2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <asm/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
139 #include <trace/events/sock.h>
142 #include <net/busy_poll.h>
144 static DEFINE_MUTEX(proto_list_mutex);
145 static LIST_HEAD(proto_list);
148 * sk_ns_capable - General socket capability test
149 * @sk: Socket to use a capability on or through
150 * @user_ns: The user namespace of the capability to use
151 * @cap: The capability to use
153 * Test to see if the opener of the socket had when the socket was
154 * created and the current process has the capability @cap in the user
155 * namespace @user_ns.
157 bool sk_ns_capable(const struct sock *sk,
158 struct user_namespace *user_ns, int cap)
160 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
161 ns_capable(user_ns, cap);
163 EXPORT_SYMBOL(sk_ns_capable);
166 * sk_capable - Socket global capability test
167 * @sk: Socket to use a capability on or through
168 * @cap: The global capability to use
170 * Test to see if the opener of the socket had when the socket was
171 * created and the current process has the capability @cap in all user
174 bool sk_capable(const struct sock *sk, int cap)
176 return sk_ns_capable(sk, &init_user_ns, cap);
178 EXPORT_SYMBOL(sk_capable);
181 * sk_net_capable - Network namespace socket capability test
182 * @sk: Socket to use a capability on or through
183 * @cap: The capability to use
185 * Test to see if the opener of the socket had when the socket was created
186 * and the current process has the capability @cap over the network namespace
187 * the socket is a member of.
189 bool sk_net_capable(const struct sock *sk, int cap)
191 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193 EXPORT_SYMBOL(sk_net_capable);
196 * Each address family might have different locking rules, so we have
197 * one slock key per address family:
199 static struct lock_class_key af_family_keys[AF_MAX];
200 static struct lock_class_key af_family_slock_keys[AF_MAX];
203 * Make lock validator output more readable. (we pre-construct these
204 * strings build-time, so that runtime initialization of socket
207 static const char *const af_family_key_strings[AF_MAX+1] = {
208 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
209 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
210 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
211 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
212 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
213 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
214 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
215 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
216 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
217 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
218 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
219 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
220 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
221 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_KCM" ,
222 "sk_lock-AF_QIPCRTR", "sk_lock-AF_MAX"
224 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
225 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
226 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
227 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
228 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
229 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
230 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
231 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
232 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
233 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
234 "slock-27" , "slock-28" , "slock-AF_CAN" ,
235 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
236 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
237 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
238 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" ,
239 "slock-AF_QIPCRTR", "slock-AF_MAX"
241 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
242 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
243 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
244 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
245 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
246 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
247 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
248 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
249 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
250 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
251 "clock-27" , "clock-28" , "clock-AF_CAN" ,
252 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
253 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
254 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
255 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" ,
256 "clock-AF_QIPCRTR", "clock-AF_MAX"
260 * sk_callback_lock locking rules are per-address-family,
261 * so split the lock classes by using a per-AF key:
263 static struct lock_class_key af_callback_keys[AF_MAX];
265 /* Take into consideration the size of the struct sk_buff overhead in the
266 * determination of these values, since that is non-constant across
267 * platforms. This makes socket queueing behavior and performance
268 * not depend upon such differences.
270 #define _SK_MEM_PACKETS 256
271 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
272 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
273 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
275 /* Run time adjustable parameters. */
276 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
277 EXPORT_SYMBOL(sysctl_wmem_max);
278 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
279 EXPORT_SYMBOL(sysctl_rmem_max);
280 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
281 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
283 /* Maximal space eaten by iovec or ancillary data plus some space */
284 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
285 EXPORT_SYMBOL(sysctl_optmem_max);
287 int sysctl_tstamp_allow_data __read_mostly = 1;
289 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
290 EXPORT_SYMBOL_GPL(memalloc_socks);
293 * sk_set_memalloc - sets %SOCK_MEMALLOC
294 * @sk: socket to set it on
296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
297 * It's the responsibility of the admin to adjust min_free_kbytes
298 * to meet the requirements
300 void sk_set_memalloc(struct sock *sk)
302 sock_set_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation |= __GFP_MEMALLOC;
304 static_key_slow_inc(&memalloc_socks);
306 EXPORT_SYMBOL_GPL(sk_set_memalloc);
308 void sk_clear_memalloc(struct sock *sk)
310 sock_reset_flag(sk, SOCK_MEMALLOC);
311 sk->sk_allocation &= ~__GFP_MEMALLOC;
312 static_key_slow_dec(&memalloc_socks);
315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
316 * progress of swapping. SOCK_MEMALLOC may be cleared while
317 * it has rmem allocations due to the last swapfile being deactivated
318 * but there is a risk that the socket is unusable due to exceeding
319 * the rmem limits. Reclaim the reserves and obey rmem limits again.
323 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
325 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
328 unsigned long pflags = current->flags;
330 /* these should have been dropped before queueing */
331 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
333 current->flags |= PF_MEMALLOC;
334 ret = sk->sk_backlog_rcv(sk, skb);
335 tsk_restore_flags(current, pflags, PF_MEMALLOC);
339 EXPORT_SYMBOL(__sk_backlog_rcv);
341 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
345 if (optlen < sizeof(tv))
347 if (copy_from_user(&tv, optval, sizeof(tv)))
349 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
353 static int warned __read_mostly;
356 if (warned < 10 && net_ratelimit()) {
358 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
359 __func__, current->comm, task_pid_nr(current));
363 *timeo_p = MAX_SCHEDULE_TIMEOUT;
364 if (tv.tv_sec == 0 && tv.tv_usec == 0)
366 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
367 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
371 static void sock_warn_obsolete_bsdism(const char *name)
374 static char warncomm[TASK_COMM_LEN];
375 if (strcmp(warncomm, current->comm) && warned < 5) {
376 strcpy(warncomm, current->comm);
377 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
383 static bool sock_needs_netstamp(const struct sock *sk)
385 switch (sk->sk_family) {
394 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
396 if (sk->sk_flags & flags) {
397 sk->sk_flags &= ~flags;
398 if (sock_needs_netstamp(sk) &&
399 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
400 net_disable_timestamp();
405 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
408 struct sk_buff_head *list = &sk->sk_receive_queue;
410 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
411 atomic_inc(&sk->sk_drops);
412 trace_sock_rcvqueue_full(sk, skb);
416 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
417 atomic_inc(&sk->sk_drops);
422 skb_set_owner_r(skb, sk);
424 /* we escape from rcu protected region, make sure we dont leak
429 spin_lock_irqsave(&list->lock, flags);
430 sock_skb_set_dropcount(sk, skb);
431 __skb_queue_tail(list, skb);
432 spin_unlock_irqrestore(&list->lock, flags);
434 if (!sock_flag(sk, SOCK_DEAD))
435 sk->sk_data_ready(sk);
438 EXPORT_SYMBOL(__sock_queue_rcv_skb);
440 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
444 err = sk_filter(sk, skb);
448 return __sock_queue_rcv_skb(sk, skb);
450 EXPORT_SYMBOL(sock_queue_rcv_skb);
452 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
453 const int nested, unsigned int trim_cap, bool refcounted)
455 int rc = NET_RX_SUCCESS;
457 if (sk_filter_trim_cap(sk, skb, trim_cap))
458 goto discard_and_relse;
462 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
463 atomic_inc(&sk->sk_drops);
464 goto discard_and_relse;
467 bh_lock_sock_nested(sk);
470 if (!sock_owned_by_user(sk)) {
472 * trylock + unlock semantics:
474 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
476 rc = sk_backlog_rcv(sk, skb);
478 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
479 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
481 atomic_inc(&sk->sk_drops);
482 goto discard_and_relse;
494 EXPORT_SYMBOL(__sk_receive_skb);
496 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
498 struct dst_entry *dst = __sk_dst_get(sk);
500 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
501 sk_tx_queue_clear(sk);
502 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
509 EXPORT_SYMBOL(__sk_dst_check);
511 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
513 struct dst_entry *dst = sk_dst_get(sk);
515 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
523 EXPORT_SYMBOL(sk_dst_check);
525 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
528 int ret = -ENOPROTOOPT;
529 #ifdef CONFIG_NETDEVICES
530 struct net *net = sock_net(sk);
531 char devname[IFNAMSIZ];
536 if (!ns_capable(net->user_ns, CAP_NET_RAW))
543 /* Bind this socket to a particular device like "eth0",
544 * as specified in the passed interface name. If the
545 * name is "" or the option length is zero the socket
548 if (optlen > IFNAMSIZ - 1)
549 optlen = IFNAMSIZ - 1;
550 memset(devname, 0, sizeof(devname));
553 if (copy_from_user(devname, optval, optlen))
557 if (devname[0] != '\0') {
558 struct net_device *dev;
561 dev = dev_get_by_name_rcu(net, devname);
563 index = dev->ifindex;
571 sk->sk_bound_dev_if = index;
583 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
584 int __user *optlen, int len)
586 int ret = -ENOPROTOOPT;
587 #ifdef CONFIG_NETDEVICES
588 struct net *net = sock_net(sk);
589 char devname[IFNAMSIZ];
591 if (sk->sk_bound_dev_if == 0) {
600 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
604 len = strlen(devname) + 1;
607 if (copy_to_user(optval, devname, len))
612 if (put_user(len, optlen))
623 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
626 sock_set_flag(sk, bit);
628 sock_reset_flag(sk, bit);
631 bool sk_mc_loop(struct sock *sk)
633 if (dev_recursion_level())
637 switch (sk->sk_family) {
639 return inet_sk(sk)->mc_loop;
640 #if IS_ENABLED(CONFIG_IPV6)
642 return inet6_sk(sk)->mc_loop;
648 EXPORT_SYMBOL(sk_mc_loop);
651 * This is meant for all protocols to use and covers goings on
652 * at the socket level. Everything here is generic.
655 int sock_setsockopt(struct socket *sock, int level, int optname,
656 char __user *optval, unsigned int optlen)
658 struct sock *sk = sock->sk;
665 * Options without arguments
668 if (optname == SO_BINDTODEVICE)
669 return sock_setbindtodevice(sk, optval, optlen);
671 if (optlen < sizeof(int))
674 if (get_user(val, (int __user *)optval))
677 valbool = val ? 1 : 0;
683 if (val && !capable(CAP_NET_ADMIN))
686 sock_valbool_flag(sk, SOCK_DBG, valbool);
689 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
692 sk->sk_reuseport = valbool;
701 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
705 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
708 /* Don't error on this BSD doesn't and if you think
709 * about it this is right. Otherwise apps have to
710 * play 'guess the biggest size' games. RCVBUF/SNDBUF
711 * are treated in BSD as hints
713 val = min_t(u32, val, sysctl_wmem_max);
715 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
716 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
717 /* Wake up sending tasks if we upped the value. */
718 sk->sk_write_space(sk);
722 if (!capable(CAP_NET_ADMIN)) {
729 /* Don't error on this BSD doesn't and if you think
730 * about it this is right. Otherwise apps have to
731 * play 'guess the biggest size' games. RCVBUF/SNDBUF
732 * are treated in BSD as hints
734 val = min_t(u32, val, sysctl_rmem_max);
736 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
738 * We double it on the way in to account for
739 * "struct sk_buff" etc. overhead. Applications
740 * assume that the SO_RCVBUF setting they make will
741 * allow that much actual data to be received on that
744 * Applications are unaware that "struct sk_buff" and
745 * other overheads allocate from the receive buffer
746 * during socket buffer allocation.
748 * And after considering the possible alternatives,
749 * returning the value we actually used in getsockopt
750 * is the most desirable behavior.
752 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
756 if (!capable(CAP_NET_ADMIN)) {
764 if (sk->sk_protocol == IPPROTO_TCP &&
765 sk->sk_type == SOCK_STREAM)
766 tcp_set_keepalive(sk, valbool);
768 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
772 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
776 sk->sk_no_check_tx = valbool;
780 if ((val >= 0 && val <= 6) ||
781 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
782 sk->sk_priority = val;
788 if (optlen < sizeof(ling)) {
789 ret = -EINVAL; /* 1003.1g */
792 if (copy_from_user(&ling, optval, sizeof(ling))) {
797 sock_reset_flag(sk, SOCK_LINGER);
799 #if (BITS_PER_LONG == 32)
800 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
801 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
804 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
805 sock_set_flag(sk, SOCK_LINGER);
810 sock_warn_obsolete_bsdism("setsockopt");
815 set_bit(SOCK_PASSCRED, &sock->flags);
817 clear_bit(SOCK_PASSCRED, &sock->flags);
823 if (optname == SO_TIMESTAMP)
824 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
826 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
827 sock_set_flag(sk, SOCK_RCVTSTAMP);
828 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
830 sock_reset_flag(sk, SOCK_RCVTSTAMP);
831 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
835 case SO_TIMESTAMPING:
836 if (val & ~SOF_TIMESTAMPING_MASK) {
841 if (val & SOF_TIMESTAMPING_OPT_ID &&
842 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
843 if (sk->sk_protocol == IPPROTO_TCP &&
844 sk->sk_type == SOCK_STREAM) {
845 if ((1 << sk->sk_state) &
846 (TCPF_CLOSE | TCPF_LISTEN)) {
850 sk->sk_tskey = tcp_sk(sk)->snd_una;
855 sk->sk_tsflags = val;
856 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
857 sock_enable_timestamp(sk,
858 SOCK_TIMESTAMPING_RX_SOFTWARE);
860 sock_disable_timestamp(sk,
861 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
867 sk->sk_rcvlowat = val ? : 1;
871 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
875 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
878 case SO_ATTACH_FILTER:
880 if (optlen == sizeof(struct sock_fprog)) {
881 struct sock_fprog fprog;
884 if (copy_from_user(&fprog, optval, sizeof(fprog)))
887 ret = sk_attach_filter(&fprog, sk);
893 if (optlen == sizeof(u32)) {
897 if (copy_from_user(&ufd, optval, sizeof(ufd)))
900 ret = sk_attach_bpf(ufd, sk);
904 case SO_ATTACH_REUSEPORT_CBPF:
906 if (optlen == sizeof(struct sock_fprog)) {
907 struct sock_fprog fprog;
910 if (copy_from_user(&fprog, optval, sizeof(fprog)))
913 ret = sk_reuseport_attach_filter(&fprog, sk);
917 case SO_ATTACH_REUSEPORT_EBPF:
919 if (optlen == sizeof(u32)) {
923 if (copy_from_user(&ufd, optval, sizeof(ufd)))
926 ret = sk_reuseport_attach_bpf(ufd, sk);
930 case SO_DETACH_FILTER:
931 ret = sk_detach_filter(sk);
935 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
938 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
943 set_bit(SOCK_PASSSEC, &sock->flags);
945 clear_bit(SOCK_PASSSEC, &sock->flags);
948 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
955 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
959 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
963 if (sock->ops->set_peek_off)
964 ret = sock->ops->set_peek_off(sk, val);
970 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
973 case SO_SELECT_ERR_QUEUE:
974 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
977 #ifdef CONFIG_NET_RX_BUSY_POLL
979 /* allow unprivileged users to decrease the value */
980 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
986 sk->sk_ll_usec = val;
991 case SO_MAX_PACING_RATE:
992 sk->sk_max_pacing_rate = val;
993 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
994 sk->sk_max_pacing_rate);
997 case SO_INCOMING_CPU:
998 sk->sk_incoming_cpu = val;
1003 dst_negative_advice(sk);
1012 EXPORT_SYMBOL(sock_setsockopt);
1015 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1016 struct ucred *ucred)
1018 ucred->pid = pid_vnr(pid);
1019 ucred->uid = ucred->gid = -1;
1021 struct user_namespace *current_ns = current_user_ns();
1023 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1024 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1028 int sock_getsockopt(struct socket *sock, int level, int optname,
1029 char __user *optval, int __user *optlen)
1031 struct sock *sk = sock->sk;
1039 int lv = sizeof(int);
1042 if (get_user(len, optlen))
1047 memset(&v, 0, sizeof(v));
1051 v.val = sock_flag(sk, SOCK_DBG);
1055 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1059 v.val = sock_flag(sk, SOCK_BROADCAST);
1063 v.val = sk->sk_sndbuf;
1067 v.val = sk->sk_rcvbuf;
1071 v.val = sk->sk_reuse;
1075 v.val = sk->sk_reuseport;
1079 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1083 v.val = sk->sk_type;
1087 v.val = sk->sk_protocol;
1091 v.val = sk->sk_family;
1095 v.val = -sock_error(sk);
1097 v.val = xchg(&sk->sk_err_soft, 0);
1101 v.val = sock_flag(sk, SOCK_URGINLINE);
1105 v.val = sk->sk_no_check_tx;
1109 v.val = sk->sk_priority;
1113 lv = sizeof(v.ling);
1114 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1115 v.ling.l_linger = sk->sk_lingertime / HZ;
1119 sock_warn_obsolete_bsdism("getsockopt");
1123 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1124 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1127 case SO_TIMESTAMPNS:
1128 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1131 case SO_TIMESTAMPING:
1132 v.val = sk->sk_tsflags;
1136 lv = sizeof(struct timeval);
1137 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1141 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1142 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1147 lv = sizeof(struct timeval);
1148 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1152 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1153 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1158 v.val = sk->sk_rcvlowat;
1166 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1171 struct ucred peercred;
1172 if (len > sizeof(peercred))
1173 len = sizeof(peercred);
1174 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1175 if (copy_to_user(optval, &peercred, len))
1184 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1188 if (copy_to_user(optval, address, len))
1193 /* Dubious BSD thing... Probably nobody even uses it, but
1194 * the UNIX standard wants it for whatever reason... -DaveM
1197 v.val = sk->sk_state == TCP_LISTEN;
1201 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1205 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1208 v.val = sk->sk_mark;
1212 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1215 case SO_WIFI_STATUS:
1216 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1220 if (!sock->ops->set_peek_off)
1223 v.val = sk->sk_peek_off;
1226 v.val = sock_flag(sk, SOCK_NOFCS);
1229 case SO_BINDTODEVICE:
1230 return sock_getbindtodevice(sk, optval, optlen, len);
1233 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1239 case SO_LOCK_FILTER:
1240 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1243 case SO_BPF_EXTENSIONS:
1244 v.val = bpf_tell_extensions();
1247 case SO_SELECT_ERR_QUEUE:
1248 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1251 #ifdef CONFIG_NET_RX_BUSY_POLL
1253 v.val = sk->sk_ll_usec;
1257 case SO_MAX_PACING_RATE:
1258 v.val = sk->sk_max_pacing_rate;
1261 case SO_INCOMING_CPU:
1262 v.val = sk->sk_incoming_cpu;
1266 /* We implement the SO_SNDLOWAT etc to not be settable
1269 return -ENOPROTOOPT;
1274 if (copy_to_user(optval, &v, len))
1277 if (put_user(len, optlen))
1283 * Initialize an sk_lock.
1285 * (We also register the sk_lock with the lock validator.)
1287 static inline void sock_lock_init(struct sock *sk)
1289 sock_lock_init_class_and_name(sk,
1290 af_family_slock_key_strings[sk->sk_family],
1291 af_family_slock_keys + sk->sk_family,
1292 af_family_key_strings[sk->sk_family],
1293 af_family_keys + sk->sk_family);
1297 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1298 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1299 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1301 static void sock_copy(struct sock *nsk, const struct sock *osk)
1303 #ifdef CONFIG_SECURITY_NETWORK
1304 void *sptr = nsk->sk_security;
1306 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1308 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1309 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1311 #ifdef CONFIG_SECURITY_NETWORK
1312 nsk->sk_security = sptr;
1313 security_sk_clone(osk, nsk);
1317 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1321 struct kmem_cache *slab;
1325 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1328 if (priority & __GFP_ZERO)
1329 sk_prot_clear_nulls(sk, prot->obj_size);
1331 sk = kmalloc(prot->obj_size, priority);
1334 kmemcheck_annotate_bitfield(sk, flags);
1336 if (security_sk_alloc(sk, family, priority))
1339 if (!try_module_get(prot->owner))
1341 sk_tx_queue_clear(sk);
1347 security_sk_free(sk);
1350 kmem_cache_free(slab, sk);
1356 static void sk_prot_free(struct proto *prot, struct sock *sk)
1358 struct kmem_cache *slab;
1359 struct module *owner;
1361 owner = prot->owner;
1364 cgroup_sk_free(&sk->sk_cgrp_data);
1365 mem_cgroup_sk_free(sk);
1366 security_sk_free(sk);
1368 kmem_cache_free(slab, sk);
1375 * sk_alloc - All socket objects are allocated here
1376 * @net: the applicable net namespace
1377 * @family: protocol family
1378 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1379 * @prot: struct proto associated with this new sock instance
1380 * @kern: is this to be a kernel socket?
1382 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1383 struct proto *prot, int kern)
1387 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1389 sk->sk_family = family;
1391 * See comment in struct sock definition to understand
1392 * why we need sk_prot_creator -acme
1394 sk->sk_prot = sk->sk_prot_creator = prot;
1396 sk->sk_net_refcnt = kern ? 0 : 1;
1397 if (likely(sk->sk_net_refcnt))
1399 sock_net_set(sk, net);
1400 atomic_set(&sk->sk_wmem_alloc, 1);
1402 mem_cgroup_sk_alloc(sk);
1403 cgroup_sk_alloc(&sk->sk_cgrp_data);
1404 sock_update_classid(&sk->sk_cgrp_data);
1405 sock_update_netprioidx(&sk->sk_cgrp_data);
1406 sk_tx_queue_clear(sk);
1411 EXPORT_SYMBOL(sk_alloc);
1413 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1414 * grace period. This is the case for UDP sockets and TCP listeners.
1416 static void __sk_destruct(struct rcu_head *head)
1418 struct sock *sk = container_of(head, struct sock, sk_rcu);
1419 struct sk_filter *filter;
1421 if (sk->sk_destruct)
1422 sk->sk_destruct(sk);
1424 filter = rcu_dereference_check(sk->sk_filter,
1425 atomic_read(&sk->sk_wmem_alloc) == 0);
1427 sk_filter_uncharge(sk, filter);
1428 RCU_INIT_POINTER(sk->sk_filter, NULL);
1431 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1433 if (atomic_read(&sk->sk_omem_alloc))
1434 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1435 __func__, atomic_read(&sk->sk_omem_alloc));
1437 if (sk->sk_frag.page) {
1438 put_page(sk->sk_frag.page);
1439 sk->sk_frag.page = NULL;
1442 if (sk->sk_peer_cred)
1443 put_cred(sk->sk_peer_cred);
1444 put_pid(sk->sk_peer_pid);
1445 if (likely(sk->sk_net_refcnt))
1446 put_net(sock_net(sk));
1447 sk_prot_free(sk->sk_prot_creator, sk);
1450 void sk_destruct(struct sock *sk)
1452 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1454 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1455 reuseport_detach_sock(sk);
1456 use_call_rcu = true;
1460 call_rcu(&sk->sk_rcu, __sk_destruct);
1462 __sk_destruct(&sk->sk_rcu);
1465 static void __sk_free(struct sock *sk)
1467 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1468 sock_diag_broadcast_destroy(sk);
1473 void sk_free(struct sock *sk)
1476 * We subtract one from sk_wmem_alloc and can know if
1477 * some packets are still in some tx queue.
1478 * If not null, sock_wfree() will call __sk_free(sk) later
1480 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1483 EXPORT_SYMBOL(sk_free);
1486 * sk_clone_lock - clone a socket, and lock its clone
1487 * @sk: the socket to clone
1488 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1490 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1492 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1495 bool is_charged = true;
1497 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1498 if (newsk != NULL) {
1499 struct sk_filter *filter;
1501 sock_copy(newsk, sk);
1503 newsk->sk_prot_creator = sk->sk_prot;
1506 if (likely(newsk->sk_net_refcnt))
1507 get_net(sock_net(newsk));
1508 sk_node_init(&newsk->sk_node);
1509 sock_lock_init(newsk);
1510 bh_lock_sock(newsk);
1511 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1512 newsk->sk_backlog.len = 0;
1514 atomic_set(&newsk->sk_rmem_alloc, 0);
1516 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1518 atomic_set(&newsk->sk_wmem_alloc, 1);
1519 atomic_set(&newsk->sk_omem_alloc, 0);
1520 skb_queue_head_init(&newsk->sk_receive_queue);
1521 skb_queue_head_init(&newsk->sk_write_queue);
1523 rwlock_init(&newsk->sk_callback_lock);
1524 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1525 af_callback_keys + newsk->sk_family,
1526 af_family_clock_key_strings[newsk->sk_family]);
1528 newsk->sk_dst_cache = NULL;
1529 newsk->sk_wmem_queued = 0;
1530 newsk->sk_forward_alloc = 0;
1531 atomic_set(&newsk->sk_drops, 0);
1532 newsk->sk_send_head = NULL;
1533 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1535 sock_reset_flag(newsk, SOCK_DONE);
1536 cgroup_sk_clone(&newsk->sk_cgrp_data);
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, sk))) {
1548 /* We need to make sure that we don't uncharge the new
1549 * socket if we couldn't charge it in the first place
1550 * as otherwise we uncharge the parent's filter.
1553 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1554 /* It is still raw copy of parent, so invalidate
1555 * destructor and make plain sk_free() */
1556 newsk->sk_destruct = NULL;
1557 bh_unlock_sock(newsk);
1562 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1565 newsk->sk_err_soft = 0;
1566 newsk->sk_priority = 0;
1567 newsk->sk_incoming_cpu = raw_smp_processor_id();
1568 atomic64_set(&newsk->sk_cookie, 0);
1570 mem_cgroup_sk_alloc(newsk);
1572 * Before updating sk_refcnt, we must commit prior changes to memory
1573 * (Documentation/RCU/rculist_nulls.txt for details)
1576 atomic_set(&newsk->sk_refcnt, 2);
1579 * Increment the counter in the same struct proto as the master
1580 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1581 * is the same as sk->sk_prot->socks, as this field was copied
1584 * This _changes_ the previous behaviour, where
1585 * tcp_create_openreq_child always was incrementing the
1586 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1587 * to be taken into account in all callers. -acme
1589 sk_refcnt_debug_inc(newsk);
1590 sk_set_socket(newsk, NULL);
1591 sk_tx_queue_clear(newsk);
1592 newsk->sk_wq = NULL;
1594 if (newsk->sk_prot->sockets_allocated)
1595 sk_sockets_allocated_inc(newsk);
1597 if (sock_needs_netstamp(sk) &&
1598 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1599 net_enable_timestamp();
1604 EXPORT_SYMBOL_GPL(sk_clone_lock);
1606 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1610 sk_dst_set(sk, dst);
1611 sk->sk_route_caps = dst->dev->features;
1612 if (sk->sk_route_caps & NETIF_F_GSO)
1613 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1614 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1615 if (sk_can_gso(sk)) {
1616 if (dst->header_len) {
1617 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1619 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1620 sk->sk_gso_max_size = dst->dev->gso_max_size;
1621 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1624 sk->sk_gso_max_segs = max_segs;
1626 EXPORT_SYMBOL_GPL(sk_setup_caps);
1629 * Simple resource managers for sockets.
1634 * Write buffer destructor automatically called from kfree_skb.
1636 void sock_wfree(struct sk_buff *skb)
1638 struct sock *sk = skb->sk;
1639 unsigned int len = skb->truesize;
1641 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1643 * Keep a reference on sk_wmem_alloc, this will be released
1644 * after sk_write_space() call
1646 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1647 sk->sk_write_space(sk);
1651 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1652 * could not do because of in-flight packets
1654 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1657 EXPORT_SYMBOL(sock_wfree);
1659 /* This variant of sock_wfree() is used by TCP,
1660 * since it sets SOCK_USE_WRITE_QUEUE.
1662 void __sock_wfree(struct sk_buff *skb)
1664 struct sock *sk = skb->sk;
1666 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1670 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1675 if (unlikely(!sk_fullsock(sk))) {
1676 skb->destructor = sock_edemux;
1681 skb->destructor = sock_wfree;
1682 skb_set_hash_from_sk(skb, sk);
1684 * We used to take a refcount on sk, but following operation
1685 * is enough to guarantee sk_free() wont free this sock until
1686 * all in-flight packets are completed
1688 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1690 EXPORT_SYMBOL(skb_set_owner_w);
1692 /* This helper is used by netem, as it can hold packets in its
1693 * delay queue. We want to allow the owner socket to send more
1694 * packets, as if they were already TX completed by a typical driver.
1695 * But we also want to keep skb->sk set because some packet schedulers
1696 * rely on it (sch_fq for example).
1698 void skb_orphan_partial(struct sk_buff *skb)
1700 if (skb_is_tcp_pure_ack(skb))
1703 if (skb->destructor == sock_wfree
1705 || skb->destructor == tcp_wfree
1708 struct sock *sk = skb->sk;
1710 if (atomic_inc_not_zero(&sk->sk_refcnt)) {
1711 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1712 skb->destructor = sock_efree;
1718 EXPORT_SYMBOL(skb_orphan_partial);
1721 * Read buffer destructor automatically called from kfree_skb.
1723 void sock_rfree(struct sk_buff *skb)
1725 struct sock *sk = skb->sk;
1726 unsigned int len = skb->truesize;
1728 atomic_sub(len, &sk->sk_rmem_alloc);
1729 sk_mem_uncharge(sk, len);
1731 EXPORT_SYMBOL(sock_rfree);
1734 * Buffer destructor for skbs that are not used directly in read or write
1735 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1737 void sock_efree(struct sk_buff *skb)
1741 EXPORT_SYMBOL(sock_efree);
1743 kuid_t sock_i_uid(struct sock *sk)
1747 read_lock_bh(&sk->sk_callback_lock);
1748 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1749 read_unlock_bh(&sk->sk_callback_lock);
1752 EXPORT_SYMBOL(sock_i_uid);
1754 unsigned long sock_i_ino(struct sock *sk)
1758 read_lock_bh(&sk->sk_callback_lock);
1759 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1760 read_unlock_bh(&sk->sk_callback_lock);
1763 EXPORT_SYMBOL(sock_i_ino);
1766 * Allocate a skb from the socket's send buffer.
1768 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1771 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1772 struct sk_buff *skb = alloc_skb(size, priority);
1774 skb_set_owner_w(skb, sk);
1780 EXPORT_SYMBOL(sock_wmalloc);
1783 * Allocate a memory block from the socket's option memory buffer.
1785 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1787 if ((unsigned int)size <= sysctl_optmem_max &&
1788 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1790 /* First do the add, to avoid the race if kmalloc
1793 atomic_add(size, &sk->sk_omem_alloc);
1794 mem = kmalloc(size, priority);
1797 atomic_sub(size, &sk->sk_omem_alloc);
1801 EXPORT_SYMBOL(sock_kmalloc);
1803 /* Free an option memory block. Note, we actually want the inline
1804 * here as this allows gcc to detect the nullify and fold away the
1805 * condition entirely.
1807 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1810 if (WARN_ON_ONCE(!mem))
1816 atomic_sub(size, &sk->sk_omem_alloc);
1819 void sock_kfree_s(struct sock *sk, void *mem, int size)
1821 __sock_kfree_s(sk, mem, size, false);
1823 EXPORT_SYMBOL(sock_kfree_s);
1825 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1827 __sock_kfree_s(sk, mem, size, true);
1829 EXPORT_SYMBOL(sock_kzfree_s);
1831 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1832 I think, these locks should be removed for datagram sockets.
1834 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1838 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1842 if (signal_pending(current))
1844 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1845 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1846 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1848 if (sk->sk_shutdown & SEND_SHUTDOWN)
1852 timeo = schedule_timeout(timeo);
1854 finish_wait(sk_sleep(sk), &wait);
1860 * Generic send/receive buffer handlers
1863 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1864 unsigned long data_len, int noblock,
1865 int *errcode, int max_page_order)
1867 struct sk_buff *skb;
1871 timeo = sock_sndtimeo(sk, noblock);
1873 err = sock_error(sk);
1878 if (sk->sk_shutdown & SEND_SHUTDOWN)
1881 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1884 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1885 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1889 if (signal_pending(current))
1891 timeo = sock_wait_for_wmem(sk, timeo);
1893 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1894 errcode, sk->sk_allocation);
1896 skb_set_owner_w(skb, sk);
1900 err = sock_intr_errno(timeo);
1905 EXPORT_SYMBOL(sock_alloc_send_pskb);
1907 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1908 int noblock, int *errcode)
1910 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1912 EXPORT_SYMBOL(sock_alloc_send_skb);
1914 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1915 struct sockcm_cookie *sockc)
1919 switch (cmsg->cmsg_type) {
1921 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1923 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1925 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1927 case SO_TIMESTAMPING:
1928 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1931 tsflags = *(u32 *)CMSG_DATA(cmsg);
1932 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1935 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1936 sockc->tsflags |= tsflags;
1938 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1940 case SCM_CREDENTIALS:
1947 EXPORT_SYMBOL(__sock_cmsg_send);
1949 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1950 struct sockcm_cookie *sockc)
1952 struct cmsghdr *cmsg;
1955 for_each_cmsghdr(cmsg, msg) {
1956 if (!CMSG_OK(msg, cmsg))
1958 if (cmsg->cmsg_level != SOL_SOCKET)
1960 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1966 EXPORT_SYMBOL(sock_cmsg_send);
1968 /* On 32bit arches, an skb frag is limited to 2^15 */
1969 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1972 * skb_page_frag_refill - check that a page_frag contains enough room
1973 * @sz: minimum size of the fragment we want to get
1974 * @pfrag: pointer to page_frag
1975 * @gfp: priority for memory allocation
1977 * Note: While this allocator tries to use high order pages, there is
1978 * no guarantee that allocations succeed. Therefore, @sz MUST be
1979 * less or equal than PAGE_SIZE.
1981 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1984 if (page_ref_count(pfrag->page) == 1) {
1988 if (pfrag->offset + sz <= pfrag->size)
1990 put_page(pfrag->page);
1994 if (SKB_FRAG_PAGE_ORDER) {
1995 /* Avoid direct reclaim but allow kswapd to wake */
1996 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1997 __GFP_COMP | __GFP_NOWARN |
1999 SKB_FRAG_PAGE_ORDER);
2000 if (likely(pfrag->page)) {
2001 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2005 pfrag->page = alloc_page(gfp);
2006 if (likely(pfrag->page)) {
2007 pfrag->size = PAGE_SIZE;
2012 EXPORT_SYMBOL(skb_page_frag_refill);
2014 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2016 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2019 sk_enter_memory_pressure(sk);
2020 sk_stream_moderate_sndbuf(sk);
2023 EXPORT_SYMBOL(sk_page_frag_refill);
2025 static void __lock_sock(struct sock *sk)
2026 __releases(&sk->sk_lock.slock)
2027 __acquires(&sk->sk_lock.slock)
2032 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2033 TASK_UNINTERRUPTIBLE);
2034 spin_unlock_bh(&sk->sk_lock.slock);
2036 spin_lock_bh(&sk->sk_lock.slock);
2037 if (!sock_owned_by_user(sk))
2040 finish_wait(&sk->sk_lock.wq, &wait);
2043 static void __release_sock(struct sock *sk)
2044 __releases(&sk->sk_lock.slock)
2045 __acquires(&sk->sk_lock.slock)
2047 struct sk_buff *skb, *next;
2049 while ((skb = sk->sk_backlog.head) != NULL) {
2050 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2052 spin_unlock_bh(&sk->sk_lock.slock);
2057 WARN_ON_ONCE(skb_dst_is_noref(skb));
2059 sk_backlog_rcv(sk, skb);
2064 } while (skb != NULL);
2066 spin_lock_bh(&sk->sk_lock.slock);
2070 * Doing the zeroing here guarantee we can not loop forever
2071 * while a wild producer attempts to flood us.
2073 sk->sk_backlog.len = 0;
2076 void __sk_flush_backlog(struct sock *sk)
2078 spin_lock_bh(&sk->sk_lock.slock);
2080 spin_unlock_bh(&sk->sk_lock.slock);
2084 * sk_wait_data - wait for data to arrive at sk_receive_queue
2085 * @sk: sock to wait on
2086 * @timeo: for how long
2087 * @skb: last skb seen on sk_receive_queue
2089 * Now socket state including sk->sk_err is changed only under lock,
2090 * hence we may omit checks after joining wait queue.
2091 * We check receive queue before schedule() only as optimization;
2092 * it is very likely that release_sock() added new data.
2094 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2099 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2100 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2101 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2102 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2103 finish_wait(sk_sleep(sk), &wait);
2106 EXPORT_SYMBOL(sk_wait_data);
2109 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2111 * @size: memory size to allocate
2112 * @kind: allocation type
2114 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2115 * rmem allocation. This function assumes that protocols which have
2116 * memory_pressure use sk_wmem_queued as write buffer accounting.
2118 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2120 struct proto *prot = sk->sk_prot;
2121 int amt = sk_mem_pages(size);
2124 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2126 allocated = sk_memory_allocated_add(sk, amt);
2128 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2129 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2130 goto suppress_allocation;
2133 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2134 sk_leave_memory_pressure(sk);
2138 /* Under pressure. */
2139 if (allocated > sk_prot_mem_limits(sk, 1))
2140 sk_enter_memory_pressure(sk);
2142 /* Over hard limit. */
2143 if (allocated > sk_prot_mem_limits(sk, 2))
2144 goto suppress_allocation;
2146 /* guarantee minimum buffer size under pressure */
2147 if (kind == SK_MEM_RECV) {
2148 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2151 } else { /* SK_MEM_SEND */
2152 if (sk->sk_type == SOCK_STREAM) {
2153 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2155 } else if (atomic_read(&sk->sk_wmem_alloc) <
2156 prot->sysctl_wmem[0])
2160 if (sk_has_memory_pressure(sk)) {
2163 if (!sk_under_memory_pressure(sk))
2165 alloc = sk_sockets_allocated_read_positive(sk);
2166 if (sk_prot_mem_limits(sk, 2) > alloc *
2167 sk_mem_pages(sk->sk_wmem_queued +
2168 atomic_read(&sk->sk_rmem_alloc) +
2169 sk->sk_forward_alloc))
2173 suppress_allocation:
2175 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2176 sk_stream_moderate_sndbuf(sk);
2178 /* Fail only if socket is _under_ its sndbuf.
2179 * In this case we cannot block, so that we have to fail.
2181 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2185 trace_sock_exceed_buf_limit(sk, prot, allocated);
2187 /* Alas. Undo changes. */
2188 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2190 sk_memory_allocated_sub(sk, amt);
2192 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2193 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2197 EXPORT_SYMBOL(__sk_mem_schedule);
2200 * __sk_mem_reclaim - reclaim memory_allocated
2202 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2204 void __sk_mem_reclaim(struct sock *sk, int amount)
2206 amount >>= SK_MEM_QUANTUM_SHIFT;
2207 sk_memory_allocated_sub(sk, amount);
2208 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2210 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2211 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2213 if (sk_under_memory_pressure(sk) &&
2214 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2215 sk_leave_memory_pressure(sk);
2217 EXPORT_SYMBOL(__sk_mem_reclaim);
2219 int sk_set_peek_off(struct sock *sk, int val)
2224 sk->sk_peek_off = val;
2227 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2230 * Set of default routines for initialising struct proto_ops when
2231 * the protocol does not support a particular function. In certain
2232 * cases where it makes no sense for a protocol to have a "do nothing"
2233 * function, some default processing is provided.
2236 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2240 EXPORT_SYMBOL(sock_no_bind);
2242 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2247 EXPORT_SYMBOL(sock_no_connect);
2249 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2253 EXPORT_SYMBOL(sock_no_socketpair);
2255 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2259 EXPORT_SYMBOL(sock_no_accept);
2261 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2266 EXPORT_SYMBOL(sock_no_getname);
2268 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2272 EXPORT_SYMBOL(sock_no_poll);
2274 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2278 EXPORT_SYMBOL(sock_no_ioctl);
2280 int sock_no_listen(struct socket *sock, int backlog)
2284 EXPORT_SYMBOL(sock_no_listen);
2286 int sock_no_shutdown(struct socket *sock, int how)
2290 EXPORT_SYMBOL(sock_no_shutdown);
2292 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2293 char __user *optval, unsigned int optlen)
2297 EXPORT_SYMBOL(sock_no_setsockopt);
2299 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2300 char __user *optval, int __user *optlen)
2304 EXPORT_SYMBOL(sock_no_getsockopt);
2306 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2310 EXPORT_SYMBOL(sock_no_sendmsg);
2312 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2317 EXPORT_SYMBOL(sock_no_recvmsg);
2319 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2321 /* Mirror missing mmap method error code */
2324 EXPORT_SYMBOL(sock_no_mmap);
2327 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2328 * various sock-based usage counts.
2330 void __receive_sock(struct file *file)
2332 struct socket *sock;
2336 * The resulting value of "error" is ignored here since we only
2337 * need to take action when the file is a socket and testing
2338 * "sock" for NULL is sufficient.
2340 sock = sock_from_file(file, &error);
2342 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2343 sock_update_classid(&sock->sk->sk_cgrp_data);
2347 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2350 struct msghdr msg = {.msg_flags = flags};
2352 char *kaddr = kmap(page);
2353 iov.iov_base = kaddr + offset;
2355 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2359 EXPORT_SYMBOL(sock_no_sendpage);
2362 * Default Socket Callbacks
2365 static void sock_def_wakeup(struct sock *sk)
2367 struct socket_wq *wq;
2370 wq = rcu_dereference(sk->sk_wq);
2371 if (skwq_has_sleeper(wq))
2372 wake_up_interruptible_all(&wq->wait);
2376 static void sock_def_error_report(struct sock *sk)
2378 struct socket_wq *wq;
2381 wq = rcu_dereference(sk->sk_wq);
2382 if (skwq_has_sleeper(wq))
2383 wake_up_interruptible_poll(&wq->wait, POLLERR);
2384 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2388 static void sock_def_readable(struct sock *sk)
2390 struct socket_wq *wq;
2393 wq = rcu_dereference(sk->sk_wq);
2394 if (skwq_has_sleeper(wq))
2395 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2396 POLLRDNORM | POLLRDBAND);
2397 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2401 static void sock_def_write_space(struct sock *sk)
2403 struct socket_wq *wq;
2407 /* Do not wake up a writer until he can make "significant"
2410 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2411 wq = rcu_dereference(sk->sk_wq);
2412 if (skwq_has_sleeper(wq))
2413 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2414 POLLWRNORM | POLLWRBAND);
2416 /* Should agree with poll, otherwise some programs break */
2417 if (sock_writeable(sk))
2418 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2424 static void sock_def_destruct(struct sock *sk)
2428 void sk_send_sigurg(struct sock *sk)
2430 if (sk->sk_socket && sk->sk_socket->file)
2431 if (send_sigurg(&sk->sk_socket->file->f_owner))
2432 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2434 EXPORT_SYMBOL(sk_send_sigurg);
2436 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2437 unsigned long expires)
2439 if (!mod_timer(timer, expires))
2442 EXPORT_SYMBOL(sk_reset_timer);
2444 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2446 if (del_timer(timer))
2449 EXPORT_SYMBOL(sk_stop_timer);
2451 void sock_init_data(struct socket *sock, struct sock *sk)
2453 skb_queue_head_init(&sk->sk_receive_queue);
2454 skb_queue_head_init(&sk->sk_write_queue);
2455 skb_queue_head_init(&sk->sk_error_queue);
2457 sk->sk_send_head = NULL;
2459 init_timer(&sk->sk_timer);
2461 sk->sk_allocation = GFP_KERNEL;
2462 sk->sk_rcvbuf = sysctl_rmem_default;
2463 sk->sk_sndbuf = sysctl_wmem_default;
2464 sk->sk_state = TCP_CLOSE;
2465 sk_set_socket(sk, sock);
2467 sock_set_flag(sk, SOCK_ZAPPED);
2470 sk->sk_type = sock->type;
2471 sk->sk_wq = sock->wq;
2476 rwlock_init(&sk->sk_callback_lock);
2477 lockdep_set_class_and_name(&sk->sk_callback_lock,
2478 af_callback_keys + sk->sk_family,
2479 af_family_clock_key_strings[sk->sk_family]);
2481 sk->sk_state_change = sock_def_wakeup;
2482 sk->sk_data_ready = sock_def_readable;
2483 sk->sk_write_space = sock_def_write_space;
2484 sk->sk_error_report = sock_def_error_report;
2485 sk->sk_destruct = sock_def_destruct;
2487 sk->sk_frag.page = NULL;
2488 sk->sk_frag.offset = 0;
2489 sk->sk_peek_off = -1;
2491 sk->sk_peer_pid = NULL;
2492 sk->sk_peer_cred = NULL;
2493 sk->sk_write_pending = 0;
2494 sk->sk_rcvlowat = 1;
2495 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2496 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2498 sk->sk_stamp = ktime_set(-1L, 0);
2499 #if BITS_PER_LONG==32
2500 seqlock_init(&sk->sk_stamp_seq);
2503 #ifdef CONFIG_NET_RX_BUSY_POLL
2505 sk->sk_ll_usec = sysctl_net_busy_read;
2508 sk->sk_max_pacing_rate = ~0U;
2509 sk->sk_pacing_rate = ~0U;
2510 sk->sk_incoming_cpu = -1;
2512 * Before updating sk_refcnt, we must commit prior changes to memory
2513 * (Documentation/RCU/rculist_nulls.txt for details)
2516 atomic_set(&sk->sk_refcnt, 1);
2517 atomic_set(&sk->sk_drops, 0);
2519 EXPORT_SYMBOL(sock_init_data);
2521 void lock_sock_nested(struct sock *sk, int subclass)
2524 spin_lock_bh(&sk->sk_lock.slock);
2525 if (sk->sk_lock.owned)
2527 sk->sk_lock.owned = 1;
2528 spin_unlock(&sk->sk_lock.slock);
2530 * The sk_lock has mutex_lock() semantics here:
2532 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2535 EXPORT_SYMBOL(lock_sock_nested);
2537 void release_sock(struct sock *sk)
2539 spin_lock_bh(&sk->sk_lock.slock);
2540 if (sk->sk_backlog.tail)
2543 /* Warning : release_cb() might need to release sk ownership,
2544 * ie call sock_release_ownership(sk) before us.
2546 if (sk->sk_prot->release_cb)
2547 sk->sk_prot->release_cb(sk);
2549 sock_release_ownership(sk);
2550 if (waitqueue_active(&sk->sk_lock.wq))
2551 wake_up(&sk->sk_lock.wq);
2552 spin_unlock_bh(&sk->sk_lock.slock);
2554 EXPORT_SYMBOL(release_sock);
2557 * lock_sock_fast - fast version of lock_sock
2560 * This version should be used for very small section, where process wont block
2561 * return false if fast path is taken
2562 * sk_lock.slock locked, owned = 0, BH disabled
2563 * return true if slow path is taken
2564 * sk_lock.slock unlocked, owned = 1, BH enabled
2566 bool lock_sock_fast(struct sock *sk)
2569 spin_lock_bh(&sk->sk_lock.slock);
2571 if (!sk->sk_lock.owned)
2573 * Note : We must disable BH
2578 sk->sk_lock.owned = 1;
2579 spin_unlock(&sk->sk_lock.slock);
2581 * The sk_lock has mutex_lock() semantics here:
2583 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2587 EXPORT_SYMBOL(lock_sock_fast);
2589 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2592 if (!sock_flag(sk, SOCK_TIMESTAMP))
2593 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2594 tv = ktime_to_timeval(sk->sk_stamp);
2595 if (tv.tv_sec == -1)
2597 if (tv.tv_sec == 0) {
2598 sk->sk_stamp = ktime_get_real();
2599 tv = ktime_to_timeval(sk->sk_stamp);
2601 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2603 EXPORT_SYMBOL(sock_get_timestamp);
2605 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2608 if (!sock_flag(sk, SOCK_TIMESTAMP))
2609 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2610 ts = ktime_to_timespec(sk->sk_stamp);
2611 if (ts.tv_sec == -1)
2613 if (ts.tv_sec == 0) {
2614 sk->sk_stamp = ktime_get_real();
2615 ts = ktime_to_timespec(sk->sk_stamp);
2617 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2619 EXPORT_SYMBOL(sock_get_timestampns);
2621 void sock_enable_timestamp(struct sock *sk, int flag)
2623 if (!sock_flag(sk, flag)) {
2624 unsigned long previous_flags = sk->sk_flags;
2626 sock_set_flag(sk, flag);
2628 * we just set one of the two flags which require net
2629 * time stamping, but time stamping might have been on
2630 * already because of the other one
2632 if (sock_needs_netstamp(sk) &&
2633 !(previous_flags & SK_FLAGS_TIMESTAMP))
2634 net_enable_timestamp();
2638 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2639 int level, int type)
2641 struct sock_exterr_skb *serr;
2642 struct sk_buff *skb;
2646 skb = sock_dequeue_err_skb(sk);
2652 msg->msg_flags |= MSG_TRUNC;
2655 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2659 sock_recv_timestamp(msg, sk, skb);
2661 serr = SKB_EXT_ERR(skb);
2662 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2664 msg->msg_flags |= MSG_ERRQUEUE;
2672 EXPORT_SYMBOL(sock_recv_errqueue);
2675 * Get a socket option on an socket.
2677 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2678 * asynchronous errors should be reported by getsockopt. We assume
2679 * this means if you specify SO_ERROR (otherwise whats the point of it).
2681 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2682 char __user *optval, int __user *optlen)
2684 struct sock *sk = sock->sk;
2686 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2688 EXPORT_SYMBOL(sock_common_getsockopt);
2690 #ifdef CONFIG_COMPAT
2691 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2692 char __user *optval, int __user *optlen)
2694 struct sock *sk = sock->sk;
2696 if (sk->sk_prot->compat_getsockopt != NULL)
2697 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2699 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2701 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2704 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2707 struct sock *sk = sock->sk;
2711 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2712 flags & ~MSG_DONTWAIT, &addr_len);
2714 msg->msg_namelen = addr_len;
2717 EXPORT_SYMBOL(sock_common_recvmsg);
2720 * Set socket options on an inet socket.
2722 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2723 char __user *optval, unsigned int optlen)
2725 struct sock *sk = sock->sk;
2727 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2729 EXPORT_SYMBOL(sock_common_setsockopt);
2731 #ifdef CONFIG_COMPAT
2732 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2733 char __user *optval, unsigned int optlen)
2735 struct sock *sk = sock->sk;
2737 if (sk->sk_prot->compat_setsockopt != NULL)
2738 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2740 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2742 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2745 void sk_common_release(struct sock *sk)
2747 if (sk->sk_prot->destroy)
2748 sk->sk_prot->destroy(sk);
2751 * Observation: when sock_common_release is called, processes have
2752 * no access to socket. But net still has.
2753 * Step one, detach it from networking:
2755 * A. Remove from hash tables.
2758 sk->sk_prot->unhash(sk);
2761 * In this point socket cannot receive new packets, but it is possible
2762 * that some packets are in flight because some CPU runs receiver and
2763 * did hash table lookup before we unhashed socket. They will achieve
2764 * receive queue and will be purged by socket destructor.
2766 * Also we still have packets pending on receive queue and probably,
2767 * our own packets waiting in device queues. sock_destroy will drain
2768 * receive queue, but transmitted packets will delay socket destruction
2769 * until the last reference will be released.
2774 xfrm_sk_free_policy(sk);
2776 sk_refcnt_debug_release(sk);
2780 EXPORT_SYMBOL(sk_common_release);
2782 #ifdef CONFIG_PROC_FS
2783 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2785 int val[PROTO_INUSE_NR];
2788 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2790 #ifdef CONFIG_NET_NS
2791 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2793 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2795 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2797 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2799 int cpu, idx = prot->inuse_idx;
2802 for_each_possible_cpu(cpu)
2803 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2805 return res >= 0 ? res : 0;
2807 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2809 static int __net_init sock_inuse_init_net(struct net *net)
2811 net->core.inuse = alloc_percpu(struct prot_inuse);
2812 return net->core.inuse ? 0 : -ENOMEM;
2815 static void __net_exit sock_inuse_exit_net(struct net *net)
2817 free_percpu(net->core.inuse);
2820 static struct pernet_operations net_inuse_ops = {
2821 .init = sock_inuse_init_net,
2822 .exit = sock_inuse_exit_net,
2825 static __init int net_inuse_init(void)
2827 if (register_pernet_subsys(&net_inuse_ops))
2828 panic("Cannot initialize net inuse counters");
2833 core_initcall(net_inuse_init);
2835 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2837 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2839 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2841 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2843 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2845 int cpu, idx = prot->inuse_idx;
2848 for_each_possible_cpu(cpu)
2849 res += per_cpu(prot_inuse, cpu).val[idx];
2851 return res >= 0 ? res : 0;
2853 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2856 static void assign_proto_idx(struct proto *prot)
2858 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2860 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2861 pr_err("PROTO_INUSE_NR exhausted\n");
2865 set_bit(prot->inuse_idx, proto_inuse_idx);
2868 static void release_proto_idx(struct proto *prot)
2870 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2871 clear_bit(prot->inuse_idx, proto_inuse_idx);
2874 static inline void assign_proto_idx(struct proto *prot)
2878 static inline void release_proto_idx(struct proto *prot)
2883 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2887 kfree(rsk_prot->slab_name);
2888 rsk_prot->slab_name = NULL;
2889 kmem_cache_destroy(rsk_prot->slab);
2890 rsk_prot->slab = NULL;
2893 static int req_prot_init(const struct proto *prot)
2895 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2900 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2902 if (!rsk_prot->slab_name)
2905 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2906 rsk_prot->obj_size, 0,
2907 prot->slab_flags, NULL);
2909 if (!rsk_prot->slab) {
2910 pr_crit("%s: Can't create request sock SLAB cache!\n",
2917 int proto_register(struct proto *prot, int alloc_slab)
2920 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2921 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2924 if (prot->slab == NULL) {
2925 pr_crit("%s: Can't create sock SLAB cache!\n",
2930 if (req_prot_init(prot))
2931 goto out_free_request_sock_slab;
2933 if (prot->twsk_prot != NULL) {
2934 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2936 if (prot->twsk_prot->twsk_slab_name == NULL)
2937 goto out_free_request_sock_slab;
2939 prot->twsk_prot->twsk_slab =
2940 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2941 prot->twsk_prot->twsk_obj_size,
2945 if (prot->twsk_prot->twsk_slab == NULL)
2946 goto out_free_timewait_sock_slab_name;
2950 mutex_lock(&proto_list_mutex);
2951 list_add(&prot->node, &proto_list);
2952 assign_proto_idx(prot);
2953 mutex_unlock(&proto_list_mutex);
2956 out_free_timewait_sock_slab_name:
2957 kfree(prot->twsk_prot->twsk_slab_name);
2958 out_free_request_sock_slab:
2959 req_prot_cleanup(prot->rsk_prot);
2961 kmem_cache_destroy(prot->slab);
2966 EXPORT_SYMBOL(proto_register);
2968 void proto_unregister(struct proto *prot)
2970 mutex_lock(&proto_list_mutex);
2971 release_proto_idx(prot);
2972 list_del(&prot->node);
2973 mutex_unlock(&proto_list_mutex);
2975 kmem_cache_destroy(prot->slab);
2978 req_prot_cleanup(prot->rsk_prot);
2980 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2981 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2982 kfree(prot->twsk_prot->twsk_slab_name);
2983 prot->twsk_prot->twsk_slab = NULL;
2986 EXPORT_SYMBOL(proto_unregister);
2988 #ifdef CONFIG_PROC_FS
2989 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2990 __acquires(proto_list_mutex)
2992 mutex_lock(&proto_list_mutex);
2993 return seq_list_start_head(&proto_list, *pos);
2996 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2998 return seq_list_next(v, &proto_list, pos);
3001 static void proto_seq_stop(struct seq_file *seq, void *v)
3002 __releases(proto_list_mutex)
3004 mutex_unlock(&proto_list_mutex);
3007 static char proto_method_implemented(const void *method)
3009 return method == NULL ? 'n' : 'y';
3011 static long sock_prot_memory_allocated(struct proto *proto)
3013 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3016 static char *sock_prot_memory_pressure(struct proto *proto)
3018 return proto->memory_pressure != NULL ?
3019 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3022 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3025 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3026 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3029 sock_prot_inuse_get(seq_file_net(seq), proto),
3030 sock_prot_memory_allocated(proto),
3031 sock_prot_memory_pressure(proto),
3033 proto->slab == NULL ? "no" : "yes",
3034 module_name(proto->owner),
3035 proto_method_implemented(proto->close),
3036 proto_method_implemented(proto->connect),
3037 proto_method_implemented(proto->disconnect),
3038 proto_method_implemented(proto->accept),
3039 proto_method_implemented(proto->ioctl),
3040 proto_method_implemented(proto->init),
3041 proto_method_implemented(proto->destroy),
3042 proto_method_implemented(proto->shutdown),
3043 proto_method_implemented(proto->setsockopt),
3044 proto_method_implemented(proto->getsockopt),
3045 proto_method_implemented(proto->sendmsg),
3046 proto_method_implemented(proto->recvmsg),
3047 proto_method_implemented(proto->sendpage),
3048 proto_method_implemented(proto->bind),
3049 proto_method_implemented(proto->backlog_rcv),
3050 proto_method_implemented(proto->hash),
3051 proto_method_implemented(proto->unhash),
3052 proto_method_implemented(proto->get_port),
3053 proto_method_implemented(proto->enter_memory_pressure));
3056 static int proto_seq_show(struct seq_file *seq, void *v)
3058 if (v == &proto_list)
3059 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3068 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3070 proto_seq_printf(seq, list_entry(v, struct proto, node));
3074 static const struct seq_operations proto_seq_ops = {
3075 .start = proto_seq_start,
3076 .next = proto_seq_next,
3077 .stop = proto_seq_stop,
3078 .show = proto_seq_show,
3081 static int proto_seq_open(struct inode *inode, struct file *file)
3083 return seq_open_net(inode, file, &proto_seq_ops,
3084 sizeof(struct seq_net_private));
3087 static const struct file_operations proto_seq_fops = {
3088 .owner = THIS_MODULE,
3089 .open = proto_seq_open,
3091 .llseek = seq_lseek,
3092 .release = seq_release_net,
3095 static __net_init int proto_init_net(struct net *net)
3097 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3103 static __net_exit void proto_exit_net(struct net *net)
3105 remove_proc_entry("protocols", net->proc_net);
3109 static __net_initdata struct pernet_operations proto_net_ops = {
3110 .init = proto_init_net,
3111 .exit = proto_exit_net,
3114 static int __init proto_init(void)
3116 return register_pernet_subsys(&proto_net_ops);
3119 subsys_initcall(proto_init);
3121 #endif /* PROC_FS */