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)
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;
496 EXPORT_SYMBOL(__sk_receive_skb);
498 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
500 struct dst_entry *dst = __sk_dst_get(sk);
502 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
503 sk_tx_queue_clear(sk);
504 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
511 EXPORT_SYMBOL(__sk_dst_check);
513 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
515 struct dst_entry *dst = sk_dst_get(sk);
517 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
525 EXPORT_SYMBOL(sk_dst_check);
527 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
530 int ret = -ENOPROTOOPT;
531 #ifdef CONFIG_NETDEVICES
532 struct net *net = sock_net(sk);
533 char devname[IFNAMSIZ];
538 if (!ns_capable(net->user_ns, CAP_NET_RAW))
545 /* Bind this socket to a particular device like "eth0",
546 * as specified in the passed interface name. If the
547 * name is "" or the option length is zero the socket
550 if (optlen > IFNAMSIZ - 1)
551 optlen = IFNAMSIZ - 1;
552 memset(devname, 0, sizeof(devname));
555 if (copy_from_user(devname, optval, optlen))
559 if (devname[0] != '\0') {
560 struct net_device *dev;
563 dev = dev_get_by_name_rcu(net, devname);
565 index = dev->ifindex;
573 sk->sk_bound_dev_if = index;
585 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
586 int __user *optlen, int len)
588 int ret = -ENOPROTOOPT;
589 #ifdef CONFIG_NETDEVICES
590 struct net *net = sock_net(sk);
591 char devname[IFNAMSIZ];
593 if (sk->sk_bound_dev_if == 0) {
602 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
606 len = strlen(devname) + 1;
609 if (copy_to_user(optval, devname, len))
614 if (put_user(len, optlen))
625 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
628 sock_set_flag(sk, bit);
630 sock_reset_flag(sk, bit);
633 bool sk_mc_loop(struct sock *sk)
635 if (dev_recursion_level())
639 switch (sk->sk_family) {
641 return inet_sk(sk)->mc_loop;
642 #if IS_ENABLED(CONFIG_IPV6)
644 return inet6_sk(sk)->mc_loop;
650 EXPORT_SYMBOL(sk_mc_loop);
653 * This is meant for all protocols to use and covers goings on
654 * at the socket level. Everything here is generic.
657 int sock_setsockopt(struct socket *sock, int level, int optname,
658 char __user *optval, unsigned int optlen)
660 struct sock *sk = sock->sk;
667 * Options without arguments
670 if (optname == SO_BINDTODEVICE)
671 return sock_setbindtodevice(sk, optval, optlen);
673 if (optlen < sizeof(int))
676 if (get_user(val, (int __user *)optval))
679 valbool = val ? 1 : 0;
685 if (val && !capable(CAP_NET_ADMIN))
688 sock_valbool_flag(sk, SOCK_DBG, valbool);
691 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
694 sk->sk_reuseport = valbool;
703 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
706 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
709 /* Don't error on this BSD doesn't and if you think
710 * about it this is right. Otherwise apps have to
711 * play 'guess the biggest size' games. RCVBUF/SNDBUF
712 * are treated in BSD as hints
714 val = min_t(u32, val, sysctl_wmem_max);
716 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
717 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
718 /* Wake up sending tasks if we upped the value. */
719 sk->sk_write_space(sk);
723 if (!capable(CAP_NET_ADMIN)) {
730 /* Don't error on this BSD doesn't and if you think
731 * about it this is right. Otherwise apps have to
732 * play 'guess the biggest size' games. RCVBUF/SNDBUF
733 * are treated in BSD as hints
735 val = min_t(u32, val, sysctl_rmem_max);
737 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
739 * We double it on the way in to account for
740 * "struct sk_buff" etc. overhead. Applications
741 * assume that the SO_RCVBUF setting they make will
742 * allow that much actual data to be received on that
745 * Applications are unaware that "struct sk_buff" and
746 * other overheads allocate from the receive buffer
747 * during socket buffer allocation.
749 * And after considering the possible alternatives,
750 * returning the value we actually used in getsockopt
751 * is the most desirable behavior.
753 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
757 if (!capable(CAP_NET_ADMIN)) {
765 if (sk->sk_protocol == IPPROTO_TCP &&
766 sk->sk_type == SOCK_STREAM)
767 tcp_set_keepalive(sk, valbool);
769 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
773 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
777 sk->sk_no_check_tx = valbool;
781 if ((val >= 0 && val <= 6) ||
782 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
783 sk->sk_priority = val;
789 if (optlen < sizeof(ling)) {
790 ret = -EINVAL; /* 1003.1g */
793 if (copy_from_user(&ling, optval, sizeof(ling))) {
798 sock_reset_flag(sk, SOCK_LINGER);
800 #if (BITS_PER_LONG == 32)
801 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
802 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
805 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
806 sock_set_flag(sk, SOCK_LINGER);
811 sock_warn_obsolete_bsdism("setsockopt");
816 set_bit(SOCK_PASSCRED, &sock->flags);
818 clear_bit(SOCK_PASSCRED, &sock->flags);
824 if (optname == SO_TIMESTAMP)
825 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
827 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
828 sock_set_flag(sk, SOCK_RCVTSTAMP);
829 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
831 sock_reset_flag(sk, SOCK_RCVTSTAMP);
832 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
836 case SO_TIMESTAMPING:
837 if (val & ~SOF_TIMESTAMPING_MASK) {
842 if (val & SOF_TIMESTAMPING_OPT_ID &&
843 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
844 if (sk->sk_protocol == IPPROTO_TCP &&
845 sk->sk_type == SOCK_STREAM) {
846 if ((1 << sk->sk_state) &
847 (TCPF_CLOSE | TCPF_LISTEN)) {
851 sk->sk_tskey = tcp_sk(sk)->snd_una;
856 sk->sk_tsflags = val;
857 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
858 sock_enable_timestamp(sk,
859 SOCK_TIMESTAMPING_RX_SOFTWARE);
861 sock_disable_timestamp(sk,
862 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
868 sk->sk_rcvlowat = val ? : 1;
872 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
876 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
879 case SO_ATTACH_FILTER:
881 if (optlen == sizeof(struct sock_fprog)) {
882 struct sock_fprog fprog;
885 if (copy_from_user(&fprog, optval, sizeof(fprog)))
888 ret = sk_attach_filter(&fprog, sk);
894 if (optlen == sizeof(u32)) {
898 if (copy_from_user(&ufd, optval, sizeof(ufd)))
901 ret = sk_attach_bpf(ufd, sk);
905 case SO_ATTACH_REUSEPORT_CBPF:
907 if (optlen == sizeof(struct sock_fprog)) {
908 struct sock_fprog fprog;
911 if (copy_from_user(&fprog, optval, sizeof(fprog)))
914 ret = sk_reuseport_attach_filter(&fprog, sk);
918 case SO_ATTACH_REUSEPORT_EBPF:
920 if (optlen == sizeof(u32)) {
924 if (copy_from_user(&ufd, optval, sizeof(ufd)))
927 ret = sk_reuseport_attach_bpf(ufd, sk);
931 case SO_DETACH_FILTER:
932 ret = sk_detach_filter(sk);
936 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
939 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
944 set_bit(SOCK_PASSSEC, &sock->flags);
946 clear_bit(SOCK_PASSSEC, &sock->flags);
949 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
956 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
960 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
964 if (sock->ops->set_peek_off)
965 ret = sock->ops->set_peek_off(sk, val);
971 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
974 case SO_SELECT_ERR_QUEUE:
975 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
978 #ifdef CONFIG_NET_RX_BUSY_POLL
980 /* allow unprivileged users to decrease the value */
981 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
987 sk->sk_ll_usec = val;
992 case SO_MAX_PACING_RATE:
993 sk->sk_max_pacing_rate = val;
994 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
995 sk->sk_max_pacing_rate);
998 case SO_INCOMING_CPU:
999 sk->sk_incoming_cpu = val;
1004 dst_negative_advice(sk);
1013 EXPORT_SYMBOL(sock_setsockopt);
1016 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1017 struct ucred *ucred)
1019 ucred->pid = pid_vnr(pid);
1020 ucred->uid = ucred->gid = -1;
1022 struct user_namespace *current_ns = current_user_ns();
1024 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1025 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1029 int sock_getsockopt(struct socket *sock, int level, int optname,
1030 char __user *optval, int __user *optlen)
1032 struct sock *sk = sock->sk;
1040 int lv = sizeof(int);
1043 if (get_user(len, optlen))
1048 memset(&v, 0, sizeof(v));
1052 v.val = sock_flag(sk, SOCK_DBG);
1056 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1060 v.val = sock_flag(sk, SOCK_BROADCAST);
1064 v.val = sk->sk_sndbuf;
1068 v.val = sk->sk_rcvbuf;
1072 v.val = sk->sk_reuse;
1076 v.val = sk->sk_reuseport;
1080 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1084 v.val = sk->sk_type;
1088 v.val = sk->sk_protocol;
1092 v.val = sk->sk_family;
1096 v.val = -sock_error(sk);
1098 v.val = xchg(&sk->sk_err_soft, 0);
1102 v.val = sock_flag(sk, SOCK_URGINLINE);
1106 v.val = sk->sk_no_check_tx;
1110 v.val = sk->sk_priority;
1114 lv = sizeof(v.ling);
1115 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1116 v.ling.l_linger = sk->sk_lingertime / HZ;
1120 sock_warn_obsolete_bsdism("getsockopt");
1124 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1125 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1128 case SO_TIMESTAMPNS:
1129 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1132 case SO_TIMESTAMPING:
1133 v.val = sk->sk_tsflags;
1137 lv = sizeof(struct timeval);
1138 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1142 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1143 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1148 lv = sizeof(struct timeval);
1149 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1153 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1154 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1159 v.val = sk->sk_rcvlowat;
1167 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1172 struct ucred peercred;
1173 if (len > sizeof(peercred))
1174 len = sizeof(peercred);
1175 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1176 if (copy_to_user(optval, &peercred, len))
1185 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1189 if (copy_to_user(optval, address, len))
1194 /* Dubious BSD thing... Probably nobody even uses it, but
1195 * the UNIX standard wants it for whatever reason... -DaveM
1198 v.val = sk->sk_state == TCP_LISTEN;
1202 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1206 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1209 v.val = sk->sk_mark;
1213 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1216 case SO_WIFI_STATUS:
1217 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1221 if (!sock->ops->set_peek_off)
1224 v.val = sk->sk_peek_off;
1227 v.val = sock_flag(sk, SOCK_NOFCS);
1230 case SO_BINDTODEVICE:
1231 return sock_getbindtodevice(sk, optval, optlen, len);
1234 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1240 case SO_LOCK_FILTER:
1241 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1244 case SO_BPF_EXTENSIONS:
1245 v.val = bpf_tell_extensions();
1248 case SO_SELECT_ERR_QUEUE:
1249 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1252 #ifdef CONFIG_NET_RX_BUSY_POLL
1254 v.val = sk->sk_ll_usec;
1258 case SO_MAX_PACING_RATE:
1259 v.val = sk->sk_max_pacing_rate;
1262 case SO_INCOMING_CPU:
1263 v.val = sk->sk_incoming_cpu;
1267 /* We implement the SO_SNDLOWAT etc to not be settable
1270 return -ENOPROTOOPT;
1275 if (copy_to_user(optval, &v, len))
1278 if (put_user(len, optlen))
1284 * Initialize an sk_lock.
1286 * (We also register the sk_lock with the lock validator.)
1288 static inline void sock_lock_init(struct sock *sk)
1290 sock_lock_init_class_and_name(sk,
1291 af_family_slock_key_strings[sk->sk_family],
1292 af_family_slock_keys + sk->sk_family,
1293 af_family_key_strings[sk->sk_family],
1294 af_family_keys + sk->sk_family);
1298 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1299 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1300 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1302 static void sock_copy(struct sock *nsk, const struct sock *osk)
1304 #ifdef CONFIG_SECURITY_NETWORK
1305 void *sptr = nsk->sk_security;
1307 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1309 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1310 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1312 #ifdef CONFIG_SECURITY_NETWORK
1313 nsk->sk_security = sptr;
1314 security_sk_clone(osk, nsk);
1318 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1320 unsigned long nulls1, nulls2;
1322 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1323 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1324 if (nulls1 > nulls2)
1325 swap(nulls1, nulls2);
1328 memset((char *)sk, 0, nulls1);
1329 memset((char *)sk + nulls1 + sizeof(void *), 0,
1330 nulls2 - nulls1 - sizeof(void *));
1331 memset((char *)sk + nulls2 + sizeof(void *), 0,
1332 size - nulls2 - sizeof(void *));
1334 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1336 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1340 struct kmem_cache *slab;
1344 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1347 if (priority & __GFP_ZERO) {
1349 prot->clear_sk(sk, prot->obj_size);
1351 sk_prot_clear_nulls(sk, prot->obj_size);
1354 sk = kmalloc(prot->obj_size, priority);
1357 kmemcheck_annotate_bitfield(sk, flags);
1359 if (security_sk_alloc(sk, family, priority))
1362 if (!try_module_get(prot->owner))
1364 sk_tx_queue_clear(sk);
1365 cgroup_sk_alloc(&sk->sk_cgrp_data);
1371 security_sk_free(sk);
1374 kmem_cache_free(slab, sk);
1380 static void sk_prot_free(struct proto *prot, struct sock *sk)
1382 struct kmem_cache *slab;
1383 struct module *owner;
1385 owner = prot->owner;
1388 cgroup_sk_free(&sk->sk_cgrp_data);
1389 security_sk_free(sk);
1391 kmem_cache_free(slab, sk);
1398 * sk_alloc - All socket objects are allocated here
1399 * @net: the applicable net namespace
1400 * @family: protocol family
1401 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1402 * @prot: struct proto associated with this new sock instance
1403 * @kern: is this to be a kernel socket?
1405 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1406 struct proto *prot, int kern)
1410 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1412 sk->sk_family = family;
1414 * See comment in struct sock definition to understand
1415 * why we need sk_prot_creator -acme
1417 sk->sk_prot = sk->sk_prot_creator = prot;
1419 sk->sk_net_refcnt = kern ? 0 : 1;
1420 if (likely(sk->sk_net_refcnt))
1422 sock_net_set(sk, net);
1423 atomic_set(&sk->sk_wmem_alloc, 1);
1425 sock_update_classid(&sk->sk_cgrp_data);
1426 sock_update_netprioidx(&sk->sk_cgrp_data);
1431 EXPORT_SYMBOL(sk_alloc);
1433 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1434 * grace period. This is the case for UDP sockets and TCP listeners.
1436 static void __sk_destruct(struct rcu_head *head)
1438 struct sock *sk = container_of(head, struct sock, sk_rcu);
1439 struct sk_filter *filter;
1441 if (sk->sk_destruct)
1442 sk->sk_destruct(sk);
1444 filter = rcu_dereference_check(sk->sk_filter,
1445 atomic_read(&sk->sk_wmem_alloc) == 0);
1447 sk_filter_uncharge(sk, filter);
1448 RCU_INIT_POINTER(sk->sk_filter, NULL);
1450 if (rcu_access_pointer(sk->sk_reuseport_cb))
1451 reuseport_detach_sock(sk);
1453 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1455 if (atomic_read(&sk->sk_omem_alloc))
1456 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1457 __func__, atomic_read(&sk->sk_omem_alloc));
1459 if (sk->sk_peer_cred)
1460 put_cred(sk->sk_peer_cred);
1461 put_pid(sk->sk_peer_pid);
1462 if (likely(sk->sk_net_refcnt))
1463 put_net(sock_net(sk));
1464 sk_prot_free(sk->sk_prot_creator, sk);
1467 void sk_destruct(struct sock *sk)
1469 if (sock_flag(sk, SOCK_RCU_FREE))
1470 call_rcu(&sk->sk_rcu, __sk_destruct);
1472 __sk_destruct(&sk->sk_rcu);
1475 static void __sk_free(struct sock *sk)
1477 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1478 sock_diag_broadcast_destroy(sk);
1483 void sk_free(struct sock *sk)
1486 * We subtract one from sk_wmem_alloc and can know if
1487 * some packets are still in some tx queue.
1488 * If not null, sock_wfree() will call __sk_free(sk) later
1490 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1493 EXPORT_SYMBOL(sk_free);
1496 * sk_clone_lock - clone a socket, and lock its clone
1497 * @sk: the socket to clone
1498 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1500 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1502 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1505 bool is_charged = true;
1507 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1508 if (newsk != NULL) {
1509 struct sk_filter *filter;
1511 sock_copy(newsk, sk);
1514 if (likely(newsk->sk_net_refcnt))
1515 get_net(sock_net(newsk));
1516 sk_node_init(&newsk->sk_node);
1517 sock_lock_init(newsk);
1518 bh_lock_sock(newsk);
1519 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1520 newsk->sk_backlog.len = 0;
1522 atomic_set(&newsk->sk_rmem_alloc, 0);
1524 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1526 atomic_set(&newsk->sk_wmem_alloc, 1);
1527 atomic_set(&newsk->sk_omem_alloc, 0);
1528 skb_queue_head_init(&newsk->sk_receive_queue);
1529 skb_queue_head_init(&newsk->sk_write_queue);
1531 rwlock_init(&newsk->sk_callback_lock);
1532 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1533 af_callback_keys + newsk->sk_family,
1534 af_family_clock_key_strings[newsk->sk_family]);
1536 newsk->sk_dst_cache = NULL;
1537 newsk->sk_wmem_queued = 0;
1538 newsk->sk_forward_alloc = 0;
1539 atomic_set(&newsk->sk_drops, 0);
1540 newsk->sk_send_head = NULL;
1541 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1543 sock_reset_flag(newsk, SOCK_DONE);
1544 skb_queue_head_init(&newsk->sk_error_queue);
1546 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1548 /* though it's an empty new sock, the charging may fail
1549 * if sysctl_optmem_max was changed between creation of
1550 * original socket and cloning
1552 is_charged = sk_filter_charge(newsk, filter);
1554 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1555 /* It is still raw copy of parent, so invalidate
1556 * destructor and make plain sk_free() */
1557 newsk->sk_destruct = NULL;
1558 bh_unlock_sock(newsk);
1563 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1566 newsk->sk_priority = 0;
1567 newsk->sk_incoming_cpu = raw_smp_processor_id();
1568 atomic64_set(&newsk->sk_cookie, 0);
1570 * Before updating sk_refcnt, we must commit prior changes to memory
1571 * (Documentation/RCU/rculist_nulls.txt for details)
1574 atomic_set(&newsk->sk_refcnt, 2);
1577 * Increment the counter in the same struct proto as the master
1578 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1579 * is the same as sk->sk_prot->socks, as this field was copied
1582 * This _changes_ the previous behaviour, where
1583 * tcp_create_openreq_child always was incrementing the
1584 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1585 * to be taken into account in all callers. -acme
1587 sk_refcnt_debug_inc(newsk);
1588 sk_set_socket(newsk, NULL);
1589 newsk->sk_wq = NULL;
1591 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1592 sock_update_memcg(newsk);
1594 if (newsk->sk_prot->sockets_allocated)
1595 sk_sockets_allocated_inc(newsk);
1597 if (sock_needs_netstamp(sk) &&
1598 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1599 net_enable_timestamp();
1604 EXPORT_SYMBOL_GPL(sk_clone_lock);
1606 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1610 sk_dst_set(sk, dst);
1611 sk->sk_route_caps = dst->dev->features;
1612 if (sk->sk_route_caps & NETIF_F_GSO)
1613 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1614 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1615 if (sk_can_gso(sk)) {
1616 if (dst->header_len) {
1617 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1619 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1620 sk->sk_gso_max_size = dst->dev->gso_max_size;
1621 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1624 sk->sk_gso_max_segs = max_segs;
1626 EXPORT_SYMBOL_GPL(sk_setup_caps);
1629 * Simple resource managers for sockets.
1634 * Write buffer destructor automatically called from kfree_skb.
1636 void sock_wfree(struct sk_buff *skb)
1638 struct sock *sk = skb->sk;
1639 unsigned int len = skb->truesize;
1641 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1643 * Keep a reference on sk_wmem_alloc, this will be released
1644 * after sk_write_space() call
1646 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1647 sk->sk_write_space(sk);
1651 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1652 * could not do because of in-flight packets
1654 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1657 EXPORT_SYMBOL(sock_wfree);
1659 /* This variant of sock_wfree() is used by TCP,
1660 * since it sets SOCK_USE_WRITE_QUEUE.
1662 void __sock_wfree(struct sk_buff *skb)
1664 struct sock *sk = skb->sk;
1666 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1670 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1675 if (unlikely(!sk_fullsock(sk))) {
1676 skb->destructor = sock_edemux;
1681 skb->destructor = sock_wfree;
1682 skb_set_hash_from_sk(skb, sk);
1684 * We used to take a refcount on sk, but following operation
1685 * is enough to guarantee sk_free() wont free this sock until
1686 * all in-flight packets are completed
1688 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1690 EXPORT_SYMBOL(skb_set_owner_w);
1692 /* This helper is used by netem, as it can hold packets in its
1693 * delay queue. We want to allow the owner socket to send more
1694 * packets, as if they were already TX completed by a typical driver.
1695 * But we also want to keep skb->sk set because some packet schedulers
1696 * rely on it (sch_fq for example). So we set skb->truesize to a small
1697 * amount (1) and decrease sk_wmem_alloc accordingly.
1699 void skb_orphan_partial(struct sk_buff *skb)
1701 /* If this skb is a TCP pure ACK or already went here,
1702 * we have nothing to do. 2 is already a very small truesize.
1704 if (skb->truesize <= 2)
1707 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1708 * so we do not completely orphan skb, but transfert all
1709 * accounted bytes but one, to avoid unexpected reorders.
1711 if (skb->destructor == sock_wfree
1713 || skb->destructor == tcp_wfree
1716 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1722 EXPORT_SYMBOL(skb_orphan_partial);
1725 * Read buffer destructor automatically called from kfree_skb.
1727 void sock_rfree(struct sk_buff *skb)
1729 struct sock *sk = skb->sk;
1730 unsigned int len = skb->truesize;
1732 atomic_sub(len, &sk->sk_rmem_alloc);
1733 sk_mem_uncharge(sk, len);
1735 EXPORT_SYMBOL(sock_rfree);
1738 * Buffer destructor for skbs that are not used directly in read or write
1739 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1741 void sock_efree(struct sk_buff *skb)
1745 EXPORT_SYMBOL(sock_efree);
1747 kuid_t sock_i_uid(struct sock *sk)
1751 read_lock_bh(&sk->sk_callback_lock);
1752 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1753 read_unlock_bh(&sk->sk_callback_lock);
1756 EXPORT_SYMBOL(sock_i_uid);
1758 unsigned long sock_i_ino(struct sock *sk)
1762 read_lock_bh(&sk->sk_callback_lock);
1763 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1764 read_unlock_bh(&sk->sk_callback_lock);
1767 EXPORT_SYMBOL(sock_i_ino);
1770 * Allocate a skb from the socket's send buffer.
1772 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1775 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1776 struct sk_buff *skb = alloc_skb(size, priority);
1778 skb_set_owner_w(skb, sk);
1784 EXPORT_SYMBOL(sock_wmalloc);
1787 * Allocate a memory block from the socket's option memory buffer.
1789 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1791 if ((unsigned int)size <= sysctl_optmem_max &&
1792 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1794 /* First do the add, to avoid the race if kmalloc
1797 atomic_add(size, &sk->sk_omem_alloc);
1798 mem = kmalloc(size, priority);
1801 atomic_sub(size, &sk->sk_omem_alloc);
1805 EXPORT_SYMBOL(sock_kmalloc);
1807 /* Free an option memory block. Note, we actually want the inline
1808 * here as this allows gcc to detect the nullify and fold away the
1809 * condition entirely.
1811 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1814 if (WARN_ON_ONCE(!mem))
1820 atomic_sub(size, &sk->sk_omem_alloc);
1823 void sock_kfree_s(struct sock *sk, void *mem, int size)
1825 __sock_kfree_s(sk, mem, size, false);
1827 EXPORT_SYMBOL(sock_kfree_s);
1829 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1831 __sock_kfree_s(sk, mem, size, true);
1833 EXPORT_SYMBOL(sock_kzfree_s);
1835 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1836 I think, these locks should be removed for datagram sockets.
1838 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1842 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1846 if (signal_pending(current))
1848 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1849 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1850 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1852 if (sk->sk_shutdown & SEND_SHUTDOWN)
1856 timeo = schedule_timeout(timeo);
1858 finish_wait(sk_sleep(sk), &wait);
1864 * Generic send/receive buffer handlers
1867 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1868 unsigned long data_len, int noblock,
1869 int *errcode, int max_page_order)
1871 struct sk_buff *skb;
1875 timeo = sock_sndtimeo(sk, noblock);
1877 err = sock_error(sk);
1882 if (sk->sk_shutdown & SEND_SHUTDOWN)
1885 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1888 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1889 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1893 if (signal_pending(current))
1895 timeo = sock_wait_for_wmem(sk, timeo);
1897 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1898 errcode, sk->sk_allocation);
1900 skb_set_owner_w(skb, sk);
1904 err = sock_intr_errno(timeo);
1909 EXPORT_SYMBOL(sock_alloc_send_pskb);
1911 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1912 int noblock, int *errcode)
1914 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1916 EXPORT_SYMBOL(sock_alloc_send_skb);
1918 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1919 struct sockcm_cookie *sockc)
1923 switch (cmsg->cmsg_type) {
1925 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1927 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1929 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1931 case SO_TIMESTAMPING:
1932 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1935 tsflags = *(u32 *)CMSG_DATA(cmsg);
1936 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1939 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1940 sockc->tsflags |= tsflags;
1942 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1944 case SCM_CREDENTIALS:
1951 EXPORT_SYMBOL(__sock_cmsg_send);
1953 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1954 struct sockcm_cookie *sockc)
1956 struct cmsghdr *cmsg;
1959 for_each_cmsghdr(cmsg, msg) {
1960 if (!CMSG_OK(msg, cmsg))
1962 if (cmsg->cmsg_level != SOL_SOCKET)
1964 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1970 EXPORT_SYMBOL(sock_cmsg_send);
1972 /* On 32bit arches, an skb frag is limited to 2^15 */
1973 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1976 * skb_page_frag_refill - check that a page_frag contains enough room
1977 * @sz: minimum size of the fragment we want to get
1978 * @pfrag: pointer to page_frag
1979 * @gfp: priority for memory allocation
1981 * Note: While this allocator tries to use high order pages, there is
1982 * no guarantee that allocations succeed. Therefore, @sz MUST be
1983 * less or equal than PAGE_SIZE.
1985 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1988 if (page_ref_count(pfrag->page) == 1) {
1992 if (pfrag->offset + sz <= pfrag->size)
1994 put_page(pfrag->page);
1998 if (SKB_FRAG_PAGE_ORDER) {
1999 /* Avoid direct reclaim but allow kswapd to wake */
2000 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2001 __GFP_COMP | __GFP_NOWARN |
2003 SKB_FRAG_PAGE_ORDER);
2004 if (likely(pfrag->page)) {
2005 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2009 pfrag->page = alloc_page(gfp);
2010 if (likely(pfrag->page)) {
2011 pfrag->size = PAGE_SIZE;
2016 EXPORT_SYMBOL(skb_page_frag_refill);
2018 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2020 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2023 sk_enter_memory_pressure(sk);
2024 sk_stream_moderate_sndbuf(sk);
2027 EXPORT_SYMBOL(sk_page_frag_refill);
2029 static void __lock_sock(struct sock *sk)
2030 __releases(&sk->sk_lock.slock)
2031 __acquires(&sk->sk_lock.slock)
2036 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2037 TASK_UNINTERRUPTIBLE);
2038 spin_unlock_bh(&sk->sk_lock.slock);
2040 spin_lock_bh(&sk->sk_lock.slock);
2041 if (!sock_owned_by_user(sk))
2044 finish_wait(&sk->sk_lock.wq, &wait);
2047 static void __release_sock(struct sock *sk)
2048 __releases(&sk->sk_lock.slock)
2049 __acquires(&sk->sk_lock.slock)
2051 struct sk_buff *skb, *next;
2053 while ((skb = sk->sk_backlog.head) != NULL) {
2054 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2056 spin_unlock_bh(&sk->sk_lock.slock);
2061 WARN_ON_ONCE(skb_dst_is_noref(skb));
2063 sk_backlog_rcv(sk, skb);
2068 } while (skb != NULL);
2070 spin_lock_bh(&sk->sk_lock.slock);
2074 * Doing the zeroing here guarantee we can not loop forever
2075 * while a wild producer attempts to flood us.
2077 sk->sk_backlog.len = 0;
2080 void __sk_flush_backlog(struct sock *sk)
2082 spin_lock_bh(&sk->sk_lock.slock);
2084 spin_unlock_bh(&sk->sk_lock.slock);
2088 * sk_wait_data - wait for data to arrive at sk_receive_queue
2089 * @sk: sock to wait on
2090 * @timeo: for how long
2091 * @skb: last skb seen on sk_receive_queue
2093 * Now socket state including sk->sk_err is changed only under lock,
2094 * hence we may omit checks after joining wait queue.
2095 * We check receive queue before schedule() only as optimization;
2096 * it is very likely that release_sock() added new data.
2098 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2103 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2104 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2105 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2106 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2107 finish_wait(sk_sleep(sk), &wait);
2110 EXPORT_SYMBOL(sk_wait_data);
2113 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2115 * @size: memory size to allocate
2116 * @kind: allocation type
2118 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2119 * rmem allocation. This function assumes that protocols which have
2120 * memory_pressure use sk_wmem_queued as write buffer accounting.
2122 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2124 struct proto *prot = sk->sk_prot;
2125 int amt = sk_mem_pages(size);
2128 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2130 allocated = sk_memory_allocated_add(sk, amt);
2132 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2133 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2134 goto suppress_allocation;
2137 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2138 sk_leave_memory_pressure(sk);
2142 /* Under pressure. */
2143 if (allocated > sk_prot_mem_limits(sk, 1))
2144 sk_enter_memory_pressure(sk);
2146 /* Over hard limit. */
2147 if (allocated > sk_prot_mem_limits(sk, 2))
2148 goto suppress_allocation;
2150 /* guarantee minimum buffer size under pressure */
2151 if (kind == SK_MEM_RECV) {
2152 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2155 } else { /* SK_MEM_SEND */
2156 if (sk->sk_type == SOCK_STREAM) {
2157 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2159 } else if (atomic_read(&sk->sk_wmem_alloc) <
2160 prot->sysctl_wmem[0])
2164 if (sk_has_memory_pressure(sk)) {
2167 if (!sk_under_memory_pressure(sk))
2169 alloc = sk_sockets_allocated_read_positive(sk);
2170 if (sk_prot_mem_limits(sk, 2) > alloc *
2171 sk_mem_pages(sk->sk_wmem_queued +
2172 atomic_read(&sk->sk_rmem_alloc) +
2173 sk->sk_forward_alloc))
2177 suppress_allocation:
2179 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2180 sk_stream_moderate_sndbuf(sk);
2182 /* Fail only if socket is _under_ its sndbuf.
2183 * In this case we cannot block, so that we have to fail.
2185 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2189 trace_sock_exceed_buf_limit(sk, prot, allocated);
2191 /* Alas. Undo changes. */
2192 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2194 sk_memory_allocated_sub(sk, amt);
2196 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2197 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2201 EXPORT_SYMBOL(__sk_mem_schedule);
2204 * __sk_mem_reclaim - reclaim memory_allocated
2206 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2208 void __sk_mem_reclaim(struct sock *sk, int amount)
2210 amount >>= SK_MEM_QUANTUM_SHIFT;
2211 sk_memory_allocated_sub(sk, amount);
2212 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
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_reclaim);
2223 int sk_set_peek_off(struct sock *sk, int val)
2228 sk->sk_peek_off = val;
2231 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2234 * Set of default routines for initialising struct proto_ops when
2235 * the protocol does not support a particular function. In certain
2236 * cases where it makes no sense for a protocol to have a "do nothing"
2237 * function, some default processing is provided.
2240 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2244 EXPORT_SYMBOL(sock_no_bind);
2246 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2251 EXPORT_SYMBOL(sock_no_connect);
2253 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2257 EXPORT_SYMBOL(sock_no_socketpair);
2259 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2263 EXPORT_SYMBOL(sock_no_accept);
2265 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2270 EXPORT_SYMBOL(sock_no_getname);
2272 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2276 EXPORT_SYMBOL(sock_no_poll);
2278 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2282 EXPORT_SYMBOL(sock_no_ioctl);
2284 int sock_no_listen(struct socket *sock, int backlog)
2288 EXPORT_SYMBOL(sock_no_listen);
2290 int sock_no_shutdown(struct socket *sock, int how)
2294 EXPORT_SYMBOL(sock_no_shutdown);
2296 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2297 char __user *optval, unsigned int optlen)
2301 EXPORT_SYMBOL(sock_no_setsockopt);
2303 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2304 char __user *optval, int __user *optlen)
2308 EXPORT_SYMBOL(sock_no_getsockopt);
2310 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2314 EXPORT_SYMBOL(sock_no_sendmsg);
2316 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2321 EXPORT_SYMBOL(sock_no_recvmsg);
2323 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2325 /* Mirror missing mmap method error code */
2328 EXPORT_SYMBOL(sock_no_mmap);
2330 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2333 struct msghdr msg = {.msg_flags = flags};
2335 char *kaddr = kmap(page);
2336 iov.iov_base = kaddr + offset;
2338 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2342 EXPORT_SYMBOL(sock_no_sendpage);
2345 * Default Socket Callbacks
2348 static void sock_def_wakeup(struct sock *sk)
2350 struct socket_wq *wq;
2353 wq = rcu_dereference(sk->sk_wq);
2354 if (skwq_has_sleeper(wq))
2355 wake_up_interruptible_all(&wq->wait);
2359 static void sock_def_error_report(struct sock *sk)
2361 struct socket_wq *wq;
2364 wq = rcu_dereference(sk->sk_wq);
2365 if (skwq_has_sleeper(wq))
2366 wake_up_interruptible_poll(&wq->wait, POLLERR);
2367 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2371 static void sock_def_readable(struct sock *sk)
2373 struct socket_wq *wq;
2376 wq = rcu_dereference(sk->sk_wq);
2377 if (skwq_has_sleeper(wq))
2378 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2379 POLLRDNORM | POLLRDBAND);
2380 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2384 static void sock_def_write_space(struct sock *sk)
2386 struct socket_wq *wq;
2390 /* Do not wake up a writer until he can make "significant"
2393 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2394 wq = rcu_dereference(sk->sk_wq);
2395 if (skwq_has_sleeper(wq))
2396 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2397 POLLWRNORM | POLLWRBAND);
2399 /* Should agree with poll, otherwise some programs break */
2400 if (sock_writeable(sk))
2401 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2407 static void sock_def_destruct(struct sock *sk)
2411 void sk_send_sigurg(struct sock *sk)
2413 if (sk->sk_socket && sk->sk_socket->file)
2414 if (send_sigurg(&sk->sk_socket->file->f_owner))
2415 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2417 EXPORT_SYMBOL(sk_send_sigurg);
2419 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2420 unsigned long expires)
2422 if (!mod_timer(timer, expires))
2425 EXPORT_SYMBOL(sk_reset_timer);
2427 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2429 if (del_timer(timer))
2432 EXPORT_SYMBOL(sk_stop_timer);
2434 void sock_init_data(struct socket *sock, struct sock *sk)
2436 skb_queue_head_init(&sk->sk_receive_queue);
2437 skb_queue_head_init(&sk->sk_write_queue);
2438 skb_queue_head_init(&sk->sk_error_queue);
2440 sk->sk_send_head = NULL;
2442 init_timer(&sk->sk_timer);
2444 sk->sk_allocation = GFP_KERNEL;
2445 sk->sk_rcvbuf = sysctl_rmem_default;
2446 sk->sk_sndbuf = sysctl_wmem_default;
2447 sk->sk_state = TCP_CLOSE;
2448 sk_set_socket(sk, sock);
2450 sock_set_flag(sk, SOCK_ZAPPED);
2453 sk->sk_type = sock->type;
2454 sk->sk_wq = sock->wq;
2459 rwlock_init(&sk->sk_callback_lock);
2460 lockdep_set_class_and_name(&sk->sk_callback_lock,
2461 af_callback_keys + sk->sk_family,
2462 af_family_clock_key_strings[sk->sk_family]);
2464 sk->sk_state_change = sock_def_wakeup;
2465 sk->sk_data_ready = sock_def_readable;
2466 sk->sk_write_space = sock_def_write_space;
2467 sk->sk_error_report = sock_def_error_report;
2468 sk->sk_destruct = sock_def_destruct;
2470 sk->sk_frag.page = NULL;
2471 sk->sk_frag.offset = 0;
2472 sk->sk_peek_off = -1;
2474 sk->sk_peer_pid = NULL;
2475 sk->sk_peer_cred = NULL;
2476 sk->sk_write_pending = 0;
2477 sk->sk_rcvlowat = 1;
2478 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2479 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2481 sk->sk_stamp = ktime_set(-1L, 0);
2483 #ifdef CONFIG_NET_RX_BUSY_POLL
2485 sk->sk_ll_usec = sysctl_net_busy_read;
2488 sk->sk_max_pacing_rate = ~0U;
2489 sk->sk_pacing_rate = ~0U;
2490 sk->sk_incoming_cpu = -1;
2492 * Before updating sk_refcnt, we must commit prior changes to memory
2493 * (Documentation/RCU/rculist_nulls.txt for details)
2496 atomic_set(&sk->sk_refcnt, 1);
2497 atomic_set(&sk->sk_drops, 0);
2499 EXPORT_SYMBOL(sock_init_data);
2501 void lock_sock_nested(struct sock *sk, int subclass)
2504 spin_lock_bh(&sk->sk_lock.slock);
2505 if (sk->sk_lock.owned)
2507 sk->sk_lock.owned = 1;
2508 spin_unlock(&sk->sk_lock.slock);
2510 * The sk_lock has mutex_lock() semantics here:
2512 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2515 EXPORT_SYMBOL(lock_sock_nested);
2517 void release_sock(struct sock *sk)
2519 spin_lock_bh(&sk->sk_lock.slock);
2520 if (sk->sk_backlog.tail)
2523 /* Warning : release_cb() might need to release sk ownership,
2524 * ie call sock_release_ownership(sk) before us.
2526 if (sk->sk_prot->release_cb)
2527 sk->sk_prot->release_cb(sk);
2529 sock_release_ownership(sk);
2530 if (waitqueue_active(&sk->sk_lock.wq))
2531 wake_up(&sk->sk_lock.wq);
2532 spin_unlock_bh(&sk->sk_lock.slock);
2534 EXPORT_SYMBOL(release_sock);
2537 * lock_sock_fast - fast version of lock_sock
2540 * This version should be used for very small section, where process wont block
2541 * return false if fast path is taken
2542 * sk_lock.slock locked, owned = 0, BH disabled
2543 * return true if slow path is taken
2544 * sk_lock.slock unlocked, owned = 1, BH enabled
2546 bool lock_sock_fast(struct sock *sk)
2549 spin_lock_bh(&sk->sk_lock.slock);
2551 if (!sk->sk_lock.owned)
2553 * Note : We must disable BH
2558 sk->sk_lock.owned = 1;
2559 spin_unlock(&sk->sk_lock.slock);
2561 * The sk_lock has mutex_lock() semantics here:
2563 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2567 EXPORT_SYMBOL(lock_sock_fast);
2569 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2572 if (!sock_flag(sk, SOCK_TIMESTAMP))
2573 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2574 tv = ktime_to_timeval(sk->sk_stamp);
2575 if (tv.tv_sec == -1)
2577 if (tv.tv_sec == 0) {
2578 sk->sk_stamp = ktime_get_real();
2579 tv = ktime_to_timeval(sk->sk_stamp);
2581 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2583 EXPORT_SYMBOL(sock_get_timestamp);
2585 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2588 if (!sock_flag(sk, SOCK_TIMESTAMP))
2589 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2590 ts = ktime_to_timespec(sk->sk_stamp);
2591 if (ts.tv_sec == -1)
2593 if (ts.tv_sec == 0) {
2594 sk->sk_stamp = ktime_get_real();
2595 ts = ktime_to_timespec(sk->sk_stamp);
2597 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2599 EXPORT_SYMBOL(sock_get_timestampns);
2601 void sock_enable_timestamp(struct sock *sk, int flag)
2603 if (!sock_flag(sk, flag)) {
2604 unsigned long previous_flags = sk->sk_flags;
2606 sock_set_flag(sk, flag);
2608 * we just set one of the two flags which require net
2609 * time stamping, but time stamping might have been on
2610 * already because of the other one
2612 if (sock_needs_netstamp(sk) &&
2613 !(previous_flags & SK_FLAGS_TIMESTAMP))
2614 net_enable_timestamp();
2618 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2619 int level, int type)
2621 struct sock_exterr_skb *serr;
2622 struct sk_buff *skb;
2626 skb = sock_dequeue_err_skb(sk);
2632 msg->msg_flags |= MSG_TRUNC;
2635 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2639 sock_recv_timestamp(msg, sk, skb);
2641 serr = SKB_EXT_ERR(skb);
2642 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2644 msg->msg_flags |= MSG_ERRQUEUE;
2652 EXPORT_SYMBOL(sock_recv_errqueue);
2655 * Get a socket option on an socket.
2657 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2658 * asynchronous errors should be reported by getsockopt. We assume
2659 * this means if you specify SO_ERROR (otherwise whats the point of it).
2661 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2662 char __user *optval, int __user *optlen)
2664 struct sock *sk = sock->sk;
2666 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2668 EXPORT_SYMBOL(sock_common_getsockopt);
2670 #ifdef CONFIG_COMPAT
2671 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2672 char __user *optval, int __user *optlen)
2674 struct sock *sk = sock->sk;
2676 if (sk->sk_prot->compat_getsockopt != NULL)
2677 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2679 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2681 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2684 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2687 struct sock *sk = sock->sk;
2691 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2692 flags & ~MSG_DONTWAIT, &addr_len);
2694 msg->msg_namelen = addr_len;
2697 EXPORT_SYMBOL(sock_common_recvmsg);
2700 * Set socket options on an inet socket.
2702 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2703 char __user *optval, unsigned int optlen)
2705 struct sock *sk = sock->sk;
2707 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2709 EXPORT_SYMBOL(sock_common_setsockopt);
2711 #ifdef CONFIG_COMPAT
2712 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2713 char __user *optval, unsigned int optlen)
2715 struct sock *sk = sock->sk;
2717 if (sk->sk_prot->compat_setsockopt != NULL)
2718 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2720 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2722 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2725 void sk_common_release(struct sock *sk)
2727 if (sk->sk_prot->destroy)
2728 sk->sk_prot->destroy(sk);
2731 * Observation: when sock_common_release is called, processes have
2732 * no access to socket. But net still has.
2733 * Step one, detach it from networking:
2735 * A. Remove from hash tables.
2738 sk->sk_prot->unhash(sk);
2741 * In this point socket cannot receive new packets, but it is possible
2742 * that some packets are in flight because some CPU runs receiver and
2743 * did hash table lookup before we unhashed socket. They will achieve
2744 * receive queue and will be purged by socket destructor.
2746 * Also we still have packets pending on receive queue and probably,
2747 * our own packets waiting in device queues. sock_destroy will drain
2748 * receive queue, but transmitted packets will delay socket destruction
2749 * until the last reference will be released.
2754 xfrm_sk_free_policy(sk);
2756 sk_refcnt_debug_release(sk);
2758 if (sk->sk_frag.page) {
2759 put_page(sk->sk_frag.page);
2760 sk->sk_frag.page = NULL;
2765 EXPORT_SYMBOL(sk_common_release);
2767 #ifdef CONFIG_PROC_FS
2768 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2770 int val[PROTO_INUSE_NR];
2773 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2775 #ifdef CONFIG_NET_NS
2776 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2778 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2780 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2782 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2784 int cpu, idx = prot->inuse_idx;
2787 for_each_possible_cpu(cpu)
2788 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2790 return res >= 0 ? res : 0;
2792 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2794 static int __net_init sock_inuse_init_net(struct net *net)
2796 net->core.inuse = alloc_percpu(struct prot_inuse);
2797 return net->core.inuse ? 0 : -ENOMEM;
2800 static void __net_exit sock_inuse_exit_net(struct net *net)
2802 free_percpu(net->core.inuse);
2805 static struct pernet_operations net_inuse_ops = {
2806 .init = sock_inuse_init_net,
2807 .exit = sock_inuse_exit_net,
2810 static __init int net_inuse_init(void)
2812 if (register_pernet_subsys(&net_inuse_ops))
2813 panic("Cannot initialize net inuse counters");
2818 core_initcall(net_inuse_init);
2820 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2822 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2824 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2826 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2828 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2830 int cpu, idx = prot->inuse_idx;
2833 for_each_possible_cpu(cpu)
2834 res += per_cpu(prot_inuse, cpu).val[idx];
2836 return res >= 0 ? res : 0;
2838 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2841 static void assign_proto_idx(struct proto *prot)
2843 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2845 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2846 pr_err("PROTO_INUSE_NR exhausted\n");
2850 set_bit(prot->inuse_idx, proto_inuse_idx);
2853 static void release_proto_idx(struct proto *prot)
2855 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2856 clear_bit(prot->inuse_idx, proto_inuse_idx);
2859 static inline void assign_proto_idx(struct proto *prot)
2863 static inline void release_proto_idx(struct proto *prot)
2868 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2872 kfree(rsk_prot->slab_name);
2873 rsk_prot->slab_name = NULL;
2874 kmem_cache_destroy(rsk_prot->slab);
2875 rsk_prot->slab = NULL;
2878 static int req_prot_init(const struct proto *prot)
2880 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2885 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2887 if (!rsk_prot->slab_name)
2890 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2891 rsk_prot->obj_size, 0,
2892 prot->slab_flags, NULL);
2894 if (!rsk_prot->slab) {
2895 pr_crit("%s: Can't create request sock SLAB cache!\n",
2902 int proto_register(struct proto *prot, int alloc_slab)
2905 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2906 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2909 if (prot->slab == NULL) {
2910 pr_crit("%s: Can't create sock SLAB cache!\n",
2915 if (req_prot_init(prot))
2916 goto out_free_request_sock_slab;
2918 if (prot->twsk_prot != NULL) {
2919 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2921 if (prot->twsk_prot->twsk_slab_name == NULL)
2922 goto out_free_request_sock_slab;
2924 prot->twsk_prot->twsk_slab =
2925 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2926 prot->twsk_prot->twsk_obj_size,
2930 if (prot->twsk_prot->twsk_slab == NULL)
2931 goto out_free_timewait_sock_slab_name;
2935 mutex_lock(&proto_list_mutex);
2936 list_add(&prot->node, &proto_list);
2937 assign_proto_idx(prot);
2938 mutex_unlock(&proto_list_mutex);
2941 out_free_timewait_sock_slab_name:
2942 kfree(prot->twsk_prot->twsk_slab_name);
2943 out_free_request_sock_slab:
2944 req_prot_cleanup(prot->rsk_prot);
2946 kmem_cache_destroy(prot->slab);
2951 EXPORT_SYMBOL(proto_register);
2953 void proto_unregister(struct proto *prot)
2955 mutex_lock(&proto_list_mutex);
2956 release_proto_idx(prot);
2957 list_del(&prot->node);
2958 mutex_unlock(&proto_list_mutex);
2960 kmem_cache_destroy(prot->slab);
2963 req_prot_cleanup(prot->rsk_prot);
2965 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2966 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2967 kfree(prot->twsk_prot->twsk_slab_name);
2968 prot->twsk_prot->twsk_slab = NULL;
2971 EXPORT_SYMBOL(proto_unregister);
2973 #ifdef CONFIG_PROC_FS
2974 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2975 __acquires(proto_list_mutex)
2977 mutex_lock(&proto_list_mutex);
2978 return seq_list_start_head(&proto_list, *pos);
2981 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2983 return seq_list_next(v, &proto_list, pos);
2986 static void proto_seq_stop(struct seq_file *seq, void *v)
2987 __releases(proto_list_mutex)
2989 mutex_unlock(&proto_list_mutex);
2992 static char proto_method_implemented(const void *method)
2994 return method == NULL ? 'n' : 'y';
2996 static long sock_prot_memory_allocated(struct proto *proto)
2998 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3001 static char *sock_prot_memory_pressure(struct proto *proto)
3003 return proto->memory_pressure != NULL ?
3004 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3007 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3010 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3011 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3014 sock_prot_inuse_get(seq_file_net(seq), proto),
3015 sock_prot_memory_allocated(proto),
3016 sock_prot_memory_pressure(proto),
3018 proto->slab == NULL ? "no" : "yes",
3019 module_name(proto->owner),
3020 proto_method_implemented(proto->close),
3021 proto_method_implemented(proto->connect),
3022 proto_method_implemented(proto->disconnect),
3023 proto_method_implemented(proto->accept),
3024 proto_method_implemented(proto->ioctl),
3025 proto_method_implemented(proto->init),
3026 proto_method_implemented(proto->destroy),
3027 proto_method_implemented(proto->shutdown),
3028 proto_method_implemented(proto->setsockopt),
3029 proto_method_implemented(proto->getsockopt),
3030 proto_method_implemented(proto->sendmsg),
3031 proto_method_implemented(proto->recvmsg),
3032 proto_method_implemented(proto->sendpage),
3033 proto_method_implemented(proto->bind),
3034 proto_method_implemented(proto->backlog_rcv),
3035 proto_method_implemented(proto->hash),
3036 proto_method_implemented(proto->unhash),
3037 proto_method_implemented(proto->get_port),
3038 proto_method_implemented(proto->enter_memory_pressure));
3041 static int proto_seq_show(struct seq_file *seq, void *v)
3043 if (v == &proto_list)
3044 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3053 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3055 proto_seq_printf(seq, list_entry(v, struct proto, node));
3059 static const struct seq_operations proto_seq_ops = {
3060 .start = proto_seq_start,
3061 .next = proto_seq_next,
3062 .stop = proto_seq_stop,
3063 .show = proto_seq_show,
3066 static int proto_seq_open(struct inode *inode, struct file *file)
3068 return seq_open_net(inode, file, &proto_seq_ops,
3069 sizeof(struct seq_net_private));
3072 static const struct file_operations proto_seq_fops = {
3073 .owner = THIS_MODULE,
3074 .open = proto_seq_open,
3076 .llseek = seq_lseek,
3077 .release = seq_release_net,
3080 static __net_init int proto_init_net(struct net *net)
3082 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3088 static __net_exit void proto_exit_net(struct net *net)
3090 remove_proc_entry("protocols", net->proc_net);
3094 static __net_initdata struct pernet_operations proto_net_ops = {
3095 .init = proto_init_net,
3096 .exit = proto_exit_net,
3099 static int __init proto_init(void)
3101 return register_pernet_subsys(&proto_net_ops);
3104 subsys_initcall(proto_init);
3106 #endif /* PROC_FS */