned = 0, BH disabled
2559 * return true if slow path is taken
2560 * sk_lock.slock unlocked, owned = 1, BH enabled
2562 bool lock_sock_fast(struct sock *sk)
2565 spin_lock_bh(&sk->sk_lock.slock);
2567 if (!sk->sk_lock.owned)
2569 * Note : We must disable BH
2574 sk->sk_lock.owned = 1;
2575 spin_unlock(&sk->sk_lock.slock);
2577 * The sk_lock has mutex_lock() semantics here:
2579 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2583 EXPORT_SYMBOL(lock_sock_fast);
2585 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2588 if (!sock_flag(sk, SOCK_TIMESTAMP))
2589 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2590 tv = ktime_to_timeval(sk->sk_stamp);
2591 if (tv.tv_sec == -1)
2593 if (tv.tv_sec == 0) {
2594 sk->sk_stamp = ktime_get_real();
2595 tv = ktime_to_timeval(sk->sk_stamp);
2597 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2599 EXPORT_SYMBOL(sock_get_timestamp);
2601 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2604 if (!sock_flag(sk, SOCK_TIMESTAMP))
2605 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2606 ts = ktime_to_timespec(sk->sk_stamp);
2607 if (ts.tv_sec == -1)
2609 if (ts.tv_sec == 0) {
2610 sk->sk_stamp = ktime_get_real();
2611 ts = ktime_to_timespec(sk->sk_stamp);
2613 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2615 EXPORT_SYMBOL(sock_get_timestampns);
2617 void sock_enable_timestamp(struct sock *sk, int flag)
2619 if (!sock_flag(sk, flag)) {
2620 unsigned long previous_flags = sk->sk_flags;
2622 sock_set_flag(sk, flag);
2624 * we just set one of the two flags which require net
2625 * time stamping, but time stamping might have been on
2626 * already because of the other one
2628 if (sock_needs_netstamp(sk) &&
2629 !(previous_flags & SK_FLAGS_TIMESTAMP))
2630 net_enable_timestamp();
2634 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2635 int level, int type)
2637 struct sock_exterr_skb *serr;
2638 struct sk_buff *skb;
2642 skb = sock_dequeue_err_skb(sk);
2648 msg->msg_flags |= MSG_TRUNC;
2651 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2655 sock_recv_timestamp(msg, sk, skb);
2657 serr = SKB_EXT_ERR(skb);
2658 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2660 msg->msg_flags |= MSG_ERRQUEUE;
2668 EXPORT_SYMBOL(sock_recv_errqueue);
2671 * Get a socket option on an socket.
2673 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2674 * asynchronous errors should be reported by getsockopt. We assume
2675 * this means if you specify SO_ERROR (otherwise whats the point of it).
2677 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2678 char __user *optval, int __user *optlen)
2680 struct sock *sk = sock->sk;
2682 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2684 EXPORT_SYMBOL(sock_common_getsockopt);
2686 #ifdef CONFIG_COMPAT
2687 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2688 char __user *optval, int __user *optlen)
2690 struct sock *sk = sock->sk;
2692 if (sk->sk_prot->compat_getsockopt != NULL)
2693 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2695 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2697 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2700 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2703 struct sock *sk = sock->sk;
2707 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2708 flags & ~MSG_DONTWAIT, &addr_len);
2710 msg->msg_namelen = addr_len;
2713 EXPORT_SYMBOL(sock_common_recvmsg);
2716 * Set socket options on an inet socket.
2718 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2719 char __user *optval, unsigned int optlen)
2721 struct sock *sk = sock->sk;
2723 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2725 EXPORT_SYMBOL(sock_common_setsockopt);
2727 #ifdef CONFIG_COMPAT
2728 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2729 char __user *optval, unsigned int optlen)
2731 struct sock *sk = sock->sk;
2733 if (sk->sk_prot->compat_setsockopt != NULL)
2734 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2736 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2738 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2741 void sk_common_release(struct sock *sk)
2743 if (sk->sk_prot->destroy)
2744 sk->sk_prot->destroy(sk);
2747 * Observation: when sock_common_release is called, processes have
2748 * no access to socket. But net still has.
2749 * Step one, detach it from networking:
2751 * A. Remove from hash tables.
2754 sk->sk_prot->unhash(sk);
2757 * In this point socket cannot receive new packets, but it is possible
2758 * that some packets are in flight because some CPU runs receiver and
2759 * did hash table lookup before we unhashed socket. They will achieve
2760 * receive queue and will be purged by socket destructor.
2762 * Also we still have packets pending on receive queue and probably,
2763 * our own packets waiting in device queues. sock_destroy will drain
2764 * receive queue, but transmitted packets will delay socket destruction
2765 * until the last reference will be released.
2770 xfrm_sk_free_policy(sk);
2772 sk_refcnt_debug_release(sk);
2774 if (sk->sk_frag.page) {
2775 put_page(sk->sk_frag.page);
2776 sk->sk_frag.page = NULL;
2781 EXPORT_SYMBOL(sk_common_release);
2783 #ifdef CONFIG_PROC_FS
2784 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2786 int val[PROTO_INUSE_NR];
2789 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2791 #ifdef CONFIG_NET_NS
2792 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2794 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2796 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2798 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2800 int cpu, idx = prot->inuse_idx;
2803 for_each_possible_cpu(cpu)
2804 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2806 return res >= 0 ? res : 0;
2808 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2810 static int __net_init sock_inuse_init_net(struct net *net)
2812 net->core.inuse = alloc_percpu(struct prot_inuse);
2813 return net->core.inuse ? 0 : -ENOMEM;
2816 static void __net_exit sock_inuse_exit_net(struct net *net)
2818 free_percpu(net->core.inuse);
2821 static struct pernet_operations net_inuse_ops = {
2822 .init = sock_inuse_init_net,
2823 .exit = sock_inuse_exit_net,
2826 static __init int net_inuse_init(void)
2828 if (register_pernet_subsys(&net_inuse_ops))
2829 panic("Cannot initialize net inuse counters");
2834 core_initcall(net_inuse_init);
2836 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2838 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2840 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2842 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2844 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2846 int cpu, idx = prot->inuse_idx;
2849 for_each_possible_cpu(cpu)
2850 res += per_cpu(prot_inuse, cpu).val[idx];
2852 return res >= 0 ? res : 0;
2854 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2857 static void assign_proto_idx(struct proto *prot)
2859 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2861 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2862 pr_err("PROTO_INUSE_NR exhausted\n");
2866 set_bit(prot->inuse_idx, proto_inuse_idx);
2869 static void release_proto_idx(struct proto *prot)
2871 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2872 clear_bit(prot->inuse_idx, proto_inuse_idx);
2875 static inline void assign_proto_idx(struct proto *prot)
2879 static inline void release_proto_idx(struct proto *prot)
2884 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2888 kfree(rsk_prot->slab_name);
2889 rsk_prot->slab_name = NULL;
2890 kmem_cache_destroy(rsk_prot->slab);
2891 rsk_prot->slab = NULL;
2894 static int req_prot_init(const struct proto *prot)
2896 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2901 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2903 if (!rsk_prot->slab_name)
2906 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2907 rsk_prot->obj_size, 0,
2908 prot->slab_flags, NULL);
2910 if (!rsk_prot->slab) {
2911 pr_crit("%s: Can't create request sock SLAB cache!\n",
2918 int proto_register(struct proto *prot, int alloc_slab)
2921 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2922 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2925 if (prot->slab == NULL) {
2926 pr_crit("%s: Can't create sock SLAB cache!\n",
2931 if (req_prot_init(prot))
2932 goto out_free_request_sock_slab;
2934 if (prot->twsk_prot != NULL) {
2935 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2937 if (prot->twsk_prot->twsk_slab_name == NULL)
2938 goto out_free_request_sock_slab;
2940 prot->twsk_prot->twsk_slab =
2941 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2942 prot->twsk_prot->twsk_obj_size,
2946 if (prot->twsk_prot->twsk_slab == NULL)
2947 goto out_free_timewait_sock_slab_name;
2951 mutex_lock(&proto_list_mutex);
2952 list_add(&prot->node, &proto_list);
2953 assign_proto_idx(prot);
2954 mutex_unlock(&proto_list_mutex);
2957 out_free_timewait_sock_slab_name:
2958 kfree(prot->twsk_prot->twsk_slab_name);
2959 out_free_request_sock_slab:
2960 req_prot_cleanup(prot->rsk_prot);
2962 kmem_cache_destroy(prot->slab);
2967 EXPORT_SYMBOL(proto_register);
2969 void proto_unregister(struct proto *prot)
2971 mutex_lock(&proto_list_mutex);
2972 release_proto_idx(prot);
2973 list_del(&prot->node);
2974 mutex_unlock(&proto_list_mutex);
2976 kmem_cache_destroy(prot->slab);
2979 req_prot_cleanup(prot->rsk_prot);
2981 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2982 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2983 kfree(prot->twsk_prot->twsk_slab_name);
2984 prot->twsk_prot->twsk_slab = NULL;
2987 EXPORT_SYMBOL(proto_unregister);
2989 #ifdef CONFIG_PROC_FS
2990 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2991 __acquires(proto_list_mutex)
2993 mutex_lock(&proto_list_mutex);
2994 return seq_list_start_head(&proto_list, *pos);
2997 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2999 return seq_list_next(v, &proto_list, pos);
3002 static void proto_seq_stop(struct seq_file *seq, void *v)
3003 __releases(proto_list_mutex)
3005 mutex_unlock(&proto_list_mutex);
3008 static char proto_method_implemented(const void *method)
3010 return method == NULL ? 'n' : 'y';
3012 static long sock_prot_memory_allocated(struct proto *proto)
3014 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3017 static char *sock_prot_memory_pressure(struct proto *proto)
3019 return proto->memory_pressure != NULL ?
3020 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3023 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3026 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3027 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3030 sock_prot_inuse_get(seq_file_net(seq), proto),
3031 sock_prot_memory_allocated(proto),
3032 sock_prot_memory_pressure(proto),
3034 proto->slab == NULL ? "no" : "yes",
3035 module_name(proto->owner),
3036 proto_method_implemented(proto->close),
3037 proto_method_implemented(proto->connect),
3038 proto_method_implemented(proto->disconnect),
3039 proto_method_implemented(proto->accept),
3040 proto_method_implemented(proto->ioctl),
3041 proto_method_implemented(proto->init),
3042 proto_method_implemented(proto->destroy),
3043 proto_method_implemented(proto->shutdown),
3044 proto_method_implemented(proto->setsockopt),
3045 proto_method_implemented(proto->getsockopt),
3046 proto_method_implemented(proto->sendmsg),
3047 proto_method_implemented(proto->recvmsg),
3048 proto_method_implemented(proto->sendpage),
3049 proto_method_implemented(proto->bind),
3050 proto_method_implemented(proto->backlog_rcv),
3051 proto_method_implemented(proto->hash),
3052 proto_method_implemented(proto->unhash),
3053 proto_method_implemented(proto->get_port),
3054 proto_method_implemented(proto->enter_memory_pressure));
3057 static int proto_seq_show(struct seq_file *seq, void *v)
3059 if (v == &proto_list)
3060 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3069 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3071 proto_seq_printf(seq, list_entry(v, struct proto, node));
3075 static const struct seq_operations proto_seq_ops = {
3076 .start = proto_seq_start,
3077 .next = proto_seq_next,
3078 .stop = proto_seq_stop,
3079 .show = proto_seq_show,
3082 static int proto_seq_open(struct inode *inode, struct file *file)
3084 return seq_open_net(inode, file, &proto_seq_ops,
3085 sizeof(struct seq_net_private));
3088 static const struct file_operations proto_seq_fops = {
3089 .owner = THIS_MODULE,
3090 .open = proto_seq_open,
3092 .llseek = seq_lseek,
3093 .release = seq_release_net,
3096 static __net_init int proto_init_net(struct net *net)
3098 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3104 static __net_exit void proto_exit_net(struct net *net)
3106 remove_proc_entry("protocols", net->proc_net);
3110 static __net_initdata struct pernet_operations proto_net_ops = {
3111 .init = proto_init_net,
3112 .exit = proto_exit_net,
3115 static int __init proto_init(void)
3117 return register_pernet_subsys(&proto_net_ops);
3120 subsys_initcall(proto_init);
3122 #endif /* PROC_FS */