2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <asm/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
138 #include <trace/events/sock.h>
144 #include <net/busy_poll.h>
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
150 * sk_ns_capable - General socket capability test
151 * @sk: Socket to use a capability on or through
152 * @user_ns: The user namespace of the capability to use
153 * @cap: The capability to use
155 * Test to see if the opener of the socket had when the socket was
156 * created and the current process has the capability @cap in the user
157 * namespace @user_ns.
159 bool sk_ns_capable(const struct sock *sk,
160 struct user_namespace *user_ns, int cap)
162 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
163 ns_capable(user_ns, cap);
165 EXPORT_SYMBOL(sk_ns_capable);
168 * sk_capable - Socket global capability test
169 * @sk: Socket to use a capability on or through
170 * @cap: The global capability to use
172 * Test to see if the opener of the socket had when the socket was
173 * created and the current process has the capability @cap in all user
176 bool sk_capable(const struct sock *sk, int cap)
178 return sk_ns_capable(sk, &init_user_ns, cap);
180 EXPORT_SYMBOL(sk_capable);
183 * sk_net_capable - Network namespace socket capability test
184 * @sk: Socket to use a capability on or through
185 * @cap: The capability to use
187 * Test to see if the opener of the socket had when the socket was created
188 * and the current process has the capability @cap over the network namespace
189 * the socket is a member of.
191 bool sk_net_capable(const struct sock *sk, int cap)
193 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195 EXPORT_SYMBOL(sk_net_capable);
198 #ifdef CONFIG_MEMCG_KMEM
199 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
204 mutex_lock(&proto_list_mutex);
205 list_for_each_entry(proto, &proto_list, node) {
206 if (proto->init_cgroup) {
207 ret = proto->init_cgroup(memcg, ss);
213 mutex_unlock(&proto_list_mutex);
216 list_for_each_entry_continue_reverse(proto, &proto_list, node)
217 if (proto->destroy_cgroup)
218 proto->destroy_cgroup(memcg);
219 mutex_unlock(&proto_list_mutex);
223 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
227 mutex_lock(&proto_list_mutex);
228 list_for_each_entry_reverse(proto, &proto_list, node)
229 if (proto->destroy_cgroup)
230 proto->destroy_cgroup(memcg);
231 mutex_unlock(&proto_list_mutex);
236 * Each address family might have different locking rules, so we have
237 * one slock key per address family:
239 static struct lock_class_key af_family_keys[AF_MAX];
240 static struct lock_class_key af_family_slock_keys[AF_MAX];
242 #if defined(CONFIG_MEMCG_KMEM)
243 struct static_key memcg_socket_limit_enabled;
244 EXPORT_SYMBOL(memcg_socket_limit_enabled);
248 * Make lock validator output more readable. (we pre-construct these
249 * strings build-time, so that runtime initialization of socket
252 static const char *const af_family_key_strings[AF_MAX+1] = {
253 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
254 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
255 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
256 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
257 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
258 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
259 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
260 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
261 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
262 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
263 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
264 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
265 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
266 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
268 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
269 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
270 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
271 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
272 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
273 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
274 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
275 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
276 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
277 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
278 "slock-27" , "slock-28" , "slock-AF_CAN" ,
279 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
280 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
281 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
282 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
284 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
285 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
286 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
287 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
288 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
289 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
290 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
291 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
292 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
293 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
294 "clock-27" , "clock-28" , "clock-AF_CAN" ,
295 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
296 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
297 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
298 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
302 * sk_callback_lock locking rules are per-address-family,
303 * so split the lock classes by using a per-AF key:
305 static struct lock_class_key af_callback_keys[AF_MAX];
307 /* Take into consideration the size of the struct sk_buff overhead in the
308 * determination of these values, since that is non-constant across
309 * platforms. This makes socket queueing behavior and performance
310 * not depend upon such differences.
312 #define _SK_MEM_PACKETS 256
313 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
314 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
317 /* Run time adjustable parameters. */
318 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
319 EXPORT_SYMBOL(sysctl_wmem_max);
320 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
321 EXPORT_SYMBOL(sysctl_rmem_max);
322 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
323 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
325 /* Maximal space eaten by iovec or ancillary data plus some space */
326 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
327 EXPORT_SYMBOL(sysctl_optmem_max);
329 int sysctl_tstamp_allow_data __read_mostly = 1;
331 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
332 EXPORT_SYMBOL_GPL(memalloc_socks);
335 * sk_set_memalloc - sets %SOCK_MEMALLOC
336 * @sk: socket to set it on
338 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
339 * It's the responsibility of the admin to adjust min_free_kbytes
340 * to meet the requirements
342 void sk_set_memalloc(struct sock *sk)
344 sock_set_flag(sk, SOCK_MEMALLOC);
345 sk->sk_allocation |= __GFP_MEMALLOC;
346 static_key_slow_inc(&memalloc_socks);
348 EXPORT_SYMBOL_GPL(sk_set_memalloc);
350 void sk_clear_memalloc(struct sock *sk)
352 sock_reset_flag(sk, SOCK_MEMALLOC);
353 sk->sk_allocation &= ~__GFP_MEMALLOC;
354 static_key_slow_dec(&memalloc_socks);
357 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
358 * progress of swapping. SOCK_MEMALLOC may be cleared while
359 * it has rmem allocations due to the last swapfile being deactivated
360 * but there is a risk that the socket is unusable due to exceeding
361 * the rmem limits. Reclaim the reserves and obey rmem limits again.
365 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
367 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
370 unsigned long pflags = current->flags;
372 /* these should have been dropped before queueing */
373 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
375 current->flags |= PF_MEMALLOC;
376 ret = sk->sk_backlog_rcv(sk, skb);
377 tsk_restore_flags(current, pflags, PF_MEMALLOC);
381 EXPORT_SYMBOL(__sk_backlog_rcv);
383 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
387 if (optlen < sizeof(tv))
389 if (copy_from_user(&tv, optval, sizeof(tv)))
391 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
395 static int warned __read_mostly;
398 if (warned < 10 && net_ratelimit()) {
400 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
401 __func__, current->comm, task_pid_nr(current));
405 *timeo_p = MAX_SCHEDULE_TIMEOUT;
406 if (tv.tv_sec == 0 && tv.tv_usec == 0)
408 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
409 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
413 static void sock_warn_obsolete_bsdism(const char *name)
416 static char warncomm[TASK_COMM_LEN];
417 if (strcmp(warncomm, current->comm) && warned < 5) {
418 strcpy(warncomm, current->comm);
419 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
425 static bool sock_needs_netstamp(const struct sock *sk)
427 switch (sk->sk_family) {
436 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
438 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
440 if (sk->sk_flags & flags) {
441 sk->sk_flags &= ~flags;
442 if (sock_needs_netstamp(sk) &&
443 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
444 net_disable_timestamp();
449 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
453 struct sk_buff_head *list = &sk->sk_receive_queue;
455 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
456 atomic_inc(&sk->sk_drops);
457 trace_sock_rcvqueue_full(sk, skb);
461 err = sk_filter(sk, skb);
465 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
466 atomic_inc(&sk->sk_drops);
471 skb_set_owner_r(skb, sk);
473 /* we escape from rcu protected region, make sure we dont leak
478 spin_lock_irqsave(&list->lock, flags);
479 sock_skb_set_dropcount(sk, skb);
480 __skb_queue_tail(list, skb);
481 spin_unlock_irqrestore(&list->lock, flags);
483 if (!sock_flag(sk, SOCK_DEAD))
484 sk->sk_data_ready(sk);
487 EXPORT_SYMBOL(sock_queue_rcv_skb);
489 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
491 int rc = NET_RX_SUCCESS;
493 if (sk_filter(sk, skb))
494 goto discard_and_relse;
498 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
499 atomic_inc(&sk->sk_drops);
500 goto discard_and_relse;
503 bh_lock_sock_nested(sk);
506 if (!sock_owned_by_user(sk)) {
508 * trylock + unlock semantics:
510 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
512 rc = sk_backlog_rcv(sk, skb);
514 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
515 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
517 atomic_inc(&sk->sk_drops);
518 goto discard_and_relse;
529 EXPORT_SYMBOL(sk_receive_skb);
531 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
533 struct dst_entry *dst = __sk_dst_get(sk);
535 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
536 sk_tx_queue_clear(sk);
537 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
544 EXPORT_SYMBOL(__sk_dst_check);
546 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
548 struct dst_entry *dst = sk_dst_get(sk);
550 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
558 EXPORT_SYMBOL(sk_dst_check);
560 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
563 int ret = -ENOPROTOOPT;
564 #ifdef CONFIG_NETDEVICES
565 struct net *net = sock_net(sk);
566 char devname[IFNAMSIZ];
571 if (!ns_capable(net->user_ns, CAP_NET_RAW))
578 /* Bind this socket to a particular device like "eth0",
579 * as specified in the passed interface name. If the
580 * name is "" or the option length is zero the socket
583 if (optlen > IFNAMSIZ - 1)
584 optlen = IFNAMSIZ - 1;
585 memset(devname, 0, sizeof(devname));
588 if (copy_from_user(devname, optval, optlen))
592 if (devname[0] != '\0') {
593 struct net_device *dev;
596 dev = dev_get_by_name_rcu(net, devname);
598 index = dev->ifindex;
606 sk->sk_bound_dev_if = index;
618 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
619 int __user *optlen, int len)
621 int ret = -ENOPROTOOPT;
622 #ifdef CONFIG_NETDEVICES
623 struct net *net = sock_net(sk);
624 char devname[IFNAMSIZ];
626 if (sk->sk_bound_dev_if == 0) {
635 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
639 len = strlen(devname) + 1;
642 if (copy_to_user(optval, devname, len))
647 if (put_user(len, optlen))
658 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
661 sock_set_flag(sk, bit);
663 sock_reset_flag(sk, bit);
666 bool sk_mc_loop(struct sock *sk)
668 if (dev_recursion_level())
672 switch (sk->sk_family) {
674 return inet_sk(sk)->mc_loop;
675 #if IS_ENABLED(CONFIG_IPV6)
677 return inet6_sk(sk)->mc_loop;
683 EXPORT_SYMBOL(sk_mc_loop);
686 * This is meant for all protocols to use and covers goings on
687 * at the socket level. Everything here is generic.
690 int sock_setsockopt(struct socket *sock, int level, int optname,
691 char __user *optval, unsigned int optlen)
693 struct sock *sk = sock->sk;
700 * Options without arguments
703 if (optname == SO_BINDTODEVICE)
704 return sock_setbindtodevice(sk, optval, optlen);
706 if (optlen < sizeof(int))
709 if (get_user(val, (int __user *)optval))
712 valbool = val ? 1 : 0;
718 if (val && !capable(CAP_NET_ADMIN))
721 sock_valbool_flag(sk, SOCK_DBG, valbool);
724 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
727 sk->sk_reuseport = valbool;
736 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
739 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
742 /* Don't error on this BSD doesn't and if you think
743 * about it this is right. Otherwise apps have to
744 * play 'guess the biggest size' games. RCVBUF/SNDBUF
745 * are treated in BSD as hints
747 val = min_t(u32, val, sysctl_wmem_max);
749 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
750 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
751 /* Wake up sending tasks if we upped the value. */
752 sk->sk_write_space(sk);
756 if (!capable(CAP_NET_ADMIN)) {
763 /* Don't error on this BSD doesn't and if you think
764 * about it this is right. Otherwise apps have to
765 * play 'guess the biggest size' games. RCVBUF/SNDBUF
766 * are treated in BSD as hints
768 val = min_t(u32, val, sysctl_rmem_max);
770 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
772 * We double it on the way in to account for
773 * "struct sk_buff" etc. overhead. Applications
774 * assume that the SO_RCVBUF setting they make will
775 * allow that much actual data to be received on that
778 * Applications are unaware that "struct sk_buff" and
779 * other overheads allocate from the receive buffer
780 * during socket buffer allocation.
782 * And after considering the possible alternatives,
783 * returning the value we actually used in getsockopt
784 * is the most desirable behavior.
786 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
790 if (!capable(CAP_NET_ADMIN)) {
798 if (sk->sk_protocol == IPPROTO_TCP &&
799 sk->sk_type == SOCK_STREAM)
800 tcp_set_keepalive(sk, valbool);
802 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
806 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
810 sk->sk_no_check_tx = valbool;
814 if ((val >= 0 && val <= 6) ||
815 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
816 sk->sk_priority = val;
822 if (optlen < sizeof(ling)) {
823 ret = -EINVAL; /* 1003.1g */
826 if (copy_from_user(&ling, optval, sizeof(ling))) {
831 sock_reset_flag(sk, SOCK_LINGER);
833 #if (BITS_PER_LONG == 32)
834 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
835 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
838 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
839 sock_set_flag(sk, SOCK_LINGER);
844 sock_warn_obsolete_bsdism("setsockopt");
849 set_bit(SOCK_PASSCRED, &sock->flags);
851 clear_bit(SOCK_PASSCRED, &sock->flags);
857 if (optname == SO_TIMESTAMP)
858 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
860 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
861 sock_set_flag(sk, SOCK_RCVTSTAMP);
862 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
864 sock_reset_flag(sk, SOCK_RCVTSTAMP);
865 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
869 case SO_TIMESTAMPING:
870 if (val & ~SOF_TIMESTAMPING_MASK) {
875 if (val & SOF_TIMESTAMPING_OPT_ID &&
876 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
877 if (sk->sk_protocol == IPPROTO_TCP) {
878 if (sk->sk_state != TCP_ESTABLISHED) {
882 sk->sk_tskey = tcp_sk(sk)->snd_una;
887 sk->sk_tsflags = val;
888 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
889 sock_enable_timestamp(sk,
890 SOCK_TIMESTAMPING_RX_SOFTWARE);
892 sock_disable_timestamp(sk,
893 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
899 sk->sk_rcvlowat = val ? : 1;
903 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
907 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
910 case SO_ATTACH_FILTER:
912 if (optlen == sizeof(struct sock_fprog)) {
913 struct sock_fprog fprog;
916 if (copy_from_user(&fprog, optval, sizeof(fprog)))
919 ret = sk_attach_filter(&fprog, sk);
925 if (optlen == sizeof(u32)) {
929 if (copy_from_user(&ufd, optval, sizeof(ufd)))
932 ret = sk_attach_bpf(ufd, sk);
936 case SO_DETACH_FILTER:
937 ret = sk_detach_filter(sk);
941 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
944 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
949 set_bit(SOCK_PASSSEC, &sock->flags);
951 clear_bit(SOCK_PASSSEC, &sock->flags);
954 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
961 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
965 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
969 if (sock->ops->set_peek_off)
970 ret = sock->ops->set_peek_off(sk, val);
976 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
979 case SO_SELECT_ERR_QUEUE:
980 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
983 #ifdef CONFIG_NET_RX_BUSY_POLL
985 /* allow unprivileged users to decrease the value */
986 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
992 sk->sk_ll_usec = val;
997 case SO_MAX_PACING_RATE:
998 sk->sk_max_pacing_rate = val;
999 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1000 sk->sk_max_pacing_rate);
1003 case SO_INCOMING_CPU:
1004 sk->sk_incoming_cpu = val;
1014 EXPORT_SYMBOL(sock_setsockopt);
1017 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1018 struct ucred *ucred)
1020 ucred->pid = pid_vnr(pid);
1021 ucred->uid = ucred->gid = -1;
1023 struct user_namespace *current_ns = current_user_ns();
1025 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1026 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1030 int sock_getsockopt(struct socket *sock, int level, int optname,
1031 char __user *optval, int __user *optlen)
1033 struct sock *sk = sock->sk;
1041 int lv = sizeof(int);
1044 if (get_user(len, optlen))
1049 memset(&v, 0, sizeof(v));
1053 v.val = sock_flag(sk, SOCK_DBG);
1057 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1061 v.val = sock_flag(sk, SOCK_BROADCAST);
1065 v.val = sk->sk_sndbuf;
1069 v.val = sk->sk_rcvbuf;
1073 v.val = sk->sk_reuse;
1077 v.val = sk->sk_reuseport;
1081 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1085 v.val = sk->sk_type;
1089 v.val = sk->sk_protocol;
1093 v.val = sk->sk_family;
1097 v.val = -sock_error(sk);
1099 v.val = xchg(&sk->sk_err_soft, 0);
1103 v.val = sock_flag(sk, SOCK_URGINLINE);
1107 v.val = sk->sk_no_check_tx;
1111 v.val = sk->sk_priority;
1115 lv = sizeof(v.ling);
1116 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1117 v.ling.l_linger = sk->sk_lingertime / HZ;
1121 sock_warn_obsolete_bsdism("getsockopt");
1125 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1126 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1129 case SO_TIMESTAMPNS:
1130 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1133 case SO_TIMESTAMPING:
1134 v.val = sk->sk_tsflags;
1138 lv = sizeof(struct timeval);
1139 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1143 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1144 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1149 lv = sizeof(struct timeval);
1150 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1154 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1155 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1160 v.val = sk->sk_rcvlowat;
1168 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1173 struct ucred peercred;
1174 if (len > sizeof(peercred))
1175 len = sizeof(peercred);
1176 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1177 if (copy_to_user(optval, &peercred, len))
1186 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1190 if (copy_to_user(optval, address, len))
1195 /* Dubious BSD thing... Probably nobody even uses it, but
1196 * the UNIX standard wants it for whatever reason... -DaveM
1199 v.val = sk->sk_state == TCP_LISTEN;
1203 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1207 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1210 v.val = sk->sk_mark;
1214 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1217 case SO_WIFI_STATUS:
1218 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1222 if (!sock->ops->set_peek_off)
1225 v.val = sk->sk_peek_off;
1228 v.val = sock_flag(sk, SOCK_NOFCS);
1231 case SO_BINDTODEVICE:
1232 return sock_getbindtodevice(sk, optval, optlen, len);
1235 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1241 case SO_LOCK_FILTER:
1242 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1245 case SO_BPF_EXTENSIONS:
1246 v.val = bpf_tell_extensions();
1249 case SO_SELECT_ERR_QUEUE:
1250 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1253 #ifdef CONFIG_NET_RX_BUSY_POLL
1255 v.val = sk->sk_ll_usec;
1259 case SO_MAX_PACING_RATE:
1260 v.val = sk->sk_max_pacing_rate;
1263 case SO_INCOMING_CPU:
1264 v.val = sk->sk_incoming_cpu;
1268 /* We implement the SO_SNDLOWAT etc to not be settable
1271 return -ENOPROTOOPT;
1276 if (copy_to_user(optval, &v, len))
1279 if (put_user(len, optlen))
1285 * Initialize an sk_lock.
1287 * (We also register the sk_lock with the lock validator.)
1289 static inline void sock_lock_init(struct sock *sk)
1291 sock_lock_init_class_and_name(sk,
1292 af_family_slock_key_strings[sk->sk_family],
1293 af_family_slock_keys + sk->sk_family,
1294 af_family_key_strings[sk->sk_family],
1295 af_family_keys + sk->sk_family);
1299 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1300 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1301 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1303 static void sock_copy(struct sock *nsk, const struct sock *osk)
1305 #ifdef CONFIG_SECURITY_NETWORK
1306 void *sptr = nsk->sk_security;
1308 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1310 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1311 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1313 #ifdef CONFIG_SECURITY_NETWORK
1314 nsk->sk_security = sptr;
1315 security_sk_clone(osk, nsk);
1319 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1321 unsigned long nulls1, nulls2;
1323 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1324 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1325 if (nulls1 > nulls2)
1326 swap(nulls1, nulls2);
1329 memset((char *)sk, 0, nulls1);
1330 memset((char *)sk + nulls1 + sizeof(void *), 0,
1331 nulls2 - nulls1 - sizeof(void *));
1332 memset((char *)sk + nulls2 + sizeof(void *), 0,
1333 size - nulls2 - sizeof(void *));
1335 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1337 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1341 struct kmem_cache *slab;
1345 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1348 if (priority & __GFP_ZERO) {
1350 prot->clear_sk(sk, prot->obj_size);
1352 sk_prot_clear_nulls(sk, prot->obj_size);
1355 sk = kmalloc(prot->obj_size, priority);
1358 kmemcheck_annotate_bitfield(sk, flags);
1360 if (security_sk_alloc(sk, family, priority))
1363 if (!try_module_get(prot->owner))
1365 sk_tx_queue_clear(sk);
1371 security_sk_free(sk);
1374 kmem_cache_free(slab, sk);
1380 static void sk_prot_free(struct proto *prot, struct sock *sk)
1382 struct kmem_cache *slab;
1383 struct module *owner;
1385 owner = prot->owner;
1388 security_sk_free(sk);
1390 kmem_cache_free(slab, sk);
1396 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1397 void sock_update_netprioidx(struct sock *sk)
1402 sk->sk_cgrp_prioidx = task_netprioidx(current);
1404 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1408 * sk_alloc - All socket objects are allocated here
1409 * @net: the applicable net namespace
1410 * @family: protocol family
1411 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1412 * @prot: struct proto associated with this new sock instance
1413 * @kern: is this to be a kernel socket?
1415 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1416 struct proto *prot, int kern)
1420 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1422 sk->sk_family = family;
1424 * See comment in struct sock definition to understand
1425 * why we need sk_prot_creator -acme
1427 sk->sk_prot = sk->sk_prot_creator = prot;
1429 sk->sk_net_refcnt = kern ? 0 : 1;
1430 if (likely(sk->sk_net_refcnt))
1432 sock_net_set(sk, net);
1433 atomic_set(&sk->sk_wmem_alloc, 1);
1435 sock_update_classid(sk);
1436 sock_update_netprioidx(sk);
1441 EXPORT_SYMBOL(sk_alloc);
1443 void sk_destruct(struct sock *sk)
1445 struct sk_filter *filter;
1447 if (sk->sk_destruct)
1448 sk->sk_destruct(sk);
1450 filter = rcu_dereference_check(sk->sk_filter,
1451 atomic_read(&sk->sk_wmem_alloc) == 0);
1453 sk_filter_uncharge(sk, filter);
1454 RCU_INIT_POINTER(sk->sk_filter, NULL);
1457 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1459 if (atomic_read(&sk->sk_omem_alloc))
1460 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1461 __func__, atomic_read(&sk->sk_omem_alloc));
1463 if (sk->sk_peer_cred)
1464 put_cred(sk->sk_peer_cred);
1465 put_pid(sk->sk_peer_pid);
1466 if (likely(sk->sk_net_refcnt))
1467 put_net(sock_net(sk));
1468 sk_prot_free(sk->sk_prot_creator, sk);
1471 static void __sk_free(struct sock *sk)
1473 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1474 sock_diag_broadcast_destroy(sk);
1479 void sk_free(struct sock *sk)
1482 * We subtract one from sk_wmem_alloc and can know if
1483 * some packets are still in some tx queue.
1484 * If not null, sock_wfree() will call __sk_free(sk) later
1486 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1489 EXPORT_SYMBOL(sk_free);
1491 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1493 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1494 sock_update_memcg(newsk);
1498 * sk_clone_lock - clone a socket, and lock its clone
1499 * @sk: the socket to clone
1500 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1502 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1504 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1507 bool is_charged = true;
1509 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1510 if (newsk != NULL) {
1511 struct sk_filter *filter;
1513 sock_copy(newsk, sk);
1516 if (likely(newsk->sk_net_refcnt))
1517 get_net(sock_net(newsk));
1518 sk_node_init(&newsk->sk_node);
1519 sock_lock_init(newsk);
1520 bh_lock_sock(newsk);
1521 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1522 newsk->sk_backlog.len = 0;
1524 atomic_set(&newsk->sk_rmem_alloc, 0);
1526 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1528 atomic_set(&newsk->sk_wmem_alloc, 1);
1529 atomic_set(&newsk->sk_omem_alloc, 0);
1530 skb_queue_head_init(&newsk->sk_receive_queue);
1531 skb_queue_head_init(&newsk->sk_write_queue);
1533 spin_lock_init(&newsk->sk_dst_lock);
1534 rwlock_init(&newsk->sk_callback_lock);
1535 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1536 af_callback_keys + newsk->sk_family,
1537 af_family_clock_key_strings[newsk->sk_family]);
1539 newsk->sk_dst_cache = NULL;
1540 newsk->sk_wmem_queued = 0;
1541 newsk->sk_forward_alloc = 0;
1542 newsk->sk_send_head = NULL;
1543 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1545 sock_reset_flag(newsk, SOCK_DONE);
1546 skb_queue_head_init(&newsk->sk_error_queue);
1548 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1550 /* though it's an empty new sock, the charging may fail
1551 * if sysctl_optmem_max was changed between creation of
1552 * original socket and cloning
1554 is_charged = sk_filter_charge(newsk, filter);
1556 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1557 /* It is still raw copy of parent, so invalidate
1558 * destructor and make plain sk_free() */
1559 newsk->sk_destruct = NULL;
1560 bh_unlock_sock(newsk);
1567 newsk->sk_priority = 0;
1568 newsk->sk_incoming_cpu = raw_smp_processor_id();
1569 atomic64_set(&newsk->sk_cookie, 0);
1571 * Before updating sk_refcnt, we must commit prior changes to memory
1572 * (Documentation/RCU/rculist_nulls.txt for details)
1575 atomic_set(&newsk->sk_refcnt, 2);
1578 * Increment the counter in the same struct proto as the master
1579 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1580 * is the same as sk->sk_prot->socks, as this field was copied
1583 * This _changes_ the previous behaviour, where
1584 * tcp_create_openreq_child always was incrementing the
1585 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1586 * to be taken into account in all callers. -acme
1588 sk_refcnt_debug_inc(newsk);
1589 sk_set_socket(newsk, NULL);
1590 newsk->sk_wq = NULL;
1592 sk_update_clone(sk, newsk);
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 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1664 if (unlikely(!sk_fullsock(sk))) {
1665 skb->destructor = sock_edemux;
1670 skb->destructor = sock_wfree;
1671 skb_set_hash_from_sk(skb, sk);
1673 * We used to take a refcount on sk, but following operation
1674 * is enough to guarantee sk_free() wont free this sock until
1675 * all in-flight packets are completed
1677 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1679 EXPORT_SYMBOL(skb_set_owner_w);
1681 void skb_orphan_partial(struct sk_buff *skb)
1683 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1684 * so we do not completely orphan skb, but transfert all
1685 * accounted bytes but one, to avoid unexpected reorders.
1687 if (skb->destructor == sock_wfree
1689 || skb->destructor == tcp_wfree
1692 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1698 EXPORT_SYMBOL(skb_orphan_partial);
1701 * Read buffer destructor automatically called from kfree_skb.
1703 void sock_rfree(struct sk_buff *skb)
1705 struct sock *sk = skb->sk;
1706 unsigned int len = skb->truesize;
1708 atomic_sub(len, &sk->sk_rmem_alloc);
1709 sk_mem_uncharge(sk, len);
1711 EXPORT_SYMBOL(sock_rfree);
1714 * Buffer destructor for skbs that are not used directly in read or write
1715 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1717 void sock_efree(struct sk_buff *skb)
1721 EXPORT_SYMBOL(sock_efree);
1723 kuid_t sock_i_uid(struct sock *sk)
1727 read_lock_bh(&sk->sk_callback_lock);
1728 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1729 read_unlock_bh(&sk->sk_callback_lock);
1732 EXPORT_SYMBOL(sock_i_uid);
1734 unsigned long sock_i_ino(struct sock *sk)
1738 read_lock_bh(&sk->sk_callback_lock);
1739 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1740 read_unlock_bh(&sk->sk_callback_lock);
1743 EXPORT_SYMBOL(sock_i_ino);
1746 * Allocate a skb from the socket's send buffer.
1748 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1751 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1752 struct sk_buff *skb = alloc_skb(size, priority);
1754 skb_set_owner_w(skb, sk);
1760 EXPORT_SYMBOL(sock_wmalloc);
1763 * Allocate a memory block from the socket's option memory buffer.
1765 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1767 if ((unsigned int)size <= sysctl_optmem_max &&
1768 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1770 /* First do the add, to avoid the race if kmalloc
1773 atomic_add(size, &sk->sk_omem_alloc);
1774 mem = kmalloc(size, priority);
1777 atomic_sub(size, &sk->sk_omem_alloc);
1781 EXPORT_SYMBOL(sock_kmalloc);
1783 /* Free an option memory block. Note, we actually want the inline
1784 * here as this allows gcc to detect the nullify and fold away the
1785 * condition entirely.
1787 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1790 if (WARN_ON_ONCE(!mem))
1796 atomic_sub(size, &sk->sk_omem_alloc);
1799 void sock_kfree_s(struct sock *sk, void *mem, int size)
1801 __sock_kfree_s(sk, mem, size, false);
1803 EXPORT_SYMBOL(sock_kfree_s);
1805 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1807 __sock_kfree_s(sk, mem, size, true);
1809 EXPORT_SYMBOL(sock_kzfree_s);
1811 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1812 I think, these locks should be removed for datagram sockets.
1814 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1818 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1822 if (signal_pending(current))
1824 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1825 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1826 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1828 if (sk->sk_shutdown & SEND_SHUTDOWN)
1832 timeo = schedule_timeout(timeo);
1834 finish_wait(sk_sleep(sk), &wait);
1840 * Generic send/receive buffer handlers
1843 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1844 unsigned long data_len, int noblock,
1845 int *errcode, int max_page_order)
1847 struct sk_buff *skb;
1851 timeo = sock_sndtimeo(sk, noblock);
1853 err = sock_error(sk);
1858 if (sk->sk_shutdown & SEND_SHUTDOWN)
1861 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1864 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1865 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1869 if (signal_pending(current))
1871 timeo = sock_wait_for_wmem(sk, timeo);
1873 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1874 errcode, sk->sk_allocation);
1876 skb_set_owner_w(skb, sk);
1880 err = sock_intr_errno(timeo);
1885 EXPORT_SYMBOL(sock_alloc_send_pskb);
1887 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1888 int noblock, int *errcode)
1890 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1892 EXPORT_SYMBOL(sock_alloc_send_skb);
1894 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1895 struct sockcm_cookie *sockc)
1897 struct cmsghdr *cmsg;
1899 for_each_cmsghdr(cmsg, msg) {
1900 if (!CMSG_OK(msg, cmsg))
1902 if (cmsg->cmsg_level != SOL_SOCKET)
1904 switch (cmsg->cmsg_type) {
1906 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1908 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1910 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1918 EXPORT_SYMBOL(sock_cmsg_send);
1920 /* On 32bit arches, an skb frag is limited to 2^15 */
1921 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1924 * skb_page_frag_refill - check that a page_frag contains enough room
1925 * @sz: minimum size of the fragment we want to get
1926 * @pfrag: pointer to page_frag
1927 * @gfp: priority for memory allocation
1929 * Note: While this allocator tries to use high order pages, there is
1930 * no guarantee that allocations succeed. Therefore, @sz MUST be
1931 * less or equal than PAGE_SIZE.
1933 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1936 if (atomic_read(&pfrag->page->_count) == 1) {
1940 if (pfrag->offset + sz <= pfrag->size)
1942 put_page(pfrag->page);
1946 if (SKB_FRAG_PAGE_ORDER) {
1947 pfrag->page = alloc_pages((gfp & ~__GFP_WAIT) | __GFP_COMP |
1948 __GFP_NOWARN | __GFP_NORETRY,
1949 SKB_FRAG_PAGE_ORDER);
1950 if (likely(pfrag->page)) {
1951 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1955 pfrag->page = alloc_page(gfp);
1956 if (likely(pfrag->page)) {
1957 pfrag->size = PAGE_SIZE;
1962 EXPORT_SYMBOL(skb_page_frag_refill);
1964 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1966 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1969 sk_enter_memory_pressure(sk);
1970 sk_stream_moderate_sndbuf(sk);
1973 EXPORT_SYMBOL(sk_page_frag_refill);
1975 static void __lock_sock(struct sock *sk)
1976 __releases(&sk->sk_lock.slock)
1977 __acquires(&sk->sk_lock.slock)
1982 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1983 TASK_UNINTERRUPTIBLE);
1984 spin_unlock_bh(&sk->sk_lock.slock);
1986 spin_lock_bh(&sk->sk_lock.slock);
1987 if (!sock_owned_by_user(sk))
1990 finish_wait(&sk->sk_lock.wq, &wait);
1993 static void __release_sock(struct sock *sk)
1994 __releases(&sk->sk_lock.slock)
1995 __acquires(&sk->sk_lock.slock)
1997 struct sk_buff *skb = sk->sk_backlog.head;
2000 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2004 struct sk_buff *next = skb->next;
2007 WARN_ON_ONCE(skb_dst_is_noref(skb));
2009 sk_backlog_rcv(sk, skb);
2012 * We are in process context here with softirqs
2013 * disabled, use cond_resched_softirq() to preempt.
2014 * This is safe to do because we've taken the backlog
2017 cond_resched_softirq();
2020 } while (skb != NULL);
2023 } while ((skb = sk->sk_backlog.head) != NULL);
2026 * Doing the zeroing here guarantee we can not loop forever
2027 * while a wild producer attempts to flood us.
2029 sk->sk_backlog.len = 0;
2033 * sk_wait_data - wait for data to arrive at sk_receive_queue
2034 * @sk: sock to wait on
2035 * @timeo: for how long
2036 * @skb: last skb seen on sk_receive_queue
2038 * Now socket state including sk->sk_err is changed only under lock,
2039 * hence we may omit checks after joining wait queue.
2040 * We check receive queue before schedule() only as optimization;
2041 * it is very likely that release_sock() added new data.
2043 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2048 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2049 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2050 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2051 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2052 finish_wait(sk_sleep(sk), &wait);
2055 EXPORT_SYMBOL(sk_wait_data);
2058 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2060 * @size: memory size to allocate
2061 * @kind: allocation type
2063 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2064 * rmem allocation. This function assumes that protocols which have
2065 * memory_pressure use sk_wmem_queued as write buffer accounting.
2067 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2069 struct proto *prot = sk->sk_prot;
2070 int amt = sk_mem_pages(size);
2072 int parent_status = UNDER_LIMIT;
2074 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2076 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2079 if (parent_status == UNDER_LIMIT &&
2080 allocated <= sk_prot_mem_limits(sk, 0)) {
2081 sk_leave_memory_pressure(sk);
2085 /* Under pressure. (we or our parents) */
2086 if ((parent_status > SOFT_LIMIT) ||
2087 allocated > sk_prot_mem_limits(sk, 1))
2088 sk_enter_memory_pressure(sk);
2090 /* Over hard limit (we or our parents) */
2091 if ((parent_status == OVER_LIMIT) ||
2092 (allocated > sk_prot_mem_limits(sk, 2)))
2093 goto suppress_allocation;
2095 /* guarantee minimum buffer size under pressure */
2096 if (kind == SK_MEM_RECV) {
2097 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2100 } else { /* SK_MEM_SEND */
2101 if (sk->sk_type == SOCK_STREAM) {
2102 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2104 } else if (atomic_read(&sk->sk_wmem_alloc) <
2105 prot->sysctl_wmem[0])
2109 if (sk_has_memory_pressure(sk)) {
2112 if (!sk_under_memory_pressure(sk))
2114 alloc = sk_sockets_allocated_read_positive(sk);
2115 if (sk_prot_mem_limits(sk, 2) > alloc *
2116 sk_mem_pages(sk->sk_wmem_queued +
2117 atomic_read(&sk->sk_rmem_alloc) +
2118 sk->sk_forward_alloc))
2122 suppress_allocation:
2124 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2125 sk_stream_moderate_sndbuf(sk);
2127 /* Fail only if socket is _under_ its sndbuf.
2128 * In this case we cannot block, so that we have to fail.
2130 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2134 trace_sock_exceed_buf_limit(sk, prot, allocated);
2136 /* Alas. Undo changes. */
2137 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2139 sk_memory_allocated_sub(sk, amt);
2143 EXPORT_SYMBOL(__sk_mem_schedule);
2146 * __sk_mem_reclaim - reclaim memory_allocated
2148 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2150 void __sk_mem_reclaim(struct sock *sk, int amount)
2152 amount >>= SK_MEM_QUANTUM_SHIFT;
2153 sk_memory_allocated_sub(sk, amount);
2154 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2156 if (sk_under_memory_pressure(sk) &&
2157 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2158 sk_leave_memory_pressure(sk);
2160 EXPORT_SYMBOL(__sk_mem_reclaim);
2164 * Set of default routines for initialising struct proto_ops when
2165 * the protocol does not support a particular function. In certain
2166 * cases where it makes no sense for a protocol to have a "do nothing"
2167 * function, some default processing is provided.
2170 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2174 EXPORT_SYMBOL(sock_no_bind);
2176 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2181 EXPORT_SYMBOL(sock_no_connect);
2183 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2187 EXPORT_SYMBOL(sock_no_socketpair);
2189 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2193 EXPORT_SYMBOL(sock_no_accept);
2195 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2200 EXPORT_SYMBOL(sock_no_getname);
2202 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2206 EXPORT_SYMBOL(sock_no_poll);
2208 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2212 EXPORT_SYMBOL(sock_no_ioctl);
2214 int sock_no_listen(struct socket *sock, int backlog)
2218 EXPORT_SYMBOL(sock_no_listen);
2220 int sock_no_shutdown(struct socket *sock, int how)
2224 EXPORT_SYMBOL(sock_no_shutdown);
2226 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2227 char __user *optval, unsigned int optlen)
2231 EXPORT_SYMBOL(sock_no_setsockopt);
2233 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2234 char __user *optval, int __user *optlen)
2238 EXPORT_SYMBOL(sock_no_getsockopt);
2240 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2244 EXPORT_SYMBOL(sock_no_sendmsg);
2246 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2251 EXPORT_SYMBOL(sock_no_recvmsg);
2253 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2255 /* Mirror missing mmap method error code */
2258 EXPORT_SYMBOL(sock_no_mmap);
2260 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2263 struct msghdr msg = {.msg_flags = flags};
2265 char *kaddr = kmap(page);
2266 iov.iov_base = kaddr + offset;
2268 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2272 EXPORT_SYMBOL(sock_no_sendpage);
2275 * Default Socket Callbacks
2278 static void sock_def_wakeup(struct sock *sk)
2280 struct socket_wq *wq;
2283 wq = rcu_dereference(sk->sk_wq);
2284 if (wq_has_sleeper(wq))
2285 wake_up_interruptible_all(&wq->wait);
2289 static void sock_def_error_report(struct sock *sk)
2291 struct socket_wq *wq;
2294 wq = rcu_dereference(sk->sk_wq);
2295 if (wq_has_sleeper(wq))
2296 wake_up_interruptible_poll(&wq->wait, POLLERR);
2297 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2301 static void sock_def_readable(struct sock *sk)
2303 struct socket_wq *wq;
2306 wq = rcu_dereference(sk->sk_wq);
2307 if (wq_has_sleeper(wq))
2308 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2309 POLLRDNORM | POLLRDBAND);
2310 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2314 static void sock_def_write_space(struct sock *sk)
2316 struct socket_wq *wq;
2320 /* Do not wake up a writer until he can make "significant"
2323 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2324 wq = rcu_dereference(sk->sk_wq);
2325 if (wq_has_sleeper(wq))
2326 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2327 POLLWRNORM | POLLWRBAND);
2329 /* Should agree with poll, otherwise some programs break */
2330 if (sock_writeable(sk))
2331 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2337 static void sock_def_destruct(struct sock *sk)
2341 void sk_send_sigurg(struct sock *sk)
2343 if (sk->sk_socket && sk->sk_socket->file)
2344 if (send_sigurg(&sk->sk_socket->file->f_owner))
2345 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2347 EXPORT_SYMBOL(sk_send_sigurg);
2349 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2350 unsigned long expires)
2352 if (!mod_timer(timer, expires))
2355 EXPORT_SYMBOL(sk_reset_timer);
2357 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2359 if (del_timer(timer))
2362 EXPORT_SYMBOL(sk_stop_timer);
2364 void sock_init_data(struct socket *sock, struct sock *sk)
2366 skb_queue_head_init(&sk->sk_receive_queue);
2367 skb_queue_head_init(&sk->sk_write_queue);
2368 skb_queue_head_init(&sk->sk_error_queue);
2370 sk->sk_send_head = NULL;
2372 init_timer(&sk->sk_timer);
2374 sk->sk_allocation = GFP_KERNEL;
2375 sk->sk_rcvbuf = sysctl_rmem_default;
2376 sk->sk_sndbuf = sysctl_wmem_default;
2377 sk->sk_state = TCP_CLOSE;
2378 sk_set_socket(sk, sock);
2380 sock_set_flag(sk, SOCK_ZAPPED);
2383 sk->sk_type = sock->type;
2384 sk->sk_wq = sock->wq;
2389 spin_lock_init(&sk->sk_dst_lock);
2390 rwlock_init(&sk->sk_callback_lock);
2391 lockdep_set_class_and_name(&sk->sk_callback_lock,
2392 af_callback_keys + sk->sk_family,
2393 af_family_clock_key_strings[sk->sk_family]);
2395 sk->sk_state_change = sock_def_wakeup;
2396 sk->sk_data_ready = sock_def_readable;
2397 sk->sk_write_space = sock_def_write_space;
2398 sk->sk_error_report = sock_def_error_report;
2399 sk->sk_destruct = sock_def_destruct;
2401 sk->sk_frag.page = NULL;
2402 sk->sk_frag.offset = 0;
2403 sk->sk_peek_off = -1;
2405 sk->sk_peer_pid = NULL;
2406 sk->sk_peer_cred = NULL;
2407 sk->sk_write_pending = 0;
2408 sk->sk_rcvlowat = 1;
2409 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2410 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2412 sk->sk_stamp = ktime_set(-1L, 0);
2414 #ifdef CONFIG_NET_RX_BUSY_POLL
2416 sk->sk_ll_usec = sysctl_net_busy_read;
2419 sk->sk_max_pacing_rate = ~0U;
2420 sk->sk_pacing_rate = ~0U;
2421 sk->sk_incoming_cpu = -1;
2423 * Before updating sk_refcnt, we must commit prior changes to memory
2424 * (Documentation/RCU/rculist_nulls.txt for details)
2427 atomic_set(&sk->sk_refcnt, 1);
2428 atomic_set(&sk->sk_drops, 0);
2430 EXPORT_SYMBOL(sock_init_data);
2432 void lock_sock_nested(struct sock *sk, int subclass)
2435 spin_lock_bh(&sk->sk_lock.slock);
2436 if (sk->sk_lock.owned)
2438 sk->sk_lock.owned = 1;
2439 spin_unlock(&sk->sk_lock.slock);
2441 * The sk_lock has mutex_lock() semantics here:
2443 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2446 EXPORT_SYMBOL(lock_sock_nested);
2448 void release_sock(struct sock *sk)
2451 * The sk_lock has mutex_unlock() semantics:
2453 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2455 spin_lock_bh(&sk->sk_lock.slock);
2456 if (sk->sk_backlog.tail)
2459 /* Warning : release_cb() might need to release sk ownership,
2460 * ie call sock_release_ownership(sk) before us.
2462 if (sk->sk_prot->release_cb)
2463 sk->sk_prot->release_cb(sk);
2465 sock_release_ownership(sk);
2466 if (waitqueue_active(&sk->sk_lock.wq))
2467 wake_up(&sk->sk_lock.wq);
2468 spin_unlock_bh(&sk->sk_lock.slock);
2470 EXPORT_SYMBOL(release_sock);
2473 * lock_sock_fast - fast version of lock_sock
2476 * This version should be used for very small section, where process wont block
2477 * return false if fast path is taken
2478 * sk_lock.slock locked, owned = 0, BH disabled
2479 * return true if slow path is taken
2480 * sk_lock.slock unlocked, owned = 1, BH enabled
2482 bool lock_sock_fast(struct sock *sk)
2485 spin_lock_bh(&sk->sk_lock.slock);
2487 if (!sk->sk_lock.owned)
2489 * Note : We must disable BH
2494 sk->sk_lock.owned = 1;
2495 spin_unlock(&sk->sk_lock.slock);
2497 * The sk_lock has mutex_lock() semantics here:
2499 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2503 EXPORT_SYMBOL(lock_sock_fast);
2505 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2508 if (!sock_flag(sk, SOCK_TIMESTAMP))
2509 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2510 tv = ktime_to_timeval(sk->sk_stamp);
2511 if (tv.tv_sec == -1)
2513 if (tv.tv_sec == 0) {
2514 sk->sk_stamp = ktime_get_real();
2515 tv = ktime_to_timeval(sk->sk_stamp);
2517 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2519 EXPORT_SYMBOL(sock_get_timestamp);
2521 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2524 if (!sock_flag(sk, SOCK_TIMESTAMP))
2525 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2526 ts = ktime_to_timespec(sk->sk_stamp);
2527 if (ts.tv_sec == -1)
2529 if (ts.tv_sec == 0) {
2530 sk->sk_stamp = ktime_get_real();
2531 ts = ktime_to_timespec(sk->sk_stamp);
2533 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2535 EXPORT_SYMBOL(sock_get_timestampns);
2537 void sock_enable_timestamp(struct sock *sk, int flag)
2539 if (!sock_flag(sk, flag)) {
2540 unsigned long previous_flags = sk->sk_flags;
2542 sock_set_flag(sk, flag);
2544 * we just set one of the two flags which require net
2545 * time stamping, but time stamping might have been on
2546 * already because of the other one
2548 if (sock_needs_netstamp(sk) &&
2549 !(previous_flags & SK_FLAGS_TIMESTAMP))
2550 net_enable_timestamp();
2554 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2555 int level, int type)
2557 struct sock_exterr_skb *serr;
2558 struct sk_buff *skb;
2562 skb = sock_dequeue_err_skb(sk);
2568 msg->msg_flags |= MSG_TRUNC;
2571 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2575 sock_recv_timestamp(msg, sk, skb);
2577 serr = SKB_EXT_ERR(skb);
2578 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2580 msg->msg_flags |= MSG_ERRQUEUE;
2588 EXPORT_SYMBOL(sock_recv_errqueue);
2591 * Get a socket option on an socket.
2593 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2594 * asynchronous errors should be reported by getsockopt. We assume
2595 * this means if you specify SO_ERROR (otherwise whats the point of it).
2597 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2598 char __user *optval, int __user *optlen)
2600 struct sock *sk = sock->sk;
2602 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2604 EXPORT_SYMBOL(sock_common_getsockopt);
2606 #ifdef CONFIG_COMPAT
2607 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2608 char __user *optval, int __user *optlen)
2610 struct sock *sk = sock->sk;
2612 if (sk->sk_prot->compat_getsockopt != NULL)
2613 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2615 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2617 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2620 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2623 struct sock *sk = sock->sk;
2627 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2628 flags & ~MSG_DONTWAIT, &addr_len);
2630 msg->msg_namelen = addr_len;
2633 EXPORT_SYMBOL(sock_common_recvmsg);
2636 * Set socket options on an inet socket.
2638 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2639 char __user *optval, unsigned int optlen)
2641 struct sock *sk = sock->sk;
2643 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2645 EXPORT_SYMBOL(sock_common_setsockopt);
2647 #ifdef CONFIG_COMPAT
2648 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2649 char __user *optval, unsigned int optlen)
2651 struct sock *sk = sock->sk;
2653 if (sk->sk_prot->compat_setsockopt != NULL)
2654 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2656 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2658 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2661 void sk_common_release(struct sock *sk)
2663 if (sk->sk_prot->destroy)
2664 sk->sk_prot->destroy(sk);
2667 * Observation: when sock_common_release is called, processes have
2668 * no access to socket. But net still has.
2669 * Step one, detach it from networking:
2671 * A. Remove from hash tables.
2674 sk->sk_prot->unhash(sk);
2677 * In this point socket cannot receive new packets, but it is possible
2678 * that some packets are in flight because some CPU runs receiver and
2679 * did hash table lookup before we unhashed socket. They will achieve
2680 * receive queue and will be purged by socket destructor.
2682 * Also we still have packets pending on receive queue and probably,
2683 * our own packets waiting in device queues. sock_destroy will drain
2684 * receive queue, but transmitted packets will delay socket destruction
2685 * until the last reference will be released.
2690 xfrm_sk_free_policy(sk);
2692 sk_refcnt_debug_release(sk);
2694 if (sk->sk_frag.page) {
2695 put_page(sk->sk_frag.page);
2696 sk->sk_frag.page = NULL;
2701 EXPORT_SYMBOL(sk_common_release);
2703 #ifdef CONFIG_PROC_FS
2704 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2706 int val[PROTO_INUSE_NR];
2709 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2711 #ifdef CONFIG_NET_NS
2712 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2714 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2716 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2718 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2720 int cpu, idx = prot->inuse_idx;
2723 for_each_possible_cpu(cpu)
2724 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2726 return res >= 0 ? res : 0;
2728 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2730 static int __net_init sock_inuse_init_net(struct net *net)
2732 net->core.inuse = alloc_percpu(struct prot_inuse);
2733 return net->core.inuse ? 0 : -ENOMEM;
2736 static void __net_exit sock_inuse_exit_net(struct net *net)
2738 free_percpu(net->core.inuse);
2741 static struct pernet_operations net_inuse_ops = {
2742 .init = sock_inuse_init_net,
2743 .exit = sock_inuse_exit_net,
2746 static __init int net_inuse_init(void)
2748 if (register_pernet_subsys(&net_inuse_ops))
2749 panic("Cannot initialize net inuse counters");
2754 core_initcall(net_inuse_init);
2756 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2758 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2760 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2762 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2764 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2766 int cpu, idx = prot->inuse_idx;
2769 for_each_possible_cpu(cpu)
2770 res += per_cpu(prot_inuse, cpu).val[idx];
2772 return res >= 0 ? res : 0;
2774 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2777 static void assign_proto_idx(struct proto *prot)
2779 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2781 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2782 pr_err("PROTO_INUSE_NR exhausted\n");
2786 set_bit(prot->inuse_idx, proto_inuse_idx);
2789 static void release_proto_idx(struct proto *prot)
2791 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2792 clear_bit(prot->inuse_idx, proto_inuse_idx);
2795 static inline void assign_proto_idx(struct proto *prot)
2799 static inline void release_proto_idx(struct proto *prot)
2804 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2808 kfree(rsk_prot->slab_name);
2809 rsk_prot->slab_name = NULL;
2810 kmem_cache_destroy(rsk_prot->slab);
2811 rsk_prot->slab = NULL;
2814 static int req_prot_init(const struct proto *prot)
2816 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2821 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2823 if (!rsk_prot->slab_name)
2826 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2827 rsk_prot->obj_size, 0,
2828 prot->slab_flags, NULL);
2830 if (!rsk_prot->slab) {
2831 pr_crit("%s: Can't create request sock SLAB cache!\n",
2838 int proto_register(struct proto *prot, int alloc_slab)
2841 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2842 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2845 if (prot->slab == NULL) {
2846 pr_crit("%s: Can't create sock SLAB cache!\n",
2851 if (req_prot_init(prot))
2852 goto out_free_request_sock_slab;
2854 if (prot->twsk_prot != NULL) {
2855 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2857 if (prot->twsk_prot->twsk_slab_name == NULL)
2858 goto out_free_request_sock_slab;
2860 prot->twsk_prot->twsk_slab =
2861 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2862 prot->twsk_prot->twsk_obj_size,
2866 if (prot->twsk_prot->twsk_slab == NULL)
2867 goto out_free_timewait_sock_slab_name;
2871 mutex_lock(&proto_list_mutex);
2872 list_add(&prot->node, &proto_list);
2873 assign_proto_idx(prot);
2874 mutex_unlock(&proto_list_mutex);
2877 out_free_timewait_sock_slab_name:
2878 kfree(prot->twsk_prot->twsk_slab_name);
2879 out_free_request_sock_slab:
2880 req_prot_cleanup(prot->rsk_prot);
2882 kmem_cache_destroy(prot->slab);
2887 EXPORT_SYMBOL(proto_register);
2889 void proto_unregister(struct proto *prot)
2891 mutex_lock(&proto_list_mutex);
2892 release_proto_idx(prot);
2893 list_del(&prot->node);
2894 mutex_unlock(&proto_list_mutex);
2896 kmem_cache_destroy(prot->slab);
2899 req_prot_cleanup(prot->rsk_prot);
2901 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2902 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2903 kfree(prot->twsk_prot->twsk_slab_name);
2904 prot->twsk_prot->twsk_slab = NULL;
2907 EXPORT_SYMBOL(proto_unregister);
2909 #ifdef CONFIG_PROC_FS
2910 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2911 __acquires(proto_list_mutex)
2913 mutex_lock(&proto_list_mutex);
2914 return seq_list_start_head(&proto_list, *pos);
2917 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2919 return seq_list_next(v, &proto_list, pos);
2922 static void proto_seq_stop(struct seq_file *seq, void *v)
2923 __releases(proto_list_mutex)
2925 mutex_unlock(&proto_list_mutex);
2928 static char proto_method_implemented(const void *method)
2930 return method == NULL ? 'n' : 'y';
2932 static long sock_prot_memory_allocated(struct proto *proto)
2934 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2937 static char *sock_prot_memory_pressure(struct proto *proto)
2939 return proto->memory_pressure != NULL ?
2940 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2943 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2946 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2947 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2950 sock_prot_inuse_get(seq_file_net(seq), proto),
2951 sock_prot_memory_allocated(proto),
2952 sock_prot_memory_pressure(proto),
2954 proto->slab == NULL ? "no" : "yes",
2955 module_name(proto->owner),
2956 proto_method_implemented(proto->close),
2957 proto_method_implemented(proto->connect),
2958 proto_method_implemented(proto->disconnect),
2959 proto_method_implemented(proto->accept),
2960 proto_method_implemented(proto->ioctl),
2961 proto_method_implemented(proto->init),
2962 proto_method_implemented(proto->destroy),
2963 proto_method_implemented(proto->shutdown),
2964 proto_method_implemented(proto->setsockopt),
2965 proto_method_implemented(proto->getsockopt),
2966 proto_method_implemented(proto->sendmsg),
2967 proto_method_implemented(proto->recvmsg),
2968 proto_method_implemented(proto->sendpage),
2969 proto_method_implemented(proto->bind),
2970 proto_method_implemented(proto->backlog_rcv),
2971 proto_method_implemented(proto->hash),
2972 proto_method_implemented(proto->unhash),
2973 proto_method_implemented(proto->get_port),
2974 proto_method_implemented(proto->enter_memory_pressure));
2977 static int proto_seq_show(struct seq_file *seq, void *v)
2979 if (v == &proto_list)
2980 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2989 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2991 proto_seq_printf(seq, list_entry(v, struct proto, node));
2995 static const struct seq_operations proto_seq_ops = {
2996 .start = proto_seq_start,
2997 .next = proto_seq_next,
2998 .stop = proto_seq_stop,
2999 .show = proto_seq_show,
3002 static int proto_seq_open(struct inode *inode, struct file *file)
3004 return seq_open_net(inode, file, &proto_seq_ops,
3005 sizeof(struct seq_net_private));
3008 static const struct file_operations proto_seq_fops = {
3009 .owner = THIS_MODULE,
3010 .open = proto_seq_open,
3012 .llseek = seq_lseek,
3013 .release = seq_release_net,
3016 static __net_init int proto_init_net(struct net *net)
3018 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3024 static __net_exit void proto_exit_net(struct net *net)
3026 remove_proc_entry("protocols", net->proc_net);
3030 static __net_initdata struct pernet_operations proto_net_ops = {
3031 .init = proto_init_net,
3032 .exit = proto_exit_net,
3035 static int __init proto_init(void)
3037 return register_pernet_subsys(&proto_net_ops);
3040 subsys_initcall(proto_init);
3042 #endif /* PROC_FS */