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);
657 * This is meant for all protocols to use and covers goings on
658 * at the socket level. Everything here is generic.
661 int sock_setsockopt(struct socket *sock, int level, int optname,
662 char __user *optval, unsigned int optlen)
664 struct sock *sk = sock->sk;
671 * Options without arguments
674 if (optname == SO_BINDTODEVICE)
675 return sock_setbindtodevice(sk, optval, optlen);
677 if (optlen < sizeof(int))
680 if (get_user(val, (int __user *)optval))
683 valbool = val ? 1 : 0;
689 if (val && !capable(CAP_NET_ADMIN))
692 sock_valbool_flag(sk, SOCK_DBG, valbool);
695 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
698 sk->sk_reuseport = valbool;
707 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
710 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
713 /* Don't error on this BSD doesn't and if you think
714 * about it this is right. Otherwise apps have to
715 * play 'guess the biggest size' games. RCVBUF/SNDBUF
716 * are treated in BSD as hints
718 val = min_t(u32, val, sysctl_wmem_max);
720 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
721 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
722 /* Wake up sending tasks if we upped the value. */
723 sk->sk_write_space(sk);
727 if (!capable(CAP_NET_ADMIN)) {
734 /* Don't error on this BSD doesn't and if you think
735 * about it this is right. Otherwise apps have to
736 * play 'guess the biggest size' games. RCVBUF/SNDBUF
737 * are treated in BSD as hints
739 val = min_t(u32, val, sysctl_rmem_max);
741 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
743 * We double it on the way in to account for
744 * "struct sk_buff" etc. overhead. Applications
745 * assume that the SO_RCVBUF setting they make will
746 * allow that much actual data to be received on that
749 * Applications are unaware that "struct sk_buff" and
750 * other overheads allocate from the receive buffer
751 * during socket buffer allocation.
753 * And after considering the possible alternatives,
754 * returning the value we actually used in getsockopt
755 * is the most desirable behavior.
757 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
761 if (!capable(CAP_NET_ADMIN)) {
769 if (sk->sk_protocol == IPPROTO_TCP &&
770 sk->sk_type == SOCK_STREAM)
771 tcp_set_keepalive(sk, valbool);
773 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
777 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
781 sk->sk_no_check_tx = valbool;
785 if ((val >= 0 && val <= 6) ||
786 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
787 sk->sk_priority = val;
793 if (optlen < sizeof(ling)) {
794 ret = -EINVAL; /* 1003.1g */
797 if (copy_from_user(&ling, optval, sizeof(ling))) {
802 sock_reset_flag(sk, SOCK_LINGER);
804 #if (BITS_PER_LONG == 32)
805 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
806 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
809 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
810 sock_set_flag(sk, SOCK_LINGER);
815 sock_warn_obsolete_bsdism("setsockopt");
820 set_bit(SOCK_PASSCRED, &sock->flags);
822 clear_bit(SOCK_PASSCRED, &sock->flags);
828 if (optname == SO_TIMESTAMP)
829 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
831 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
832 sock_set_flag(sk, SOCK_RCVTSTAMP);
833 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
835 sock_reset_flag(sk, SOCK_RCVTSTAMP);
836 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
840 case SO_TIMESTAMPING:
841 if (val & ~SOF_TIMESTAMPING_MASK) {
846 if (val & SOF_TIMESTAMPING_OPT_ID &&
847 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
848 if (sk->sk_protocol == IPPROTO_TCP) {
849 if (sk->sk_state != TCP_ESTABLISHED) {
853 sk->sk_tskey = tcp_sk(sk)->snd_una;
858 sk->sk_tsflags = val;
859 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
860 sock_enable_timestamp(sk,
861 SOCK_TIMESTAMPING_RX_SOFTWARE);
863 sock_disable_timestamp(sk,
864 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
870 sk->sk_rcvlowat = val ? : 1;
874 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
878 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
881 case SO_ATTACH_FILTER:
883 if (optlen == sizeof(struct sock_fprog)) {
884 struct sock_fprog fprog;
887 if (copy_from_user(&fprog, optval, sizeof(fprog)))
890 ret = sk_attach_filter(&fprog, sk);
896 if (optlen == sizeof(u32)) {
900 if (copy_from_user(&ufd, optval, sizeof(ufd)))
903 ret = sk_attach_bpf(ufd, sk);
907 case SO_DETACH_FILTER:
908 ret = sk_detach_filter(sk);
912 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
915 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
920 set_bit(SOCK_PASSSEC, &sock->flags);
922 clear_bit(SOCK_PASSSEC, &sock->flags);
925 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
931 /* We implement the SO_SNDLOWAT etc to
932 not be settable (1003.1g 5.3) */
934 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
938 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
942 if (sock->ops->set_peek_off)
943 ret = sock->ops->set_peek_off(sk, val);
949 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
952 case SO_SELECT_ERR_QUEUE:
953 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
956 #ifdef CONFIG_NET_RX_BUSY_POLL
958 /* allow unprivileged users to decrease the value */
959 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
965 sk->sk_ll_usec = val;
970 case SO_MAX_PACING_RATE:
971 sk->sk_max_pacing_rate = val;
972 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
973 sk->sk_max_pacing_rate);
983 EXPORT_SYMBOL(sock_setsockopt);
986 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
989 ucred->pid = pid_vnr(pid);
990 ucred->uid = ucred->gid = -1;
992 struct user_namespace *current_ns = current_user_ns();
994 ucred->uid = from_kuid_munged(current_ns, cred->euid);
995 ucred->gid = from_kgid_munged(current_ns, cred->egid);
999 int sock_getsockopt(struct socket *sock, int level, int optname,
1000 char __user *optval, int __user *optlen)
1002 struct sock *sk = sock->sk;
1010 int lv = sizeof(int);
1013 if (get_user(len, optlen))
1018 memset(&v, 0, sizeof(v));
1022 v.val = sock_flag(sk, SOCK_DBG);
1026 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1030 v.val = sock_flag(sk, SOCK_BROADCAST);
1034 v.val = sk->sk_sndbuf;
1038 v.val = sk->sk_rcvbuf;
1042 v.val = sk->sk_reuse;
1046 v.val = sk->sk_reuseport;
1050 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1054 v.val = sk->sk_type;
1058 v.val = sk->sk_protocol;
1062 v.val = sk->sk_family;
1066 v.val = -sock_error(sk);
1068 v.val = xchg(&sk->sk_err_soft, 0);
1072 v.val = sock_flag(sk, SOCK_URGINLINE);
1076 v.val = sk->sk_no_check_tx;
1080 v.val = sk->sk_priority;
1084 lv = sizeof(v.ling);
1085 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1086 v.ling.l_linger = sk->sk_lingertime / HZ;
1090 sock_warn_obsolete_bsdism("getsockopt");
1094 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1095 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1098 case SO_TIMESTAMPNS:
1099 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1102 case SO_TIMESTAMPING:
1103 v.val = sk->sk_tsflags;
1107 lv = sizeof(struct timeval);
1108 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1112 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1113 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1118 lv = sizeof(struct timeval);
1119 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1123 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1124 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1129 v.val = sk->sk_rcvlowat;
1137 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1142 struct ucred peercred;
1143 if (len > sizeof(peercred))
1144 len = sizeof(peercred);
1145 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1146 if (copy_to_user(optval, &peercred, len))
1155 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1159 if (copy_to_user(optval, address, len))
1164 /* Dubious BSD thing... Probably nobody even uses it, but
1165 * the UNIX standard wants it for whatever reason... -DaveM
1168 v.val = sk->sk_state == TCP_LISTEN;
1172 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1176 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1179 v.val = sk->sk_mark;
1183 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1186 case SO_WIFI_STATUS:
1187 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1191 if (!sock->ops->set_peek_off)
1194 v.val = sk->sk_peek_off;
1197 v.val = sock_flag(sk, SOCK_NOFCS);
1200 case SO_BINDTODEVICE:
1201 return sock_getbindtodevice(sk, optval, optlen, len);
1204 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1210 case SO_LOCK_FILTER:
1211 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1214 case SO_BPF_EXTENSIONS:
1215 v.val = bpf_tell_extensions();
1218 case SO_SELECT_ERR_QUEUE:
1219 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1222 #ifdef CONFIG_NET_RX_BUSY_POLL
1224 v.val = sk->sk_ll_usec;
1228 case SO_MAX_PACING_RATE:
1229 v.val = sk->sk_max_pacing_rate;
1232 case SO_INCOMING_CPU:
1233 v.val = sk->sk_incoming_cpu;
1237 return -ENOPROTOOPT;
1242 if (copy_to_user(optval, &v, len))
1245 if (put_user(len, optlen))
1251 * Initialize an sk_lock.
1253 * (We also register the sk_lock with the lock validator.)
1255 static inline void sock_lock_init(struct sock *sk)
1257 sock_lock_init_class_and_name(sk,
1258 af_family_slock_key_strings[sk->sk_family],
1259 af_family_slock_keys + sk->sk_family,
1260 af_family_key_strings[sk->sk_family],
1261 af_family_keys + sk->sk_family);
1265 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1266 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1267 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1269 static void sock_copy(struct sock *nsk, const struct sock *osk)
1271 #ifdef CONFIG_SECURITY_NETWORK
1272 void *sptr = nsk->sk_security;
1274 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1276 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1277 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1279 #ifdef CONFIG_SECURITY_NETWORK
1280 nsk->sk_security = sptr;
1281 security_sk_clone(osk, nsk);
1285 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1287 unsigned long nulls1, nulls2;
1289 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1290 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1291 if (nulls1 > nulls2)
1292 swap(nulls1, nulls2);
1295 memset((char *)sk, 0, nulls1);
1296 memset((char *)sk + nulls1 + sizeof(void *), 0,
1297 nulls2 - nulls1 - sizeof(void *));
1298 memset((char *)sk + nulls2 + sizeof(void *), 0,
1299 size - nulls2 - sizeof(void *));
1301 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1303 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1307 struct kmem_cache *slab;
1311 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1314 if (priority & __GFP_ZERO) {
1316 prot->clear_sk(sk, prot->obj_size);
1318 sk_prot_clear_nulls(sk, prot->obj_size);
1321 sk = kmalloc(prot->obj_size, priority);
1324 kmemcheck_annotate_bitfield(sk, flags);
1326 if (security_sk_alloc(sk, family, priority))
1329 if (!try_module_get(prot->owner))
1331 sk_tx_queue_clear(sk);
1337 security_sk_free(sk);
1340 kmem_cache_free(slab, sk);
1346 static void sk_prot_free(struct proto *prot, struct sock *sk)
1348 struct kmem_cache *slab;
1349 struct module *owner;
1351 owner = prot->owner;
1354 security_sk_free(sk);
1356 kmem_cache_free(slab, sk);
1362 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1363 void sock_update_netprioidx(struct sock *sk)
1368 sk->sk_cgrp_prioidx = task_netprioidx(current);
1370 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1374 * sk_alloc - All socket objects are allocated here
1375 * @net: the applicable net namespace
1376 * @family: protocol family
1377 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1378 * @prot: struct proto associated with this new sock instance
1380 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1385 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1387 sk->sk_family = family;
1389 * See comment in struct sock definition to understand
1390 * why we need sk_prot_creator -acme
1392 sk->sk_prot = sk->sk_prot_creator = prot;
1394 sock_net_set(sk, get_net(net));
1395 atomic_set(&sk->sk_wmem_alloc, 1);
1397 sock_update_classid(sk);
1398 sock_update_netprioidx(sk);
1403 EXPORT_SYMBOL(sk_alloc);
1405 static void __sk_free(struct sock *sk)
1407 struct sk_filter *filter;
1409 if (sk->sk_destruct)
1410 sk->sk_destruct(sk);
1412 filter = rcu_dereference_check(sk->sk_filter,
1413 atomic_read(&sk->sk_wmem_alloc) == 0);
1415 sk_filter_uncharge(sk, filter);
1416 RCU_INIT_POINTER(sk->sk_filter, NULL);
1419 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1421 if (atomic_read(&sk->sk_omem_alloc))
1422 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1423 __func__, atomic_read(&sk->sk_omem_alloc));
1425 if (sk->sk_peer_cred)
1426 put_cred(sk->sk_peer_cred);
1427 put_pid(sk->sk_peer_pid);
1428 put_net(sock_net(sk));
1429 sk_prot_free(sk->sk_prot_creator, sk);
1432 void sk_free(struct sock *sk)
1435 * We subtract one from sk_wmem_alloc and can know if
1436 * some packets are still in some tx queue.
1437 * If not null, sock_wfree() will call __sk_free(sk) later
1439 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1442 EXPORT_SYMBOL(sk_free);
1445 * Last sock_put should drop reference to sk->sk_net. It has already
1446 * been dropped in sk_change_net. Taking reference to stopping namespace
1448 * Take reference to a socket to remove it from hash _alive_ and after that
1449 * destroy it in the context of init_net.
1451 void sk_release_kernel(struct sock *sk)
1453 if (sk == NULL || sk->sk_socket == NULL)
1457 sock_release(sk->sk_socket);
1458 release_net(sock_net(sk));
1459 sock_net_set(sk, get_net(&init_net));
1462 EXPORT_SYMBOL(sk_release_kernel);
1464 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1466 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1467 sock_update_memcg(newsk);
1471 * sk_clone_lock - clone a socket, and lock its clone
1472 * @sk: the socket to clone
1473 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1475 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1477 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1480 bool is_charged = true;
1482 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1483 if (newsk != NULL) {
1484 struct sk_filter *filter;
1486 sock_copy(newsk, sk);
1489 get_net(sock_net(newsk));
1490 sk_node_init(&newsk->sk_node);
1491 sock_lock_init(newsk);
1492 bh_lock_sock(newsk);
1493 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1494 newsk->sk_backlog.len = 0;
1496 atomic_set(&newsk->sk_rmem_alloc, 0);
1498 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1500 atomic_set(&newsk->sk_wmem_alloc, 1);
1501 atomic_set(&newsk->sk_omem_alloc, 0);
1502 skb_queue_head_init(&newsk->sk_receive_queue);
1503 skb_queue_head_init(&newsk->sk_write_queue);
1505 spin_lock_init(&newsk->sk_dst_lock);
1506 rwlock_init(&newsk->sk_callback_lock);
1507 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1508 af_callback_keys + newsk->sk_family,
1509 af_family_clock_key_strings[newsk->sk_family]);
1511 newsk->sk_dst_cache = NULL;
1512 newsk->sk_wmem_queued = 0;
1513 newsk->sk_forward_alloc = 0;
1514 newsk->sk_send_head = NULL;
1515 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1517 sock_reset_flag(newsk, SOCK_DONE);
1518 skb_queue_head_init(&newsk->sk_error_queue);
1520 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1522 /* though it's an empty new sock, the charging may fail
1523 * if sysctl_optmem_max was changed between creation of
1524 * original socket and cloning
1526 is_charged = sk_filter_charge(newsk, filter);
1528 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1529 /* It is still raw copy of parent, so invalidate
1530 * destructor and make plain sk_free() */
1531 newsk->sk_destruct = NULL;
1532 bh_unlock_sock(newsk);
1539 newsk->sk_priority = 0;
1540 newsk->sk_incoming_cpu = raw_smp_processor_id();
1542 * Before updating sk_refcnt, we must commit prior changes to memory
1543 * (Documentation/RCU/rculist_nulls.txt for details)
1546 atomic_set(&newsk->sk_refcnt, 2);
1549 * Increment the counter in the same struct proto as the master
1550 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1551 * is the same as sk->sk_prot->socks, as this field was copied
1554 * This _changes_ the previous behaviour, where
1555 * tcp_create_openreq_child always was incrementing the
1556 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1557 * to be taken into account in all callers. -acme
1559 sk_refcnt_debug_inc(newsk);
1560 sk_set_socket(newsk, NULL);
1561 newsk->sk_wq = NULL;
1563 sk_update_clone(sk, newsk);
1565 if (newsk->sk_prot->sockets_allocated)
1566 sk_sockets_allocated_inc(newsk);
1568 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1569 net_enable_timestamp();
1574 EXPORT_SYMBOL_GPL(sk_clone_lock);
1576 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1578 __sk_dst_set(sk, dst);
1579 sk->sk_route_caps = dst->dev->features;
1580 if (sk->sk_route_caps & NETIF_F_GSO)
1581 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1582 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1583 if (sk_can_gso(sk)) {
1584 if (dst->header_len) {
1585 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1587 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1588 sk->sk_gso_max_size = dst->dev->gso_max_size;
1589 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1593 EXPORT_SYMBOL_GPL(sk_setup_caps);
1596 * Simple resource managers for sockets.
1601 * Write buffer destructor automatically called from kfree_skb.
1603 void sock_wfree(struct sk_buff *skb)
1605 struct sock *sk = skb->sk;
1606 unsigned int len = skb->truesize;
1608 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1610 * Keep a reference on sk_wmem_alloc, this will be released
1611 * after sk_write_space() call
1613 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1614 sk->sk_write_space(sk);
1618 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1619 * could not do because of in-flight packets
1621 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1624 EXPORT_SYMBOL(sock_wfree);
1626 void skb_orphan_partial(struct sk_buff *skb)
1628 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1629 * so we do not completely orphan skb, but transfert all
1630 * accounted bytes but one, to avoid unexpected reorders.
1632 if (skb->destructor == sock_wfree
1634 || skb->destructor == tcp_wfree
1637 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1643 EXPORT_SYMBOL(skb_orphan_partial);
1646 * Read buffer destructor automatically called from kfree_skb.
1648 void sock_rfree(struct sk_buff *skb)
1650 struct sock *sk = skb->sk;
1651 unsigned int len = skb->truesize;
1653 atomic_sub(len, &sk->sk_rmem_alloc);
1654 sk_mem_uncharge(sk, len);
1656 EXPORT_SYMBOL(sock_rfree);
1658 void sock_efree(struct sk_buff *skb)
1662 EXPORT_SYMBOL(sock_efree);
1665 void sock_edemux(struct sk_buff *skb)
1667 struct sock *sk = skb->sk;
1669 if (sk->sk_state == TCP_TIME_WAIT)
1670 inet_twsk_put(inet_twsk(sk));
1674 EXPORT_SYMBOL(sock_edemux);
1677 kuid_t sock_i_uid(struct sock *sk)
1681 read_lock_bh(&sk->sk_callback_lock);
1682 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1683 read_unlock_bh(&sk->sk_callback_lock);
1686 EXPORT_SYMBOL(sock_i_uid);
1688 unsigned long sock_i_ino(struct sock *sk)
1692 read_lock_bh(&sk->sk_callback_lock);
1693 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1694 read_unlock_bh(&sk->sk_callback_lock);
1697 EXPORT_SYMBOL(sock_i_ino);
1700 * Allocate a skb from the socket's send buffer.
1702 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1705 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1706 struct sk_buff *skb = alloc_skb(size, priority);
1708 skb_set_owner_w(skb, sk);
1714 EXPORT_SYMBOL(sock_wmalloc);
1717 * Allocate a memory block from the socket's option memory buffer.
1719 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1721 if ((unsigned int)size <= sysctl_optmem_max &&
1722 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1724 /* First do the add, to avoid the race if kmalloc
1727 atomic_add(size, &sk->sk_omem_alloc);
1728 mem = kmalloc(size, priority);
1731 atomic_sub(size, &sk->sk_omem_alloc);
1735 EXPORT_SYMBOL(sock_kmalloc);
1737 /* Free an option memory block. Note, we actually want the inline
1738 * here as this allows gcc to detect the nullify and fold away the
1739 * condition entirely.
1741 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1744 if (WARN_ON_ONCE(!mem))
1750 atomic_sub(size, &sk->sk_omem_alloc);
1753 void sock_kfree_s(struct sock *sk, void *mem, int size)
1755 __sock_kfree_s(sk, mem, size, false);
1757 EXPORT_SYMBOL(sock_kfree_s);
1759 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1761 __sock_kfree_s(sk, mem, size, true);
1763 EXPORT_SYMBOL(sock_kzfree_s);
1765 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1766 I think, these locks should be removed for datagram sockets.
1768 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1772 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1776 if (signal_pending(current))
1778 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1779 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1780 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1782 if (sk->sk_shutdown & SEND_SHUTDOWN)
1786 timeo = schedule_timeout(timeo);
1788 finish_wait(sk_sleep(sk), &wait);
1794 * Generic send/receive buffer handlers
1797 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1798 unsigned long data_len, int noblock,
1799 int *errcode, int max_page_order)
1801 struct sk_buff *skb;
1805 timeo = sock_sndtimeo(sk, noblock);
1807 err = sock_error(sk);
1812 if (sk->sk_shutdown & SEND_SHUTDOWN)
1815 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1818 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1819 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1823 if (signal_pending(current))
1825 timeo = sock_wait_for_wmem(sk, timeo);
1827 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1828 errcode, sk->sk_allocation);
1830 skb_set_owner_w(skb, sk);
1834 err = sock_intr_errno(timeo);
1839 EXPORT_SYMBOL(sock_alloc_send_pskb);
1841 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1842 int noblock, int *errcode)
1844 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1846 EXPORT_SYMBOL(sock_alloc_send_skb);
1848 /* On 32bit arches, an skb frag is limited to 2^15 */
1849 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1852 * skb_page_frag_refill - check that a page_frag contains enough room
1853 * @sz: minimum size of the fragment we want to get
1854 * @pfrag: pointer to page_frag
1855 * @gfp: priority for memory allocation
1857 * Note: While this allocator tries to use high order pages, there is
1858 * no guarantee that allocations succeed. Therefore, @sz MUST be
1859 * less or equal than PAGE_SIZE.
1861 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1864 if (atomic_read(&pfrag->page->_count) == 1) {
1868 if (pfrag->offset + sz <= pfrag->size)
1870 put_page(pfrag->page);
1874 if (SKB_FRAG_PAGE_ORDER) {
1875 pfrag->page = alloc_pages(gfp | __GFP_COMP |
1876 __GFP_NOWARN | __GFP_NORETRY,
1877 SKB_FRAG_PAGE_ORDER);
1878 if (likely(pfrag->page)) {
1879 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1883 pfrag->page = alloc_page(gfp);
1884 if (likely(pfrag->page)) {
1885 pfrag->size = PAGE_SIZE;
1890 EXPORT_SYMBOL(skb_page_frag_refill);
1892 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1894 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1897 sk_enter_memory_pressure(sk);
1898 sk_stream_moderate_sndbuf(sk);
1901 EXPORT_SYMBOL(sk_page_frag_refill);
1903 static void __lock_sock(struct sock *sk)
1904 __releases(&sk->sk_lock.slock)
1905 __acquires(&sk->sk_lock.slock)
1910 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1911 TASK_UNINTERRUPTIBLE);
1912 spin_unlock_bh(&sk->sk_lock.slock);
1914 spin_lock_bh(&sk->sk_lock.slock);
1915 if (!sock_owned_by_user(sk))
1918 finish_wait(&sk->sk_lock.wq, &wait);
1921 static void __release_sock(struct sock *sk)
1922 __releases(&sk->sk_lock.slock)
1923 __acquires(&sk->sk_lock.slock)
1925 struct sk_buff *skb = sk->sk_backlog.head;
1928 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1932 struct sk_buff *next = skb->next;
1935 WARN_ON_ONCE(skb_dst_is_noref(skb));
1937 sk_backlog_rcv(sk, skb);
1940 * We are in process context here with softirqs
1941 * disabled, use cond_resched_softirq() to preempt.
1942 * This is safe to do because we've taken the backlog
1945 cond_resched_softirq();
1948 } while (skb != NULL);
1951 } while ((skb = sk->sk_backlog.head) != NULL);
1954 * Doing the zeroing here guarantee we can not loop forever
1955 * while a wild producer attempts to flood us.
1957 sk->sk_backlog.len = 0;
1961 * sk_wait_data - wait for data to arrive at sk_receive_queue
1962 * @sk: sock to wait on
1963 * @timeo: for how long
1965 * Now socket state including sk->sk_err is changed only under lock,
1966 * hence we may omit checks after joining wait queue.
1967 * We check receive queue before schedule() only as optimization;
1968 * it is very likely that release_sock() added new data.
1970 int sk_wait_data(struct sock *sk, long *timeo)
1975 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1976 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1977 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1978 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1979 finish_wait(sk_sleep(sk), &wait);
1982 EXPORT_SYMBOL(sk_wait_data);
1985 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1987 * @size: memory size to allocate
1988 * @kind: allocation type
1990 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1991 * rmem allocation. This function assumes that protocols which have
1992 * memory_pressure use sk_wmem_queued as write buffer accounting.
1994 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1996 struct proto *prot = sk->sk_prot;
1997 int amt = sk_mem_pages(size);
1999 int parent_status = UNDER_LIMIT;
2001 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2003 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2006 if (parent_status == UNDER_LIMIT &&
2007 allocated <= sk_prot_mem_limits(sk, 0)) {
2008 sk_leave_memory_pressure(sk);
2012 /* Under pressure. (we or our parents) */
2013 if ((parent_status > SOFT_LIMIT) ||
2014 allocated > sk_prot_mem_limits(sk, 1))
2015 sk_enter_memory_pressure(sk);
2017 /* Over hard limit (we or our parents) */
2018 if ((parent_status == OVER_LIMIT) ||
2019 (allocated > sk_prot_mem_limits(sk, 2)))
2020 goto suppress_allocation;
2022 /* guarantee minimum buffer size under pressure */
2023 if (kind == SK_MEM_RECV) {
2024 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2027 } else { /* SK_MEM_SEND */
2028 if (sk->sk_type == SOCK_STREAM) {
2029 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2031 } else if (atomic_read(&sk->sk_wmem_alloc) <
2032 prot->sysctl_wmem[0])
2036 if (sk_has_memory_pressure(sk)) {
2039 if (!sk_under_memory_pressure(sk))
2041 alloc = sk_sockets_allocated_read_positive(sk);
2042 if (sk_prot_mem_limits(sk, 2) > alloc *
2043 sk_mem_pages(sk->sk_wmem_queued +
2044 atomic_read(&sk->sk_rmem_alloc) +
2045 sk->sk_forward_alloc))
2049 suppress_allocation:
2051 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2052 sk_stream_moderate_sndbuf(sk);
2054 /* Fail only if socket is _under_ its sndbuf.
2055 * In this case we cannot block, so that we have to fail.
2057 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2061 trace_sock_exceed_buf_limit(sk, prot, allocated);
2063 /* Alas. Undo changes. */
2064 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2066 sk_memory_allocated_sub(sk, amt);
2070 EXPORT_SYMBOL(__sk_mem_schedule);
2073 * __sk_reclaim - reclaim memory_allocated
2076 void __sk_mem_reclaim(struct sock *sk)
2078 sk_memory_allocated_sub(sk,
2079 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2080 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2082 if (sk_under_memory_pressure(sk) &&
2083 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2084 sk_leave_memory_pressure(sk);
2086 EXPORT_SYMBOL(__sk_mem_reclaim);
2090 * Set of default routines for initialising struct proto_ops when
2091 * the protocol does not support a particular function. In certain
2092 * cases where it makes no sense for a protocol to have a "do nothing"
2093 * function, some default processing is provided.
2096 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2100 EXPORT_SYMBOL(sock_no_bind);
2102 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2107 EXPORT_SYMBOL(sock_no_connect);
2109 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2113 EXPORT_SYMBOL(sock_no_socketpair);
2115 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2119 EXPORT_SYMBOL(sock_no_accept);
2121 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2126 EXPORT_SYMBOL(sock_no_getname);
2128 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2132 EXPORT_SYMBOL(sock_no_poll);
2134 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2138 EXPORT_SYMBOL(sock_no_ioctl);
2140 int sock_no_listen(struct socket *sock, int backlog)
2144 EXPORT_SYMBOL(sock_no_listen);
2146 int sock_no_shutdown(struct socket *sock, int how)
2150 EXPORT_SYMBOL(sock_no_shutdown);
2152 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2153 char __user *optval, unsigned int optlen)
2157 EXPORT_SYMBOL(sock_no_setsockopt);
2159 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2160 char __user *optval, int __user *optlen)
2164 EXPORT_SYMBOL(sock_no_getsockopt);
2166 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2170 EXPORT_SYMBOL(sock_no_sendmsg);
2172 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2177 EXPORT_SYMBOL(sock_no_recvmsg);
2179 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2181 /* Mirror missing mmap method error code */
2184 EXPORT_SYMBOL(sock_no_mmap);
2186 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2189 struct msghdr msg = {.msg_flags = flags};
2191 char *kaddr = kmap(page);
2192 iov.iov_base = kaddr + offset;
2194 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2198 EXPORT_SYMBOL(sock_no_sendpage);
2201 * Default Socket Callbacks
2204 static void sock_def_wakeup(struct sock *sk)
2206 struct socket_wq *wq;
2209 wq = rcu_dereference(sk->sk_wq);
2210 if (wq_has_sleeper(wq))
2211 wake_up_interruptible_all(&wq->wait);
2215 static void sock_def_error_report(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_poll(&wq->wait, POLLERR);
2223 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2227 static void sock_def_readable(struct sock *sk)
2229 struct socket_wq *wq;
2232 wq = rcu_dereference(sk->sk_wq);
2233 if (wq_has_sleeper(wq))
2234 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2235 POLLRDNORM | POLLRDBAND);
2236 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2240 static void sock_def_write_space(struct sock *sk)
2242 struct socket_wq *wq;
2246 /* Do not wake up a writer until he can make "significant"
2249 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2250 wq = rcu_dereference(sk->sk_wq);
2251 if (wq_has_sleeper(wq))
2252 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2253 POLLWRNORM | POLLWRBAND);
2255 /* Should agree with poll, otherwise some programs break */
2256 if (sock_writeable(sk))
2257 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2263 static void sock_def_destruct(struct sock *sk)
2265 kfree(sk->sk_protinfo);
2268 void sk_send_sigurg(struct sock *sk)
2270 if (sk->sk_socket && sk->sk_socket->file)
2271 if (send_sigurg(&sk->sk_socket->file->f_owner))
2272 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2274 EXPORT_SYMBOL(sk_send_sigurg);
2276 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2277 unsigned long expires)
2279 if (!mod_timer(timer, expires))
2282 EXPORT_SYMBOL(sk_reset_timer);
2284 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2286 if (del_timer(timer))
2289 EXPORT_SYMBOL(sk_stop_timer);
2291 void sock_init_data(struct socket *sock, struct sock *sk)
2293 skb_queue_head_init(&sk->sk_receive_queue);
2294 skb_queue_head_init(&sk->sk_write_queue);
2295 skb_queue_head_init(&sk->sk_error_queue);
2297 sk->sk_send_head = NULL;
2299 init_timer(&sk->sk_timer);
2301 sk->sk_allocation = GFP_KERNEL;
2302 sk->sk_rcvbuf = sysctl_rmem_default;
2303 sk->sk_sndbuf = sysctl_wmem_default;
2304 sk->sk_state = TCP_CLOSE;
2305 sk_set_socket(sk, sock);
2307 sock_set_flag(sk, SOCK_ZAPPED);
2310 sk->sk_type = sock->type;
2311 sk->sk_wq = sock->wq;
2316 spin_lock_init(&sk->sk_dst_lock);
2317 rwlock_init(&sk->sk_callback_lock);
2318 lockdep_set_class_and_name(&sk->sk_callback_lock,
2319 af_callback_keys + sk->sk_family,
2320 af_family_clock_key_strings[sk->sk_family]);
2322 sk->sk_state_change = sock_def_wakeup;
2323 sk->sk_data_ready = sock_def_readable;
2324 sk->sk_write_space = sock_def_write_space;
2325 sk->sk_error_report = sock_def_error_report;
2326 sk->sk_destruct = sock_def_destruct;
2328 sk->sk_frag.page = NULL;
2329 sk->sk_frag.offset = 0;
2330 sk->sk_peek_off = -1;
2332 sk->sk_peer_pid = NULL;
2333 sk->sk_peer_cred = NULL;
2334 sk->sk_write_pending = 0;
2335 sk->sk_rcvlowat = 1;
2336 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2337 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2339 sk->sk_stamp = ktime_set(-1L, 0);
2341 #ifdef CONFIG_NET_RX_BUSY_POLL
2343 sk->sk_ll_usec = sysctl_net_busy_read;
2346 sk->sk_max_pacing_rate = ~0U;
2347 sk->sk_pacing_rate = ~0U;
2349 * Before updating sk_refcnt, we must commit prior changes to memory
2350 * (Documentation/RCU/rculist_nulls.txt for details)
2353 atomic_set(&sk->sk_refcnt, 1);
2354 atomic_set(&sk->sk_drops, 0);
2356 EXPORT_SYMBOL(sock_init_data);
2358 void lock_sock_nested(struct sock *sk, int subclass)
2361 spin_lock_bh(&sk->sk_lock.slock);
2362 if (sk->sk_lock.owned)
2364 sk->sk_lock.owned = 1;
2365 spin_unlock(&sk->sk_lock.slock);
2367 * The sk_lock has mutex_lock() semantics here:
2369 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2372 EXPORT_SYMBOL(lock_sock_nested);
2374 void release_sock(struct sock *sk)
2377 * The sk_lock has mutex_unlock() semantics:
2379 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2381 spin_lock_bh(&sk->sk_lock.slock);
2382 if (sk->sk_backlog.tail)
2385 /* Warning : release_cb() might need to release sk ownership,
2386 * ie call sock_release_ownership(sk) before us.
2388 if (sk->sk_prot->release_cb)
2389 sk->sk_prot->release_cb(sk);
2391 sock_release_ownership(sk);
2392 if (waitqueue_active(&sk->sk_lock.wq))
2393 wake_up(&sk->sk_lock.wq);
2394 spin_unlock_bh(&sk->sk_lock.slock);
2396 EXPORT_SYMBOL(release_sock);
2399 * lock_sock_fast - fast version of lock_sock
2402 * This version should be used for very small section, where process wont block
2403 * return false if fast path is taken
2404 * sk_lock.slock locked, owned = 0, BH disabled
2405 * return true if slow path is taken
2406 * sk_lock.slock unlocked, owned = 1, BH enabled
2408 bool lock_sock_fast(struct sock *sk)
2411 spin_lock_bh(&sk->sk_lock.slock);
2413 if (!sk->sk_lock.owned)
2415 * Note : We must disable BH
2420 sk->sk_lock.owned = 1;
2421 spin_unlock(&sk->sk_lock.slock);
2423 * The sk_lock has mutex_lock() semantics here:
2425 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2429 EXPORT_SYMBOL(lock_sock_fast);
2431 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2434 if (!sock_flag(sk, SOCK_TIMESTAMP))
2435 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2436 tv = ktime_to_timeval(sk->sk_stamp);
2437 if (tv.tv_sec == -1)
2439 if (tv.tv_sec == 0) {
2440 sk->sk_stamp = ktime_get_real();
2441 tv = ktime_to_timeval(sk->sk_stamp);
2443 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2445 EXPORT_SYMBOL(sock_get_timestamp);
2447 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2450 if (!sock_flag(sk, SOCK_TIMESTAMP))
2451 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2452 ts = ktime_to_timespec(sk->sk_stamp);
2453 if (ts.tv_sec == -1)
2455 if (ts.tv_sec == 0) {
2456 sk->sk_stamp = ktime_get_real();
2457 ts = ktime_to_timespec(sk->sk_stamp);
2459 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2461 EXPORT_SYMBOL(sock_get_timestampns);
2463 void sock_enable_timestamp(struct sock *sk, int flag)
2465 if (!sock_flag(sk, flag)) {
2466 unsigned long previous_flags = sk->sk_flags;
2468 sock_set_flag(sk, flag);
2470 * we just set one of the two flags which require net
2471 * time stamping, but time stamping might have been on
2472 * already because of the other one
2474 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2475 net_enable_timestamp();
2479 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2480 int level, int type)
2482 struct sock_exterr_skb *serr;
2483 struct sk_buff *skb;
2487 skb = sock_dequeue_err_skb(sk);
2493 msg->msg_flags |= MSG_TRUNC;
2496 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2500 sock_recv_timestamp(msg, sk, skb);
2502 serr = SKB_EXT_ERR(skb);
2503 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2505 msg->msg_flags |= MSG_ERRQUEUE;
2513 EXPORT_SYMBOL(sock_recv_errqueue);
2516 * Get a socket option on an socket.
2518 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2519 * asynchronous errors should be reported by getsockopt. We assume
2520 * this means if you specify SO_ERROR (otherwise whats the point of it).
2522 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2523 char __user *optval, int __user *optlen)
2525 struct sock *sk = sock->sk;
2527 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2529 EXPORT_SYMBOL(sock_common_getsockopt);
2531 #ifdef CONFIG_COMPAT
2532 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2533 char __user *optval, int __user *optlen)
2535 struct sock *sk = sock->sk;
2537 if (sk->sk_prot->compat_getsockopt != NULL)
2538 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2540 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2542 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2545 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2548 struct sock *sk = sock->sk;
2552 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2553 flags & ~MSG_DONTWAIT, &addr_len);
2555 msg->msg_namelen = addr_len;
2558 EXPORT_SYMBOL(sock_common_recvmsg);
2561 * Set socket options on an inet socket.
2563 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2564 char __user *optval, unsigned int optlen)
2566 struct sock *sk = sock->sk;
2568 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2570 EXPORT_SYMBOL(sock_common_setsockopt);
2572 #ifdef CONFIG_COMPAT
2573 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2574 char __user *optval, unsigned int optlen)
2576 struct sock *sk = sock->sk;
2578 if (sk->sk_prot->compat_setsockopt != NULL)
2579 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2581 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2583 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2586 void sk_common_release(struct sock *sk)
2588 if (sk->sk_prot->destroy)
2589 sk->sk_prot->destroy(sk);
2592 * Observation: when sock_common_release is called, processes have
2593 * no access to socket. But net still has.
2594 * Step one, detach it from networking:
2596 * A. Remove from hash tables.
2599 sk->sk_prot->unhash(sk);
2602 * In this point socket cannot receive new packets, but it is possible
2603 * that some packets are in flight because some CPU runs receiver and
2604 * did hash table lookup before we unhashed socket. They will achieve
2605 * receive queue and will be purged by socket destructor.
2607 * Also we still have packets pending on receive queue and probably,
2608 * our own packets waiting in device queues. sock_destroy will drain
2609 * receive queue, but transmitted packets will delay socket destruction
2610 * until the last reference will be released.
2615 xfrm_sk_free_policy(sk);
2617 sk_refcnt_debug_release(sk);
2619 if (sk->sk_frag.page) {
2620 put_page(sk->sk_frag.page);
2621 sk->sk_frag.page = NULL;
2626 EXPORT_SYMBOL(sk_common_release);
2628 #ifdef CONFIG_PROC_FS
2629 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2631 int val[PROTO_INUSE_NR];
2634 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2636 #ifdef CONFIG_NET_NS
2637 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2639 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2641 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2643 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2645 int cpu, idx = prot->inuse_idx;
2648 for_each_possible_cpu(cpu)
2649 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2651 return res >= 0 ? res : 0;
2653 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2655 static int __net_init sock_inuse_init_net(struct net *net)
2657 net->core.inuse = alloc_percpu(struct prot_inuse);
2658 return net->core.inuse ? 0 : -ENOMEM;
2661 static void __net_exit sock_inuse_exit_net(struct net *net)
2663 free_percpu(net->core.inuse);
2666 static struct pernet_operations net_inuse_ops = {
2667 .init = sock_inuse_init_net,
2668 .exit = sock_inuse_exit_net,
2671 static __init int net_inuse_init(void)
2673 if (register_pernet_subsys(&net_inuse_ops))
2674 panic("Cannot initialize net inuse counters");
2679 core_initcall(net_inuse_init);
2681 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2683 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2685 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2687 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2689 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2691 int cpu, idx = prot->inuse_idx;
2694 for_each_possible_cpu(cpu)
2695 res += per_cpu(prot_inuse, cpu).val[idx];
2697 return res >= 0 ? res : 0;
2699 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2702 static void assign_proto_idx(struct proto *prot)
2704 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2706 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2707 pr_err("PROTO_INUSE_NR exhausted\n");
2711 set_bit(prot->inuse_idx, proto_inuse_idx);
2714 static void release_proto_idx(struct proto *prot)
2716 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2717 clear_bit(prot->inuse_idx, proto_inuse_idx);
2720 static inline void assign_proto_idx(struct proto *prot)
2724 static inline void release_proto_idx(struct proto *prot)
2729 int proto_register(struct proto *prot, int alloc_slab)
2732 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2733 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2736 if (prot->slab == NULL) {
2737 pr_crit("%s: Can't create sock SLAB cache!\n",
2742 if (prot->rsk_prot != NULL) {
2743 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2744 if (prot->rsk_prot->slab_name == NULL)
2745 goto out_free_sock_slab;
2747 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2748 prot->rsk_prot->obj_size, 0,
2749 SLAB_HWCACHE_ALIGN, NULL);
2751 if (prot->rsk_prot->slab == NULL) {
2752 pr_crit("%s: Can't create request sock SLAB cache!\n",
2754 goto out_free_request_sock_slab_name;
2758 if (prot->twsk_prot != NULL) {
2759 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2761 if (prot->twsk_prot->twsk_slab_name == NULL)
2762 goto out_free_request_sock_slab;
2764 prot->twsk_prot->twsk_slab =
2765 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2766 prot->twsk_prot->twsk_obj_size,
2768 SLAB_HWCACHE_ALIGN |
2771 if (prot->twsk_prot->twsk_slab == NULL)
2772 goto out_free_timewait_sock_slab_name;
2776 mutex_lock(&proto_list_mutex);
2777 list_add(&prot->node, &proto_list);
2778 assign_proto_idx(prot);
2779 mutex_unlock(&proto_list_mutex);
2782 out_free_timewait_sock_slab_name:
2783 kfree(prot->twsk_prot->twsk_slab_name);
2784 out_free_request_sock_slab:
2785 if (prot->rsk_prot && prot->rsk_prot->slab) {
2786 kmem_cache_destroy(prot->rsk_prot->slab);
2787 prot->rsk_prot->slab = NULL;
2789 out_free_request_sock_slab_name:
2791 kfree(prot->rsk_prot->slab_name);
2793 kmem_cache_destroy(prot->slab);
2798 EXPORT_SYMBOL(proto_register);
2800 void proto_unregister(struct proto *prot)
2802 mutex_lock(&proto_list_mutex);
2803 release_proto_idx(prot);
2804 list_del(&prot->node);
2805 mutex_unlock(&proto_list_mutex);
2807 if (prot->slab != NULL) {
2808 kmem_cache_destroy(prot->slab);
2812 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2813 kmem_cache_destroy(prot->rsk_prot->slab);
2814 kfree(prot->rsk_prot->slab_name);
2815 prot->rsk_prot->slab = NULL;
2818 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2819 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2820 kfree(prot->twsk_prot->twsk_slab_name);
2821 prot->twsk_prot->twsk_slab = NULL;
2824 EXPORT_SYMBOL(proto_unregister);
2826 #ifdef CONFIG_PROC_FS
2827 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2828 __acquires(proto_list_mutex)
2830 mutex_lock(&proto_list_mutex);
2831 return seq_list_start_head(&proto_list, *pos);
2834 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2836 return seq_list_next(v, &proto_list, pos);
2839 static void proto_seq_stop(struct seq_file *seq, void *v)
2840 __releases(proto_list_mutex)
2842 mutex_unlock(&proto_list_mutex);
2845 static char proto_method_implemented(const void *method)
2847 return method == NULL ? 'n' : 'y';
2849 static long sock_prot_memory_allocated(struct proto *proto)
2851 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2854 static char *sock_prot_memory_pressure(struct proto *proto)
2856 return proto->memory_pressure != NULL ?
2857 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2860 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2863 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2864 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2867 sock_prot_inuse_get(seq_file_net(seq), proto),
2868 sock_prot_memory_allocated(proto),
2869 sock_prot_memory_pressure(proto),
2871 proto->slab == NULL ? "no" : "yes",
2872 module_name(proto->owner),
2873 proto_method_implemented(proto->close),
2874 proto_method_implemented(proto->connect),
2875 proto_method_implemented(proto->disconnect),
2876 proto_method_implemented(proto->accept),
2877 proto_method_implemented(proto->ioctl),
2878 proto_method_implemented(proto->init),
2879 proto_method_implemented(proto->destroy),
2880 proto_method_implemented(proto->shutdown),
2881 proto_method_implemented(proto->setsockopt),
2882 proto_method_implemented(proto->getsockopt),
2883 proto_method_implemented(proto->sendmsg),
2884 proto_method_implemented(proto->recvmsg),
2885 proto_method_implemented(proto->sendpage),
2886 proto_method_implemented(proto->bind),
2887 proto_method_implemented(proto->backlog_rcv),
2888 proto_method_implemented(proto->hash),
2889 proto_method_implemented(proto->unhash),
2890 proto_method_implemented(proto->get_port),
2891 proto_method_implemented(proto->enter_memory_pressure));
2894 static int proto_seq_show(struct seq_file *seq, void *v)
2896 if (v == &proto_list)
2897 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2906 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2908 proto_seq_printf(seq, list_entry(v, struct proto, node));
2912 static const struct seq_operations proto_seq_ops = {
2913 .start = proto_seq_start,
2914 .next = proto_seq_next,
2915 .stop = proto_seq_stop,
2916 .show = proto_seq_show,
2919 static int proto_seq_open(struct inode *inode, struct file *file)
2921 return seq_open_net(inode, file, &proto_seq_ops,
2922 sizeof(struct seq_net_private));
2925 static const struct file_operations proto_seq_fops = {
2926 .owner = THIS_MODULE,
2927 .open = proto_seq_open,
2929 .llseek = seq_lseek,
2930 .release = seq_release_net,
2933 static __net_init int proto_init_net(struct net *net)
2935 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2941 static __net_exit void proto_exit_net(struct net *net)
2943 remove_proc_entry("protocols", net->proc_net);
2947 static __net_initdata struct pernet_operations proto_net_ops = {
2948 .init = proto_init_net,
2949 .exit = proto_exit_net,
2952 static int __init proto_init(void)
2954 return register_pernet_subsys(&proto_net_ops);
2957 subsys_initcall(proto_init);
2959 #endif /* PROC_FS */