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>
138 #include <trace/events/sock.h>
144 #include <net/busy_poll.h>
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
150 * sk_ns_capable - General socket capability test
151 * @sk: Socket to use a capability on or through
152 * @user_ns: The user namespace of the capability to use
153 * @cap: The capability to use
155 * Test to see if the opener of the socket had when the socket was
156 * created and the current process has the capability @cap in the user
157 * namespace @user_ns.
159 bool sk_ns_capable(const struct sock *sk,
160 struct user_namespace *user_ns, int cap)
162 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
163 ns_capable(user_ns, cap);
165 EXPORT_SYMBOL(sk_ns_capable);
168 * sk_capable - Socket global capability test
169 * @sk: Socket to use a capability on or through
170 * @cap: The global capability to use
172 * Test to see if the opener of the socket had when the socket was
173 * created and the current process has the capability @cap in all user
176 bool sk_capable(const struct sock *sk, int cap)
178 return sk_ns_capable(sk, &init_user_ns, cap);
180 EXPORT_SYMBOL(sk_capable);
183 * sk_net_capable - Network namespace socket capability test
184 * @sk: Socket to use a capability on or through
185 * @cap: The capability to use
187 * Test to see if the opener of the socket had when the socket was created
188 * and the current process has the capability @cap over the network namespace
189 * the socket is a member of.
191 bool sk_net_capable(const struct sock *sk, int cap)
193 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195 EXPORT_SYMBOL(sk_net_capable);
198 #ifdef CONFIG_MEMCG_KMEM
199 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
204 mutex_lock(&proto_list_mutex);
205 list_for_each_entry(proto, &proto_list, node) {
206 if (proto->init_cgroup) {
207 ret = proto->init_cgroup(memcg, ss);
213 mutex_unlock(&proto_list_mutex);
216 list_for_each_entry_continue_reverse(proto, &proto_list, node)
217 if (proto->destroy_cgroup)
218 proto->destroy_cgroup(memcg);
219 mutex_unlock(&proto_list_mutex);
223 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
227 mutex_lock(&proto_list_mutex);
228 list_for_each_entry_reverse(proto, &proto_list, node)
229 if (proto->destroy_cgroup)
230 proto->destroy_cgroup(memcg);
231 mutex_unlock(&proto_list_mutex);
236 * Each address family might have different locking rules, so we have
237 * one slock key per address family:
239 static struct lock_class_key af_family_keys[AF_MAX];
240 static struct lock_class_key af_family_slock_keys[AF_MAX];
242 #if defined(CONFIG_MEMCG_KMEM)
243 struct static_key memcg_socket_limit_enabled;
244 EXPORT_SYMBOL(memcg_socket_limit_enabled);
248 * Make lock validator output more readable. (we pre-construct these
249 * strings build-time, so that runtime initialization of socket
252 static const char *const af_family_key_strings[AF_MAX+1] = {
253 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
254 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
255 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
256 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
257 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
258 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
259 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
260 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
261 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
262 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
263 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
264 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
265 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
266 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
268 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
269 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
270 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
271 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
272 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
273 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
274 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
275 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
276 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
277 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
278 "slock-27" , "slock-28" , "slock-AF_CAN" ,
279 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
280 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
281 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
282 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
284 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
285 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
286 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
287 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
288 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
289 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
290 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
291 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
292 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
293 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
294 "clock-27" , "clock-28" , "clock-AF_CAN" ,
295 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
296 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
297 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
298 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
302 * sk_callback_lock locking rules are per-address-family,
303 * so split the lock classes by using a per-AF key:
305 static struct lock_class_key af_callback_keys[AF_MAX];
307 /* Take into consideration the size of the struct sk_buff overhead in the
308 * determination of these values, since that is non-constant across
309 * platforms. This makes socket queueing behavior and performance
310 * not depend upon such differences.
312 #define _SK_MEM_PACKETS 256
313 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
314 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
317 /* Run time adjustable parameters. */
318 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
319 EXPORT_SYMBOL(sysctl_wmem_max);
320 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
321 EXPORT_SYMBOL(sysctl_rmem_max);
322 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
323 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
325 /* Maximal space eaten by iovec or ancillary data plus some space */
326 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
327 EXPORT_SYMBOL(sysctl_optmem_max);
329 int sysctl_tstamp_allow_data __read_mostly = 1;
331 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
332 EXPORT_SYMBOL_GPL(memalloc_socks);
335 * sk_set_memalloc - sets %SOCK_MEMALLOC
336 * @sk: socket to set it on
338 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
339 * It's the responsibility of the admin to adjust min_free_kbytes
340 * to meet the requirements
342 void sk_set_memalloc(struct sock *sk)
344 sock_set_flag(sk, SOCK_MEMALLOC);
345 sk->sk_allocation |= __GFP_MEMALLOC;
346 static_key_slow_inc(&memalloc_socks);
348 EXPORT_SYMBOL_GPL(sk_set_memalloc);
350 void sk_clear_memalloc(struct sock *sk)
352 sock_reset_flag(sk, SOCK_MEMALLOC);
353 sk->sk_allocation &= ~__GFP_MEMALLOC;
354 static_key_slow_dec(&memalloc_socks);
357 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
358 * progress of swapping. SOCK_MEMALLOC may be cleared while
359 * it has rmem allocations due to the last swapfile being deactivated
360 * but there is a risk that the socket is unusable due to exceeding
361 * the rmem limits. Reclaim the reserves and obey rmem limits again.
365 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
367 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
370 unsigned long pflags = current->flags;
372 /* these should have been dropped before queueing */
373 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
375 current->flags |= PF_MEMALLOC;
376 ret = sk->sk_backlog_rcv(sk, skb);
377 tsk_restore_flags(current, pflags, PF_MEMALLOC);
381 EXPORT_SYMBOL(__sk_backlog_rcv);
383 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
387 if (optlen < sizeof(tv))
389 if (copy_from_user(&tv, optval, sizeof(tv)))
391 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
395 static int warned __read_mostly;
398 if (warned < 10 && net_ratelimit()) {
400 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
401 __func__, current->comm, task_pid_nr(current));
405 *timeo_p = MAX_SCHEDULE_TIMEOUT;
406 if (tv.tv_sec == 0 && tv.tv_usec == 0)
408 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
409 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
413 static void sock_warn_obsolete_bsdism(const char *name)
416 static char warncomm[TASK_COMM_LEN];
417 if (strcmp(warncomm, current->comm) && warned < 5) {
418 strcpy(warncomm, current->comm);
419 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
425 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
427 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
429 if (sk->sk_flags & flags) {
430 sk->sk_flags &= ~flags;
431 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
432 net_disable_timestamp();
437 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
441 struct sk_buff_head *list = &sk->sk_receive_queue;
443 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
444 atomic_inc(&sk->sk_drops);
445 trace_sock_rcvqueue_full(sk, skb);
449 err = sk_filter(sk, skb);
453 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
454 atomic_inc(&sk->sk_drops);
459 skb_set_owner_r(skb, sk);
461 /* we escape from rcu protected region, make sure we dont leak
466 spin_lock_irqsave(&list->lock, flags);
467 sock_skb_set_dropcount(sk, skb);
468 __skb_queue_tail(list, skb);
469 spin_unlock_irqrestore(&list->lock, flags);
471 if (!sock_flag(sk, SOCK_DEAD))
472 sk->sk_data_ready(sk);
475 EXPORT_SYMBOL(sock_queue_rcv_skb);
477 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
479 int rc = NET_RX_SUCCESS;
481 if (sk_filter(sk, skb))
482 goto discard_and_relse;
486 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
487 atomic_inc(&sk->sk_drops);
488 goto discard_and_relse;
491 bh_lock_sock_nested(sk);
494 if (!sock_owned_by_user(sk)) {
496 * trylock + unlock semantics:
498 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
500 rc = sk_backlog_rcv(sk, skb);
502 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
503 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
505 atomic_inc(&sk->sk_drops);
506 goto discard_and_relse;
517 EXPORT_SYMBOL(sk_receive_skb);
519 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
521 struct dst_entry *dst = __sk_dst_get(sk);
523 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
524 sk_tx_queue_clear(sk);
525 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
532 EXPORT_SYMBOL(__sk_dst_check);
534 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
536 struct dst_entry *dst = sk_dst_get(sk);
538 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
546 EXPORT_SYMBOL(sk_dst_check);
548 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
551 int ret = -ENOPROTOOPT;
552 #ifdef CONFIG_NETDEVICES
553 struct net *net = sock_net(sk);
554 char devname[IFNAMSIZ];
559 if (!ns_capable(net->user_ns, CAP_NET_RAW))
566 /* Bind this socket to a particular device like "eth0",
567 * as specified in the passed interface name. If the
568 * name is "" or the option length is zero the socket
571 if (optlen > IFNAMSIZ - 1)
572 optlen = IFNAMSIZ - 1;
573 memset(devname, 0, sizeof(devname));
576 if (copy_from_user(devname, optval, optlen))
580 if (devname[0] != '\0') {
581 struct net_device *dev;
584 dev = dev_get_by_name_rcu(net, devname);
586 index = dev->ifindex;
594 sk->sk_bound_dev_if = index;
606 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
607 int __user *optlen, int len)
609 int ret = -ENOPROTOOPT;
610 #ifdef CONFIG_NETDEVICES
611 struct net *net = sock_net(sk);
612 char devname[IFNAMSIZ];
614 if (sk->sk_bound_dev_if == 0) {
623 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
627 len = strlen(devname) + 1;
630 if (copy_to_user(optval, devname, len))
635 if (put_user(len, optlen))
646 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
649 sock_set_flag(sk, bit);
651 sock_reset_flag(sk, bit);
654 bool sk_mc_loop(struct sock *sk)
656 if (dev_recursion_level())
660 switch (sk->sk_family) {
662 return inet_sk(sk)->mc_loop;
663 #if IS_ENABLED(CONFIG_IPV6)
665 return inet6_sk(sk)->mc_loop;
671 EXPORT_SYMBOL(sk_mc_loop);
674 * This is meant for all protocols to use and covers goings on
675 * at the socket level. Everything here is generic.
678 int sock_setsockopt(struct socket *sock, int level, int optname,
679 char __user *optval, unsigned int optlen)
681 struct sock *sk = sock->sk;
688 * Options without arguments
691 if (optname == SO_BINDTODEVICE)
692 return sock_setbindtodevice(sk, optval, optlen);
694 if (optlen < sizeof(int))
697 if (get_user(val, (int __user *)optval))
700 valbool = val ? 1 : 0;
706 if (val && !capable(CAP_NET_ADMIN))
709 sock_valbool_flag(sk, SOCK_DBG, valbool);
712 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
715 sk->sk_reuseport = valbool;
724 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
727 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
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_wmem_max);
737 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
738 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
739 /* Wake up sending tasks if we upped the value. */
740 sk->sk_write_space(sk);
744 if (!capable(CAP_NET_ADMIN)) {
751 /* Don't error on this BSD doesn't and if you think
752 * about it this is right. Otherwise apps have to
753 * play 'guess the biggest size' games. RCVBUF/SNDBUF
754 * are treated in BSD as hints
756 val = min_t(u32, val, sysctl_rmem_max);
758 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
760 * We double it on the way in to account for
761 * "struct sk_buff" etc. overhead. Applications
762 * assume that the SO_RCVBUF setting they make will
763 * allow that much actual data to be received on that
766 * Applications are unaware that "struct sk_buff" and
767 * other overheads allocate from the receive buffer
768 * during socket buffer allocation.
770 * And after considering the possible alternatives,
771 * returning the value we actually used in getsockopt
772 * is the most desirable behavior.
774 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
778 if (!capable(CAP_NET_ADMIN)) {
786 if (sk->sk_protocol == IPPROTO_TCP &&
787 sk->sk_type == SOCK_STREAM)
788 tcp_set_keepalive(sk, valbool);
790 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
794 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
798 sk->sk_no_check_tx = valbool;
802 if ((val >= 0 && val <= 6) ||
803 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
804 sk->sk_priority = val;
810 if (optlen < sizeof(ling)) {
811 ret = -EINVAL; /* 1003.1g */
814 if (copy_from_user(&ling, optval, sizeof(ling))) {
819 sock_reset_flag(sk, SOCK_LINGER);
821 #if (BITS_PER_LONG == 32)
822 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
823 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
826 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
827 sock_set_flag(sk, SOCK_LINGER);
832 sock_warn_obsolete_bsdism("setsockopt");
837 set_bit(SOCK_PASSCRED, &sock->flags);
839 clear_bit(SOCK_PASSCRED, &sock->flags);
845 if (optname == SO_TIMESTAMP)
846 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
848 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
849 sock_set_flag(sk, SOCK_RCVTSTAMP);
850 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
852 sock_reset_flag(sk, SOCK_RCVTSTAMP);
853 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
857 case SO_TIMESTAMPING:
858 if (val & ~SOF_TIMESTAMPING_MASK) {
863 if (val & SOF_TIMESTAMPING_OPT_ID &&
864 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
865 if (sk->sk_protocol == IPPROTO_TCP) {
866 if (sk->sk_state != TCP_ESTABLISHED) {
870 sk->sk_tskey = tcp_sk(sk)->snd_una;
875 sk->sk_tsflags = val;
876 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
877 sock_enable_timestamp(sk,
878 SOCK_TIMESTAMPING_RX_SOFTWARE);
880 sock_disable_timestamp(sk,
881 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
887 sk->sk_rcvlowat = val ? : 1;
891 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
895 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
898 case SO_ATTACH_FILTER:
900 if (optlen == sizeof(struct sock_fprog)) {
901 struct sock_fprog fprog;
904 if (copy_from_user(&fprog, optval, sizeof(fprog)))
907 ret = sk_attach_filter(&fprog, sk);
913 if (optlen == sizeof(u32)) {
917 if (copy_from_user(&ufd, optval, sizeof(ufd)))
920 ret = sk_attach_bpf(ufd, sk);
924 case SO_DETACH_FILTER:
925 ret = sk_detach_filter(sk);
929 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
932 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
937 set_bit(SOCK_PASSSEC, &sock->flags);
939 clear_bit(SOCK_PASSSEC, &sock->flags);
942 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
949 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
953 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
957 if (sock->ops->set_peek_off)
958 ret = sock->ops->set_peek_off(sk, val);
964 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
967 case SO_SELECT_ERR_QUEUE:
968 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
971 #ifdef CONFIG_NET_RX_BUSY_POLL
973 /* allow unprivileged users to decrease the value */
974 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
980 sk->sk_ll_usec = val;
985 case SO_MAX_PACING_RATE:
986 sk->sk_max_pacing_rate = val;
987 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
988 sk->sk_max_pacing_rate);
991 case SO_INCOMING_CPU:
992 sk->sk_incoming_cpu = val;
1002 EXPORT_SYMBOL(sock_setsockopt);
1005 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1006 struct ucred *ucred)
1008 ucred->pid = pid_vnr(pid);
1009 ucred->uid = ucred->gid = -1;
1011 struct user_namespace *current_ns = current_user_ns();
1013 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1014 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1018 int sock_getsockopt(struct socket *sock, int level, int optname,
1019 char __user *optval, int __user *optlen)
1021 struct sock *sk = sock->sk;
1029 int lv = sizeof(int);
1032 if (get_user(len, optlen))
1037 memset(&v, 0, sizeof(v));
1041 v.val = sock_flag(sk, SOCK_DBG);
1045 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1049 v.val = sock_flag(sk, SOCK_BROADCAST);
1053 v.val = sk->sk_sndbuf;
1057 v.val = sk->sk_rcvbuf;
1061 v.val = sk->sk_reuse;
1065 v.val = sk->sk_reuseport;
1069 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1073 v.val = sk->sk_type;
1077 v.val = sk->sk_protocol;
1081 v.val = sk->sk_family;
1085 v.val = -sock_error(sk);
1087 v.val = xchg(&sk->sk_err_soft, 0);
1091 v.val = sock_flag(sk, SOCK_URGINLINE);
1095 v.val = sk->sk_no_check_tx;
1099 v.val = sk->sk_priority;
1103 lv = sizeof(v.ling);
1104 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1105 v.ling.l_linger = sk->sk_lingertime / HZ;
1109 sock_warn_obsolete_bsdism("getsockopt");
1113 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1114 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1117 case SO_TIMESTAMPNS:
1118 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1121 case SO_TIMESTAMPING:
1122 v.val = sk->sk_tsflags;
1126 lv = sizeof(struct timeval);
1127 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1131 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1132 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1137 lv = sizeof(struct timeval);
1138 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1142 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1143 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1148 v.val = sk->sk_rcvlowat;
1156 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1161 struct ucred peercred;
1162 if (len > sizeof(peercred))
1163 len = sizeof(peercred);
1164 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1165 if (copy_to_user(optval, &peercred, len))
1174 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1178 if (copy_to_user(optval, address, len))
1183 /* Dubious BSD thing... Probably nobody even uses it, but
1184 * the UNIX standard wants it for whatever reason... -DaveM
1187 v.val = sk->sk_state == TCP_LISTEN;
1191 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1195 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1198 v.val = sk->sk_mark;
1202 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1205 case SO_WIFI_STATUS:
1206 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1210 if (!sock->ops->set_peek_off)
1213 v.val = sk->sk_peek_off;
1216 v.val = sock_flag(sk, SOCK_NOFCS);
1219 case SO_BINDTODEVICE:
1220 return sock_getbindtodevice(sk, optval, optlen, len);
1223 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1229 case SO_LOCK_FILTER:
1230 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1233 case SO_BPF_EXTENSIONS:
1234 v.val = bpf_tell_extensions();
1237 case SO_SELECT_ERR_QUEUE:
1238 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1241 #ifdef CONFIG_NET_RX_BUSY_POLL
1243 v.val = sk->sk_ll_usec;
1247 case SO_MAX_PACING_RATE:
1248 v.val = sk->sk_max_pacing_rate;
1251 case SO_INCOMING_CPU:
1252 v.val = sk->sk_incoming_cpu;
1256 /* We implement the SO_SNDLOWAT etc to not be settable
1259 return -ENOPROTOOPT;
1264 if (copy_to_user(optval, &v, len))
1267 if (put_user(len, optlen))
1273 * Initialize an sk_lock.
1275 * (We also register the sk_lock with the lock validator.)
1277 static inline void sock_lock_init(struct sock *sk)
1279 sock_lock_init_class_and_name(sk,
1280 af_family_slock_key_strings[sk->sk_family],
1281 af_family_slock_keys + sk->sk_family,
1282 af_family_key_strings[sk->sk_family],
1283 af_family_keys + sk->sk_family);
1287 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1288 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1289 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1291 static void sock_copy(struct sock *nsk, const struct sock *osk)
1293 #ifdef CONFIG_SECURITY_NETWORK
1294 void *sptr = nsk->sk_security;
1296 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1298 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1299 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1301 #ifdef CONFIG_SECURITY_NETWORK
1302 nsk->sk_security = sptr;
1303 security_sk_clone(osk, nsk);
1307 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1309 unsigned long nulls1, nulls2;
1311 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1312 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1313 if (nulls1 > nulls2)
1314 swap(nulls1, nulls2);
1317 memset((char *)sk, 0, nulls1);
1318 memset((char *)sk + nulls1 + sizeof(void *), 0,
1319 nulls2 - nulls1 - sizeof(void *));
1320 memset((char *)sk + nulls2 + sizeof(void *), 0,
1321 size - nulls2 - sizeof(void *));
1323 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1325 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1329 struct kmem_cache *slab;
1333 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1336 if (priority & __GFP_ZERO) {
1338 prot->clear_sk(sk, prot->obj_size);
1340 sk_prot_clear_nulls(sk, prot->obj_size);
1343 sk = kmalloc(prot->obj_size, priority);
1346 kmemcheck_annotate_bitfield(sk, flags);
1348 if (security_sk_alloc(sk, family, priority))
1351 if (!try_module_get(prot->owner))
1353 sk_tx_queue_clear(sk);
1359 security_sk_free(sk);
1362 kmem_cache_free(slab, sk);
1368 static void sk_prot_free(struct proto *prot, struct sock *sk)
1370 struct kmem_cache *slab;
1371 struct module *owner;
1373 owner = prot->owner;
1376 security_sk_free(sk);
1378 kmem_cache_free(slab, sk);
1384 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1385 void sock_update_netprioidx(struct sock *sk)
1390 sk->sk_cgrp_prioidx = task_netprioidx(current);
1392 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1396 * sk_alloc - All socket objects are allocated here
1397 * @net: the applicable net namespace
1398 * @family: protocol family
1399 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1400 * @prot: struct proto associated with this new sock instance
1401 * @kern: is this to be a kernel socket?
1403 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1404 struct proto *prot, int kern)
1408 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1410 sk->sk_family = family;
1412 * See comment in struct sock definition to understand
1413 * why we need sk_prot_creator -acme
1415 sk->sk_prot = sk->sk_prot_creator = prot;
1417 sk->sk_net_refcnt = kern ? 0 : 1;
1418 if (likely(sk->sk_net_refcnt))
1420 sock_net_set(sk, net);
1421 atomic_set(&sk->sk_wmem_alloc, 1);
1423 sock_update_classid(sk);
1424 sock_update_netprioidx(sk);
1429 EXPORT_SYMBOL(sk_alloc);
1431 void sk_destruct(struct sock *sk)
1433 struct sk_filter *filter;
1435 if (sk->sk_destruct)
1436 sk->sk_destruct(sk);
1438 filter = rcu_dereference_check(sk->sk_filter,
1439 atomic_read(&sk->sk_wmem_alloc) == 0);
1441 sk_filter_uncharge(sk, filter);
1442 RCU_INIT_POINTER(sk->sk_filter, NULL);
1445 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1447 if (atomic_read(&sk->sk_omem_alloc))
1448 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1449 __func__, atomic_read(&sk->sk_omem_alloc));
1451 if (sk->sk_peer_cred)
1452 put_cred(sk->sk_peer_cred);
1453 put_pid(sk->sk_peer_pid);
1454 if (likely(sk->sk_net_refcnt))
1455 put_net(sock_net(sk));
1456 sk_prot_free(sk->sk_prot_creator, sk);
1459 static void __sk_free(struct sock *sk)
1461 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1462 sock_diag_broadcast_destroy(sk);
1467 void sk_free(struct sock *sk)
1470 * We subtract one from sk_wmem_alloc and can know if
1471 * some packets are still in some tx queue.
1472 * If not null, sock_wfree() will call __sk_free(sk) later
1474 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1477 EXPORT_SYMBOL(sk_free);
1479 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1481 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1482 sock_update_memcg(newsk);
1486 * sk_clone_lock - clone a socket, and lock its clone
1487 * @sk: the socket to clone
1488 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1490 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1492 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1495 bool is_charged = true;
1497 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1498 if (newsk != NULL) {
1499 struct sk_filter *filter;
1501 sock_copy(newsk, sk);
1504 if (likely(newsk->sk_net_refcnt))
1505 get_net(sock_net(newsk));
1506 sk_node_init(&newsk->sk_node);
1507 sock_lock_init(newsk);
1508 bh_lock_sock(newsk);
1509 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1510 newsk->sk_backlog.len = 0;
1512 atomic_set(&newsk->sk_rmem_alloc, 0);
1514 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1516 atomic_set(&newsk->sk_wmem_alloc, 1);
1517 atomic_set(&newsk->sk_omem_alloc, 0);
1518 skb_queue_head_init(&newsk->sk_receive_queue);
1519 skb_queue_head_init(&newsk->sk_write_queue);
1521 spin_lock_init(&newsk->sk_dst_lock);
1522 rwlock_init(&newsk->sk_callback_lock);
1523 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1524 af_callback_keys + newsk->sk_family,
1525 af_family_clock_key_strings[newsk->sk_family]);
1527 newsk->sk_dst_cache = NULL;
1528 newsk->sk_wmem_queued = 0;
1529 newsk->sk_forward_alloc = 0;
1530 newsk->sk_send_head = NULL;
1531 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1533 sock_reset_flag(newsk, SOCK_DONE);
1534 skb_queue_head_init(&newsk->sk_error_queue);
1536 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1538 /* though it's an empty new sock, the charging may fail
1539 * if sysctl_optmem_max was changed between creation of
1540 * original socket and cloning
1542 is_charged = sk_filter_charge(newsk, filter);
1544 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1545 /* It is still raw copy of parent, so invalidate
1546 * destructor and make plain sk_free() */
1547 newsk->sk_destruct = NULL;
1548 bh_unlock_sock(newsk);
1555 newsk->sk_priority = 0;
1556 newsk->sk_incoming_cpu = raw_smp_processor_id();
1557 atomic64_set(&newsk->sk_cookie, 0);
1559 * Before updating sk_refcnt, we must commit prior changes to memory
1560 * (Documentation/RCU/rculist_nulls.txt for details)
1563 atomic_set(&newsk->sk_refcnt, 2);
1566 * Increment the counter in the same struct proto as the master
1567 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1568 * is the same as sk->sk_prot->socks, as this field was copied
1571 * This _changes_ the previous behaviour, where
1572 * tcp_create_openreq_child always was incrementing the
1573 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1574 * to be taken into account in all callers. -acme
1576 sk_refcnt_debug_inc(newsk);
1577 sk_set_socket(newsk, NULL);
1578 newsk->sk_wq = NULL;
1580 sk_update_clone(sk, newsk);
1582 if (newsk->sk_prot->sockets_allocated)
1583 sk_sockets_allocated_inc(newsk);
1585 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1586 net_enable_timestamp();
1591 EXPORT_SYMBOL_GPL(sk_clone_lock);
1593 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1597 __sk_dst_set(sk, dst);
1598 sk->sk_route_caps = dst->dev->features;
1599 if (sk->sk_route_caps & NETIF_F_GSO)
1600 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1601 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1602 if (sk_can_gso(sk)) {
1603 if (dst->header_len) {
1604 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1606 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1607 sk->sk_gso_max_size = dst->dev->gso_max_size;
1608 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1611 sk->sk_gso_max_segs = max_segs;
1613 EXPORT_SYMBOL_GPL(sk_setup_caps);
1616 * Simple resource managers for sockets.
1621 * Write buffer destructor automatically called from kfree_skb.
1623 void sock_wfree(struct sk_buff *skb)
1625 struct sock *sk = skb->sk;
1626 unsigned int len = skb->truesize;
1628 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1630 * Keep a reference on sk_wmem_alloc, this will be released
1631 * after sk_write_space() call
1633 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1634 sk->sk_write_space(sk);
1638 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1639 * could not do because of in-flight packets
1641 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1644 EXPORT_SYMBOL(sock_wfree);
1646 void skb_orphan_partial(struct sk_buff *skb)
1648 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1649 * so we do not completely orphan skb, but transfert all
1650 * accounted bytes but one, to avoid unexpected reorders.
1652 if (skb->destructor == sock_wfree
1654 || skb->destructor == tcp_wfree
1657 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1663 EXPORT_SYMBOL(skb_orphan_partial);
1666 * Read buffer destructor automatically called from kfree_skb.
1668 void sock_rfree(struct sk_buff *skb)
1670 struct sock *sk = skb->sk;
1671 unsigned int len = skb->truesize;
1673 atomic_sub(len, &sk->sk_rmem_alloc);
1674 sk_mem_uncharge(sk, len);
1676 EXPORT_SYMBOL(sock_rfree);
1679 * Buffer destructor for skbs that are not used directly in read or write
1680 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1682 void sock_efree(struct sk_buff *skb)
1686 EXPORT_SYMBOL(sock_efree);
1688 kuid_t sock_i_uid(struct sock *sk)
1692 read_lock_bh(&sk->sk_callback_lock);
1693 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1694 read_unlock_bh(&sk->sk_callback_lock);
1697 EXPORT_SYMBOL(sock_i_uid);
1699 unsigned long sock_i_ino(struct sock *sk)
1703 read_lock_bh(&sk->sk_callback_lock);
1704 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1705 read_unlock_bh(&sk->sk_callback_lock);
1708 EXPORT_SYMBOL(sock_i_ino);
1711 * Allocate a skb from the socket's send buffer.
1713 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1716 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1717 struct sk_buff *skb = alloc_skb(size, priority);
1719 skb_set_owner_w(skb, sk);
1725 EXPORT_SYMBOL(sock_wmalloc);
1728 * Allocate a memory block from the socket's option memory buffer.
1730 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1732 if ((unsigned int)size <= sysctl_optmem_max &&
1733 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1735 /* First do the add, to avoid the race if kmalloc
1738 atomic_add(size, &sk->sk_omem_alloc);
1739 mem = kmalloc(size, priority);
1742 atomic_sub(size, &sk->sk_omem_alloc);
1746 EXPORT_SYMBOL(sock_kmalloc);
1748 /* Free an option memory block. Note, we actually want the inline
1749 * here as this allows gcc to detect the nullify and fold away the
1750 * condition entirely.
1752 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1755 if (WARN_ON_ONCE(!mem))
1761 atomic_sub(size, &sk->sk_omem_alloc);
1764 void sock_kfree_s(struct sock *sk, void *mem, int size)
1766 __sock_kfree_s(sk, mem, size, false);
1768 EXPORT_SYMBOL(sock_kfree_s);
1770 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1772 __sock_kfree_s(sk, mem, size, true);
1774 EXPORT_SYMBOL(sock_kzfree_s);
1776 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1777 I think, these locks should be removed for datagram sockets.
1779 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1783 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1787 if (signal_pending(current))
1789 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1790 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1791 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1793 if (sk->sk_shutdown & SEND_SHUTDOWN)
1797 timeo = schedule_timeout(timeo);
1799 finish_wait(sk_sleep(sk), &wait);
1805 * Generic send/receive buffer handlers
1808 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1809 unsigned long data_len, int noblock,
1810 int *errcode, int max_page_order)
1812 struct sk_buff *skb;
1816 timeo = sock_sndtimeo(sk, noblock);
1818 err = sock_error(sk);
1823 if (sk->sk_shutdown & SEND_SHUTDOWN)
1826 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1829 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1830 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1834 if (signal_pending(current))
1836 timeo = sock_wait_for_wmem(sk, timeo);
1838 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1839 errcode, sk->sk_allocation);
1841 skb_set_owner_w(skb, sk);
1845 err = sock_intr_errno(timeo);
1850 EXPORT_SYMBOL(sock_alloc_send_pskb);
1852 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1853 int noblock, int *errcode)
1855 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1857 EXPORT_SYMBOL(sock_alloc_send_skb);
1859 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1860 struct sockcm_cookie *sockc)
1862 struct cmsghdr *cmsg;
1864 for_each_cmsghdr(cmsg, msg) {
1865 if (!CMSG_OK(msg, cmsg))
1867 if (cmsg->cmsg_level != SOL_SOCKET)
1869 switch (cmsg->cmsg_type) {
1871 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1873 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1875 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1883 EXPORT_SYMBOL(sock_cmsg_send);
1885 /* On 32bit arches, an skb frag is limited to 2^15 */
1886 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1889 * skb_page_frag_refill - check that a page_frag contains enough room
1890 * @sz: minimum size of the fragment we want to get
1891 * @pfrag: pointer to page_frag
1892 * @gfp: priority for memory allocation
1894 * Note: While this allocator tries to use high order pages, there is
1895 * no guarantee that allocations succeed. Therefore, @sz MUST be
1896 * less or equal than PAGE_SIZE.
1898 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1901 if (atomic_read(&pfrag->page->_count) == 1) {
1905 if (pfrag->offset + sz <= pfrag->size)
1907 put_page(pfrag->page);
1911 if (SKB_FRAG_PAGE_ORDER) {
1912 pfrag->page = alloc_pages((gfp & ~__GFP_WAIT) | __GFP_COMP |
1913 __GFP_NOWARN | __GFP_NORETRY,
1914 SKB_FRAG_PAGE_ORDER);
1915 if (likely(pfrag->page)) {
1916 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1920 pfrag->page = alloc_page(gfp);
1921 if (likely(pfrag->page)) {
1922 pfrag->size = PAGE_SIZE;
1927 EXPORT_SYMBOL(skb_page_frag_refill);
1929 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1931 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1934 sk_enter_memory_pressure(sk);
1935 sk_stream_moderate_sndbuf(sk);
1938 EXPORT_SYMBOL(sk_page_frag_refill);
1940 static void __lock_sock(struct sock *sk)
1941 __releases(&sk->sk_lock.slock)
1942 __acquires(&sk->sk_lock.slock)
1947 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1948 TASK_UNINTERRUPTIBLE);
1949 spin_unlock_bh(&sk->sk_lock.slock);
1951 spin_lock_bh(&sk->sk_lock.slock);
1952 if (!sock_owned_by_user(sk))
1955 finish_wait(&sk->sk_lock.wq, &wait);
1958 static void __release_sock(struct sock *sk)
1959 __releases(&sk->sk_lock.slock)
1960 __acquires(&sk->sk_lock.slock)
1962 struct sk_buff *skb = sk->sk_backlog.head;
1965 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1969 struct sk_buff *next = skb->next;
1972 WARN_ON_ONCE(skb_dst_is_noref(skb));
1974 sk_backlog_rcv(sk, skb);
1977 * We are in process context here with softirqs
1978 * disabled, use cond_resched_softirq() to preempt.
1979 * This is safe to do because we've taken the backlog
1982 cond_resched_softirq();
1985 } while (skb != NULL);
1988 } while ((skb = sk->sk_backlog.head) != NULL);
1991 * Doing the zeroing here guarantee we can not loop forever
1992 * while a wild producer attempts to flood us.
1994 sk->sk_backlog.len = 0;
1998 * sk_wait_data - wait for data to arrive at sk_receive_queue
1999 * @sk: sock to wait on
2000 * @timeo: for how long
2001 * @skb: last skb seen on sk_receive_queue
2003 * Now socket state including sk->sk_err is changed only under lock,
2004 * hence we may omit checks after joining wait queue.
2005 * We check receive queue before schedule() only as optimization;
2006 * it is very likely that release_sock() added new data.
2008 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2013 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2014 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2015 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2016 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2017 finish_wait(sk_sleep(sk), &wait);
2020 EXPORT_SYMBOL(sk_wait_data);
2023 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2025 * @size: memory size to allocate
2026 * @kind: allocation type
2028 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2029 * rmem allocation. This function assumes that protocols which have
2030 * memory_pressure use sk_wmem_queued as write buffer accounting.
2032 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2034 struct proto *prot = sk->sk_prot;
2035 int amt = sk_mem_pages(size);
2037 int parent_status = UNDER_LIMIT;
2039 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2041 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2044 if (parent_status == UNDER_LIMIT &&
2045 allocated <= sk_prot_mem_limits(sk, 0)) {
2046 sk_leave_memory_pressure(sk);
2050 /* Under pressure. (we or our parents) */
2051 if ((parent_status > SOFT_LIMIT) ||
2052 allocated > sk_prot_mem_limits(sk, 1))
2053 sk_enter_memory_pressure(sk);
2055 /* Over hard limit (we or our parents) */
2056 if ((parent_status == OVER_LIMIT) ||
2057 (allocated > sk_prot_mem_limits(sk, 2)))
2058 goto suppress_allocation;
2060 /* guarantee minimum buffer size under pressure */
2061 if (kind == SK_MEM_RECV) {
2062 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2065 } else { /* SK_MEM_SEND */
2066 if (sk->sk_type == SOCK_STREAM) {
2067 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2069 } else if (atomic_read(&sk->sk_wmem_alloc) <
2070 prot->sysctl_wmem[0])
2074 if (sk_has_memory_pressure(sk)) {
2077 if (!sk_under_memory_pressure(sk))
2079 alloc = sk_sockets_allocated_read_positive(sk);
2080 if (sk_prot_mem_limits(sk, 2) > alloc *
2081 sk_mem_pages(sk->sk_wmem_queued +
2082 atomic_read(&sk->sk_rmem_alloc) +
2083 sk->sk_forward_alloc))
2087 suppress_allocation:
2089 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2090 sk_stream_moderate_sndbuf(sk);
2092 /* Fail only if socket is _under_ its sndbuf.
2093 * In this case we cannot block, so that we have to fail.
2095 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2099 trace_sock_exceed_buf_limit(sk, prot, allocated);
2101 /* Alas. Undo changes. */
2102 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2104 sk_memory_allocated_sub(sk, amt);
2108 EXPORT_SYMBOL(__sk_mem_schedule);
2111 * __sk_mem_reclaim - reclaim memory_allocated
2113 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2115 void __sk_mem_reclaim(struct sock *sk, int amount)
2117 amount >>= SK_MEM_QUANTUM_SHIFT;
2118 sk_memory_allocated_sub(sk, amount);
2119 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2121 if (sk_under_memory_pressure(sk) &&
2122 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2123 sk_leave_memory_pressure(sk);
2125 EXPORT_SYMBOL(__sk_mem_reclaim);
2129 * Set of default routines for initialising struct proto_ops when
2130 * the protocol does not support a particular function. In certain
2131 * cases where it makes no sense for a protocol to have a "do nothing"
2132 * function, some default processing is provided.
2135 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2139 EXPORT_SYMBOL(sock_no_bind);
2141 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2146 EXPORT_SYMBOL(sock_no_connect);
2148 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2152 EXPORT_SYMBOL(sock_no_socketpair);
2154 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2158 EXPORT_SYMBOL(sock_no_accept);
2160 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2165 EXPORT_SYMBOL(sock_no_getname);
2167 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2171 EXPORT_SYMBOL(sock_no_poll);
2173 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2177 EXPORT_SYMBOL(sock_no_ioctl);
2179 int sock_no_listen(struct socket *sock, int backlog)
2183 EXPORT_SYMBOL(sock_no_listen);
2185 int sock_no_shutdown(struct socket *sock, int how)
2189 EXPORT_SYMBOL(sock_no_shutdown);
2191 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2192 char __user *optval, unsigned int optlen)
2196 EXPORT_SYMBOL(sock_no_setsockopt);
2198 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2199 char __user *optval, int __user *optlen)
2203 EXPORT_SYMBOL(sock_no_getsockopt);
2205 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2209 EXPORT_SYMBOL(sock_no_sendmsg);
2211 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2216 EXPORT_SYMBOL(sock_no_recvmsg);
2218 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2220 /* Mirror missing mmap method error code */
2223 EXPORT_SYMBOL(sock_no_mmap);
2225 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2228 struct msghdr msg = {.msg_flags = flags};
2230 char *kaddr = kmap(page);
2231 iov.iov_base = kaddr + offset;
2233 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2237 EXPORT_SYMBOL(sock_no_sendpage);
2240 * Default Socket Callbacks
2243 static void sock_def_wakeup(struct sock *sk)
2245 struct socket_wq *wq;
2248 wq = rcu_dereference(sk->sk_wq);
2249 if (wq_has_sleeper(wq))
2250 wake_up_interruptible_all(&wq->wait);
2254 static void sock_def_error_report(struct sock *sk)
2256 struct socket_wq *wq;
2259 wq = rcu_dereference(sk->sk_wq);
2260 if (wq_has_sleeper(wq))
2261 wake_up_interruptible_poll(&wq->wait, POLLERR);
2262 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2266 static void sock_def_readable(struct sock *sk)
2268 struct socket_wq *wq;
2271 wq = rcu_dereference(sk->sk_wq);
2272 if (wq_has_sleeper(wq))
2273 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2274 POLLRDNORM | POLLRDBAND);
2275 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2279 static void sock_def_write_space(struct sock *sk)
2281 struct socket_wq *wq;
2285 /* Do not wake up a writer until he can make "significant"
2288 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2289 wq = rcu_dereference(sk->sk_wq);
2290 if (wq_has_sleeper(wq))
2291 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2292 POLLWRNORM | POLLWRBAND);
2294 /* Should agree with poll, otherwise some programs break */
2295 if (sock_writeable(sk))
2296 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2302 static void sock_def_destruct(struct sock *sk)
2306 void sk_send_sigurg(struct sock *sk)
2308 if (sk->sk_socket && sk->sk_socket->file)
2309 if (send_sigurg(&sk->sk_socket->file->f_owner))
2310 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2312 EXPORT_SYMBOL(sk_send_sigurg);
2314 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2315 unsigned long expires)
2317 if (!mod_timer(timer, expires))
2320 EXPORT_SYMBOL(sk_reset_timer);
2322 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2324 if (del_timer(timer))
2327 EXPORT_SYMBOL(sk_stop_timer);
2329 void sock_init_data(struct socket *sock, struct sock *sk)
2331 skb_queue_head_init(&sk->sk_receive_queue);
2332 skb_queue_head_init(&sk->sk_write_queue);
2333 skb_queue_head_init(&sk->sk_error_queue);
2335 sk->sk_send_head = NULL;
2337 init_timer(&sk->sk_timer);
2339 sk->sk_allocation = GFP_KERNEL;
2340 sk->sk_rcvbuf = sysctl_rmem_default;
2341 sk->sk_sndbuf = sysctl_wmem_default;
2342 sk->sk_state = TCP_CLOSE;
2343 sk_set_socket(sk, sock);
2345 sock_set_flag(sk, SOCK_ZAPPED);
2348 sk->sk_type = sock->type;
2349 sk->sk_wq = sock->wq;
2354 spin_lock_init(&sk->sk_dst_lock);
2355 rwlock_init(&sk->sk_callback_lock);
2356 lockdep_set_class_and_name(&sk->sk_callback_lock,
2357 af_callback_keys + sk->sk_family,
2358 af_family_clock_key_strings[sk->sk_family]);
2360 sk->sk_state_change = sock_def_wakeup;
2361 sk->sk_data_ready = sock_def_readable;
2362 sk->sk_write_space = sock_def_write_space;
2363 sk->sk_error_report = sock_def_error_report;
2364 sk->sk_destruct = sock_def_destruct;
2366 sk->sk_frag.page = NULL;
2367 sk->sk_frag.offset = 0;
2368 sk->sk_peek_off = -1;
2370 sk->sk_peer_pid = NULL;
2371 sk->sk_peer_cred = NULL;
2372 sk->sk_write_pending = 0;
2373 sk->sk_rcvlowat = 1;
2374 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2375 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2377 sk->sk_stamp = ktime_set(-1L, 0);
2379 #ifdef CONFIG_NET_RX_BUSY_POLL
2381 sk->sk_ll_usec = sysctl_net_busy_read;
2384 sk->sk_max_pacing_rate = ~0U;
2385 sk->sk_pacing_rate = ~0U;
2386 sk->sk_incoming_cpu = -1;
2388 * Before updating sk_refcnt, we must commit prior changes to memory
2389 * (Documentation/RCU/rculist_nulls.txt for details)
2392 atomic_set(&sk->sk_refcnt, 1);
2393 atomic_set(&sk->sk_drops, 0);
2395 EXPORT_SYMBOL(sock_init_data);
2397 void lock_sock_nested(struct sock *sk, int subclass)
2400 spin_lock_bh(&sk->sk_lock.slock);
2401 if (sk->sk_lock.owned)
2403 sk->sk_lock.owned = 1;
2404 spin_unlock(&sk->sk_lock.slock);
2406 * The sk_lock has mutex_lock() semantics here:
2408 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2411 EXPORT_SYMBOL(lock_sock_nested);
2413 void release_sock(struct sock *sk)
2416 * The sk_lock has mutex_unlock() semantics:
2418 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2420 spin_lock_bh(&sk->sk_lock.slock);
2421 if (sk->sk_backlog.tail)
2424 /* Warning : release_cb() might need to release sk ownership,
2425 * ie call sock_release_ownership(sk) before us.
2427 if (sk->sk_prot->release_cb)
2428 sk->sk_prot->release_cb(sk);
2430 sock_release_ownership(sk);
2431 if (waitqueue_active(&sk->sk_lock.wq))
2432 wake_up(&sk->sk_lock.wq);
2433 spin_unlock_bh(&sk->sk_lock.slock);
2435 EXPORT_SYMBOL(release_sock);
2438 * lock_sock_fast - fast version of lock_sock
2441 * This version should be used for very small section, where process wont block
2442 * return false if fast path is taken
2443 * sk_lock.slock locked, owned = 0, BH disabled
2444 * return true if slow path is taken
2445 * sk_lock.slock unlocked, owned = 1, BH enabled
2447 bool lock_sock_fast(struct sock *sk)
2450 spin_lock_bh(&sk->sk_lock.slock);
2452 if (!sk->sk_lock.owned)
2454 * Note : We must disable BH
2459 sk->sk_lock.owned = 1;
2460 spin_unlock(&sk->sk_lock.slock);
2462 * The sk_lock has mutex_lock() semantics here:
2464 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2468 EXPORT_SYMBOL(lock_sock_fast);
2470 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2473 if (!sock_flag(sk, SOCK_TIMESTAMP))
2474 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2475 tv = ktime_to_timeval(sk->sk_stamp);
2476 if (tv.tv_sec == -1)
2478 if (tv.tv_sec == 0) {
2479 sk->sk_stamp = ktime_get_real();
2480 tv = ktime_to_timeval(sk->sk_stamp);
2482 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2484 EXPORT_SYMBOL(sock_get_timestamp);
2486 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2489 if (!sock_flag(sk, SOCK_TIMESTAMP))
2490 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2491 ts = ktime_to_timespec(sk->sk_stamp);
2492 if (ts.tv_sec == -1)
2494 if (ts.tv_sec == 0) {
2495 sk->sk_stamp = ktime_get_real();
2496 ts = ktime_to_timespec(sk->sk_stamp);
2498 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2500 EXPORT_SYMBOL(sock_get_timestampns);
2502 void sock_enable_timestamp(struct sock *sk, int flag)
2504 if (!sock_flag(sk, flag)) {
2505 unsigned long previous_flags = sk->sk_flags;
2507 sock_set_flag(sk, flag);
2509 * we just set one of the two flags which require net
2510 * time stamping, but time stamping might have been on
2511 * already because of the other one
2513 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2514 net_enable_timestamp();
2518 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2519 int level, int type)
2521 struct sock_exterr_skb *serr;
2522 struct sk_buff *skb;
2526 skb = sock_dequeue_err_skb(sk);
2532 msg->msg_flags |= MSG_TRUNC;
2535 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2539 sock_recv_timestamp(msg, sk, skb);
2541 serr = SKB_EXT_ERR(skb);
2542 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2544 msg->msg_flags |= MSG_ERRQUEUE;
2552 EXPORT_SYMBOL(sock_recv_errqueue);
2555 * Get a socket option on an socket.
2557 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2558 * asynchronous errors should be reported by getsockopt. We assume
2559 * this means if you specify SO_ERROR (otherwise whats the point of it).
2561 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2562 char __user *optval, int __user *optlen)
2564 struct sock *sk = sock->sk;
2566 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2568 EXPORT_SYMBOL(sock_common_getsockopt);
2570 #ifdef CONFIG_COMPAT
2571 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2572 char __user *optval, int __user *optlen)
2574 struct sock *sk = sock->sk;
2576 if (sk->sk_prot->compat_getsockopt != NULL)
2577 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2579 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2581 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2584 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2587 struct sock *sk = sock->sk;
2591 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2592 flags & ~MSG_DONTWAIT, &addr_len);
2594 msg->msg_namelen = addr_len;
2597 EXPORT_SYMBOL(sock_common_recvmsg);
2600 * Set socket options on an inet socket.
2602 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2603 char __user *optval, unsigned int optlen)
2605 struct sock *sk = sock->sk;
2607 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2609 EXPORT_SYMBOL(sock_common_setsockopt);
2611 #ifdef CONFIG_COMPAT
2612 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2613 char __user *optval, unsigned int optlen)
2615 struct sock *sk = sock->sk;
2617 if (sk->sk_prot->compat_setsockopt != NULL)
2618 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2620 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2622 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2625 void sk_common_release(struct sock *sk)
2627 if (sk->sk_prot->destroy)
2628 sk->sk_prot->destroy(sk);
2631 * Observation: when sock_common_release is called, processes have
2632 * no access to socket. But net still has.
2633 * Step one, detach it from networking:
2635 * A. Remove from hash tables.
2638 sk->sk_prot->unhash(sk);
2641 * In this point socket cannot receive new packets, but it is possible
2642 * that some packets are in flight because some CPU runs receiver and
2643 * did hash table lookup before we unhashed socket. They will achieve
2644 * receive queue and will be purged by socket destructor.
2646 * Also we still have packets pending on receive queue and probably,
2647 * our own packets waiting in device queues. sock_destroy will drain
2648 * receive queue, but transmitted packets will delay socket destruction
2649 * until the last reference will be released.
2654 xfrm_sk_free_policy(sk);
2656 sk_refcnt_debug_release(sk);
2658 if (sk->sk_frag.page) {
2659 put_page(sk->sk_frag.page);
2660 sk->sk_frag.page = NULL;
2665 EXPORT_SYMBOL(sk_common_release);
2667 #ifdef CONFIG_PROC_FS
2668 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2670 int val[PROTO_INUSE_NR];
2673 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2675 #ifdef CONFIG_NET_NS
2676 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2678 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2680 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2682 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2684 int cpu, idx = prot->inuse_idx;
2687 for_each_possible_cpu(cpu)
2688 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2690 return res >= 0 ? res : 0;
2692 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2694 static int __net_init sock_inuse_init_net(struct net *net)
2696 net->core.inuse = alloc_percpu(struct prot_inuse);
2697 return net->core.inuse ? 0 : -ENOMEM;
2700 static void __net_exit sock_inuse_exit_net(struct net *net)
2702 free_percpu(net->core.inuse);
2705 static struct pernet_operations net_inuse_ops = {
2706 .init = sock_inuse_init_net,
2707 .exit = sock_inuse_exit_net,
2710 static __init int net_inuse_init(void)
2712 if (register_pernet_subsys(&net_inuse_ops))
2713 panic("Cannot initialize net inuse counters");
2718 core_initcall(net_inuse_init);
2720 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2722 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2724 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2726 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2728 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2730 int cpu, idx = prot->inuse_idx;
2733 for_each_possible_cpu(cpu)
2734 res += per_cpu(prot_inuse, cpu).val[idx];
2736 return res >= 0 ? res : 0;
2738 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2741 static void assign_proto_idx(struct proto *prot)
2743 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2745 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2746 pr_err("PROTO_INUSE_NR exhausted\n");
2750 set_bit(prot->inuse_idx, proto_inuse_idx);
2753 static void release_proto_idx(struct proto *prot)
2755 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2756 clear_bit(prot->inuse_idx, proto_inuse_idx);
2759 static inline void assign_proto_idx(struct proto *prot)
2763 static inline void release_proto_idx(struct proto *prot)
2768 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2772 kfree(rsk_prot->slab_name);
2773 rsk_prot->slab_name = NULL;
2774 kmem_cache_destroy(rsk_prot->slab);
2775 rsk_prot->slab = NULL;
2778 static int req_prot_init(const struct proto *prot)
2780 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2785 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2787 if (!rsk_prot->slab_name)
2790 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2791 rsk_prot->obj_size, 0,
2792 prot->slab_flags, NULL);
2794 if (!rsk_prot->slab) {
2795 pr_crit("%s: Can't create request sock SLAB cache!\n",
2802 int proto_register(struct proto *prot, int alloc_slab)
2805 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2806 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2809 if (prot->slab == NULL) {
2810 pr_crit("%s: Can't create sock SLAB cache!\n",
2815 if (req_prot_init(prot))
2816 goto out_free_request_sock_slab;
2818 if (prot->twsk_prot != NULL) {
2819 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2821 if (prot->twsk_prot->twsk_slab_name == NULL)
2822 goto out_free_request_sock_slab;
2824 prot->twsk_prot->twsk_slab =
2825 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2826 prot->twsk_prot->twsk_obj_size,
2830 if (prot->twsk_prot->twsk_slab == NULL)
2831 goto out_free_timewait_sock_slab_name;
2835 mutex_lock(&proto_list_mutex);
2836 list_add(&prot->node, &proto_list);
2837 assign_proto_idx(prot);
2838 mutex_unlock(&proto_list_mutex);
2841 out_free_timewait_sock_slab_name:
2842 kfree(prot->twsk_prot->twsk_slab_name);
2843 out_free_request_sock_slab:
2844 req_prot_cleanup(prot->rsk_prot);
2846 kmem_cache_destroy(prot->slab);
2851 EXPORT_SYMBOL(proto_register);
2853 void proto_unregister(struct proto *prot)
2855 mutex_lock(&proto_list_mutex);
2856 release_proto_idx(prot);
2857 list_del(&prot->node);
2858 mutex_unlock(&proto_list_mutex);
2860 kmem_cache_destroy(prot->slab);
2863 req_prot_cleanup(prot->rsk_prot);
2865 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2866 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2867 kfree(prot->twsk_prot->twsk_slab_name);
2868 prot->twsk_prot->twsk_slab = NULL;
2871 EXPORT_SYMBOL(proto_unregister);
2873 #ifdef CONFIG_PROC_FS
2874 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2875 __acquires(proto_list_mutex)
2877 mutex_lock(&proto_list_mutex);
2878 return seq_list_start_head(&proto_list, *pos);
2881 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2883 return seq_list_next(v, &proto_list, pos);
2886 static void proto_seq_stop(struct seq_file *seq, void *v)
2887 __releases(proto_list_mutex)
2889 mutex_unlock(&proto_list_mutex);
2892 static char proto_method_implemented(const void *method)
2894 return method == NULL ? 'n' : 'y';
2896 static long sock_prot_memory_allocated(struct proto *proto)
2898 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2901 static char *sock_prot_memory_pressure(struct proto *proto)
2903 return proto->memory_pressure != NULL ?
2904 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2907 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2910 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2911 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2914 sock_prot_inuse_get(seq_file_net(seq), proto),
2915 sock_prot_memory_allocated(proto),
2916 sock_prot_memory_pressure(proto),
2918 proto->slab == NULL ? "no" : "yes",
2919 module_name(proto->owner),
2920 proto_method_implemented(proto->close),
2921 proto_method_implemented(proto->connect),
2922 proto_method_implemented(proto->disconnect),
2923 proto_method_implemented(proto->accept),
2924 proto_method_implemented(proto->ioctl),
2925 proto_method_implemented(proto->init),
2926 proto_method_implemented(proto->destroy),
2927 proto_method_implemented(proto->shutdown),
2928 proto_method_implemented(proto->setsockopt),
2929 proto_method_implemented(proto->getsockopt),
2930 proto_method_implemented(proto->sendmsg),
2931 proto_method_implemented(proto->recvmsg),
2932 proto_method_implemented(proto->sendpage),
2933 proto_method_implemented(proto->bind),
2934 proto_method_implemented(proto->backlog_rcv),
2935 proto_method_implemented(proto->hash),
2936 proto_method_implemented(proto->unhash),
2937 proto_method_implemented(proto->get_port),
2938 proto_method_implemented(proto->enter_memory_pressure));
2941 static int proto_seq_show(struct seq_file *seq, void *v)
2943 if (v == &proto_list)
2944 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2953 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2955 proto_seq_printf(seq, list_entry(v, struct proto, node));
2959 static const struct seq_operations proto_seq_ops = {
2960 .start = proto_seq_start,
2961 .next = proto_seq_next,
2962 .stop = proto_seq_stop,
2963 .show = proto_seq_show,
2966 static int proto_seq_open(struct inode *inode, struct file *file)
2968 return seq_open_net(inode, file, &proto_seq_ops,
2969 sizeof(struct seq_net_private));
2972 static const struct file_operations proto_seq_fops = {
2973 .owner = THIS_MODULE,
2974 .open = proto_seq_open,
2976 .llseek = seq_lseek,
2977 .release = seq_release_net,
2980 static __net_init int proto_init_net(struct net *net)
2982 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2988 static __net_exit void proto_exit_net(struct net *net)
2990 remove_proc_entry("protocols", net->proc_net);
2994 static __net_initdata struct pernet_operations proto_net_ops = {
2995 .init = proto_init_net,
2996 .exit = proto_exit_net,
2999 static int __init proto_init(void)
3001 return register_pernet_subsys(&proto_net_ops);
3004 subsys_initcall(proto_init);
3006 #endif /* PROC_FS */