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.
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>
135 #include <linux/filter.h>
137 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
149 * sk_ns_capable - General socket capability test
150 * @sk: Socket to use a capability on or through
151 * @user_ns: The user namespace of the capability to use
152 * @cap: The capability to use
154 * Test to see if the opener of the socket had when the socket was
155 * created and the current process has the capability @cap in the user
156 * namespace @user_ns.
158 bool sk_ns_capable(const struct sock *sk,
159 struct user_namespace *user_ns, int cap)
161 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
162 ns_capable(user_ns, cap);
164 EXPORT_SYMBOL(sk_ns_capable);
167 * sk_capable - Socket global capability test
168 * @sk: Socket to use a capability on or through
169 * @cap: The global capability to use
171 * Test to see if the opener of the socket had when the socket was
172 * created and the current process has the capability @cap in all user
175 bool sk_capable(const struct sock *sk, int cap)
177 return sk_ns_capable(sk, &init_user_ns, cap);
179 EXPORT_SYMBOL(sk_capable);
182 * sk_net_capable - Network namespace socket capability test
183 * @sk: Socket to use a capability on or through
184 * @cap: The capability to use
186 * Test to see if the opener of the socket had when the socket was created
187 * and the current process has the capability @cap over the network namespace
188 * the socket is a member of.
190 bool sk_net_capable(const struct sock *sk, int cap)
192 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
194 EXPORT_SYMBOL(sk_net_capable);
197 #ifdef CONFIG_MEMCG_KMEM
198 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
203 mutex_lock(&proto_list_mutex);
204 list_for_each_entry(proto, &proto_list, node) {
205 if (proto->init_cgroup) {
206 ret = proto->init_cgroup(memcg, ss);
212 mutex_unlock(&proto_list_mutex);
215 list_for_each_entry_continue_reverse(proto, &proto_list, node)
216 if (proto->destroy_cgroup)
217 proto->destroy_cgroup(memcg);
218 mutex_unlock(&proto_list_mutex);
222 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
226 mutex_lock(&proto_list_mutex);
227 list_for_each_entry_reverse(proto, &proto_list, node)
228 if (proto->destroy_cgroup)
229 proto->destroy_cgroup(memcg);
230 mutex_unlock(&proto_list_mutex);
235 * Each address family might have different locking rules, so we have
236 * one slock key per address family:
238 static struct lock_class_key af_family_keys[AF_MAX];
239 static struct lock_class_key af_family_slock_keys[AF_MAX];
241 #if defined(CONFIG_MEMCG_KMEM)
242 struct static_key memcg_socket_limit_enabled;
243 EXPORT_SYMBOL(memcg_socket_limit_enabled);
247 * Make lock validator output more readable. (we pre-construct these
248 * strings build-time, so that runtime initialization of socket
251 static const char *const af_family_key_strings[AF_MAX+1] = {
252 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
253 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
254 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
255 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
256 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
257 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
258 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
259 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
260 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
261 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
262 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
263 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
264 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
265 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
267 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
268 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
269 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
270 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
271 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
272 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
273 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
274 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
275 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
276 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
277 "slock-27" , "slock-28" , "slock-AF_CAN" ,
278 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
279 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
280 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
281 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
283 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
284 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
285 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
286 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
287 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
288 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
289 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
290 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
291 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
292 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
293 "clock-27" , "clock-28" , "clock-AF_CAN" ,
294 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
295 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
296 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
297 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
301 * sk_callback_lock locking rules are per-address-family,
302 * so split the lock classes by using a per-AF key:
304 static struct lock_class_key af_callback_keys[AF_MAX];
306 /* Take into consideration the size of the struct sk_buff overhead in the
307 * determination of these values, since that is non-constant across
308 * platforms. This makes socket queueing behavior and performance
309 * not depend upon such differences.
311 #define _SK_MEM_PACKETS 256
312 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
313 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
314 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
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 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
331 EXPORT_SYMBOL_GPL(memalloc_socks);
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_key_slow_inc(&memalloc_socks);
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_key_slow_dec(&memalloc_socks);
356 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
357 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
358 * it has rmem allocations there is a risk that the user of the
359 * socket cannot make forward progress due to exceeding the rmem
360 * limits. By rights, sk_clear_memalloc() should only be called
361 * on sockets being torn down but warn and reset the accounting if
362 * that assumption breaks.
364 if (WARN_ON(sk->sk_forward_alloc))
367 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
369 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
372 unsigned long pflags = current->flags;
374 /* these should have been dropped before queueing */
375 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
377 current->flags |= PF_MEMALLOC;
378 ret = sk->sk_backlog_rcv(sk, skb);
379 tsk_restore_flags(current, pflags, PF_MEMALLOC);
383 EXPORT_SYMBOL(__sk_backlog_rcv);
385 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
389 if (optlen < sizeof(tv))
391 if (copy_from_user(&tv, optval, sizeof(tv)))
393 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
397 static int warned __read_mostly;
400 if (warned < 10 && net_ratelimit()) {
402 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
403 __func__, current->comm, task_pid_nr(current));
407 *timeo_p = MAX_SCHEDULE_TIMEOUT;
408 if (tv.tv_sec == 0 && tv.tv_usec == 0)
410 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
411 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
415 static void sock_warn_obsolete_bsdism(const char *name)
418 static char warncomm[TASK_COMM_LEN];
419 if (strcmp(warncomm, current->comm) && warned < 5) {
420 strcpy(warncomm, current->comm);
421 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
427 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
429 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
431 if (sk->sk_flags & flags) {
432 sk->sk_flags &= ~flags;
433 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
434 net_disable_timestamp();
439 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
443 struct sk_buff_head *list = &sk->sk_receive_queue;
445 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
446 atomic_inc(&sk->sk_drops);
447 trace_sock_rcvqueue_full(sk, skb);
451 err = sk_filter(sk, skb);
455 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
456 atomic_inc(&sk->sk_drops);
461 skb_set_owner_r(skb, sk);
463 /* we escape from rcu protected region, make sure we dont leak
468 spin_lock_irqsave(&list->lock, flags);
469 sock_skb_set_dropcount(sk, skb);
470 __skb_queue_tail(list, skb);
471 spin_unlock_irqrestore(&list->lock, flags);
473 if (!sock_flag(sk, SOCK_DEAD))
474 sk->sk_data_ready(sk);
477 EXPORT_SYMBOL(sock_queue_rcv_skb);
479 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
481 int rc = NET_RX_SUCCESS;
483 if (sk_filter(sk, skb))
484 goto discard_and_relse;
488 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
489 atomic_inc(&sk->sk_drops);
490 goto discard_and_relse;
493 bh_lock_sock_nested(sk);
496 if (!sock_owned_by_user(sk)) {
498 * trylock + unlock semantics:
500 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
502 rc = sk_backlog_rcv(sk, skb);
504 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
505 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
507 atomic_inc(&sk->sk_drops);
508 goto discard_and_relse;
519 EXPORT_SYMBOL(sk_receive_skb);
521 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
523 struct dst_entry *dst = __sk_dst_get(sk);
525 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
526 sk_tx_queue_clear(sk);
527 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
534 EXPORT_SYMBOL(__sk_dst_check);
536 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
538 struct dst_entry *dst = sk_dst_get(sk);
540 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
548 EXPORT_SYMBOL(sk_dst_check);
550 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
553 int ret = -ENOPROTOOPT;
554 #ifdef CONFIG_NETDEVICES
555 struct net *net = sock_net(sk);
556 char devname[IFNAMSIZ];
561 if (!ns_capable(net->user_ns, CAP_NET_RAW))
568 /* Bind this socket to a particular device like "eth0",
569 * as specified in the passed interface name. If the
570 * name is "" or the option length is zero the socket
573 if (optlen > IFNAMSIZ - 1)
574 optlen = IFNAMSIZ - 1;
575 memset(devname, 0, sizeof(devname));
578 if (copy_from_user(devname, optval, optlen))
582 if (devname[0] != '\0') {
583 struct net_device *dev;
586 dev = dev_get_by_name_rcu(net, devname);
588 index = dev->ifindex;
596 sk->sk_bound_dev_if = index;
608 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
609 int __user *optlen, int len)
611 int ret = -ENOPROTOOPT;
612 #ifdef CONFIG_NETDEVICES
613 struct net *net = sock_net(sk);
614 char devname[IFNAMSIZ];
616 if (sk->sk_bound_dev_if == 0) {
625 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
629 len = strlen(devname) + 1;
632 if (copy_to_user(optval, devname, len))
637 if (put_user(len, optlen))
648 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
651 sock_set_flag(sk, bit);
653 sock_reset_flag(sk, bit);
656 bool sk_mc_loop(struct sock *sk)
658 if (dev_recursion_level())
662 switch (sk->sk_family) {
664 return inet_sk(sk)->mc_loop;
665 #if IS_ENABLED(CONFIG_IPV6)
667 return inet6_sk(sk)->mc_loop;
673 EXPORT_SYMBOL(sk_mc_loop);
676 * This is meant for all protocols to use and covers goings on
677 * at the socket level. Everything here is generic.
680 int sock_setsockopt(struct socket *sock, int level, int optname,
681 char __user *optval, unsigned int optlen)
683 struct sock *sk = sock->sk;
690 * Options without arguments
693 if (optname == SO_BINDTODEVICE)
694 return sock_setbindtodevice(sk, optval, optlen);
696 if (optlen < sizeof(int))
699 if (get_user(val, (int __user *)optval))
702 valbool = val ? 1 : 0;
708 if (val && !capable(CAP_NET_ADMIN))
711 sock_valbool_flag(sk, SOCK_DBG, valbool);
714 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
717 sk->sk_reuseport = valbool;
726 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
729 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
732 /* Don't error on this BSD doesn't and if you think
733 * about it this is right. Otherwise apps have to
734 * play 'guess the biggest size' games. RCVBUF/SNDBUF
735 * are treated in BSD as hints
737 val = min_t(u32, val, sysctl_wmem_max);
739 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
740 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
741 /* Wake up sending tasks if we upped the value. */
742 sk->sk_write_space(sk);
746 if (!capable(CAP_NET_ADMIN)) {
753 /* Don't error on this BSD doesn't and if you think
754 * about it this is right. Otherwise apps have to
755 * play 'guess the biggest size' games. RCVBUF/SNDBUF
756 * are treated in BSD as hints
758 val = min_t(u32, val, sysctl_rmem_max);
760 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
762 * We double it on the way in to account for
763 * "struct sk_buff" etc. overhead. Applications
764 * assume that the SO_RCVBUF setting they make will
765 * allow that much actual data to be received on that
768 * Applications are unaware that "struct sk_buff" and
769 * other overheads allocate from the receive buffer
770 * during socket buffer allocation.
772 * And after considering the possible alternatives,
773 * returning the value we actually used in getsockopt
774 * is the most desirable behavior.
776 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
780 if (!capable(CAP_NET_ADMIN)) {
788 if (sk->sk_protocol == IPPROTO_TCP &&
789 sk->sk_type == SOCK_STREAM)
790 tcp_set_keepalive(sk, valbool);
792 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
796 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
800 sk->sk_no_check_tx = valbool;
804 if ((val >= 0 && val <= 6) ||
805 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
806 sk->sk_priority = val;
812 if (optlen < sizeof(ling)) {
813 ret = -EINVAL; /* 1003.1g */
816 if (copy_from_user(&ling, optval, sizeof(ling))) {
821 sock_reset_flag(sk, SOCK_LINGER);
823 #if (BITS_PER_LONG == 32)
824 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
825 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
828 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
829 sock_set_flag(sk, SOCK_LINGER);
834 sock_warn_obsolete_bsdism("setsockopt");
839 set_bit(SOCK_PASSCRED, &sock->flags);
841 clear_bit(SOCK_PASSCRED, &sock->flags);
847 if (optname == SO_TIMESTAMP)
848 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
850 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
851 sock_set_flag(sk, SOCK_RCVTSTAMP);
852 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
854 sock_reset_flag(sk, SOCK_RCVTSTAMP);
855 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
859 case SO_TIMESTAMPING:
860 if (val & ~SOF_TIMESTAMPING_MASK) {
865 if (val & SOF_TIMESTAMPING_OPT_ID &&
866 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
867 if (sk->sk_protocol == IPPROTO_TCP) {
868 if (sk->sk_state != TCP_ESTABLISHED) {
872 sk->sk_tskey = tcp_sk(sk)->snd_una;
877 sk->sk_tsflags = val;
878 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
879 sock_enable_timestamp(sk,
880 SOCK_TIMESTAMPING_RX_SOFTWARE);
882 sock_disable_timestamp(sk,
883 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
889 sk->sk_rcvlowat = val ? : 1;
893 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
897 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
900 case SO_ATTACH_FILTER:
902 if (optlen == sizeof(struct sock_fprog)) {
903 struct sock_fprog fprog;
906 if (copy_from_user(&fprog, optval, sizeof(fprog)))
909 ret = sk_attach_filter(&fprog, sk);
915 if (optlen == sizeof(u32)) {
919 if (copy_from_user(&ufd, optval, sizeof(ufd)))
922 ret = sk_attach_bpf(ufd, sk);
926 case SO_DETACH_FILTER:
927 ret = sk_detach_filter(sk);
931 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
934 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
939 set_bit(SOCK_PASSSEC, &sock->flags);
941 clear_bit(SOCK_PASSSEC, &sock->flags);
944 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
951 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
955 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
959 if (sock->ops->set_peek_off)
960 ret = sock->ops->set_peek_off(sk, val);
966 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
969 case SO_SELECT_ERR_QUEUE:
970 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
973 #ifdef CONFIG_NET_RX_BUSY_POLL
975 /* allow unprivileged users to decrease the value */
976 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
982 sk->sk_ll_usec = val;
987 case SO_MAX_PACING_RATE:
988 sk->sk_max_pacing_rate = val;
989 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
990 sk->sk_max_pacing_rate);
1000 EXPORT_SYMBOL(sock_setsockopt);
1003 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1004 struct ucred *ucred)
1006 ucred->pid = pid_vnr(pid);
1007 ucred->uid = ucred->gid = -1;
1009 struct user_namespace *current_ns = current_user_ns();
1011 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1012 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1016 int sock_getsockopt(struct socket *sock, int level, int optname,
1017 char __user *optval, int __user *optlen)
1019 struct sock *sk = sock->sk;
1027 int lv = sizeof(int);
1030 if (get_user(len, optlen))
1035 memset(&v, 0, sizeof(v));
1039 v.val = sock_flag(sk, SOCK_DBG);
1043 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1047 v.val = sock_flag(sk, SOCK_BROADCAST);
1051 v.val = sk->sk_sndbuf;
1055 v.val = sk->sk_rcvbuf;
1059 v.val = sk->sk_reuse;
1063 v.val = sk->sk_reuseport;
1067 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1071 v.val = sk->sk_type;
1075 v.val = sk->sk_protocol;
1079 v.val = sk->sk_family;
1083 v.val = -sock_error(sk);
1085 v.val = xchg(&sk->sk_err_soft, 0);
1089 v.val = sock_flag(sk, SOCK_URGINLINE);
1093 v.val = sk->sk_no_check_tx;
1097 v.val = sk->sk_priority;
1101 lv = sizeof(v.ling);
1102 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1103 v.ling.l_linger = sk->sk_lingertime / HZ;
1107 sock_warn_obsolete_bsdism("getsockopt");
1111 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1112 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1115 case SO_TIMESTAMPNS:
1116 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1119 case SO_TIMESTAMPING:
1120 v.val = sk->sk_tsflags;
1124 lv = sizeof(struct timeval);
1125 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1129 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1130 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1135 lv = sizeof(struct timeval);
1136 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1140 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1141 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1146 v.val = sk->sk_rcvlowat;
1154 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1159 struct ucred peercred;
1160 if (len > sizeof(peercred))
1161 len = sizeof(peercred);
1162 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1163 if (copy_to_user(optval, &peercred, len))
1172 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1176 if (copy_to_user(optval, address, len))
1181 /* Dubious BSD thing... Probably nobody even uses it, but
1182 * the UNIX standard wants it for whatever reason... -DaveM
1185 v.val = sk->sk_state == TCP_LISTEN;
1189 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1193 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1196 v.val = sk->sk_mark;
1200 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1203 case SO_WIFI_STATUS:
1204 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1208 if (!sock->ops->set_peek_off)
1211 v.val = sk->sk_peek_off;
1214 v.val = sock_flag(sk, SOCK_NOFCS);
1217 case SO_BINDTODEVICE:
1218 return sock_getbindtodevice(sk, optval, optlen, len);
1221 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1227 case SO_LOCK_FILTER:
1228 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1231 case SO_BPF_EXTENSIONS:
1232 v.val = bpf_tell_extensions();
1235 case SO_SELECT_ERR_QUEUE:
1236 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1239 #ifdef CONFIG_NET_RX_BUSY_POLL
1241 v.val = sk->sk_ll_usec;
1245 case SO_MAX_PACING_RATE:
1246 v.val = sk->sk_max_pacing_rate;
1249 case SO_INCOMING_CPU:
1250 v.val = sk->sk_incoming_cpu;
1254 /* We implement the SO_SNDLOWAT etc to not be settable
1257 return -ENOPROTOOPT;
1262 if (copy_to_user(optval, &v, len))
1265 if (put_user(len, optlen))
1271 * Initialize an sk_lock.
1273 * (We also register the sk_lock with the lock validator.)
1275 static inline void sock_lock_init(struct sock *sk)
1277 sock_lock_init_class_and_name(sk,
1278 af_family_slock_key_strings[sk->sk_family],
1279 af_family_slock_keys + sk->sk_family,
1280 af_family_key_strings[sk->sk_family],
1281 af_family_keys + sk->sk_family);
1285 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1286 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1287 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1289 static void sock_copy(struct sock *nsk, const struct sock *osk)
1291 #ifdef CONFIG_SECURITY_NETWORK
1292 void *sptr = nsk->sk_security;
1294 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1296 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1297 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1299 #ifdef CONFIG_SECURITY_NETWORK
1300 nsk->sk_security = sptr;
1301 security_sk_clone(osk, nsk);
1305 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1307 unsigned long nulls1, nulls2;
1309 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1310 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1311 if (nulls1 > nulls2)
1312 swap(nulls1, nulls2);
1315 memset((char *)sk, 0, nulls1);
1316 memset((char *)sk + nulls1 + sizeof(void *), 0,
1317 nulls2 - nulls1 - sizeof(void *));
1318 memset((char *)sk + nulls2 + sizeof(void *), 0,
1319 size - nulls2 - sizeof(void *));
1321 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1323 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1327 struct kmem_cache *slab;
1331 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1334 if (priority & __GFP_ZERO) {
1336 prot->clear_sk(sk, prot->obj_size);
1338 sk_prot_clear_nulls(sk, prot->obj_size);
1341 sk = kmalloc(prot->obj_size, priority);
1344 kmemcheck_annotate_bitfield(sk, flags);
1346 if (security_sk_alloc(sk, family, priority))
1349 if (!try_module_get(prot->owner))
1351 sk_tx_queue_clear(sk);
1357 security_sk_free(sk);
1360 kmem_cache_free(slab, sk);
1366 static void sk_prot_free(struct proto *prot, struct sock *sk)
1368 struct kmem_cache *slab;
1369 struct module *owner;
1371 owner = prot->owner;
1374 security_sk_free(sk);
1376 kmem_cache_free(slab, sk);
1382 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1383 void sock_update_netprioidx(struct sock *sk)
1388 sk->sk_cgrp_prioidx = task_netprioidx(current);
1390 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1394 * sk_alloc - All socket objects are allocated here
1395 * @net: the applicable net namespace
1396 * @family: protocol family
1397 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1398 * @prot: struct proto associated with this new sock instance
1400 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1405 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1407 sk->sk_family = family;
1409 * See comment in struct sock definition to understand
1410 * why we need sk_prot_creator -acme
1412 sk->sk_prot = sk->sk_prot_creator = prot;
1414 sock_net_set(sk, get_net(net));
1415 atomic_set(&sk->sk_wmem_alloc, 1);
1417 sock_update_classid(sk);
1418 sock_update_netprioidx(sk);
1423 EXPORT_SYMBOL(sk_alloc);
1425 static void __sk_free(struct sock *sk)
1427 struct sk_filter *filter;
1429 if (sk->sk_destruct)
1430 sk->sk_destruct(sk);
1432 filter = rcu_dereference_check(sk->sk_filter,
1433 atomic_read(&sk->sk_wmem_alloc) == 0);
1435 sk_filter_uncharge(sk, filter);
1436 RCU_INIT_POINTER(sk->sk_filter, NULL);
1439 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1441 if (atomic_read(&sk->sk_omem_alloc))
1442 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1443 __func__, atomic_read(&sk->sk_omem_alloc));
1445 if (sk->sk_peer_cred)
1446 put_cred(sk->sk_peer_cred);
1447 put_pid(sk->sk_peer_pid);
1448 put_net(sock_net(sk));
1449 sk_prot_free(sk->sk_prot_creator, sk);
1452 void sk_free(struct sock *sk)
1455 * We subtract one from sk_wmem_alloc and can know if
1456 * some packets are still in some tx queue.
1457 * If not null, sock_wfree() will call __sk_free(sk) later
1459 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1462 EXPORT_SYMBOL(sk_free);
1465 * Last sock_put should drop reference to sk->sk_net. It has already
1466 * been dropped in sk_change_net. Taking reference to stopping namespace
1468 * Take reference to a socket to remove it from hash _alive_ and after that
1469 * destroy it in the context of init_net.
1471 void sk_release_kernel(struct sock *sk)
1473 if (sk == NULL || sk->sk_socket == NULL)
1477 sock_net_set(sk, get_net(&init_net));
1478 sock_release(sk->sk_socket);
1481 EXPORT_SYMBOL(sk_release_kernel);
1483 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1485 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1486 sock_update_memcg(newsk);
1490 * sk_clone_lock - clone a socket, and lock its clone
1491 * @sk: the socket to clone
1492 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1494 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1496 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1499 bool is_charged = true;
1501 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1502 if (newsk != NULL) {
1503 struct sk_filter *filter;
1505 sock_copy(newsk, sk);
1508 get_net(sock_net(newsk));
1509 sk_node_init(&newsk->sk_node);
1510 sock_lock_init(newsk);
1511 bh_lock_sock(newsk);
1512 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1513 newsk->sk_backlog.len = 0;
1515 atomic_set(&newsk->sk_rmem_alloc, 0);
1517 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1519 atomic_set(&newsk->sk_wmem_alloc, 1);
1520 atomic_set(&newsk->sk_omem_alloc, 0);
1521 skb_queue_head_init(&newsk->sk_receive_queue);
1522 skb_queue_head_init(&newsk->sk_write_queue);
1524 spin_lock_init(&newsk->sk_dst_lock);
1525 rwlock_init(&newsk->sk_callback_lock);
1526 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1527 af_callback_keys + newsk->sk_family,
1528 af_family_clock_key_strings[newsk->sk_family]);
1530 newsk->sk_dst_cache = NULL;
1531 newsk->sk_wmem_queued = 0;
1532 newsk->sk_forward_alloc = 0;
1533 newsk->sk_send_head = NULL;
1534 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1536 sock_reset_flag(newsk, SOCK_DONE);
1537 skb_queue_head_init(&newsk->sk_error_queue);
1539 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1541 /* though it's an empty new sock, the charging may fail
1542 * if sysctl_optmem_max was changed between creation of
1543 * original socket and cloning
1545 is_charged = sk_filter_charge(newsk, filter);
1547 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1548 /* It is still raw copy of parent, so invalidate
1549 * destructor and make plain sk_free() */
1550 newsk->sk_destruct = NULL;
1551 bh_unlock_sock(newsk);
1558 newsk->sk_priority = 0;
1559 newsk->sk_incoming_cpu = raw_smp_processor_id();
1560 atomic64_set(&newsk->sk_cookie, 0);
1562 * Before updating sk_refcnt, we must commit prior changes to memory
1563 * (Documentation/RCU/rculist_nulls.txt for details)
1566 atomic_set(&newsk->sk_refcnt, 2);
1569 * Increment the counter in the same struct proto as the master
1570 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1571 * is the same as sk->sk_prot->socks, as this field was copied
1574 * This _changes_ the previous behaviour, where
1575 * tcp_create_openreq_child always was incrementing the
1576 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1577 * to be taken into account in all callers. -acme
1579 sk_refcnt_debug_inc(newsk);
1580 sk_set_socket(newsk, NULL);
1581 newsk->sk_wq = NULL;
1583 sk_update_clone(sk, newsk);
1585 if (newsk->sk_prot->sockets_allocated)
1586 sk_sockets_allocated_inc(newsk);
1588 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1589 net_enable_timestamp();
1594 EXPORT_SYMBOL_GPL(sk_clone_lock);
1596 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1598 __sk_dst_set(sk, dst);
1599 sk->sk_route_caps = dst->dev->features;
1600 if (sk->sk_route_caps & NETIF_F_GSO)
1601 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1602 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1603 if (sk_can_gso(sk)) {
1604 if (dst->header_len) {
1605 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1607 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1608 sk->sk_gso_max_size = dst->dev->gso_max_size;
1609 sk->sk_gso_max_segs = dst->dev->gso_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 /* On 32bit arches, an skb frag is limited to 2^15 */
1860 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1863 * skb_page_frag_refill - check that a page_frag contains enough room
1864 * @sz: minimum size of the fragment we want to get
1865 * @pfrag: pointer to page_frag
1866 * @gfp: priority for memory allocation
1868 * Note: While this allocator tries to use high order pages, there is
1869 * no guarantee that allocations succeed. Therefore, @sz MUST be
1870 * less or equal than PAGE_SIZE.
1872 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1875 if (atomic_read(&pfrag->page->_count) == 1) {
1879 if (pfrag->offset + sz <= pfrag->size)
1881 put_page(pfrag->page);
1885 if (SKB_FRAG_PAGE_ORDER) {
1886 pfrag->page = alloc_pages(gfp | __GFP_COMP |
1887 __GFP_NOWARN | __GFP_NORETRY,
1888 SKB_FRAG_PAGE_ORDER);
1889 if (likely(pfrag->page)) {
1890 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1894 pfrag->page = alloc_page(gfp);
1895 if (likely(pfrag->page)) {
1896 pfrag->size = PAGE_SIZE;
1901 EXPORT_SYMBOL(skb_page_frag_refill);
1903 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1905 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1908 sk_enter_memory_pressure(sk);
1909 sk_stream_moderate_sndbuf(sk);
1912 EXPORT_SYMBOL(sk_page_frag_refill);
1914 static void __lock_sock(struct sock *sk)
1915 __releases(&sk->sk_lock.slock)
1916 __acquires(&sk->sk_lock.slock)
1921 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1922 TASK_UNINTERRUPTIBLE);
1923 spin_unlock_bh(&sk->sk_lock.slock);
1925 spin_lock_bh(&sk->sk_lock.slock);
1926 if (!sock_owned_by_user(sk))
1929 finish_wait(&sk->sk_lock.wq, &wait);
1932 static void __release_sock(struct sock *sk)
1933 __releases(&sk->sk_lock.slock)
1934 __acquires(&sk->sk_lock.slock)
1936 struct sk_buff *skb = sk->sk_backlog.head;
1939 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1943 struct sk_buff *next = skb->next;
1946 WARN_ON_ONCE(skb_dst_is_noref(skb));
1948 sk_backlog_rcv(sk, skb);
1951 * We are in process context here with softirqs
1952 * disabled, use cond_resched_softirq() to preempt.
1953 * This is safe to do because we've taken the backlog
1956 cond_resched_softirq();
1959 } while (skb != NULL);
1962 } while ((skb = sk->sk_backlog.head) != NULL);
1965 * Doing the zeroing here guarantee we can not loop forever
1966 * while a wild producer attempts to flood us.
1968 sk->sk_backlog.len = 0;
1972 * sk_wait_data - wait for data to arrive at sk_receive_queue
1973 * @sk: sock to wait on
1974 * @timeo: for how long
1976 * Now socket state including sk->sk_err is changed only under lock,
1977 * hence we may omit checks after joining wait queue.
1978 * We check receive queue before schedule() only as optimization;
1979 * it is very likely that release_sock() added new data.
1981 int sk_wait_data(struct sock *sk, long *timeo)
1986 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1987 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1988 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1989 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1990 finish_wait(sk_sleep(sk), &wait);
1993 EXPORT_SYMBOL(sk_wait_data);
1996 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1998 * @size: memory size to allocate
1999 * @kind: allocation type
2001 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2002 * rmem allocation. This function assumes that protocols which have
2003 * memory_pressure use sk_wmem_queued as write buffer accounting.
2005 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2007 struct proto *prot = sk->sk_prot;
2008 int amt = sk_mem_pages(size);
2010 int parent_status = UNDER_LIMIT;
2012 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2014 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2017 if (parent_status == UNDER_LIMIT &&
2018 allocated <= sk_prot_mem_limits(sk, 0)) {
2019 sk_leave_memory_pressure(sk);
2023 /* Under pressure. (we or our parents) */
2024 if ((parent_status > SOFT_LIMIT) ||
2025 allocated > sk_prot_mem_limits(sk, 1))
2026 sk_enter_memory_pressure(sk);
2028 /* Over hard limit (we or our parents) */
2029 if ((parent_status == OVER_LIMIT) ||
2030 (allocated > sk_prot_mem_limits(sk, 2)))
2031 goto suppress_allocation;
2033 /* guarantee minimum buffer size under pressure */
2034 if (kind == SK_MEM_RECV) {
2035 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2038 } else { /* SK_MEM_SEND */
2039 if (sk->sk_type == SOCK_STREAM) {
2040 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2042 } else if (atomic_read(&sk->sk_wmem_alloc) <
2043 prot->sysctl_wmem[0])
2047 if (sk_has_memory_pressure(sk)) {
2050 if (!sk_under_memory_pressure(sk))
2052 alloc = sk_sockets_allocated_read_positive(sk);
2053 if (sk_prot_mem_limits(sk, 2) > alloc *
2054 sk_mem_pages(sk->sk_wmem_queued +
2055 atomic_read(&sk->sk_rmem_alloc) +
2056 sk->sk_forward_alloc))
2060 suppress_allocation:
2062 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2063 sk_stream_moderate_sndbuf(sk);
2065 /* Fail only if socket is _under_ its sndbuf.
2066 * In this case we cannot block, so that we have to fail.
2068 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2072 trace_sock_exceed_buf_limit(sk, prot, allocated);
2074 /* Alas. Undo changes. */
2075 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2077 sk_memory_allocated_sub(sk, amt);
2081 EXPORT_SYMBOL(__sk_mem_schedule);
2084 * __sk_reclaim - reclaim memory_allocated
2087 void __sk_mem_reclaim(struct sock *sk)
2089 sk_memory_allocated_sub(sk,
2090 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2091 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2093 if (sk_under_memory_pressure(sk) &&
2094 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2095 sk_leave_memory_pressure(sk);
2097 EXPORT_SYMBOL(__sk_mem_reclaim);
2101 * Set of default routines for initialising struct proto_ops when
2102 * the protocol does not support a particular function. In certain
2103 * cases where it makes no sense for a protocol to have a "do nothing"
2104 * function, some default processing is provided.
2107 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2111 EXPORT_SYMBOL(sock_no_bind);
2113 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2118 EXPORT_SYMBOL(sock_no_connect);
2120 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2124 EXPORT_SYMBOL(sock_no_socketpair);
2126 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2130 EXPORT_SYMBOL(sock_no_accept);
2132 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2137 EXPORT_SYMBOL(sock_no_getname);
2139 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2143 EXPORT_SYMBOL(sock_no_poll);
2145 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2149 EXPORT_SYMBOL(sock_no_ioctl);
2151 int sock_no_listen(struct socket *sock, int backlog)
2155 EXPORT_SYMBOL(sock_no_listen);
2157 int sock_no_shutdown(struct socket *sock, int how)
2161 EXPORT_SYMBOL(sock_no_shutdown);
2163 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2164 char __user *optval, unsigned int optlen)
2168 EXPORT_SYMBOL(sock_no_setsockopt);
2170 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2171 char __user *optval, int __user *optlen)
2175 EXPORT_SYMBOL(sock_no_getsockopt);
2177 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2181 EXPORT_SYMBOL(sock_no_sendmsg);
2183 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2188 EXPORT_SYMBOL(sock_no_recvmsg);
2190 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2192 /* Mirror missing mmap method error code */
2195 EXPORT_SYMBOL(sock_no_mmap);
2197 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2200 struct msghdr msg = {.msg_flags = flags};
2202 char *kaddr = kmap(page);
2203 iov.iov_base = kaddr + offset;
2205 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2209 EXPORT_SYMBOL(sock_no_sendpage);
2212 * Default Socket Callbacks
2215 static void sock_def_wakeup(struct sock *sk)
2217 struct socket_wq *wq;
2220 wq = rcu_dereference(sk->sk_wq);
2221 if (wq_has_sleeper(wq))
2222 wake_up_interruptible_all(&wq->wait);
2226 static void sock_def_error_report(struct sock *sk)
2228 struct socket_wq *wq;
2231 wq = rcu_dereference(sk->sk_wq);
2232 if (wq_has_sleeper(wq))
2233 wake_up_interruptible_poll(&wq->wait, POLLERR);
2234 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2238 static void sock_def_readable(struct sock *sk)
2240 struct socket_wq *wq;
2243 wq = rcu_dereference(sk->sk_wq);
2244 if (wq_has_sleeper(wq))
2245 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2246 POLLRDNORM | POLLRDBAND);
2247 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2251 static void sock_def_write_space(struct sock *sk)
2253 struct socket_wq *wq;
2257 /* Do not wake up a writer until he can make "significant"
2260 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2261 wq = rcu_dereference(sk->sk_wq);
2262 if (wq_has_sleeper(wq))
2263 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2264 POLLWRNORM | POLLWRBAND);
2266 /* Should agree with poll, otherwise some programs break */
2267 if (sock_writeable(sk))
2268 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2274 static void sock_def_destruct(struct sock *sk)
2276 kfree(sk->sk_protinfo);
2279 void sk_send_sigurg(struct sock *sk)
2281 if (sk->sk_socket && sk->sk_socket->file)
2282 if (send_sigurg(&sk->sk_socket->file->f_owner))
2283 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2285 EXPORT_SYMBOL(sk_send_sigurg);
2287 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2288 unsigned long expires)
2290 if (!mod_timer(timer, expires))
2293 EXPORT_SYMBOL(sk_reset_timer);
2295 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2297 if (del_timer(timer))
2300 EXPORT_SYMBOL(sk_stop_timer);
2302 void sock_init_data(struct socket *sock, struct sock *sk)
2304 skb_queue_head_init(&sk->sk_receive_queue);
2305 skb_queue_head_init(&sk->sk_write_queue);
2306 skb_queue_head_init(&sk->sk_error_queue);
2308 sk->sk_send_head = NULL;
2310 init_timer(&sk->sk_timer);
2312 sk->sk_allocation = GFP_KERNEL;
2313 sk->sk_rcvbuf = sysctl_rmem_default;
2314 sk->sk_sndbuf = sysctl_wmem_default;
2315 sk->sk_state = TCP_CLOSE;
2316 sk_set_socket(sk, sock);
2318 sock_set_flag(sk, SOCK_ZAPPED);
2321 sk->sk_type = sock->type;
2322 sk->sk_wq = sock->wq;
2327 spin_lock_init(&sk->sk_dst_lock);
2328 rwlock_init(&sk->sk_callback_lock);
2329 lockdep_set_class_and_name(&sk->sk_callback_lock,
2330 af_callback_keys + sk->sk_family,
2331 af_family_clock_key_strings[sk->sk_family]);
2333 sk->sk_state_change = sock_def_wakeup;
2334 sk->sk_data_ready = sock_def_readable;
2335 sk->sk_write_space = sock_def_write_space;
2336 sk->sk_error_report = sock_def_error_report;
2337 sk->sk_destruct = sock_def_destruct;
2339 sk->sk_frag.page = NULL;
2340 sk->sk_frag.offset = 0;
2341 sk->sk_peek_off = -1;
2343 sk->sk_peer_pid = NULL;
2344 sk->sk_peer_cred = NULL;
2345 sk->sk_write_pending = 0;
2346 sk->sk_rcvlowat = 1;
2347 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2348 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2350 sk->sk_stamp = ktime_set(-1L, 0);
2352 #ifdef CONFIG_NET_RX_BUSY_POLL
2354 sk->sk_ll_usec = sysctl_net_busy_read;
2357 sk->sk_max_pacing_rate = ~0U;
2358 sk->sk_pacing_rate = ~0U;
2360 * Before updating sk_refcnt, we must commit prior changes to memory
2361 * (Documentation/RCU/rculist_nulls.txt for details)
2364 atomic_set(&sk->sk_refcnt, 1);
2365 atomic_set(&sk->sk_drops, 0);
2367 EXPORT_SYMBOL(sock_init_data);
2369 void lock_sock_nested(struct sock *sk, int subclass)
2372 spin_lock_bh(&sk->sk_lock.slock);
2373 if (sk->sk_lock.owned)
2375 sk->sk_lock.owned = 1;
2376 spin_unlock(&sk->sk_lock.slock);
2378 * The sk_lock has mutex_lock() semantics here:
2380 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2383 EXPORT_SYMBOL(lock_sock_nested);
2385 void release_sock(struct sock *sk)
2388 * The sk_lock has mutex_unlock() semantics:
2390 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2392 spin_lock_bh(&sk->sk_lock.slock);
2393 if (sk->sk_backlog.tail)
2396 /* Warning : release_cb() might need to release sk ownership,
2397 * ie call sock_release_ownership(sk) before us.
2399 if (sk->sk_prot->release_cb)
2400 sk->sk_prot->release_cb(sk);
2402 sock_release_ownership(sk);
2403 if (waitqueue_active(&sk->sk_lock.wq))
2404 wake_up(&sk->sk_lock.wq);
2405 spin_unlock_bh(&sk->sk_lock.slock);
2407 EXPORT_SYMBOL(release_sock);
2410 * lock_sock_fast - fast version of lock_sock
2413 * This version should be used for very small section, where process wont block
2414 * return false if fast path is taken
2415 * sk_lock.slock locked, owned = 0, BH disabled
2416 * return true if slow path is taken
2417 * sk_lock.slock unlocked, owned = 1, BH enabled
2419 bool lock_sock_fast(struct sock *sk)
2422 spin_lock_bh(&sk->sk_lock.slock);
2424 if (!sk->sk_lock.owned)
2426 * Note : We must disable BH
2431 sk->sk_lock.owned = 1;
2432 spin_unlock(&sk->sk_lock.slock);
2434 * The sk_lock has mutex_lock() semantics here:
2436 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2440 EXPORT_SYMBOL(lock_sock_fast);
2442 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2445 if (!sock_flag(sk, SOCK_TIMESTAMP))
2446 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2447 tv = ktime_to_timeval(sk->sk_stamp);
2448 if (tv.tv_sec == -1)
2450 if (tv.tv_sec == 0) {
2451 sk->sk_stamp = ktime_get_real();
2452 tv = ktime_to_timeval(sk->sk_stamp);
2454 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2456 EXPORT_SYMBOL(sock_get_timestamp);
2458 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2461 if (!sock_flag(sk, SOCK_TIMESTAMP))
2462 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2463 ts = ktime_to_timespec(sk->sk_stamp);
2464 if (ts.tv_sec == -1)
2466 if (ts.tv_sec == 0) {
2467 sk->sk_stamp = ktime_get_real();
2468 ts = ktime_to_timespec(sk->sk_stamp);
2470 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2472 EXPORT_SYMBOL(sock_get_timestampns);
2474 void sock_enable_timestamp(struct sock *sk, int flag)
2476 if (!sock_flag(sk, flag)) {
2477 unsigned long previous_flags = sk->sk_flags;
2479 sock_set_flag(sk, flag);
2481 * we just set one of the two flags which require net
2482 * time stamping, but time stamping might have been on
2483 * already because of the other one
2485 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2486 net_enable_timestamp();
2490 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2491 int level, int type)
2493 struct sock_exterr_skb *serr;
2494 struct sk_buff *skb;
2498 skb = sock_dequeue_err_skb(sk);
2504 msg->msg_flags |= MSG_TRUNC;
2507 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2511 sock_recv_timestamp(msg, sk, skb);
2513 serr = SKB_EXT_ERR(skb);
2514 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2516 msg->msg_flags |= MSG_ERRQUEUE;
2524 EXPORT_SYMBOL(sock_recv_errqueue);
2527 * Get a socket option on an socket.
2529 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2530 * asynchronous errors should be reported by getsockopt. We assume
2531 * this means if you specify SO_ERROR (otherwise whats the point of it).
2533 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2534 char __user *optval, int __user *optlen)
2536 struct sock *sk = sock->sk;
2538 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2540 EXPORT_SYMBOL(sock_common_getsockopt);
2542 #ifdef CONFIG_COMPAT
2543 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2544 char __user *optval, int __user *optlen)
2546 struct sock *sk = sock->sk;
2548 if (sk->sk_prot->compat_getsockopt != NULL)
2549 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2551 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2553 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2556 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2559 struct sock *sk = sock->sk;
2563 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2564 flags & ~MSG_DONTWAIT, &addr_len);
2566 msg->msg_namelen = addr_len;
2569 EXPORT_SYMBOL(sock_common_recvmsg);
2572 * Set socket options on an inet socket.
2574 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2575 char __user *optval, unsigned int optlen)
2577 struct sock *sk = sock->sk;
2579 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2581 EXPORT_SYMBOL(sock_common_setsockopt);
2583 #ifdef CONFIG_COMPAT
2584 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2585 char __user *optval, unsigned int optlen)
2587 struct sock *sk = sock->sk;
2589 if (sk->sk_prot->compat_setsockopt != NULL)
2590 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2592 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2594 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2597 void sk_common_release(struct sock *sk)
2599 if (sk->sk_prot->destroy)
2600 sk->sk_prot->destroy(sk);
2603 * Observation: when sock_common_release is called, processes have
2604 * no access to socket. But net still has.
2605 * Step one, detach it from networking:
2607 * A. Remove from hash tables.
2610 sk->sk_prot->unhash(sk);
2613 * In this point socket cannot receive new packets, but it is possible
2614 * that some packets are in flight because some CPU runs receiver and
2615 * did hash table lookup before we unhashed socket. They will achieve
2616 * receive queue and will be purged by socket destructor.
2618 * Also we still have packets pending on receive queue and probably,
2619 * our own packets waiting in device queues. sock_destroy will drain
2620 * receive queue, but transmitted packets will delay socket destruction
2621 * until the last reference will be released.
2626 xfrm_sk_free_policy(sk);
2628 sk_refcnt_debug_release(sk);
2630 if (sk->sk_frag.page) {
2631 put_page(sk->sk_frag.page);
2632 sk->sk_frag.page = NULL;
2637 EXPORT_SYMBOL(sk_common_release);
2639 #ifdef CONFIG_PROC_FS
2640 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2642 int val[PROTO_INUSE_NR];
2645 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2647 #ifdef CONFIG_NET_NS
2648 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2650 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2652 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2654 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2656 int cpu, idx = prot->inuse_idx;
2659 for_each_possible_cpu(cpu)
2660 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2662 return res >= 0 ? res : 0;
2664 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2666 static int __net_init sock_inuse_init_net(struct net *net)
2668 net->core.inuse = alloc_percpu(struct prot_inuse);
2669 return net->core.inuse ? 0 : -ENOMEM;
2672 static void __net_exit sock_inuse_exit_net(struct net *net)
2674 free_percpu(net->core.inuse);
2677 static struct pernet_operations net_inuse_ops = {
2678 .init = sock_inuse_init_net,
2679 .exit = sock_inuse_exit_net,
2682 static __init int net_inuse_init(void)
2684 if (register_pernet_subsys(&net_inuse_ops))
2685 panic("Cannot initialize net inuse counters");
2690 core_initcall(net_inuse_init);
2692 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2694 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2696 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2698 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2700 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2702 int cpu, idx = prot->inuse_idx;
2705 for_each_possible_cpu(cpu)
2706 res += per_cpu(prot_inuse, cpu).val[idx];
2708 return res >= 0 ? res : 0;
2710 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2713 static void assign_proto_idx(struct proto *prot)
2715 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2717 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2718 pr_err("PROTO_INUSE_NR exhausted\n");
2722 set_bit(prot->inuse_idx, proto_inuse_idx);
2725 static void release_proto_idx(struct proto *prot)
2727 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2728 clear_bit(prot->inuse_idx, proto_inuse_idx);
2731 static inline void assign_proto_idx(struct proto *prot)
2735 static inline void release_proto_idx(struct proto *prot)
2740 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2744 kfree(rsk_prot->slab_name);
2745 rsk_prot->slab_name = NULL;
2746 if (rsk_prot->slab) {
2747 kmem_cache_destroy(rsk_prot->slab);
2748 rsk_prot->slab = NULL;
2752 static int req_prot_init(const struct proto *prot)
2754 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2759 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2761 if (!rsk_prot->slab_name)
2764 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2765 rsk_prot->obj_size, 0,
2768 if (!rsk_prot->slab) {
2769 pr_crit("%s: Can't create request sock SLAB cache!\n",
2776 int proto_register(struct proto *prot, int alloc_slab)
2779 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2780 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2783 if (prot->slab == NULL) {
2784 pr_crit("%s: Can't create sock SLAB cache!\n",
2789 if (req_prot_init(prot))
2790 goto out_free_request_sock_slab;
2792 if (prot->twsk_prot != NULL) {
2793 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2795 if (prot->twsk_prot->twsk_slab_name == NULL)
2796 goto out_free_request_sock_slab;
2798 prot->twsk_prot->twsk_slab =
2799 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2800 prot->twsk_prot->twsk_obj_size,
2804 if (prot->twsk_prot->twsk_slab == NULL)
2805 goto out_free_timewait_sock_slab_name;
2809 mutex_lock(&proto_list_mutex);
2810 list_add(&prot->node, &proto_list);
2811 assign_proto_idx(prot);
2812 mutex_unlock(&proto_list_mutex);
2815 out_free_timewait_sock_slab_name:
2816 kfree(prot->twsk_prot->twsk_slab_name);
2817 out_free_request_sock_slab:
2818 req_prot_cleanup(prot->rsk_prot);
2820 kmem_cache_destroy(prot->slab);
2825 EXPORT_SYMBOL(proto_register);
2827 void proto_unregister(struct proto *prot)
2829 mutex_lock(&proto_list_mutex);
2830 release_proto_idx(prot);
2831 list_del(&prot->node);
2832 mutex_unlock(&proto_list_mutex);
2834 if (prot->slab != NULL) {
2835 kmem_cache_destroy(prot->slab);
2839 req_prot_cleanup(prot->rsk_prot);
2841 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2842 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2843 kfree(prot->twsk_prot->twsk_slab_name);
2844 prot->twsk_prot->twsk_slab = NULL;
2847 EXPORT_SYMBOL(proto_unregister);
2849 #ifdef CONFIG_PROC_FS
2850 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2851 __acquires(proto_list_mutex)
2853 mutex_lock(&proto_list_mutex);
2854 return seq_list_start_head(&proto_list, *pos);
2857 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2859 return seq_list_next(v, &proto_list, pos);
2862 static void proto_seq_stop(struct seq_file *seq, void *v)
2863 __releases(proto_list_mutex)
2865 mutex_unlock(&proto_list_mutex);
2868 static char proto_method_implemented(const void *method)
2870 return method == NULL ? 'n' : 'y';
2872 static long sock_prot_memory_allocated(struct proto *proto)
2874 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2877 static char *sock_prot_memory_pressure(struct proto *proto)
2879 return proto->memory_pressure != NULL ?
2880 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2883 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2886 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2887 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2890 sock_prot_inuse_get(seq_file_net(seq), proto),
2891 sock_prot_memory_allocated(proto),
2892 sock_prot_memory_pressure(proto),
2894 proto->slab == NULL ? "no" : "yes",
2895 module_name(proto->owner),
2896 proto_method_implemented(proto->close),
2897 proto_method_implemented(proto->connect),
2898 proto_method_implemented(proto->disconnect),
2899 proto_method_implemented(proto->accept),
2900 proto_method_implemented(proto->ioctl),
2901 proto_method_implemented(proto->init),
2902 proto_method_implemented(proto->destroy),
2903 proto_method_implemented(proto->shutdown),
2904 proto_method_implemented(proto->setsockopt),
2905 proto_method_implemented(proto->getsockopt),
2906 proto_method_implemented(proto->sendmsg),
2907 proto_method_implemented(proto->recvmsg),
2908 proto_method_implemented(proto->sendpage),
2909 proto_method_implemented(proto->bind),
2910 proto_method_implemented(proto->backlog_rcv),
2911 proto_method_implemented(proto->hash),
2912 proto_method_implemented(proto->unhash),
2913 proto_method_implemented(proto->get_port),
2914 proto_method_implemented(proto->enter_memory_pressure));
2917 static int proto_seq_show(struct seq_file *seq, void *v)
2919 if (v == &proto_list)
2920 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2929 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2931 proto_seq_printf(seq, list_entry(v, struct proto, node));
2935 static const struct seq_operations proto_seq_ops = {
2936 .start = proto_seq_start,
2937 .next = proto_seq_next,
2938 .stop = proto_seq_stop,
2939 .show = proto_seq_show,
2942 static int proto_seq_open(struct inode *inode, struct file *file)
2944 return seq_open_net(inode, file, &proto_seq_ops,
2945 sizeof(struct seq_net_private));
2948 static const struct file_operations proto_seq_fops = {
2949 .owner = THIS_MODULE,
2950 .open = proto_seq_open,
2952 .llseek = seq_lseek,
2953 .release = seq_release_net,
2956 static __net_init int proto_init_net(struct net *net)
2958 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2964 static __net_exit void proto_exit_net(struct net *net)
2966 remove_proc_entry("protocols", net->proc_net);
2970 static __net_initdata struct pernet_operations proto_net_ops = {
2971 .init = proto_init_net,
2972 .exit = proto_exit_net,
2975 static int __init proto_init(void)
2977 return register_pernet_subsys(&proto_net_ops);
2980 subsys_initcall(proto_init);
2982 #endif /* PROC_FS */