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/sched/mm.h>
106 #include <linux/timer.h>
107 #include <linux/string.h>
108 #include <linux/sockios.h>
109 #include <linux/net.h>
110 #include <linux/mm.h>
111 #include <linux/slab.h>
112 #include <linux/interrupt.h>
113 #include <linux/poll.h>
114 #include <linux/tcp.h>
115 #include <linux/init.h>
116 #include <linux/highmem.h>
117 #include <linux/user_namespace.h>
118 #include <linux/static_key.h>
119 #include <linux/memcontrol.h>
120 #include <linux/prefetch.h>
122 #include <linux/uaccess.h>
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <net/net_namespace.h>
128 #include <net/request_sock.h>
129 #include <net/sock.h>
130 #include <linux/net_tstamp.h>
131 #include <net/xfrm.h>
132 #include <linux/ipsec.h>
133 #include <net/cls_cgroup.h>
134 #include <net/netprio_cgroup.h>
135 #include <linux/sock_diag.h>
137 #include <linux/filter.h>
138 #include <net/sock_reuseport.h>
140 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
148 static void sock_inuse_add(struct net *net, int val);
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 and separate keys for internal and
203 static struct lock_class_key af_family_keys[AF_MAX];
204 static struct lock_class_key af_family_kern_keys[AF_MAX];
205 static struct lock_class_key af_family_slock_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
209 * Make lock validator output more readable. (we pre-construct these
210 * strings build-time, so that runtime initialization of socket
214 #define _sock_locks(x) \
215 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
216 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
217 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
218 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
219 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
220 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
221 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
222 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
223 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
224 x "27" , x "28" , x "AF_CAN" , \
225 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
226 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
227 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
228 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
229 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
232 static const char *const af_family_key_strings[AF_MAX+1] = {
233 _sock_locks("sk_lock-")
235 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
236 _sock_locks("slock-")
238 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
239 _sock_locks("clock-")
242 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
243 _sock_locks("k-sk_lock-")
245 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-slock-")
248 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
249 _sock_locks("k-clock-")
251 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
252 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" ,
253 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK",
254 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" ,
255 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" ,
256 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" ,
257 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" ,
258 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" ,
259 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" ,
260 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" ,
261 "rlock-27" , "rlock-28" , "rlock-AF_CAN" ,
262 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" ,
263 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" ,
264 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" ,
265 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" ,
266 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_XDP" ,
269 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
270 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" ,
271 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK",
272 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" ,
273 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" ,
274 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" ,
275 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" ,
276 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" ,
277 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" ,
278 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" ,
279 "wlock-27" , "wlock-28" , "wlock-AF_CAN" ,
280 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" ,
281 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" ,
282 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" ,
283 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" ,
284 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_XDP" ,
287 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
288 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" ,
289 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK",
290 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" ,
291 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" ,
292 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" ,
293 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" ,
294 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" ,
295 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" ,
296 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" ,
297 "elock-27" , "elock-28" , "elock-AF_CAN" ,
298 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" ,
299 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" ,
300 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" ,
301 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" ,
302 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_XDP" ,
307 * sk_callback_lock and sk queues locking rules are per-address-family,
308 * so split the lock classes by using a per-AF key:
310 static struct lock_class_key af_callback_keys[AF_MAX];
311 static struct lock_class_key af_rlock_keys[AF_MAX];
312 static struct lock_class_key af_wlock_keys[AF_MAX];
313 static struct lock_class_key af_elock_keys[AF_MAX];
314 static struct lock_class_key af_kern_callback_keys[AF_MAX];
316 /* Run time adjustable parameters. */
317 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
318 EXPORT_SYMBOL(sysctl_wmem_max);
319 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
320 EXPORT_SYMBOL(sysctl_rmem_max);
321 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
322 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
324 /* Maximal space eaten by iovec or ancillary data plus some space */
325 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
326 EXPORT_SYMBOL(sysctl_optmem_max);
328 int sysctl_tstamp_allow_data __read_mostly = 1;
330 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
331 EXPORT_SYMBOL_GPL(memalloc_socks_key);
334 * sk_set_memalloc - sets %SOCK_MEMALLOC
335 * @sk: socket to set it on
337 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
338 * It's the responsibility of the admin to adjust min_free_kbytes
339 * to meet the requirements
341 void sk_set_memalloc(struct sock *sk)
343 sock_set_flag(sk, SOCK_MEMALLOC);
344 sk->sk_allocation |= __GFP_MEMALLOC;
345 static_branch_inc(&memalloc_socks_key);
347 EXPORT_SYMBOL_GPL(sk_set_memalloc);
349 void sk_clear_memalloc(struct sock *sk)
351 sock_reset_flag(sk, SOCK_MEMALLOC);
352 sk->sk_allocation &= ~__GFP_MEMALLOC;
353 static_branch_dec(&memalloc_socks_key);
356 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
357 * progress of swapping. SOCK_MEMALLOC may be cleared while
358 * it has rmem allocations due to the last swapfile being deactivated
359 * but there is a risk that the socket is unusable due to exceeding
360 * the rmem limits. Reclaim the reserves and obey rmem limits again.
364 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
366 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
369 unsigned int noreclaim_flag;
371 /* these should have been dropped before queueing */
372 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
374 noreclaim_flag = memalloc_noreclaim_save();
375 ret = sk->sk_backlog_rcv(sk, skb);
376 memalloc_noreclaim_restore(noreclaim_flag);
380 EXPORT_SYMBOL(__sk_backlog_rcv);
382 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
386 if (optlen < sizeof(tv))
388 if (copy_from_user(&tv, optval, sizeof(tv)))
390 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
394 static int warned __read_mostly;
397 if (warned < 10 && net_ratelimit()) {
399 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
400 __func__, current->comm, task_pid_nr(current));
404 *timeo_p = MAX_SCHEDULE_TIMEOUT;
405 if (tv.tv_sec == 0 && tv.tv_usec == 0)
407 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
408 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
412 static void sock_warn_obsolete_bsdism(const char *name)
415 static char warncomm[TASK_COMM_LEN];
416 if (strcmp(warncomm, current->comm) && warned < 5) {
417 strcpy(warncomm, current->comm);
418 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
424 static bool sock_needs_netstamp(const struct sock *sk)
426 switch (sk->sk_family) {
435 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
437 if (sk->sk_flags & flags) {
438 sk->sk_flags &= ~flags;
439 if (sock_needs_netstamp(sk) &&
440 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
441 net_disable_timestamp();
446 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
449 struct sk_buff_head *list = &sk->sk_receive_queue;
451 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
452 atomic_inc(&sk->sk_drops);
453 trace_sock_rcvqueue_full(sk, skb);
457 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
458 atomic_inc(&sk->sk_drops);
463 skb_set_owner_r(skb, sk);
465 /* we escape from rcu protected region, make sure we dont leak
470 spin_lock_irqsave(&list->lock, flags);
471 sock_skb_set_dropcount(sk, skb);
472 __skb_queue_tail(list, skb);
473 spin_unlock_irqrestore(&list->lock, flags);
475 if (!sock_flag(sk, SOCK_DEAD))
476 sk->sk_data_ready(sk);
479 EXPORT_SYMBOL(__sock_queue_rcv_skb);
481 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
485 err = sk_filter(sk, skb);
489 return __sock_queue_rcv_skb(sk, skb);
491 EXPORT_SYMBOL(sock_queue_rcv_skb);
493 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
494 const int nested, unsigned int trim_cap, bool refcounted)
496 int rc = NET_RX_SUCCESS;
498 if (sk_filter_trim_cap(sk, skb, trim_cap))
499 goto discard_and_relse;
503 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
504 atomic_inc(&sk->sk_drops);
505 goto discard_and_relse;
508 bh_lock_sock_nested(sk);
511 if (!sock_owned_by_user(sk)) {
513 * trylock + unlock semantics:
515 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
517 rc = sk_backlog_rcv(sk, skb);
519 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
520 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
522 atomic_inc(&sk->sk_drops);
523 goto discard_and_relse;
535 EXPORT_SYMBOL(__sk_receive_skb);
537 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
539 struct dst_entry *dst = __sk_dst_get(sk);
541 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
542 sk_tx_queue_clear(sk);
543 sk->sk_dst_pending_confirm = 0;
544 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
551 EXPORT_SYMBOL(__sk_dst_check);
553 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
555 struct dst_entry *dst = sk_dst_get(sk);
557 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
565 EXPORT_SYMBOL(sk_dst_check);
567 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
570 int ret = -ENOPROTOOPT;
571 #ifdef CONFIG_NETDEVICES
572 struct net *net = sock_net(sk);
573 char devname[IFNAMSIZ];
578 if (!ns_capable(net->user_ns, CAP_NET_RAW))
585 /* Bind this socket to a particular device like "eth0",
586 * as specified in the passed interface name. If the
587 * name is "" or the option length is zero the socket
590 if (optlen > IFNAMSIZ - 1)
591 optlen = IFNAMSIZ - 1;
592 memset(devname, 0, sizeof(devname));
595 if (copy_from_user(devname, optval, optlen))
599 if (devname[0] != '\0') {
600 struct net_device *dev;
603 dev = dev_get_by_name_rcu(net, devname);
605 index = dev->ifindex;
613 sk->sk_bound_dev_if = index;
625 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
626 int __user *optlen, int len)
628 int ret = -ENOPROTOOPT;
629 #ifdef CONFIG_NETDEVICES
630 struct net *net = sock_net(sk);
631 char devname[IFNAMSIZ];
633 if (sk->sk_bound_dev_if == 0) {
642 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
646 len = strlen(devname) + 1;
649 if (copy_to_user(optval, devname, len))
654 if (put_user(len, optlen))
665 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
668 sock_set_flag(sk, bit);
670 sock_reset_flag(sk, bit);
673 bool sk_mc_loop(struct sock *sk)
675 if (dev_recursion_level())
679 switch (sk->sk_family) {
681 return inet_sk(sk)->mc_loop;
682 #if IS_ENABLED(CONFIG_IPV6)
684 return inet6_sk(sk)->mc_loop;
690 EXPORT_SYMBOL(sk_mc_loop);
693 * This is meant for all protocols to use and covers goings on
694 * at the socket level. Everything here is generic.
697 int sock_setsockopt(struct socket *sock, int level, int optname,
698 char __user *optval, unsigned int optlen)
700 struct sock *sk = sock->sk;
707 * Options without arguments
710 if (optname == SO_BINDTODEVICE)
711 return sock_setbindtodevice(sk, optval, optlen);
713 if (optlen < sizeof(int))
716 if (get_user(val, (int __user *)optval))
719 valbool = val ? 1 : 0;
725 if (val && !capable(CAP_NET_ADMIN))
728 sock_valbool_flag(sk, SOCK_DBG, valbool);
731 val = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
732 if ((sk->sk_family == PF_INET || sk->sk_family == PF_INET6) &&
733 inet_sk(sk)->inet_num &&
734 (sk->sk_reuse != val)) {
735 ret = (sk->sk_state == TCP_ESTABLISHED) ? -EISCONN : -EUCLEAN;
741 if ((sk->sk_family == PF_INET || sk->sk_family == PF_INET6) &&
742 inet_sk(sk)->inet_num &&
743 (sk->sk_reuseport != valbool)) {
744 ret = (sk->sk_state == TCP_ESTABLISHED) ? -EISCONN : -EUCLEAN;
747 sk->sk_reuseport = valbool;
756 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
759 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
762 /* Don't error on this BSD doesn't and if you think
763 * about it this is right. Otherwise apps have to
764 * play 'guess the biggest size' games. RCVBUF/SNDBUF
765 * are treated in BSD as hints
767 val = min_t(u32, val, sysctl_wmem_max);
769 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
770 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
771 /* Wake up sending tasks if we upped the value. */
772 sk->sk_write_space(sk);
776 if (!capable(CAP_NET_ADMIN)) {
783 /* Don't error on this BSD doesn't and if you think
784 * about it this is right. Otherwise apps have to
785 * play 'guess the biggest size' games. RCVBUF/SNDBUF
786 * are treated in BSD as hints
788 val = min_t(u32, val, sysctl_rmem_max);
790 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
792 * We double it on the way in to account for
793 * "struct sk_buff" etc. overhead. Applications
794 * assume that the SO_RCVBUF setting they make will
795 * allow that much actual data to be received on that
798 * Applications are unaware that "struct sk_buff" and
799 * other overheads allocate from the receive buffer
800 * during socket buffer allocation.
802 * And after considering the possible alternatives,
803 * returning the value we actually used in getsockopt
804 * is the most desirable behavior.
806 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
810 if (!capable(CAP_NET_ADMIN)) {
817 if (sk->sk_prot->keepalive)
818 sk->sk_prot->keepalive(sk, valbool);
819 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
823 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
827 sk->sk_no_check_tx = valbool;
831 if ((val >= 0 && val <= 6) ||
832 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
833 sk->sk_priority = val;
839 if (optlen < sizeof(ling)) {
840 ret = -EINVAL; /* 1003.1g */
843 if (copy_from_user(&ling, optval, sizeof(ling))) {
848 sock_reset_flag(sk, SOCK_LINGER);
850 #if (BITS_PER_LONG == 32)
851 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
852 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
855 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
856 sock_set_flag(sk, SOCK_LINGER);
861 sock_warn_obsolete_bsdism("setsockopt");
866 set_bit(SOCK_PASSCRED, &sock->flags);
868 clear_bit(SOCK_PASSCRED, &sock->flags);
874 if (optname == SO_TIMESTAMP)
875 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
877 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
878 sock_set_flag(sk, SOCK_RCVTSTAMP);
879 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
881 sock_reset_flag(sk, SOCK_RCVTSTAMP);
882 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
886 case SO_TIMESTAMPING:
887 if (val & ~SOF_TIMESTAMPING_MASK) {
892 if (val & SOF_TIMESTAMPING_OPT_ID &&
893 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
894 if (sk->sk_protocol == IPPROTO_TCP &&
895 sk->sk_type == SOCK_STREAM) {
896 if ((1 << sk->sk_state) &
897 (TCPF_CLOSE | TCPF_LISTEN)) {
901 sk->sk_tskey = tcp_sk(sk)->snd_una;
907 if (val & SOF_TIMESTAMPING_OPT_STATS &&
908 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
913 sk->sk_tsflags = val;
914 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
915 sock_enable_timestamp(sk,
916 SOCK_TIMESTAMPING_RX_SOFTWARE);
918 sock_disable_timestamp(sk,
919 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
925 if (sock->ops->set_rcvlowat)
926 ret = sock->ops->set_rcvlowat(sk, val);
928 sk->sk_rcvlowat = val ? : 1;
932 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
936 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
939 case SO_ATTACH_FILTER:
941 if (optlen == sizeof(struct sock_fprog)) {
942 struct sock_fprog fprog;
945 if (copy_from_user(&fprog, optval, sizeof(fprog)))
948 ret = sk_attach_filter(&fprog, sk);
954 if (optlen == sizeof(u32)) {
958 if (copy_from_user(&ufd, optval, sizeof(ufd)))
961 ret = sk_attach_bpf(ufd, sk);
965 case SO_ATTACH_REUSEPORT_CBPF:
967 if (optlen == sizeof(struct sock_fprog)) {
968 struct sock_fprog fprog;
971 if (copy_from_user(&fprog, optval, sizeof(fprog)))
974 ret = sk_reuseport_attach_filter(&fprog, sk);
978 case SO_ATTACH_REUSEPORT_EBPF:
980 if (optlen == sizeof(u32)) {
984 if (copy_from_user(&ufd, optval, sizeof(ufd)))
987 ret = sk_reuseport_attach_bpf(ufd, sk);
991 case SO_DETACH_FILTER:
992 ret = sk_detach_filter(sk);
996 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
999 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1004 set_bit(SOCK_PASSSEC, &sock->flags);
1006 clear_bit(SOCK_PASSSEC, &sock->flags);
1009 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1016 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1019 case SO_WIFI_STATUS:
1020 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1024 if (sock->ops->set_peek_off)
1025 ret = sock->ops->set_peek_off(sk, val);
1031 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1034 case SO_SELECT_ERR_QUEUE:
1035 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1038 #ifdef CONFIG_NET_RX_BUSY_POLL
1040 /* allow unprivileged users to decrease the value */
1041 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1047 sk->sk_ll_usec = val;
1052 case SO_MAX_PACING_RATE:
1054 cmpxchg(&sk->sk_pacing_status,
1057 sk->sk_max_pacing_rate = val;
1058 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1059 sk->sk_max_pacing_rate);
1062 case SO_INCOMING_CPU:
1063 sk->sk_incoming_cpu = val;
1068 dst_negative_advice(sk);
1072 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1073 if (sk->sk_protocol != IPPROTO_TCP)
1075 } else if (sk->sk_family != PF_RDS) {
1079 if (val < 0 || val > 1)
1082 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1093 EXPORT_SYMBOL(sock_setsockopt);
1096 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1097 struct ucred *ucred)
1099 ucred->pid = pid_vnr(pid);
1100 ucred->uid = ucred->gid = -1;
1102 struct user_namespace *current_ns = current_user_ns();
1104 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1105 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1109 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1111 struct user_namespace *user_ns = current_user_ns();
1114 for (i = 0; i < src->ngroups; i++)
1115 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1121 int sock_getsockopt(struct socket *sock, int level, int optname,
1122 char __user *optval, int __user *optlen)
1124 struct sock *sk = sock->sk;
1133 int lv = sizeof(int);
1136 if (get_user(len, optlen))
1141 memset(&v, 0, sizeof(v));
1145 v.val = sock_flag(sk, SOCK_DBG);
1149 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1153 v.val = sock_flag(sk, SOCK_BROADCAST);
1157 v.val = sk->sk_sndbuf;
1161 v.val = sk->sk_rcvbuf;
1165 v.val = sk->sk_reuse;
1169 v.val = sk->sk_reuseport;
1173 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1177 v.val = sk->sk_type;
1181 v.val = sk->sk_protocol;
1185 v.val = sk->sk_family;
1189 v.val = -sock_error(sk);
1191 v.val = xchg(&sk->sk_err_soft, 0);
1195 v.val = sock_flag(sk, SOCK_URGINLINE);
1199 v.val = sk->sk_no_check_tx;
1203 v.val = sk->sk_priority;
1207 lv = sizeof(v.ling);
1208 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1209 v.ling.l_linger = sk->sk_lingertime / HZ;
1213 sock_warn_obsolete_bsdism("getsockopt");
1217 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1218 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1221 case SO_TIMESTAMPNS:
1222 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1225 case SO_TIMESTAMPING:
1226 v.val = sk->sk_tsflags;
1230 lv = sizeof(struct timeval);
1231 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1235 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1236 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1241 lv = sizeof(struct timeval);
1242 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1246 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1247 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1252 v.val = sk->sk_rcvlowat;
1260 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1265 struct ucred peercred;
1266 if (len > sizeof(peercred))
1267 len = sizeof(peercred);
1268 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1269 if (copy_to_user(optval, &peercred, len))
1278 if (!sk->sk_peer_cred)
1281 n = sk->sk_peer_cred->group_info->ngroups;
1282 if (len < n * sizeof(gid_t)) {
1283 len = n * sizeof(gid_t);
1284 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1286 len = n * sizeof(gid_t);
1288 ret = groups_to_user((gid_t __user *)optval,
1289 sk->sk_peer_cred->group_info);
1299 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1304 if (copy_to_user(optval, address, len))
1309 /* Dubious BSD thing... Probably nobody even uses it, but
1310 * the UNIX standard wants it for whatever reason... -DaveM
1313 v.val = sk->sk_state == TCP_LISTEN;
1317 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1321 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1324 v.val = sk->sk_mark;
1328 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1331 case SO_WIFI_STATUS:
1332 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1336 if (!sock->ops->set_peek_off)
1339 v.val = sk->sk_peek_off;
1342 v.val = sock_flag(sk, SOCK_NOFCS);
1345 case SO_BINDTODEVICE:
1346 return sock_getbindtodevice(sk, optval, optlen, len);
1349 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1355 case SO_LOCK_FILTER:
1356 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1359 case SO_BPF_EXTENSIONS:
1360 v.val = bpf_tell_extensions();
1363 case SO_SELECT_ERR_QUEUE:
1364 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1367 #ifdef CONFIG_NET_RX_BUSY_POLL
1369 v.val = sk->sk_ll_usec;
1373 case SO_MAX_PACING_RATE:
1374 v.val = sk->sk_max_pacing_rate;
1377 case SO_INCOMING_CPU:
1378 v.val = sk->sk_incoming_cpu;
1383 u32 meminfo[SK_MEMINFO_VARS];
1385 if (get_user(len, optlen))
1388 sk_get_meminfo(sk, meminfo);
1390 len = min_t(unsigned int, len, sizeof(meminfo));
1391 if (copy_to_user(optval, &meminfo, len))
1397 #ifdef CONFIG_NET_RX_BUSY_POLL
1398 case SO_INCOMING_NAPI_ID:
1399 v.val = READ_ONCE(sk->sk_napi_id);
1401 /* aggregate non-NAPI IDs down to 0 */
1402 if (v.val < MIN_NAPI_ID)
1412 v.val64 = sock_gen_cookie(sk);
1416 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1420 /* We implement the SO_SNDLOWAT etc to not be settable
1423 return -ENOPROTOOPT;
1428 if (copy_to_user(optval, &v, len))
1431 if (put_user(len, optlen))
1437 * Initialize an sk_lock.
1439 * (We also register the sk_lock with the lock validator.)
1441 static inline void sock_lock_init(struct sock *sk)
1443 if (sk->sk_kern_sock)
1444 sock_lock_init_class_and_name(
1446 af_family_kern_slock_key_strings[sk->sk_family],
1447 af_family_kern_slock_keys + sk->sk_family,
1448 af_family_kern_key_strings[sk->sk_family],
1449 af_family_kern_keys + sk->sk_family);
1451 sock_lock_init_class_and_name(
1453 af_family_slock_key_strings[sk->sk_family],
1454 af_family_slock_keys + sk->sk_family,
1455 af_family_key_strings[sk->sk_family],
1456 af_family_keys + sk->sk_family);
1460 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1461 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1462 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1464 static void sock_copy(struct sock *nsk, const struct sock *osk)
1466 #ifdef CONFIG_SECURITY_NETWORK
1467 void *sptr = nsk->sk_security;
1469 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1471 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1472 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1474 #ifdef CONFIG_SECURITY_NETWORK
1475 nsk->sk_security = sptr;
1476 security_sk_clone(osk, nsk);
1480 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1484 struct kmem_cache *slab;
1488 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1491 if (priority & __GFP_ZERO)
1492 sk_prot_clear_nulls(sk, prot->obj_size);
1494 sk = kmalloc(prot->obj_size, priority);
1497 if (security_sk_alloc(sk, family, priority))
1500 if (!try_module_get(prot->owner))
1502 sk_tx_queue_clear(sk);
1508 security_sk_free(sk);
1511 kmem_cache_free(slab, sk);
1517 static void sk_prot_free(struct proto *prot, struct sock *sk)
1519 struct kmem_cache *slab;
1520 struct module *owner;
1522 owner = prot->owner;
1525 cgroup_sk_free(&sk->sk_cgrp_data);
1526 mem_cgroup_sk_free(sk);
1527 security_sk_free(sk);
1529 kmem_cache_free(slab, sk);
1536 * sk_alloc - All socket objects are allocated here
1537 * @net: the applicable net namespace
1538 * @family: protocol family
1539 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1540 * @prot: struct proto associated with this new sock instance
1541 * @kern: is this to be a kernel socket?
1543 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1544 struct proto *prot, int kern)
1548 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1550 sk->sk_family = family;
1552 * See comment in struct sock definition to understand
1553 * why we need sk_prot_creator -acme
1555 sk->sk_prot = sk->sk_prot_creator = prot;
1556 sk->sk_kern_sock = kern;
1558 sk->sk_net_refcnt = kern ? 0 : 1;
1559 if (likely(sk->sk_net_refcnt)) {
1561 sock_inuse_add(net, 1);
1564 sock_net_set(sk, net);
1565 refcount_set(&sk->sk_wmem_alloc, 1);
1567 mem_cgroup_sk_alloc(sk);
1568 cgroup_sk_alloc(&sk->sk_cgrp_data);
1569 sock_update_classid(&sk->sk_cgrp_data);
1570 sock_update_netprioidx(&sk->sk_cgrp_data);
1575 EXPORT_SYMBOL(sk_alloc);
1577 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1578 * grace period. This is the case for UDP sockets and TCP listeners.
1580 static void __sk_destruct(struct rcu_head *head)
1582 struct sock *sk = container_of(head, struct sock, sk_rcu);
1583 struct sk_filter *filter;
1585 if (sk->sk_destruct)
1586 sk->sk_destruct(sk);
1588 filter = rcu_dereference_check(sk->sk_filter,
1589 refcount_read(&sk->sk_wmem_alloc) == 0);
1591 sk_filter_uncharge(sk, filter);
1592 RCU_INIT_POINTER(sk->sk_filter, NULL);
1594 if (rcu_access_pointer(sk->sk_reuseport_cb))
1595 reuseport_detach_sock(sk);
1597 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1599 if (atomic_read(&sk->sk_omem_alloc))
1600 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1601 __func__, atomic_read(&sk->sk_omem_alloc));
1603 if (sk->sk_frag.page) {
1604 put_page(sk->sk_frag.page);
1605 sk->sk_frag.page = NULL;
1608 if (sk->sk_peer_cred)
1609 put_cred(sk->sk_peer_cred);
1610 put_pid(sk->sk_peer_pid);
1611 if (likely(sk->sk_net_refcnt))
1612 put_net(sock_net(sk));
1613 sk_prot_free(sk->sk_prot_creator, sk);
1616 void sk_destruct(struct sock *sk)
1618 if (sock_flag(sk, SOCK_RCU_FREE))
1619 call_rcu(&sk->sk_rcu, __sk_destruct);
1621 __sk_destruct(&sk->sk_rcu);
1624 static void __sk_free(struct sock *sk)
1626 if (likely(sk->sk_net_refcnt))
1627 sock_inuse_add(sock_net(sk), -1);
1629 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1630 sock_diag_broadcast_destroy(sk);
1635 void sk_free(struct sock *sk)
1638 * We subtract one from sk_wmem_alloc and can know if
1639 * some packets are still in some tx queue.
1640 * If not null, sock_wfree() will call __sk_free(sk) later
1642 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1645 EXPORT_SYMBOL(sk_free);
1647 static void sk_init_common(struct sock *sk)
1649 skb_queue_head_init(&sk->sk_receive_queue);
1650 skb_queue_head_init(&sk->sk_write_queue);
1651 skb_queue_head_init(&sk->sk_error_queue);
1653 rwlock_init(&sk->sk_callback_lock);
1654 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1655 af_rlock_keys + sk->sk_family,
1656 af_family_rlock_key_strings[sk->sk_family]);
1657 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1658 af_wlock_keys + sk->sk_family,
1659 af_family_wlock_key_strings[sk->sk_family]);
1660 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1661 af_elock_keys + sk->sk_family,
1662 af_family_elock_key_strings[sk->sk_family]);
1663 lockdep_set_class_and_name(&sk->sk_callback_lock,
1664 af_callback_keys + sk->sk_family,
1665 af_family_clock_key_strings[sk->sk_family]);
1669 * sk_clone_lock - clone a socket, and lock its clone
1670 * @sk: the socket to clone
1671 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1673 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1675 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1678 bool is_charged = true;
1680 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1681 if (newsk != NULL) {
1682 struct sk_filter *filter;
1684 sock_copy(newsk, sk);
1686 newsk->sk_prot_creator = sk->sk_prot;
1689 if (likely(newsk->sk_net_refcnt))
1690 get_net(sock_net(newsk));
1691 sk_node_init(&newsk->sk_node);
1692 sock_lock_init(newsk);
1693 bh_lock_sock(newsk);
1694 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1695 newsk->sk_backlog.len = 0;
1697 atomic_set(&newsk->sk_rmem_alloc, 0);
1699 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1701 refcount_set(&newsk->sk_wmem_alloc, 1);
1702 atomic_set(&newsk->sk_omem_alloc, 0);
1703 sk_init_common(newsk);
1705 newsk->sk_dst_cache = NULL;
1706 newsk->sk_dst_pending_confirm = 0;
1707 newsk->sk_wmem_queued = 0;
1708 newsk->sk_forward_alloc = 0;
1709 atomic_set(&newsk->sk_drops, 0);
1710 newsk->sk_send_head = NULL;
1711 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1712 atomic_set(&newsk->sk_zckey, 0);
1714 sock_reset_flag(newsk, SOCK_DONE);
1715 mem_cgroup_sk_alloc(newsk);
1716 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1719 filter = rcu_dereference(sk->sk_filter);
1721 /* though it's an empty new sock, the charging may fail
1722 * if sysctl_optmem_max was changed between creation of
1723 * original socket and cloning
1725 is_charged = sk_filter_charge(newsk, filter);
1726 RCU_INIT_POINTER(newsk->sk_filter, filter);
1729 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1730 /* We need to make sure that we don't uncharge the new
1731 * socket if we couldn't charge it in the first place
1732 * as otherwise we uncharge the parent's filter.
1735 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1736 sk_free_unlock_clone(newsk);
1740 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1743 newsk->sk_err_soft = 0;
1744 newsk->sk_priority = 0;
1745 newsk->sk_incoming_cpu = raw_smp_processor_id();
1746 atomic64_set(&newsk->sk_cookie, 0);
1747 if (likely(newsk->sk_net_refcnt))
1748 sock_inuse_add(sock_net(newsk), 1);
1751 * Before updating sk_refcnt, we must commit prior changes to memory
1752 * (Documentation/RCU/rculist_nulls.txt for details)
1755 refcount_set(&newsk->sk_refcnt, 2);
1758 * Increment the counter in the same struct proto as the master
1759 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1760 * is the same as sk->sk_prot->socks, as this field was copied
1763 * This _changes_ the previous behaviour, where
1764 * tcp_create_openreq_child always was incrementing the
1765 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1766 * to be taken into account in all callers. -acme
1768 sk_refcnt_debug_inc(newsk);
1769 sk_set_socket(newsk, NULL);
1770 newsk->sk_wq = NULL;
1772 if (newsk->sk_prot->sockets_allocated)
1773 sk_sockets_allocated_inc(newsk);
1775 if (sock_needs_netstamp(sk) &&
1776 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1777 net_enable_timestamp();
1782 EXPORT_SYMBOL_GPL(sk_clone_lock);
1784 void sk_free_unlock_clone(struct sock *sk)
1786 /* It is still raw copy of parent, so invalidate
1787 * destructor and make plain sk_free() */
1788 sk->sk_destruct = NULL;
1792 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1794 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1798 sk_dst_set(sk, dst);
1799 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1800 if (sk->sk_route_caps & NETIF_F_GSO)
1801 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1802 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1803 if (sk_can_gso(sk)) {
1804 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1805 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1807 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1808 sk->sk_gso_max_size = dst->dev->gso_max_size;
1809 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1812 sk->sk_gso_max_segs = max_segs;
1814 EXPORT_SYMBOL_GPL(sk_setup_caps);
1817 * Simple resource managers for sockets.
1822 * Write buffer destructor automatically called from kfree_skb.
1824 void sock_wfree(struct sk_buff *skb)
1826 struct sock *sk = skb->sk;
1827 unsigned int len = skb->truesize;
1829 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1831 * Keep a reference on sk_wmem_alloc, this will be released
1832 * after sk_write_space() call
1834 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1835 sk->sk_write_space(sk);
1839 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1840 * could not do because of in-flight packets
1842 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1845 EXPORT_SYMBOL(sock_wfree);
1847 /* This variant of sock_wfree() is used by TCP,
1848 * since it sets SOCK_USE_WRITE_QUEUE.
1850 void __sock_wfree(struct sk_buff *skb)
1852 struct sock *sk = skb->sk;
1854 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1858 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1863 if (unlikely(!sk_fullsock(sk))) {
1864 skb->destructor = sock_edemux;
1869 skb->destructor = sock_wfree;
1870 skb_set_hash_from_sk(skb, sk);
1872 * We used to take a refcount on sk, but following operation
1873 * is enough to guarantee sk_free() wont free this sock until
1874 * all in-flight packets are completed
1876 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1878 EXPORT_SYMBOL(skb_set_owner_w);
1880 /* This helper is used by netem, as it can hold packets in its
1881 * delay queue. We want to allow the owner socket to send more
1882 * packets, as if they were already TX completed by a typical driver.
1883 * But we also want to keep skb->sk set because some packet schedulers
1884 * rely on it (sch_fq for example).
1886 void skb_orphan_partial(struct sk_buff *skb)
1888 if (skb_is_tcp_pure_ack(skb))
1891 if (skb->destructor == sock_wfree
1893 || skb->destructor == tcp_wfree
1896 struct sock *sk = skb->sk;
1898 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1899 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
1900 skb->destructor = sock_efree;
1906 EXPORT_SYMBOL(skb_orphan_partial);
1909 * Read buffer destructor automatically called from kfree_skb.
1911 void sock_rfree(struct sk_buff *skb)
1913 struct sock *sk = skb->sk;
1914 unsigned int len = skb->truesize;
1916 atomic_sub(len, &sk->sk_rmem_alloc);
1917 sk_mem_uncharge(sk, len);
1919 EXPORT_SYMBOL(sock_rfree);
1922 * Buffer destructor for skbs that are not used directly in read or write
1923 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1925 void sock_efree(struct sk_buff *skb)
1929 EXPORT_SYMBOL(sock_efree);
1931 kuid_t sock_i_uid(struct sock *sk)
1935 read_lock_bh(&sk->sk_callback_lock);
1936 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1937 read_unlock_bh(&sk->sk_callback_lock);
1940 EXPORT_SYMBOL(sock_i_uid);
1942 unsigned long sock_i_ino(struct sock *sk)
1946 read_lock_bh(&sk->sk_callback_lock);
1947 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1948 read_unlock_bh(&sk->sk_callback_lock);
1951 EXPORT_SYMBOL(sock_i_ino);
1954 * Allocate a skb from the socket's send buffer.
1956 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1959 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1960 struct sk_buff *skb = alloc_skb(size, priority);
1962 skb_set_owner_w(skb, sk);
1968 EXPORT_SYMBOL(sock_wmalloc);
1970 static void sock_ofree(struct sk_buff *skb)
1972 struct sock *sk = skb->sk;
1974 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
1977 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1980 struct sk_buff *skb;
1982 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
1983 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
1987 skb = alloc_skb(size, priority);
1991 atomic_add(skb->truesize, &sk->sk_omem_alloc);
1993 skb->destructor = sock_ofree;
1998 * Allocate a memory block from the socket's option memory buffer.
2000 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2002 if ((unsigned int)size <= sysctl_optmem_max &&
2003 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2005 /* First do the add, to avoid the race if kmalloc
2008 atomic_add(size, &sk->sk_omem_alloc);
2009 mem = kmalloc(size, priority);
2012 atomic_sub(size, &sk->sk_omem_alloc);
2016 EXPORT_SYMBOL(sock_kmalloc);
2018 /* Free an option memory block. Note, we actually want the inline
2019 * here as this allows gcc to detect the nullify and fold away the
2020 * condition entirely.
2022 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2025 if (WARN_ON_ONCE(!mem))
2031 atomic_sub(size, &sk->sk_omem_alloc);
2034 void sock_kfree_s(struct sock *sk, void *mem, int size)
2036 __sock_kfree_s(sk, mem, size, false);
2038 EXPORT_SYMBOL(sock_kfree_s);
2040 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2042 __sock_kfree_s(sk, mem, size, true);
2044 EXPORT_SYMBOL(sock_kzfree_s);
2046 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2047 I think, these locks should be removed for datagram sockets.
2049 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2053 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2057 if (signal_pending(current))
2059 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2060 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2061 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2063 if (sk->sk_shutdown & SEND_SHUTDOWN)
2067 timeo = schedule_timeout(timeo);
2069 finish_wait(sk_sleep(sk), &wait);
2075 * Generic send/receive buffer handlers
2078 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2079 unsigned long data_len, int noblock,
2080 int *errcode, int max_page_order)
2082 struct sk_buff *skb;
2086 timeo = sock_sndtimeo(sk, noblock);
2088 err = sock_error(sk);
2093 if (sk->sk_shutdown & SEND_SHUTDOWN)
2096 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2099 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2100 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2104 if (signal_pending(current))
2106 timeo = sock_wait_for_wmem(sk, timeo);
2108 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2109 errcode, sk->sk_allocation);
2111 skb_set_owner_w(skb, sk);
2115 err = sock_intr_errno(timeo);
2120 EXPORT_SYMBOL(sock_alloc_send_pskb);
2122 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2123 int noblock, int *errcode)
2125 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2127 EXPORT_SYMBOL(sock_alloc_send_skb);
2129 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2130 struct sockcm_cookie *sockc)
2134 switch (cmsg->cmsg_type) {
2136 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2138 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2140 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2142 case SO_TIMESTAMPING:
2143 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2146 tsflags = *(u32 *)CMSG_DATA(cmsg);
2147 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2150 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2151 sockc->tsflags |= tsflags;
2153 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2155 case SCM_CREDENTIALS:
2162 EXPORT_SYMBOL(__sock_cmsg_send);
2164 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2165 struct sockcm_cookie *sockc)
2167 struct cmsghdr *cmsg;
2170 for_each_cmsghdr(cmsg, msg) {
2171 if (!CMSG_OK(msg, cmsg))
2173 if (cmsg->cmsg_level != SOL_SOCKET)
2175 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2181 EXPORT_SYMBOL(sock_cmsg_send);
2183 static void sk_enter_memory_pressure(struct sock *sk)
2185 if (!sk->sk_prot->enter_memory_pressure)
2188 sk->sk_prot->enter_memory_pressure(sk);
2191 static void sk_leave_memory_pressure(struct sock *sk)
2193 if (sk->sk_prot->leave_memory_pressure) {
2194 sk->sk_prot->leave_memory_pressure(sk);
2196 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2198 if (memory_pressure && *memory_pressure)
2199 *memory_pressure = 0;
2203 /* On 32bit arches, an skb frag is limited to 2^15 */
2204 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2207 * skb_page_frag_refill - check that a page_frag contains enough room
2208 * @sz: minimum size of the fragment we want to get
2209 * @pfrag: pointer to page_frag
2210 * @gfp: priority for memory allocation
2212 * Note: While this allocator tries to use high order pages, there is
2213 * no guarantee that allocations succeed. Therefore, @sz MUST be
2214 * less or equal than PAGE_SIZE.
2216 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2219 if (page_ref_count(pfrag->page) == 1) {
2223 if (pfrag->offset + sz <= pfrag->size)
2225 put_page(pfrag->page);
2229 if (SKB_FRAG_PAGE_ORDER) {
2230 /* Avoid direct reclaim but allow kswapd to wake */
2231 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2232 __GFP_COMP | __GFP_NOWARN |
2234 SKB_FRAG_PAGE_ORDER);
2235 if (likely(pfrag->page)) {
2236 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2240 pfrag->page = alloc_page(gfp);
2241 if (likely(pfrag->page)) {
2242 pfrag->size = PAGE_SIZE;
2247 EXPORT_SYMBOL(skb_page_frag_refill);
2249 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2251 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2254 sk_enter_memory_pressure(sk);
2255 sk_stream_moderate_sndbuf(sk);
2258 EXPORT_SYMBOL(sk_page_frag_refill);
2260 int sk_alloc_sg(struct sock *sk, int len, struct scatterlist *sg,
2261 int sg_start, int *sg_curr_index, unsigned int *sg_curr_size,
2264 int sg_curr = *sg_curr_index, use = 0, rc = 0;
2265 unsigned int size = *sg_curr_size;
2266 struct page_frag *pfrag;
2267 struct scatterlist *sge;
2270 pfrag = sk_page_frag(sk);
2273 unsigned int orig_offset;
2275 if (!sk_page_frag_refill(sk, pfrag)) {
2280 use = min_t(int, len, pfrag->size - pfrag->offset);
2282 if (!sk_wmem_schedule(sk, use)) {
2287 sk_mem_charge(sk, use);
2289 orig_offset = pfrag->offset;
2290 pfrag->offset += use;
2292 sge = sg + sg_curr - 1;
2293 if (sg_curr > first_coalesce && sg_page(sg) == pfrag->page &&
2294 sg->offset + sg->length == orig_offset) {
2299 sg_set_page(sge, pfrag->page, use, orig_offset);
2300 get_page(pfrag->page);
2303 if (sg_curr == MAX_SKB_FRAGS)
2306 if (sg_curr == sg_start) {
2315 *sg_curr_size = size;
2316 *sg_curr_index = sg_curr;
2319 EXPORT_SYMBOL(sk_alloc_sg);
2321 static void __lock_sock(struct sock *sk)
2322 __releases(&sk->sk_lock.slock)
2323 __acquires(&sk->sk_lock.slock)
2328 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2329 TASK_UNINTERRUPTIBLE);
2330 spin_unlock_bh(&sk->sk_lock.slock);
2332 spin_lock_bh(&sk->sk_lock.slock);
2333 if (!sock_owned_by_user(sk))
2336 finish_wait(&sk->sk_lock.wq, &wait);
2339 static void __release_sock(struct sock *sk)
2340 __releases(&sk->sk_lock.slock)
2341 __acquires(&sk->sk_lock.slock)
2343 struct sk_buff *skb, *next;
2345 while ((skb = sk->sk_backlog.head) != NULL) {
2346 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2348 spin_unlock_bh(&sk->sk_lock.slock);
2353 WARN_ON_ONCE(skb_dst_is_noref(skb));
2355 sk_backlog_rcv(sk, skb);
2360 } while (skb != NULL);
2362 spin_lock_bh(&sk->sk_lock.slock);
2366 * Doing the zeroing here guarantee we can not loop forever
2367 * while a wild producer attempts to flood us.
2369 sk->sk_backlog.len = 0;
2372 void __sk_flush_backlog(struct sock *sk)
2374 spin_lock_bh(&sk->sk_lock.slock);
2376 spin_unlock_bh(&sk->sk_lock.slock);
2380 * sk_wait_data - wait for data to arrive at sk_receive_queue
2381 * @sk: sock to wait on
2382 * @timeo: for how long
2383 * @skb: last skb seen on sk_receive_queue
2385 * Now socket state including sk->sk_err is changed only under lock,
2386 * hence we may omit checks after joining wait queue.
2387 * We check receive queue before schedule() only as optimization;
2388 * it is very likely that release_sock() added new data.
2390 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2392 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2395 add_wait_queue(sk_sleep(sk), &wait);
2396 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2397 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2398 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2399 remove_wait_queue(sk_sleep(sk), &wait);
2402 EXPORT_SYMBOL(sk_wait_data);
2405 * __sk_mem_raise_allocated - increase memory_allocated
2407 * @size: memory size to allocate
2408 * @amt: pages to allocate
2409 * @kind: allocation type
2411 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2413 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2415 struct proto *prot = sk->sk_prot;
2416 long allocated = sk_memory_allocated_add(sk, amt);
2418 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2419 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2420 goto suppress_allocation;
2423 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2424 sk_leave_memory_pressure(sk);
2428 /* Under pressure. */
2429 if (allocated > sk_prot_mem_limits(sk, 1))
2430 sk_enter_memory_pressure(sk);
2432 /* Over hard limit. */
2433 if (allocated > sk_prot_mem_limits(sk, 2))
2434 goto suppress_allocation;
2436 /* guarantee minimum buffer size under pressure */
2437 if (kind == SK_MEM_RECV) {
2438 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2441 } else { /* SK_MEM_SEND */
2442 int wmem0 = sk_get_wmem0(sk, prot);
2444 if (sk->sk_type == SOCK_STREAM) {
2445 if (sk->sk_wmem_queued < wmem0)
2447 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2452 if (sk_has_memory_pressure(sk)) {
2455 if (!sk_under_memory_pressure(sk))
2457 alloc = sk_sockets_allocated_read_positive(sk);
2458 if (sk_prot_mem_limits(sk, 2) > alloc *
2459 sk_mem_pages(sk->sk_wmem_queued +
2460 atomic_read(&sk->sk_rmem_alloc) +
2461 sk->sk_forward_alloc))
2465 suppress_allocation:
2467 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2468 sk_stream_moderate_sndbuf(sk);
2470 /* Fail only if socket is _under_ its sndbuf.
2471 * In this case we cannot block, so that we have to fail.
2473 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2477 trace_sock_exceed_buf_limit(sk, prot, allocated);
2479 sk_memory_allocated_sub(sk, amt);
2481 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2482 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2486 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2489 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2491 * @size: memory size to allocate
2492 * @kind: allocation type
2494 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2495 * rmem allocation. This function assumes that protocols which have
2496 * memory_pressure use sk_wmem_queued as write buffer accounting.
2498 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2500 int ret, amt = sk_mem_pages(size);
2502 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2503 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2505 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2508 EXPORT_SYMBOL(__sk_mem_schedule);
2511 * __sk_mem_reduce_allocated - reclaim memory_allocated
2513 * @amount: number of quanta
2515 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2517 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2519 sk_memory_allocated_sub(sk, amount);
2521 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2522 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2524 if (sk_under_memory_pressure(sk) &&
2525 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2526 sk_leave_memory_pressure(sk);
2528 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2531 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2533 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2535 void __sk_mem_reclaim(struct sock *sk, int amount)
2537 amount >>= SK_MEM_QUANTUM_SHIFT;
2538 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2539 __sk_mem_reduce_allocated(sk, amount);
2541 EXPORT_SYMBOL(__sk_mem_reclaim);
2543 int sk_set_peek_off(struct sock *sk, int val)
2545 sk->sk_peek_off = val;
2548 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2551 * Set of default routines for initialising struct proto_ops when
2552 * the protocol does not support a particular function. In certain
2553 * cases where it makes no sense for a protocol to have a "do nothing"
2554 * function, some default processing is provided.
2557 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2561 EXPORT_SYMBOL(sock_no_bind);
2563 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2568 EXPORT_SYMBOL(sock_no_connect);
2570 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2574 EXPORT_SYMBOL(sock_no_socketpair);
2576 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2581 EXPORT_SYMBOL(sock_no_accept);
2583 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2588 EXPORT_SYMBOL(sock_no_getname);
2590 __poll_t sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2594 EXPORT_SYMBOL(sock_no_poll);
2596 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2600 EXPORT_SYMBOL(sock_no_ioctl);
2602 int sock_no_listen(struct socket *sock, int backlog)
2606 EXPORT_SYMBOL(sock_no_listen);
2608 int sock_no_shutdown(struct socket *sock, int how)
2612 EXPORT_SYMBOL(sock_no_shutdown);
2614 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2615 char __user *optval, unsigned int optlen)
2619 EXPORT_SYMBOL(sock_no_setsockopt);
2621 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2622 char __user *optval, int __user *optlen)
2626 EXPORT_SYMBOL(sock_no_getsockopt);
2628 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2632 EXPORT_SYMBOL(sock_no_sendmsg);
2634 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2638 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2640 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2645 EXPORT_SYMBOL(sock_no_recvmsg);
2647 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2649 /* Mirror missing mmap method error code */
2652 EXPORT_SYMBOL(sock_no_mmap);
2654 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2657 struct msghdr msg = {.msg_flags = flags};
2659 char *kaddr = kmap(page);
2660 iov.iov_base = kaddr + offset;
2662 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2666 EXPORT_SYMBOL(sock_no_sendpage);
2668 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2669 int offset, size_t size, int flags)
2672 struct msghdr msg = {.msg_flags = flags};
2674 char *kaddr = kmap(page);
2676 iov.iov_base = kaddr + offset;
2678 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2682 EXPORT_SYMBOL(sock_no_sendpage_locked);
2685 * Default Socket Callbacks
2688 static void sock_def_wakeup(struct sock *sk)
2690 struct socket_wq *wq;
2693 wq = rcu_dereference(sk->sk_wq);
2694 if (skwq_has_sleeper(wq))
2695 wake_up_interruptible_all(&wq->wait);
2699 static void sock_def_error_report(struct sock *sk)
2701 struct socket_wq *wq;
2704 wq = rcu_dereference(sk->sk_wq);
2705 if (skwq_has_sleeper(wq))
2706 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2707 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2711 static void sock_def_readable(struct sock *sk)
2713 struct socket_wq *wq;
2716 wq = rcu_dereference(sk->sk_wq);
2717 if (skwq_has_sleeper(wq))
2718 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2719 EPOLLRDNORM | EPOLLRDBAND);
2720 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2724 static void sock_def_write_space(struct sock *sk)
2726 struct socket_wq *wq;
2730 /* Do not wake up a writer until he can make "significant"
2733 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2734 wq = rcu_dereference(sk->sk_wq);
2735 if (skwq_has_sleeper(wq))
2736 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2737 EPOLLWRNORM | EPOLLWRBAND);
2739 /* Should agree with poll, otherwise some programs break */
2740 if (sock_writeable(sk))
2741 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2747 static void sock_def_destruct(struct sock *sk)
2751 void sk_send_sigurg(struct sock *sk)
2753 if (sk->sk_socket && sk->sk_socket->file)
2754 if (send_sigurg(&sk->sk_socket->file->f_owner))
2755 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2757 EXPORT_SYMBOL(sk_send_sigurg);
2759 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2760 unsigned long expires)
2762 if (!mod_timer(timer, expires))
2765 EXPORT_SYMBOL(sk_reset_timer);
2767 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2769 if (del_timer(timer))
2772 EXPORT_SYMBOL(sk_stop_timer);
2774 void sock_init_data(struct socket *sock, struct sock *sk)
2777 sk->sk_send_head = NULL;
2779 timer_setup(&sk->sk_timer, NULL, 0);
2781 sk->sk_allocation = GFP_KERNEL;
2782 sk->sk_rcvbuf = sysctl_rmem_default;
2783 sk->sk_sndbuf = sysctl_wmem_default;
2784 sk->sk_state = TCP_CLOSE;
2785 sk_set_socket(sk, sock);
2787 sock_set_flag(sk, SOCK_ZAPPED);
2790 sk->sk_type = sock->type;
2791 sk->sk_wq = sock->wq;
2793 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2796 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2799 rwlock_init(&sk->sk_callback_lock);
2800 if (sk->sk_kern_sock)
2801 lockdep_set_class_and_name(
2802 &sk->sk_callback_lock,
2803 af_kern_callback_keys + sk->sk_family,
2804 af_family_kern_clock_key_strings[sk->sk_family]);
2806 lockdep_set_class_and_name(
2807 &sk->sk_callback_lock,
2808 af_callback_keys + sk->sk_family,
2809 af_family_clock_key_strings[sk->sk_family]);
2811 sk->sk_state_change = sock_def_wakeup;
2812 sk->sk_data_ready = sock_def_readable;
2813 sk->sk_write_space = sock_def_write_space;
2814 sk->sk_error_report = sock_def_error_report;
2815 sk->sk_destruct = sock_def_destruct;
2817 sk->sk_frag.page = NULL;
2818 sk->sk_frag.offset = 0;
2819 sk->sk_peek_off = -1;
2821 sk->sk_peer_pid = NULL;
2822 sk->sk_peer_cred = NULL;
2823 sk->sk_write_pending = 0;
2824 sk->sk_rcvlowat = 1;
2825 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2826 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2828 sk->sk_stamp = SK_DEFAULT_STAMP;
2829 atomic_set(&sk->sk_zckey, 0);
2831 #ifdef CONFIG_NET_RX_BUSY_POLL
2833 sk->sk_ll_usec = sysctl_net_busy_read;
2836 sk->sk_max_pacing_rate = ~0U;
2837 sk->sk_pacing_rate = ~0U;
2838 sk->sk_pacing_shift = 10;
2839 sk->sk_incoming_cpu = -1;
2841 * Before updating sk_refcnt, we must commit prior changes to memory
2842 * (Documentation/RCU/rculist_nulls.txt for details)
2845 refcount_set(&sk->sk_refcnt, 1);
2846 atomic_set(&sk->sk_drops, 0);
2848 EXPORT_SYMBOL(sock_init_data);
2850 void lock_sock_nested(struct sock *sk, int subclass)
2853 spin_lock_bh(&sk->sk_lock.slock);
2854 if (sk->sk_lock.owned)
2856 sk->sk_lock.owned = 1;
2857 spin_unlock(&sk->sk_lock.slock);
2859 * The sk_lock has mutex_lock() semantics here:
2861 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2864 EXPORT_SYMBOL(lock_sock_nested);
2866 void release_sock(struct sock *sk)
2868 spin_lock_bh(&sk->sk_lock.slock);
2869 if (sk->sk_backlog.tail)
2872 /* Warning : release_cb() might need to release sk ownership,
2873 * ie call sock_release_ownership(sk) before us.
2875 if (sk->sk_prot->release_cb)
2876 sk->sk_prot->release_cb(sk);
2878 sock_release_ownership(sk);
2879 if (waitqueue_active(&sk->sk_lock.wq))
2880 wake_up(&sk->sk_lock.wq);
2881 spin_unlock_bh(&sk->sk_lock.slock);
2883 EXPORT_SYMBOL(release_sock);
2886 * lock_sock_fast - fast version of lock_sock
2889 * This version should be used for very small section, where process wont block
2890 * return false if fast path is taken:
2892 * sk_lock.slock locked, owned = 0, BH disabled
2894 * return true if slow path is taken:
2896 * sk_lock.slock unlocked, owned = 1, BH enabled
2898 bool lock_sock_fast(struct sock *sk)
2901 spin_lock_bh(&sk->sk_lock.slock);
2903 if (!sk->sk_lock.owned)
2905 * Note : We must disable BH
2910 sk->sk_lock.owned = 1;
2911 spin_unlock(&sk->sk_lock.slock);
2913 * The sk_lock has mutex_lock() semantics here:
2915 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2919 EXPORT_SYMBOL(lock_sock_fast);
2921 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2924 if (!sock_flag(sk, SOCK_TIMESTAMP))
2925 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2926 tv = ktime_to_timeval(sk->sk_stamp);
2927 if (tv.tv_sec == -1)
2929 if (tv.tv_sec == 0) {
2930 sk->sk_stamp = ktime_get_real();
2931 tv = ktime_to_timeval(sk->sk_stamp);
2933 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2935 EXPORT_SYMBOL(sock_get_timestamp);
2937 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2940 if (!sock_flag(sk, SOCK_TIMESTAMP))
2941 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2942 ts = ktime_to_timespec(sk->sk_stamp);
2943 if (ts.tv_sec == -1)
2945 if (ts.tv_sec == 0) {
2946 sk->sk_stamp = ktime_get_real();
2947 ts = ktime_to_timespec(sk->sk_stamp);
2949 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2951 EXPORT_SYMBOL(sock_get_timestampns);
2953 void sock_enable_timestamp(struct sock *sk, int flag)
2955 if (!sock_flag(sk, flag)) {
2956 unsigned long previous_flags = sk->sk_flags;
2958 sock_set_flag(sk, flag);
2960 * we just set one of the two flags which require net
2961 * time stamping, but time stamping might have been on
2962 * already because of the other one
2964 if (sock_needs_netstamp(sk) &&
2965 !(previous_flags & SK_FLAGS_TIMESTAMP))
2966 net_enable_timestamp();
2970 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2971 int level, int type)
2973 struct sock_exterr_skb *serr;
2974 struct sk_buff *skb;
2978 skb = sock_dequeue_err_skb(sk);
2984 msg->msg_flags |= MSG_TRUNC;
2987 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2991 sock_recv_timestamp(msg, sk, skb);
2993 serr = SKB_EXT_ERR(skb);
2994 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2996 msg->msg_flags |= MSG_ERRQUEUE;
3004 EXPORT_SYMBOL(sock_recv_errqueue);
3007 * Get a socket option on an socket.
3009 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3010 * asynchronous errors should be reported by getsockopt. We assume
3011 * this means if you specify SO_ERROR (otherwise whats the point of it).
3013 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3014 char __user *optval, int __user *optlen)
3016 struct sock *sk = sock->sk;
3018 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3020 EXPORT_SYMBOL(sock_common_getsockopt);
3022 #ifdef CONFIG_COMPAT
3023 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3024 char __user *optval, int __user *optlen)
3026 struct sock *sk = sock->sk;
3028 if (sk->sk_prot->compat_getsockopt != NULL)
3029 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3031 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3033 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3036 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3039 struct sock *sk = sock->sk;
3043 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3044 flags & ~MSG_DONTWAIT, &addr_len);
3046 msg->msg_namelen = addr_len;
3049 EXPORT_SYMBOL(sock_common_recvmsg);
3052 * Set socket options on an inet socket.
3054 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3055 char __user *optval, unsigned int optlen)
3057 struct sock *sk = sock->sk;
3059 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3061 EXPORT_SYMBOL(sock_common_setsockopt);
3063 #ifdef CONFIG_COMPAT
3064 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3065 char __user *optval, unsigned int optlen)
3067 struct sock *sk = sock->sk;
3069 if (sk->sk_prot->compat_setsockopt != NULL)
3070 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3072 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3074 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3077 void sk_common_release(struct sock *sk)
3079 if (sk->sk_prot->destroy)
3080 sk->sk_prot->destroy(sk);
3083 * Observation: when sock_common_release is called, processes have
3084 * no access to socket. But net still has.
3085 * Step one, detach it from networking:
3087 * A. Remove from hash tables.
3090 sk->sk_prot->unhash(sk);
3093 * In this point socket cannot receive new packets, but it is possible
3094 * that some packets are in flight because some CPU runs receiver and
3095 * did hash table lookup before we unhashed socket. They will achieve
3096 * receive queue and will be purged by socket destructor.
3098 * Also we still have packets pending on receive queue and probably,
3099 * our own packets waiting in device queues. sock_destroy will drain
3100 * receive queue, but transmitted packets will delay socket destruction
3101 * until the last reference will be released.
3106 xfrm_sk_free_policy(sk);
3108 sk_refcnt_debug_release(sk);
3112 EXPORT_SYMBOL(sk_common_release);
3114 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3116 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3118 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3119 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3120 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3121 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3122 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3123 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3124 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3125 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3126 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3129 #ifdef CONFIG_PROC_FS
3130 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3132 int val[PROTO_INUSE_NR];
3135 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3137 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3139 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3141 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3143 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3145 int cpu, idx = prot->inuse_idx;
3148 for_each_possible_cpu(cpu)
3149 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3151 return res >= 0 ? res : 0;
3153 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3155 static void sock_inuse_add(struct net *net, int val)
3157 this_cpu_add(*net->core.sock_inuse, val);
3160 int sock_inuse_get(struct net *net)
3164 for_each_possible_cpu(cpu)
3165 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3170 EXPORT_SYMBOL_GPL(sock_inuse_get);
3172 static int __net_init sock_inuse_init_net(struct net *net)
3174 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3175 if (net->core.prot_inuse == NULL)
3178 net->core.sock_inuse = alloc_percpu(int);
3179 if (net->core.sock_inuse == NULL)
3185 free_percpu(net->core.prot_inuse);
3189 static void __net_exit sock_inuse_exit_net(struct net *net)
3191 free_percpu(net->core.prot_inuse);
3192 free_percpu(net->core.sock_inuse);
3195 static struct pernet_operations net_inuse_ops = {
3196 .init = sock_inuse_init_net,
3197 .exit = sock_inuse_exit_net,
3200 static __init int net_inuse_init(void)
3202 if (register_pernet_subsys(&net_inuse_ops))
3203 panic("Cannot initialize net inuse counters");
3208 core_initcall(net_inuse_init);
3210 static void assign_proto_idx(struct proto *prot)
3212 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3214 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3215 pr_err("PROTO_INUSE_NR exhausted\n");
3219 set_bit(prot->inuse_idx, proto_inuse_idx);
3222 static void release_proto_idx(struct proto *prot)
3224 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3225 clear_bit(prot->inuse_idx, proto_inuse_idx);
3228 static inline void assign_proto_idx(struct proto *prot)
3232 static inline void release_proto_idx(struct proto *prot)
3236 static void sock_inuse_add(struct net *net, int val)
3241 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3245 kfree(rsk_prot->slab_name);
3246 rsk_prot->slab_name = NULL;
3247 kmem_cache_destroy(rsk_prot->slab);
3248 rsk_prot->slab = NULL;
3251 static int req_prot_init(const struct proto *prot)
3253 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3258 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3260 if (!rsk_prot->slab_name)
3263 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3264 rsk_prot->obj_size, 0,
3265 prot->slab_flags, NULL);
3267 if (!rsk_prot->slab) {
3268 pr_crit("%s: Can't create request sock SLAB cache!\n",
3275 int proto_register(struct proto *prot, int alloc_slab)
3278 prot->slab = kmem_cache_create_usercopy(prot->name,
3280 SLAB_HWCACHE_ALIGN | prot->slab_flags,
3281 prot->useroffset, prot->usersize,
3284 if (prot->slab == NULL) {
3285 pr_crit("%s: Can't create sock SLAB cache!\n",
3290 if (req_prot_init(prot))
3291 goto out_free_request_sock_slab;
3293 if (prot->twsk_prot != NULL) {
3294 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3296 if (prot->twsk_prot->twsk_slab_name == NULL)
3297 goto out_free_request_sock_slab;
3299 prot->twsk_prot->twsk_slab =
3300 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3301 prot->twsk_prot->twsk_obj_size,
3305 if (prot->twsk_prot->twsk_slab == NULL)
3306 goto out_free_timewait_sock_slab_name;
3310 mutex_lock(&proto_list_mutex);
3311 list_add(&prot->node, &proto_list);
3312 assign_proto_idx(prot);
3313 mutex_unlock(&proto_list_mutex);
3316 out_free_timewait_sock_slab_name:
3317 kfree(prot->twsk_prot->twsk_slab_name);
3318 out_free_request_sock_slab:
3319 req_prot_cleanup(prot->rsk_prot);
3321 kmem_cache_destroy(prot->slab);
3326 EXPORT_SYMBOL(proto_register);
3328 void proto_unregister(struct proto *prot)
3330 mutex_lock(&proto_list_mutex);
3331 release_proto_idx(prot);
3332 list_del(&prot->node);
3333 mutex_unlock(&proto_list_mutex);
3335 kmem_cache_destroy(prot->slab);
3338 req_prot_cleanup(prot->rsk_prot);
3340 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3341 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3342 kfree(prot->twsk_prot->twsk_slab_name);
3343 prot->twsk_prot->twsk_slab = NULL;
3346 EXPORT_SYMBOL(proto_unregister);
3348 int sock_load_diag_module(int family, int protocol)
3351 if (!sock_is_registered(family))
3354 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3355 NETLINK_SOCK_DIAG, family);
3359 if (family == AF_INET &&
3360 !rcu_access_pointer(inet_protos[protocol]))
3364 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3365 NETLINK_SOCK_DIAG, family, protocol);
3367 EXPORT_SYMBOL(sock_load_diag_module);
3369 #ifdef CONFIG_PROC_FS
3370 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3371 __acquires(proto_list_mutex)
3373 mutex_lock(&proto_list_mutex);
3374 return seq_list_start_head(&proto_list, *pos);
3377 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3379 return seq_list_next(v, &proto_list, pos);
3382 static void proto_seq_stop(struct seq_file *seq, void *v)
3383 __releases(proto_list_mutex)
3385 mutex_unlock(&proto_list_mutex);
3388 static char proto_method_implemented(const void *method)
3390 return method == NULL ? 'n' : 'y';
3392 static long sock_prot_memory_allocated(struct proto *proto)
3394 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3397 static char *sock_prot_memory_pressure(struct proto *proto)
3399 return proto->memory_pressure != NULL ?
3400 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3403 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3406 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3407 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3410 sock_prot_inuse_get(seq_file_net(seq), proto),
3411 sock_prot_memory_allocated(proto),
3412 sock_prot_memory_pressure(proto),
3414 proto->slab == NULL ? "no" : "yes",
3415 module_name(proto->owner),
3416 proto_method_implemented(proto->close),
3417 proto_method_implemented(proto->connect),
3418 proto_method_implemented(proto->disconnect),
3419 proto_method_implemented(proto->accept),
3420 proto_method_implemented(proto->ioctl),
3421 proto_method_implemented(proto->init),
3422 proto_method_implemented(proto->destroy),
3423 proto_method_implemented(proto->shutdown),
3424 proto_method_implemented(proto->setsockopt),
3425 proto_method_implemented(proto->getsockopt),
3426 proto_method_implemented(proto->sendmsg),
3427 proto_method_implemented(proto->recvmsg),
3428 proto_method_implemented(proto->sendpage),
3429 proto_method_implemented(proto->bind),
3430 proto_method_implemented(proto->backlog_rcv),
3431 proto_method_implemented(proto->hash),
3432 proto_method_implemented(proto->unhash),
3433 proto_method_implemented(proto->get_port),
3434 proto_method_implemented(proto->enter_memory_pressure));
3437 static int proto_seq_show(struct seq_file *seq, void *v)
3439 if (v == &proto_list)
3440 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3449 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3451 proto_seq_printf(seq, list_entry(v, struct proto, node));
3455 static const struct seq_operations proto_seq_ops = {
3456 .start = proto_seq_start,
3457 .next = proto_seq_next,
3458 .stop = proto_seq_stop,
3459 .show = proto_seq_show,
3462 static int proto_seq_open(struct inode *inode, struct file *file)
3464 return seq_open_net(inode, file, &proto_seq_ops,
3465 sizeof(struct seq_net_private));
3468 static const struct file_operations proto_seq_fops = {
3469 .open = proto_seq_open,
3471 .llseek = seq_lseek,
3472 .release = seq_release_net,
3475 static __net_init int proto_init_net(struct net *net)
3477 if (!proc_create("protocols", 0444, net->proc_net, &proto_seq_fops))
3483 static __net_exit void proto_exit_net(struct net *net)
3485 remove_proc_entry("protocols", net->proc_net);
3489 static __net_initdata struct pernet_operations proto_net_ops = {
3490 .init = proto_init_net,
3491 .exit = proto_exit_net,
3494 static int __init proto_init(void)
3496 return register_pernet_subsys(&proto_net_ops);
3499 subsys_initcall(proto_init);
3501 #endif /* PROC_FS */
3503 #ifdef CONFIG_NET_RX_BUSY_POLL
3504 bool sk_busy_loop_end(void *p, unsigned long start_time)
3506 struct sock *sk = p;
3508 return !skb_queue_empty(&sk->sk_receive_queue) ||
3509 sk_busy_loop_timeout(sk, start_time);
3511 EXPORT_SYMBOL(sk_busy_loop_end);
3512 #endif /* CONFIG_NET_RX_BUSY_POLL */