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>
135 #include <linux/filter.h>
137 #include <trace/events/sock.h>
143 #include <net/busy_poll.h>
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
149 * sk_ns_capable - General socket capability test
150 * @sk: Socket to use a capability on or through
151 * @user_ns: The user namespace of the capability to use
152 * @cap: The capability to use
154 * Test to see if the opener of the socket had when the socket was
155 * created and the current process has the capability @cap in the user
156 * namespace @user_ns.
158 bool sk_ns_capable(const struct sock *sk,
159 struct user_namespace *user_ns, int cap)
161 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
162 ns_capable(user_ns, cap);
164 EXPORT_SYMBOL(sk_ns_capable);
167 * sk_capable - Socket global capability test
168 * @sk: Socket to use a capability on or through
169 * @cap: The global capbility to use
171 * Test to see if the opener of the socket had when the socket was
172 * created and the current process has the capability @cap in all user
175 bool sk_capable(const struct sock *sk, int cap)
177 return sk_ns_capable(sk, &init_user_ns, cap);
179 EXPORT_SYMBOL(sk_capable);
182 * sk_net_capable - Network namespace socket capability test
183 * @sk: Socket to use a capability on or through
184 * @cap: The capability to use
186 * Test to see if the opener of the socket had when the socke was created
187 * and the current process has the capability @cap over the network namespace
188 * the socket is a member of.
190 bool sk_net_capable(const struct sock *sk, int cap)
192 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
194 EXPORT_SYMBOL(sk_net_capable);
197 #ifdef CONFIG_MEMCG_KMEM
198 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
203 mutex_lock(&proto_list_mutex);
204 list_for_each_entry(proto, &proto_list, node) {
205 if (proto->init_cgroup) {
206 ret = proto->init_cgroup(memcg, ss);
212 mutex_unlock(&proto_list_mutex);
215 list_for_each_entry_continue_reverse(proto, &proto_list, node)
216 if (proto->destroy_cgroup)
217 proto->destroy_cgroup(memcg);
218 mutex_unlock(&proto_list_mutex);
222 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
226 mutex_lock(&proto_list_mutex);
227 list_for_each_entry_reverse(proto, &proto_list, node)
228 if (proto->destroy_cgroup)
229 proto->destroy_cgroup(memcg);
230 mutex_unlock(&proto_list_mutex);
235 * Each address family might have different locking rules, so we have
236 * one slock key per address family:
238 static struct lock_class_key af_family_keys[AF_MAX];
239 static struct lock_class_key af_family_slock_keys[AF_MAX];
241 #if defined(CONFIG_MEMCG_KMEM)
242 struct static_key memcg_socket_limit_enabled;
243 EXPORT_SYMBOL(memcg_socket_limit_enabled);
247 * Make lock validator output more readable. (we pre-construct these
248 * strings build-time, so that runtime initialization of socket
251 static const char *const af_family_key_strings[AF_MAX+1] = {
252 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
253 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
254 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
255 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
256 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
257 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
258 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
259 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
260 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
261 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
262 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
263 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
264 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
265 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
267 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
268 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
269 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
270 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
271 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
272 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
273 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
274 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
275 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
276 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
277 "slock-27" , "slock-28" , "slock-AF_CAN" ,
278 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
279 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
280 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
281 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
283 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
284 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
285 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
286 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
287 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
288 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
289 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
290 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
291 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
292 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
293 "clock-27" , "clock-28" , "clock-AF_CAN" ,
294 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
295 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
296 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
297 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
301 * sk_callback_lock locking rules are per-address-family,
302 * so split the lock classes by using a per-AF key:
304 static struct lock_class_key af_callback_keys[AF_MAX];
306 /* Take into consideration the size of the struct sk_buff overhead in the
307 * determination of these values, since that is non-constant across
308 * platforms. This makes socket queueing behavior and performance
309 * not depend upon such differences.
311 #define _SK_MEM_PACKETS 256
312 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
313 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
314 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
316 /* Run time adjustable parameters. */
317 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
318 EXPORT_SYMBOL(sysctl_wmem_max);
319 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
320 EXPORT_SYMBOL(sysctl_rmem_max);
321 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
322 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
324 /* Maximal space eaten by iovec or ancillary data plus some space */
325 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
326 EXPORT_SYMBOL(sysctl_optmem_max);
328 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
329 EXPORT_SYMBOL_GPL(memalloc_socks);
332 * sk_set_memalloc - sets %SOCK_MEMALLOC
333 * @sk: socket to set it on
335 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
336 * It's the responsibility of the admin to adjust min_free_kbytes
337 * to meet the requirements
339 void sk_set_memalloc(struct sock *sk)
341 sock_set_flag(sk, SOCK_MEMALLOC);
342 sk->sk_allocation |= __GFP_MEMALLOC;
343 static_key_slow_inc(&memalloc_socks);
345 EXPORT_SYMBOL_GPL(sk_set_memalloc);
347 void sk_clear_memalloc(struct sock *sk)
349 sock_reset_flag(sk, SOCK_MEMALLOC);
350 sk->sk_allocation &= ~__GFP_MEMALLOC;
351 static_key_slow_dec(&memalloc_socks);
354 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
355 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
356 * it has rmem allocations there is a risk that the user of the
357 * socket cannot make forward progress due to exceeding the rmem
358 * limits. By rights, sk_clear_memalloc() should only be called
359 * on sockets being torn down but warn and reset the accounting if
360 * that assumption breaks.
362 if (WARN_ON(sk->sk_forward_alloc))
365 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
367 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
370 unsigned long pflags = current->flags;
372 /* these should have been dropped before queueing */
373 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
375 current->flags |= PF_MEMALLOC;
376 ret = sk->sk_backlog_rcv(sk, skb);
377 tsk_restore_flags(current, pflags, PF_MEMALLOC);
381 EXPORT_SYMBOL(__sk_backlog_rcv);
383 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
387 if (optlen < sizeof(tv))
389 if (copy_from_user(&tv, optval, sizeof(tv)))
391 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
395 static int warned __read_mostly;
398 if (warned < 10 && net_ratelimit()) {
400 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
401 __func__, current->comm, task_pid_nr(current));
405 *timeo_p = MAX_SCHEDULE_TIMEOUT;
406 if (tv.tv_sec == 0 && tv.tv_usec == 0)
408 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
409 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
413 static void sock_warn_obsolete_bsdism(const char *name)
416 static char warncomm[TASK_COMM_LEN];
417 if (strcmp(warncomm, current->comm) && warned < 5) {
418 strcpy(warncomm, current->comm);
419 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
425 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
427 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
429 if (sk->sk_flags & flags) {
430 sk->sk_flags &= ~flags;
431 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
432 net_disable_timestamp();
437 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
442 struct sk_buff_head *list = &sk->sk_receive_queue;
444 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
445 atomic_inc(&sk->sk_drops);
446 trace_sock_rcvqueue_full(sk, skb);
450 err = sk_filter(sk, skb);
454 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
455 atomic_inc(&sk->sk_drops);
460 skb_set_owner_r(skb, sk);
462 /* Cache the SKB length before we tack it onto the receive
463 * queue. Once it is added it no longer belongs to us and
464 * may be freed by other threads of control pulling packets
469 /* we escape from rcu protected region, make sure we dont leak
474 spin_lock_irqsave(&list->lock, flags);
475 skb->dropcount = atomic_read(&sk->sk_drops);
476 __skb_queue_tail(list, skb);
477 spin_unlock_irqrestore(&list->lock, flags);
479 if (!sock_flag(sk, SOCK_DEAD))
480 sk->sk_data_ready(sk);
483 EXPORT_SYMBOL(sock_queue_rcv_skb);
485 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
487 int rc = NET_RX_SUCCESS;
489 if (sk_filter(sk, skb))
490 goto discard_and_relse;
494 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
495 atomic_inc(&sk->sk_drops);
496 goto discard_and_relse;
499 bh_lock_sock_nested(sk);
502 if (!sock_owned_by_user(sk)) {
504 * trylock + unlock semantics:
506 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
508 rc = sk_backlog_rcv(sk, skb);
510 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
511 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
513 atomic_inc(&sk->sk_drops);
514 goto discard_and_relse;
525 EXPORT_SYMBOL(sk_receive_skb);
527 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
529 struct dst_entry *dst = __sk_dst_get(sk);
531 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
532 sk_tx_queue_clear(sk);
533 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
540 EXPORT_SYMBOL(__sk_dst_check);
542 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
544 struct dst_entry *dst = sk_dst_get(sk);
546 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
554 EXPORT_SYMBOL(sk_dst_check);
556 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
559 int ret = -ENOPROTOOPT;
560 #ifdef CONFIG_NETDEVICES
561 struct net *net = sock_net(sk);
562 char devname[IFNAMSIZ];
567 if (!ns_capable(net->user_ns, CAP_NET_RAW))
574 /* Bind this socket to a particular device like "eth0",
575 * as specified in the passed interface name. If the
576 * name is "" or the option length is zero the socket
579 if (optlen > IFNAMSIZ - 1)
580 optlen = IFNAMSIZ - 1;
581 memset(devname, 0, sizeof(devname));
584 if (copy_from_user(devname, optval, optlen))
588 if (devname[0] != '\0') {
589 struct net_device *dev;
592 dev = dev_get_by_name_rcu(net, devname);
594 index = dev->ifindex;
602 sk->sk_bound_dev_if = index;
614 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
615 int __user *optlen, int len)
617 int ret = -ENOPROTOOPT;
618 #ifdef CONFIG_NETDEVICES
619 struct net *net = sock_net(sk);
620 char devname[IFNAMSIZ];
622 if (sk->sk_bound_dev_if == 0) {
631 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
635 len = strlen(devname) + 1;
638 if (copy_to_user(optval, devname, len))
643 if (put_user(len, optlen))
654 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
657 sock_set_flag(sk, bit);
659 sock_reset_flag(sk, bit);
663 * This is meant for all protocols to use and covers goings on
664 * at the socket level. Everything here is generic.
667 int sock_setsockopt(struct socket *sock, int level, int optname,
668 char __user *optval, unsigned int optlen)
670 struct sock *sk = sock->sk;
677 * Options without arguments
680 if (optname == SO_BINDTODEVICE)
681 return sock_setbindtodevice(sk, optval, optlen);
683 if (optlen < sizeof(int))
686 if (get_user(val, (int __user *)optval))
689 valbool = val ? 1 : 0;
695 if (val && !capable(CAP_NET_ADMIN))
698 sock_valbool_flag(sk, SOCK_DBG, valbool);
701 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
704 sk->sk_reuseport = valbool;
713 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
716 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
719 /* Don't error on this BSD doesn't and if you think
720 * about it this is right. Otherwise apps have to
721 * play 'guess the biggest size' games. RCVBUF/SNDBUF
722 * are treated in BSD as hints
724 val = min_t(u32, val, sysctl_wmem_max);
726 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
727 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
728 /* Wake up sending tasks if we upped the value. */
729 sk->sk_write_space(sk);
733 if (!capable(CAP_NET_ADMIN)) {
740 /* Don't error on this BSD doesn't and if you think
741 * about it this is right. Otherwise apps have to
742 * play 'guess the biggest size' games. RCVBUF/SNDBUF
743 * are treated in BSD as hints
745 val = min_t(u32, val, sysctl_rmem_max);
747 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
749 * We double it on the way in to account for
750 * "struct sk_buff" etc. overhead. Applications
751 * assume that the SO_RCVBUF setting they make will
752 * allow that much actual data to be received on that
755 * Applications are unaware that "struct sk_buff" and
756 * other overheads allocate from the receive buffer
757 * during socket buffer allocation.
759 * And after considering the possible alternatives,
760 * returning the value we actually used in getsockopt
761 * is the most desirable behavior.
763 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
767 if (!capable(CAP_NET_ADMIN)) {
775 if (sk->sk_protocol == IPPROTO_TCP &&
776 sk->sk_type == SOCK_STREAM)
777 tcp_set_keepalive(sk, valbool);
779 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
783 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
787 sk->sk_no_check_tx = valbool;
791 if ((val >= 0 && val <= 6) ||
792 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
793 sk->sk_priority = val;
799 if (optlen < sizeof(ling)) {
800 ret = -EINVAL; /* 1003.1g */
803 if (copy_from_user(&ling, optval, sizeof(ling))) {
808 sock_reset_flag(sk, SOCK_LINGER);
810 #if (BITS_PER_LONG == 32)
811 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
812 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
815 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
816 sock_set_flag(sk, SOCK_LINGER);
821 sock_warn_obsolete_bsdism("setsockopt");
826 set_bit(SOCK_PASSCRED, &sock->flags);
828 clear_bit(SOCK_PASSCRED, &sock->flags);
834 if (optname == SO_TIMESTAMP)
835 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
837 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
838 sock_set_flag(sk, SOCK_RCVTSTAMP);
839 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
841 sock_reset_flag(sk, SOCK_RCVTSTAMP);
842 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
846 case SO_TIMESTAMPING:
847 if (val & ~SOF_TIMESTAMPING_MASK) {
851 if (val & SOF_TIMESTAMPING_OPT_ID &&
852 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID))
854 sk->sk_tsflags = val;
855 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
856 sock_enable_timestamp(sk,
857 SOCK_TIMESTAMPING_RX_SOFTWARE);
859 sock_disable_timestamp(sk,
860 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
866 sk->sk_rcvlowat = val ? : 1;
870 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
874 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
877 case SO_ATTACH_FILTER:
879 if (optlen == sizeof(struct sock_fprog)) {
880 struct sock_fprog fprog;
883 if (copy_from_user(&fprog, optval, sizeof(fprog)))
886 ret = sk_attach_filter(&fprog, sk);
890 case SO_DETACH_FILTER:
891 ret = sk_detach_filter(sk);
895 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
898 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
903 set_bit(SOCK_PASSSEC, &sock->flags);
905 clear_bit(SOCK_PASSSEC, &sock->flags);
908 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
914 /* We implement the SO_SNDLOWAT etc to
915 not be settable (1003.1g 5.3) */
917 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
921 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
925 if (sock->ops->set_peek_off)
926 ret = sock->ops->set_peek_off(sk, val);
932 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
935 case SO_SELECT_ERR_QUEUE:
936 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
939 #ifdef CONFIG_NET_RX_BUSY_POLL
941 /* allow unprivileged users to decrease the value */
942 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
948 sk->sk_ll_usec = val;
953 case SO_MAX_PACING_RATE:
954 sk->sk_max_pacing_rate = val;
955 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
956 sk->sk_max_pacing_rate);
966 EXPORT_SYMBOL(sock_setsockopt);
969 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
972 ucred->pid = pid_vnr(pid);
973 ucred->uid = ucred->gid = -1;
975 struct user_namespace *current_ns = current_user_ns();
977 ucred->uid = from_kuid_munged(current_ns, cred->euid);
978 ucred->gid = from_kgid_munged(current_ns, cred->egid);
982 int sock_getsockopt(struct socket *sock, int level, int optname,
983 char __user *optval, int __user *optlen)
985 struct sock *sk = sock->sk;
993 int lv = sizeof(int);
996 if (get_user(len, optlen))
1001 memset(&v, 0, sizeof(v));
1005 v.val = sock_flag(sk, SOCK_DBG);
1009 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1013 v.val = sock_flag(sk, SOCK_BROADCAST);
1017 v.val = sk->sk_sndbuf;
1021 v.val = sk->sk_rcvbuf;
1025 v.val = sk->sk_reuse;
1029 v.val = sk->sk_reuseport;
1033 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1037 v.val = sk->sk_type;
1041 v.val = sk->sk_protocol;
1045 v.val = sk->sk_family;
1049 v.val = -sock_error(sk);
1051 v.val = xchg(&sk->sk_err_soft, 0);
1055 v.val = sock_flag(sk, SOCK_URGINLINE);
1059 v.val = sk->sk_no_check_tx;
1063 v.val = sk->sk_priority;
1067 lv = sizeof(v.ling);
1068 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1069 v.ling.l_linger = sk->sk_lingertime / HZ;
1073 sock_warn_obsolete_bsdism("getsockopt");
1077 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1078 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1081 case SO_TIMESTAMPNS:
1082 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1085 case SO_TIMESTAMPING:
1086 v.val = sk->sk_tsflags;
1090 lv = sizeof(struct timeval);
1091 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1095 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1096 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1101 lv = sizeof(struct timeval);
1102 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1106 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1107 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1112 v.val = sk->sk_rcvlowat;
1120 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1125 struct ucred peercred;
1126 if (len > sizeof(peercred))
1127 len = sizeof(peercred);
1128 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1129 if (copy_to_user(optval, &peercred, len))
1138 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1142 if (copy_to_user(optval, address, len))
1147 /* Dubious BSD thing... Probably nobody even uses it, but
1148 * the UNIX standard wants it for whatever reason... -DaveM
1151 v.val = sk->sk_state == TCP_LISTEN;
1155 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1159 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1162 v.val = sk->sk_mark;
1166 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1169 case SO_WIFI_STATUS:
1170 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1174 if (!sock->ops->set_peek_off)
1177 v.val = sk->sk_peek_off;
1180 v.val = sock_flag(sk, SOCK_NOFCS);
1183 case SO_BINDTODEVICE:
1184 return sock_getbindtodevice(sk, optval, optlen, len);
1187 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1193 case SO_LOCK_FILTER:
1194 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1197 case SO_BPF_EXTENSIONS:
1198 v.val = bpf_tell_extensions();
1201 case SO_SELECT_ERR_QUEUE:
1202 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1205 #ifdef CONFIG_NET_RX_BUSY_POLL
1207 v.val = sk->sk_ll_usec;
1211 case SO_MAX_PACING_RATE:
1212 v.val = sk->sk_max_pacing_rate;
1216 return -ENOPROTOOPT;
1221 if (copy_to_user(optval, &v, len))
1224 if (put_user(len, optlen))
1230 * Initialize an sk_lock.
1232 * (We also register the sk_lock with the lock validator.)
1234 static inline void sock_lock_init(struct sock *sk)
1236 sock_lock_init_class_and_name(sk,
1237 af_family_slock_key_strings[sk->sk_family],
1238 af_family_slock_keys + sk->sk_family,
1239 af_family_key_strings[sk->sk_family],
1240 af_family_keys + sk->sk_family);
1244 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1245 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1246 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1248 static void sock_copy(struct sock *nsk, const struct sock *osk)
1250 #ifdef CONFIG_SECURITY_NETWORK
1251 void *sptr = nsk->sk_security;
1253 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1255 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1256 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1258 #ifdef CONFIG_SECURITY_NETWORK
1259 nsk->sk_security = sptr;
1260 security_sk_clone(osk, nsk);
1264 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1266 unsigned long nulls1, nulls2;
1268 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1269 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1270 if (nulls1 > nulls2)
1271 swap(nulls1, nulls2);
1274 memset((char *)sk, 0, nulls1);
1275 memset((char *)sk + nulls1 + sizeof(void *), 0,
1276 nulls2 - nulls1 - sizeof(void *));
1277 memset((char *)sk + nulls2 + sizeof(void *), 0,
1278 size - nulls2 - sizeof(void *));
1280 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1282 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1286 struct kmem_cache *slab;
1290 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1293 if (priority & __GFP_ZERO) {
1295 prot->clear_sk(sk, prot->obj_size);
1297 sk_prot_clear_nulls(sk, prot->obj_size);
1300 sk = kmalloc(prot->obj_size, priority);
1303 kmemcheck_annotate_bitfield(sk, flags);
1305 if (security_sk_alloc(sk, family, priority))
1308 if (!try_module_get(prot->owner))
1310 sk_tx_queue_clear(sk);
1316 security_sk_free(sk);
1319 kmem_cache_free(slab, sk);
1325 static void sk_prot_free(struct proto *prot, struct sock *sk)
1327 struct kmem_cache *slab;
1328 struct module *owner;
1330 owner = prot->owner;
1333 security_sk_free(sk);
1335 kmem_cache_free(slab, sk);
1341 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1342 void sock_update_netprioidx(struct sock *sk)
1347 sk->sk_cgrp_prioidx = task_netprioidx(current);
1349 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1353 * sk_alloc - All socket objects are allocated here
1354 * @net: the applicable net namespace
1355 * @family: protocol family
1356 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1357 * @prot: struct proto associated with this new sock instance
1359 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1364 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1366 sk->sk_family = family;
1368 * See comment in struct sock definition to understand
1369 * why we need sk_prot_creator -acme
1371 sk->sk_prot = sk->sk_prot_creator = prot;
1373 sock_net_set(sk, get_net(net));
1374 atomic_set(&sk->sk_wmem_alloc, 1);
1376 sock_update_classid(sk);
1377 sock_update_netprioidx(sk);
1382 EXPORT_SYMBOL(sk_alloc);
1384 static void __sk_free(struct sock *sk)
1386 struct sk_filter *filter;
1388 if (sk->sk_destruct)
1389 sk->sk_destruct(sk);
1391 filter = rcu_dereference_check(sk->sk_filter,
1392 atomic_read(&sk->sk_wmem_alloc) == 0);
1394 sk_filter_uncharge(sk, filter);
1395 RCU_INIT_POINTER(sk->sk_filter, NULL);
1398 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1400 if (atomic_read(&sk->sk_omem_alloc))
1401 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1402 __func__, atomic_read(&sk->sk_omem_alloc));
1404 if (sk->sk_peer_cred)
1405 put_cred(sk->sk_peer_cred);
1406 put_pid(sk->sk_peer_pid);
1407 put_net(sock_net(sk));
1408 sk_prot_free(sk->sk_prot_creator, sk);
1411 void sk_free(struct sock *sk)
1414 * We subtract one from sk_wmem_alloc and can know if
1415 * some packets are still in some tx queue.
1416 * If not null, sock_wfree() will call __sk_free(sk) later
1418 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1421 EXPORT_SYMBOL(sk_free);
1424 * Last sock_put should drop reference to sk->sk_net. It has already
1425 * been dropped in sk_change_net. Taking reference to stopping namespace
1427 * Take reference to a socket to remove it from hash _alive_ and after that
1428 * destroy it in the context of init_net.
1430 void sk_release_kernel(struct sock *sk)
1432 if (sk == NULL || sk->sk_socket == NULL)
1436 sock_release(sk->sk_socket);
1437 release_net(sock_net(sk));
1438 sock_net_set(sk, get_net(&init_net));
1441 EXPORT_SYMBOL(sk_release_kernel);
1443 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1445 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1446 sock_update_memcg(newsk);
1450 * sk_clone_lock - clone a socket, and lock its clone
1451 * @sk: the socket to clone
1452 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1454 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1456 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1459 bool is_charged = true;
1461 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1462 if (newsk != NULL) {
1463 struct sk_filter *filter;
1465 sock_copy(newsk, sk);
1468 get_net(sock_net(newsk));
1469 sk_node_init(&newsk->sk_node);
1470 sock_lock_init(newsk);
1471 bh_lock_sock(newsk);
1472 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1473 newsk->sk_backlog.len = 0;
1475 atomic_set(&newsk->sk_rmem_alloc, 0);
1477 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1479 atomic_set(&newsk->sk_wmem_alloc, 1);
1480 atomic_set(&newsk->sk_omem_alloc, 0);
1481 skb_queue_head_init(&newsk->sk_receive_queue);
1482 skb_queue_head_init(&newsk->sk_write_queue);
1483 #ifdef CONFIG_NET_DMA
1484 skb_queue_head_init(&newsk->sk_async_wait_queue);
1487 spin_lock_init(&newsk->sk_dst_lock);
1488 rwlock_init(&newsk->sk_callback_lock);
1489 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1490 af_callback_keys + newsk->sk_family,
1491 af_family_clock_key_strings[newsk->sk_family]);
1493 newsk->sk_dst_cache = NULL;
1494 newsk->sk_wmem_queued = 0;
1495 newsk->sk_forward_alloc = 0;
1496 newsk->sk_send_head = NULL;
1497 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1499 sock_reset_flag(newsk, SOCK_DONE);
1500 skb_queue_head_init(&newsk->sk_error_queue);
1502 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1504 /* though it's an empty new sock, the charging may fail
1505 * if sysctl_optmem_max was changed between creation of
1506 * original socket and cloning
1508 is_charged = sk_filter_charge(newsk, filter);
1510 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1511 /* It is still raw copy of parent, so invalidate
1512 * destructor and make plain sk_free() */
1513 newsk->sk_destruct = NULL;
1514 bh_unlock_sock(newsk);
1521 newsk->sk_priority = 0;
1523 * Before updating sk_refcnt, we must commit prior changes to memory
1524 * (Documentation/RCU/rculist_nulls.txt for details)
1527 atomic_set(&newsk->sk_refcnt, 2);
1530 * Increment the counter in the same struct proto as the master
1531 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1532 * is the same as sk->sk_prot->socks, as this field was copied
1535 * This _changes_ the previous behaviour, where
1536 * tcp_create_openreq_child always was incrementing the
1537 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1538 * to be taken into account in all callers. -acme
1540 sk_refcnt_debug_inc(newsk);
1541 sk_set_socket(newsk, NULL);
1542 newsk->sk_wq = NULL;
1544 sk_update_clone(sk, newsk);
1546 if (newsk->sk_prot->sockets_allocated)
1547 sk_sockets_allocated_inc(newsk);
1549 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1550 net_enable_timestamp();
1555 EXPORT_SYMBOL_GPL(sk_clone_lock);
1557 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1559 __sk_dst_set(sk, dst);
1560 sk->sk_route_caps = dst->dev->features;
1561 if (sk->sk_route_caps & NETIF_F_GSO)
1562 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1563 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1564 if (sk_can_gso(sk)) {
1565 if (dst->header_len) {
1566 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1568 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1569 sk->sk_gso_max_size = dst->dev->gso_max_size;
1570 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1574 EXPORT_SYMBOL_GPL(sk_setup_caps);
1577 * Simple resource managers for sockets.
1582 * Write buffer destructor automatically called from kfree_skb.
1584 void sock_wfree(struct sk_buff *skb)
1586 struct sock *sk = skb->sk;
1587 unsigned int len = skb->truesize;
1589 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1591 * Keep a reference on sk_wmem_alloc, this will be released
1592 * after sk_write_space() call
1594 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1595 sk->sk_write_space(sk);
1599 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1600 * could not do because of in-flight packets
1602 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1605 EXPORT_SYMBOL(sock_wfree);
1607 void skb_orphan_partial(struct sk_buff *skb)
1609 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1610 * so we do not completely orphan skb, but transfert all
1611 * accounted bytes but one, to avoid unexpected reorders.
1613 if (skb->destructor == sock_wfree
1615 || skb->destructor == tcp_wfree
1618 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1624 EXPORT_SYMBOL(skb_orphan_partial);
1627 * Read buffer destructor automatically called from kfree_skb.
1629 void sock_rfree(struct sk_buff *skb)
1631 struct sock *sk = skb->sk;
1632 unsigned int len = skb->truesize;
1634 atomic_sub(len, &sk->sk_rmem_alloc);
1635 sk_mem_uncharge(sk, len);
1637 EXPORT_SYMBOL(sock_rfree);
1639 void sock_edemux(struct sk_buff *skb)
1641 struct sock *sk = skb->sk;
1644 if (sk->sk_state == TCP_TIME_WAIT)
1645 inet_twsk_put(inet_twsk(sk));
1650 EXPORT_SYMBOL(sock_edemux);
1652 kuid_t sock_i_uid(struct sock *sk)
1656 read_lock_bh(&sk->sk_callback_lock);
1657 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1658 read_unlock_bh(&sk->sk_callback_lock);
1661 EXPORT_SYMBOL(sock_i_uid);
1663 unsigned long sock_i_ino(struct sock *sk)
1667 read_lock_bh(&sk->sk_callback_lock);
1668 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1669 read_unlock_bh(&sk->sk_callback_lock);
1672 EXPORT_SYMBOL(sock_i_ino);
1675 * Allocate a skb from the socket's send buffer.
1677 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1680 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1681 struct sk_buff *skb = alloc_skb(size, priority);
1683 skb_set_owner_w(skb, sk);
1689 EXPORT_SYMBOL(sock_wmalloc);
1692 * Allocate a memory block from the socket's option memory buffer.
1694 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1696 if ((unsigned int)size <= sysctl_optmem_max &&
1697 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1699 /* First do the add, to avoid the race if kmalloc
1702 atomic_add(size, &sk->sk_omem_alloc);
1703 mem = kmalloc(size, priority);
1706 atomic_sub(size, &sk->sk_omem_alloc);
1710 EXPORT_SYMBOL(sock_kmalloc);
1713 * Free an option memory block.
1715 void sock_kfree_s(struct sock *sk, void *mem, int size)
1718 atomic_sub(size, &sk->sk_omem_alloc);
1720 EXPORT_SYMBOL(sock_kfree_s);
1722 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1723 I think, these locks should be removed for datagram sockets.
1725 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1729 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1733 if (signal_pending(current))
1735 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1736 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1737 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1739 if (sk->sk_shutdown & SEND_SHUTDOWN)
1743 timeo = schedule_timeout(timeo);
1745 finish_wait(sk_sleep(sk), &wait);
1751 * Generic send/receive buffer handlers
1754 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1755 unsigned long data_len, int noblock,
1756 int *errcode, int max_page_order)
1758 struct sk_buff *skb = NULL;
1759 unsigned long chunk;
1763 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1768 if (npages > MAX_SKB_FRAGS)
1771 timeo = sock_sndtimeo(sk, noblock);
1773 err = sock_error(sk);
1778 if (sk->sk_shutdown & SEND_SHUTDOWN)
1781 if (atomic_read(&sk->sk_wmem_alloc) >= sk->sk_sndbuf) {
1782 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1783 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1787 if (signal_pending(current))
1789 timeo = sock_wait_for_wmem(sk, timeo);
1794 gfp_mask = sk->sk_allocation;
1795 if (gfp_mask & __GFP_WAIT)
1796 gfp_mask |= __GFP_REPEAT;
1798 skb = alloc_skb(header_len, gfp_mask);
1802 skb->truesize += data_len;
1804 for (i = 0; npages > 0; i++) {
1805 int order = max_page_order;
1808 if (npages >= 1 << order) {
1809 page = alloc_pages(sk->sk_allocation |
1819 page = alloc_page(sk->sk_allocation);
1823 chunk = min_t(unsigned long, data_len,
1824 PAGE_SIZE << order);
1825 skb_fill_page_desc(skb, i, page, 0, chunk);
1827 npages -= 1 << order;
1831 skb_set_owner_w(skb, sk);
1835 err = sock_intr_errno(timeo);
1841 EXPORT_SYMBOL(sock_alloc_send_pskb);
1843 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1844 int noblock, int *errcode)
1846 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1848 EXPORT_SYMBOL(sock_alloc_send_skb);
1850 /* On 32bit arches, an skb frag is limited to 2^15 */
1851 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1854 * skb_page_frag_refill - check that a page_frag contains enough room
1855 * @sz: minimum size of the fragment we want to get
1856 * @pfrag: pointer to page_frag
1857 * @prio: priority for memory allocation
1859 * Note: While this allocator tries to use high order pages, there is
1860 * no guarantee that allocations succeed. Therefore, @sz MUST be
1861 * less or equal than PAGE_SIZE.
1863 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio)
1868 if (atomic_read(&pfrag->page->_count) == 1) {
1872 if (pfrag->offset + sz <= pfrag->size)
1874 put_page(pfrag->page);
1877 order = SKB_FRAG_PAGE_ORDER;
1882 gfp |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY;
1883 pfrag->page = alloc_pages(gfp, order);
1884 if (likely(pfrag->page)) {
1886 pfrag->size = PAGE_SIZE << order;
1889 } while (--order >= 0);
1893 EXPORT_SYMBOL(skb_page_frag_refill);
1895 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1897 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1900 sk_enter_memory_pressure(sk);
1901 sk_stream_moderate_sndbuf(sk);
1904 EXPORT_SYMBOL(sk_page_frag_refill);
1906 static void __lock_sock(struct sock *sk)
1907 __releases(&sk->sk_lock.slock)
1908 __acquires(&sk->sk_lock.slock)
1913 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1914 TASK_UNINTERRUPTIBLE);
1915 spin_unlock_bh(&sk->sk_lock.slock);
1917 spin_lock_bh(&sk->sk_lock.slock);
1918 if (!sock_owned_by_user(sk))
1921 finish_wait(&sk->sk_lock.wq, &wait);
1924 static void __release_sock(struct sock *sk)
1925 __releases(&sk->sk_lock.slock)
1926 __acquires(&sk->sk_lock.slock)
1928 struct sk_buff *skb = sk->sk_backlog.head;
1931 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1935 struct sk_buff *next = skb->next;
1938 WARN_ON_ONCE(skb_dst_is_noref(skb));
1940 sk_backlog_rcv(sk, skb);
1943 * We are in process context here with softirqs
1944 * disabled, use cond_resched_softirq() to preempt.
1945 * This is safe to do because we've taken the backlog
1948 cond_resched_softirq();
1951 } while (skb != NULL);
1954 } while ((skb = sk->sk_backlog.head) != NULL);
1957 * Doing the zeroing here guarantee we can not loop forever
1958 * while a wild producer attempts to flood us.
1960 sk->sk_backlog.len = 0;
1964 * sk_wait_data - wait for data to arrive at sk_receive_queue
1965 * @sk: sock to wait on
1966 * @timeo: for how long
1968 * Now socket state including sk->sk_err is changed only under lock,
1969 * hence we may omit checks after joining wait queue.
1970 * We check receive queue before schedule() only as optimization;
1971 * it is very likely that release_sock() added new data.
1973 int sk_wait_data(struct sock *sk, long *timeo)
1978 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1979 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1980 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1981 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1982 finish_wait(sk_sleep(sk), &wait);
1985 EXPORT_SYMBOL(sk_wait_data);
1988 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1990 * @size: memory size to allocate
1991 * @kind: allocation type
1993 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1994 * rmem allocation. This function assumes that protocols which have
1995 * memory_pressure use sk_wmem_queued as write buffer accounting.
1997 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1999 struct proto *prot = sk->sk_prot;
2000 int amt = sk_mem_pages(size);
2002 int parent_status = UNDER_LIMIT;
2004 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2006 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2009 if (parent_status == UNDER_LIMIT &&
2010 allocated <= sk_prot_mem_limits(sk, 0)) {
2011 sk_leave_memory_pressure(sk);
2015 /* Under pressure. (we or our parents) */
2016 if ((parent_status > SOFT_LIMIT) ||
2017 allocated > sk_prot_mem_limits(sk, 1))
2018 sk_enter_memory_pressure(sk);
2020 /* Over hard limit (we or our parents) */
2021 if ((parent_status == OVER_LIMIT) ||
2022 (allocated > sk_prot_mem_limits(sk, 2)))
2023 goto suppress_allocation;
2025 /* guarantee minimum buffer size under pressure */
2026 if (kind == SK_MEM_RECV) {
2027 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2030 } else { /* SK_MEM_SEND */
2031 if (sk->sk_type == SOCK_STREAM) {
2032 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2034 } else if (atomic_read(&sk->sk_wmem_alloc) <
2035 prot->sysctl_wmem[0])
2039 if (sk_has_memory_pressure(sk)) {
2042 if (!sk_under_memory_pressure(sk))
2044 alloc = sk_sockets_allocated_read_positive(sk);
2045 if (sk_prot_mem_limits(sk, 2) > alloc *
2046 sk_mem_pages(sk->sk_wmem_queued +
2047 atomic_read(&sk->sk_rmem_alloc) +
2048 sk->sk_forward_alloc))
2052 suppress_allocation:
2054 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2055 sk_stream_moderate_sndbuf(sk);
2057 /* Fail only if socket is _under_ its sndbuf.
2058 * In this case we cannot block, so that we have to fail.
2060 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2064 trace_sock_exceed_buf_limit(sk, prot, allocated);
2066 /* Alas. Undo changes. */
2067 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2069 sk_memory_allocated_sub(sk, amt);
2073 EXPORT_SYMBOL(__sk_mem_schedule);
2076 * __sk_reclaim - reclaim memory_allocated
2079 void __sk_mem_reclaim(struct sock *sk)
2081 sk_memory_allocated_sub(sk,
2082 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2083 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2085 if (sk_under_memory_pressure(sk) &&
2086 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2087 sk_leave_memory_pressure(sk);
2089 EXPORT_SYMBOL(__sk_mem_reclaim);
2093 * Set of default routines for initialising struct proto_ops when
2094 * the protocol does not support a particular function. In certain
2095 * cases where it makes no sense for a protocol to have a "do nothing"
2096 * function, some default processing is provided.
2099 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2103 EXPORT_SYMBOL(sock_no_bind);
2105 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2110 EXPORT_SYMBOL(sock_no_connect);
2112 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2116 EXPORT_SYMBOL(sock_no_socketpair);
2118 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2122 EXPORT_SYMBOL(sock_no_accept);
2124 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2129 EXPORT_SYMBOL(sock_no_getname);
2131 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2135 EXPORT_SYMBOL(sock_no_poll);
2137 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2141 EXPORT_SYMBOL(sock_no_ioctl);
2143 int sock_no_listen(struct socket *sock, int backlog)
2147 EXPORT_SYMBOL(sock_no_listen);
2149 int sock_no_shutdown(struct socket *sock, int how)
2153 EXPORT_SYMBOL(sock_no_shutdown);
2155 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2156 char __user *optval, unsigned int optlen)
2160 EXPORT_SYMBOL(sock_no_setsockopt);
2162 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2163 char __user *optval, int __user *optlen)
2167 EXPORT_SYMBOL(sock_no_getsockopt);
2169 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2174 EXPORT_SYMBOL(sock_no_sendmsg);
2176 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2177 size_t len, int flags)
2181 EXPORT_SYMBOL(sock_no_recvmsg);
2183 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2185 /* Mirror missing mmap method error code */
2188 EXPORT_SYMBOL(sock_no_mmap);
2190 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2193 struct msghdr msg = {.msg_flags = flags};
2195 char *kaddr = kmap(page);
2196 iov.iov_base = kaddr + offset;
2198 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2202 EXPORT_SYMBOL(sock_no_sendpage);
2205 * Default Socket Callbacks
2208 static void sock_def_wakeup(struct sock *sk)
2210 struct socket_wq *wq;
2213 wq = rcu_dereference(sk->sk_wq);
2214 if (wq_has_sleeper(wq))
2215 wake_up_interruptible_all(&wq->wait);
2219 static void sock_def_error_report(struct sock *sk)
2221 struct socket_wq *wq;
2224 wq = rcu_dereference(sk->sk_wq);
2225 if (wq_has_sleeper(wq))
2226 wake_up_interruptible_poll(&wq->wait, POLLERR);
2227 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2231 static void sock_def_readable(struct sock *sk)
2233 struct socket_wq *wq;
2236 wq = rcu_dereference(sk->sk_wq);
2237 if (wq_has_sleeper(wq))
2238 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2239 POLLRDNORM | POLLRDBAND);
2240 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2244 static void sock_def_write_space(struct sock *sk)
2246 struct socket_wq *wq;
2250 /* Do not wake up a writer until he can make "significant"
2253 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2254 wq = rcu_dereference(sk->sk_wq);
2255 if (wq_has_sleeper(wq))
2256 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2257 POLLWRNORM | POLLWRBAND);
2259 /* Should agree with poll, otherwise some programs break */
2260 if (sock_writeable(sk))
2261 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2267 static void sock_def_destruct(struct sock *sk)
2269 kfree(sk->sk_protinfo);
2272 void sk_send_sigurg(struct sock *sk)
2274 if (sk->sk_socket && sk->sk_socket->file)
2275 if (send_sigurg(&sk->sk_socket->file->f_owner))
2276 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2278 EXPORT_SYMBOL(sk_send_sigurg);
2280 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2281 unsigned long expires)
2283 if (!mod_timer(timer, expires))
2286 EXPORT_SYMBOL(sk_reset_timer);
2288 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2290 if (del_timer(timer))
2293 EXPORT_SYMBOL(sk_stop_timer);
2295 void sock_init_data(struct socket *sock, struct sock *sk)
2297 skb_queue_head_init(&sk->sk_receive_queue);
2298 skb_queue_head_init(&sk->sk_write_queue);
2299 skb_queue_head_init(&sk->sk_error_queue);
2300 #ifdef CONFIG_NET_DMA
2301 skb_queue_head_init(&sk->sk_async_wait_queue);
2304 sk->sk_send_head = NULL;
2306 init_timer(&sk->sk_timer);
2308 sk->sk_allocation = GFP_KERNEL;
2309 sk->sk_rcvbuf = sysctl_rmem_default;
2310 sk->sk_sndbuf = sysctl_wmem_default;
2311 sk->sk_state = TCP_CLOSE;
2312 sk_set_socket(sk, sock);
2314 sock_set_flag(sk, SOCK_ZAPPED);
2317 sk->sk_type = sock->type;
2318 sk->sk_wq = sock->wq;
2323 spin_lock_init(&sk->sk_dst_lock);
2324 rwlock_init(&sk->sk_callback_lock);
2325 lockdep_set_class_and_name(&sk->sk_callback_lock,
2326 af_callback_keys + sk->sk_family,
2327 af_family_clock_key_strings[sk->sk_family]);
2329 sk->sk_state_change = sock_def_wakeup;
2330 sk->sk_data_ready = sock_def_readable;
2331 sk->sk_write_space = sock_def_write_space;
2332 sk->sk_error_report = sock_def_error_report;
2333 sk->sk_destruct = sock_def_destruct;
2335 sk->sk_frag.page = NULL;
2336 sk->sk_frag.offset = 0;
2337 sk->sk_peek_off = -1;
2339 sk->sk_peer_pid = NULL;
2340 sk->sk_peer_cred = NULL;
2341 sk->sk_write_pending = 0;
2342 sk->sk_rcvlowat = 1;
2343 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2344 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2346 sk->sk_stamp = ktime_set(-1L, 0);
2348 #ifdef CONFIG_NET_RX_BUSY_POLL
2350 sk->sk_ll_usec = sysctl_net_busy_read;
2353 sk->sk_max_pacing_rate = ~0U;
2354 sk->sk_pacing_rate = ~0U;
2356 * Before updating sk_refcnt, we must commit prior changes to memory
2357 * (Documentation/RCU/rculist_nulls.txt for details)
2360 atomic_set(&sk->sk_refcnt, 1);
2361 atomic_set(&sk->sk_drops, 0);
2363 EXPORT_SYMBOL(sock_init_data);
2365 void lock_sock_nested(struct sock *sk, int subclass)
2368 spin_lock_bh(&sk->sk_lock.slock);
2369 if (sk->sk_lock.owned)
2371 sk->sk_lock.owned = 1;
2372 spin_unlock(&sk->sk_lock.slock);
2374 * The sk_lock has mutex_lock() semantics here:
2376 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2379 EXPORT_SYMBOL(lock_sock_nested);
2381 void release_sock(struct sock *sk)
2384 * The sk_lock has mutex_unlock() semantics:
2386 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2388 spin_lock_bh(&sk->sk_lock.slock);
2389 if (sk->sk_backlog.tail)
2392 /* Warning : release_cb() might need to release sk ownership,
2393 * ie call sock_release_ownership(sk) before us.
2395 if (sk->sk_prot->release_cb)
2396 sk->sk_prot->release_cb(sk);
2398 sock_release_ownership(sk);
2399 if (waitqueue_active(&sk->sk_lock.wq))
2400 wake_up(&sk->sk_lock.wq);
2401 spin_unlock_bh(&sk->sk_lock.slock);
2403 EXPORT_SYMBOL(release_sock);
2406 * lock_sock_fast - fast version of lock_sock
2409 * This version should be used for very small section, where process wont block
2410 * return false if fast path is taken
2411 * sk_lock.slock locked, owned = 0, BH disabled
2412 * return true if slow path is taken
2413 * sk_lock.slock unlocked, owned = 1, BH enabled
2415 bool lock_sock_fast(struct sock *sk)
2418 spin_lock_bh(&sk->sk_lock.slock);
2420 if (!sk->sk_lock.owned)
2422 * Note : We must disable BH
2427 sk->sk_lock.owned = 1;
2428 spin_unlock(&sk->sk_lock.slock);
2430 * The sk_lock has mutex_lock() semantics here:
2432 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2436 EXPORT_SYMBOL(lock_sock_fast);
2438 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2441 if (!sock_flag(sk, SOCK_TIMESTAMP))
2442 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2443 tv = ktime_to_timeval(sk->sk_stamp);
2444 if (tv.tv_sec == -1)
2446 if (tv.tv_sec == 0) {
2447 sk->sk_stamp = ktime_get_real();
2448 tv = ktime_to_timeval(sk->sk_stamp);
2450 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2452 EXPORT_SYMBOL(sock_get_timestamp);
2454 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2457 if (!sock_flag(sk, SOCK_TIMESTAMP))
2458 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2459 ts = ktime_to_timespec(sk->sk_stamp);
2460 if (ts.tv_sec == -1)
2462 if (ts.tv_sec == 0) {
2463 sk->sk_stamp = ktime_get_real();
2464 ts = ktime_to_timespec(sk->sk_stamp);
2466 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2468 EXPORT_SYMBOL(sock_get_timestampns);
2470 void sock_enable_timestamp(struct sock *sk, int flag)
2472 if (!sock_flag(sk, flag)) {
2473 unsigned long previous_flags = sk->sk_flags;
2475 sock_set_flag(sk, flag);
2477 * we just set one of the two flags which require net
2478 * time stamping, but time stamping might have been on
2479 * already because of the other one
2481 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2482 net_enable_timestamp();
2486 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2487 int level, int type)
2489 struct sock_exterr_skb *serr;
2490 struct sk_buff *skb, *skb2;
2494 skb = skb_dequeue(&sk->sk_error_queue);
2500 msg->msg_flags |= MSG_TRUNC;
2503 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2507 sock_recv_timestamp(msg, sk, skb);
2509 serr = SKB_EXT_ERR(skb);
2510 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2512 msg->msg_flags |= MSG_ERRQUEUE;
2515 /* Reset and regenerate socket error */
2516 spin_lock_bh(&sk->sk_error_queue.lock);
2518 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2519 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2520 spin_unlock_bh(&sk->sk_error_queue.lock);
2521 sk->sk_error_report(sk);
2523 spin_unlock_bh(&sk->sk_error_queue.lock);
2530 EXPORT_SYMBOL(sock_recv_errqueue);
2533 * Get a socket option on an socket.
2535 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2536 * asynchronous errors should be reported by getsockopt. We assume
2537 * this means if you specify SO_ERROR (otherwise whats the point of it).
2539 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2540 char __user *optval, int __user *optlen)
2542 struct sock *sk = sock->sk;
2544 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2546 EXPORT_SYMBOL(sock_common_getsockopt);
2548 #ifdef CONFIG_COMPAT
2549 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2550 char __user *optval, int __user *optlen)
2552 struct sock *sk = sock->sk;
2554 if (sk->sk_prot->compat_getsockopt != NULL)
2555 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2557 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2559 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2562 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2563 struct msghdr *msg, size_t size, int flags)
2565 struct sock *sk = sock->sk;
2569 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2570 flags & ~MSG_DONTWAIT, &addr_len);
2572 msg->msg_namelen = addr_len;
2575 EXPORT_SYMBOL(sock_common_recvmsg);
2578 * Set socket options on an inet socket.
2580 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2581 char __user *optval, unsigned int optlen)
2583 struct sock *sk = sock->sk;
2585 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2587 EXPORT_SYMBOL(sock_common_setsockopt);
2589 #ifdef CONFIG_COMPAT
2590 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2591 char __user *optval, unsigned int optlen)
2593 struct sock *sk = sock->sk;
2595 if (sk->sk_prot->compat_setsockopt != NULL)
2596 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2598 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2600 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2603 void sk_common_release(struct sock *sk)
2605 if (sk->sk_prot->destroy)
2606 sk->sk_prot->destroy(sk);
2609 * Observation: when sock_common_release is called, processes have
2610 * no access to socket. But net still has.
2611 * Step one, detach it from networking:
2613 * A. Remove from hash tables.
2616 sk->sk_prot->unhash(sk);
2619 * In this point socket cannot receive new packets, but it is possible
2620 * that some packets are in flight because some CPU runs receiver and
2621 * did hash table lookup before we unhashed socket. They will achieve
2622 * receive queue and will be purged by socket destructor.
2624 * Also we still have packets pending on receive queue and probably,
2625 * our own packets waiting in device queues. sock_destroy will drain
2626 * receive queue, but transmitted packets will delay socket destruction
2627 * until the last reference will be released.
2632 xfrm_sk_free_policy(sk);
2634 sk_refcnt_debug_release(sk);
2636 if (sk->sk_frag.page) {
2637 put_page(sk->sk_frag.page);
2638 sk->sk_frag.page = NULL;
2643 EXPORT_SYMBOL(sk_common_release);
2645 #ifdef CONFIG_PROC_FS
2646 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2648 int val[PROTO_INUSE_NR];
2651 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2653 #ifdef CONFIG_NET_NS
2654 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2656 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2658 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2660 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2662 int cpu, idx = prot->inuse_idx;
2665 for_each_possible_cpu(cpu)
2666 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2668 return res >= 0 ? res : 0;
2670 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2672 static int __net_init sock_inuse_init_net(struct net *net)
2674 net->core.inuse = alloc_percpu(struct prot_inuse);
2675 return net->core.inuse ? 0 : -ENOMEM;
2678 static void __net_exit sock_inuse_exit_net(struct net *net)
2680 free_percpu(net->core.inuse);
2683 static struct pernet_operations net_inuse_ops = {
2684 .init = sock_inuse_init_net,
2685 .exit = sock_inuse_exit_net,
2688 static __init int net_inuse_init(void)
2690 if (register_pernet_subsys(&net_inuse_ops))
2691 panic("Cannot initialize net inuse counters");
2696 core_initcall(net_inuse_init);
2698 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2700 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2702 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2704 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2706 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2708 int cpu, idx = prot->inuse_idx;
2711 for_each_possible_cpu(cpu)
2712 res += per_cpu(prot_inuse, cpu).val[idx];
2714 return res >= 0 ? res : 0;
2716 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2719 static void assign_proto_idx(struct proto *prot)
2721 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2723 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2724 pr_err("PROTO_INUSE_NR exhausted\n");
2728 set_bit(prot->inuse_idx, proto_inuse_idx);
2731 static void release_proto_idx(struct proto *prot)
2733 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2734 clear_bit(prot->inuse_idx, proto_inuse_idx);
2737 static inline void assign_proto_idx(struct proto *prot)
2741 static inline void release_proto_idx(struct proto *prot)
2746 int proto_register(struct proto *prot, int alloc_slab)
2749 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2750 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2753 if (prot->slab == NULL) {
2754 pr_crit("%s: Can't create sock SLAB cache!\n",
2759 if (prot->rsk_prot != NULL) {
2760 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2761 if (prot->rsk_prot->slab_name == NULL)
2762 goto out_free_sock_slab;
2764 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2765 prot->rsk_prot->obj_size, 0,
2766 SLAB_HWCACHE_ALIGN, NULL);
2768 if (prot->rsk_prot->slab == NULL) {
2769 pr_crit("%s: Can't create request sock SLAB cache!\n",
2771 goto out_free_request_sock_slab_name;
2775 if (prot->twsk_prot != NULL) {
2776 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2778 if (prot->twsk_prot->twsk_slab_name == NULL)
2779 goto out_free_request_sock_slab;
2781 prot->twsk_prot->twsk_slab =
2782 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2783 prot->twsk_prot->twsk_obj_size,
2785 SLAB_HWCACHE_ALIGN |
2788 if (prot->twsk_prot->twsk_slab == NULL)
2789 goto out_free_timewait_sock_slab_name;
2793 mutex_lock(&proto_list_mutex);
2794 list_add(&prot->node, &proto_list);
2795 assign_proto_idx(prot);
2796 mutex_unlock(&proto_list_mutex);
2799 out_free_timewait_sock_slab_name:
2800 kfree(prot->twsk_prot->twsk_slab_name);
2801 out_free_request_sock_slab:
2802 if (prot->rsk_prot && prot->rsk_prot->slab) {
2803 kmem_cache_destroy(prot->rsk_prot->slab);
2804 prot->rsk_prot->slab = NULL;
2806 out_free_request_sock_slab_name:
2808 kfree(prot->rsk_prot->slab_name);
2810 kmem_cache_destroy(prot->slab);
2815 EXPORT_SYMBOL(proto_register);
2817 void proto_unregister(struct proto *prot)
2819 mutex_lock(&proto_list_mutex);
2820 release_proto_idx(prot);
2821 list_del(&prot->node);
2822 mutex_unlock(&proto_list_mutex);
2824 if (prot->slab != NULL) {
2825 kmem_cache_destroy(prot->slab);
2829 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2830 kmem_cache_destroy(prot->rsk_prot->slab);
2831 kfree(prot->rsk_prot->slab_name);
2832 prot->rsk_prot->slab = NULL;
2835 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2836 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2837 kfree(prot->twsk_prot->twsk_slab_name);
2838 prot->twsk_prot->twsk_slab = NULL;
2841 EXPORT_SYMBOL(proto_unregister);
2843 #ifdef CONFIG_PROC_FS
2844 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2845 __acquires(proto_list_mutex)
2847 mutex_lock(&proto_list_mutex);
2848 return seq_list_start_head(&proto_list, *pos);
2851 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2853 return seq_list_next(v, &proto_list, pos);
2856 static void proto_seq_stop(struct seq_file *seq, void *v)
2857 __releases(proto_list_mutex)
2859 mutex_unlock(&proto_list_mutex);
2862 static char proto_method_implemented(const void *method)
2864 return method == NULL ? 'n' : 'y';
2866 static long sock_prot_memory_allocated(struct proto *proto)
2868 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2871 static char *sock_prot_memory_pressure(struct proto *proto)
2873 return proto->memory_pressure != NULL ?
2874 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2877 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2880 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2881 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2884 sock_prot_inuse_get(seq_file_net(seq), proto),
2885 sock_prot_memory_allocated(proto),
2886 sock_prot_memory_pressure(proto),
2888 proto->slab == NULL ? "no" : "yes",
2889 module_name(proto->owner),
2890 proto_method_implemented(proto->close),
2891 proto_method_implemented(proto->connect),
2892 proto_method_implemented(proto->disconnect),
2893 proto_method_implemented(proto->accept),
2894 proto_method_implemented(proto->ioctl),
2895 proto_method_implemented(proto->init),
2896 proto_method_implemented(proto->destroy),
2897 proto_method_implemented(proto->shutdown),
2898 proto_method_implemented(proto->setsockopt),
2899 proto_method_implemented(proto->getsockopt),
2900 proto_method_implemented(proto->sendmsg),
2901 proto_method_implemented(proto->recvmsg),
2902 proto_method_implemented(proto->sendpage),
2903 proto_method_implemented(proto->bind),
2904 proto_method_implemented(proto->backlog_rcv),
2905 proto_method_implemented(proto->hash),
2906 proto_method_implemented(proto->unhash),
2907 proto_method_implemented(proto->get_port),
2908 proto_method_implemented(proto->enter_memory_pressure));
2911 static int proto_seq_show(struct seq_file *seq, void *v)
2913 if (v == &proto_list)
2914 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2923 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2925 proto_seq_printf(seq, list_entry(v, struct proto, node));
2929 static const struct seq_operations proto_seq_ops = {
2930 .start = proto_seq_start,
2931 .next = proto_seq_next,
2932 .stop = proto_seq_stop,
2933 .show = proto_seq_show,
2936 static int proto_seq_open(struct inode *inode, struct file *file)
2938 return seq_open_net(inode, file, &proto_seq_ops,
2939 sizeof(struct seq_net_private));
2942 static const struct file_operations proto_seq_fops = {
2943 .owner = THIS_MODULE,
2944 .open = proto_seq_open,
2946 .llseek = seq_lseek,
2947 .release = seq_release_net,
2950 static __net_init int proto_init_net(struct net *net)
2952 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2958 static __net_exit void proto_exit_net(struct net *net)
2960 remove_proc_entry("protocols", net->proc_net);
2964 static __net_initdata struct pernet_operations proto_net_ops = {
2965 .init = proto_init_net,
2966 .exit = proto_exit_net,
2969 static int __init proto_init(void)
2971 return register_pernet_subsys(&proto_net_ops);
2974 subsys_initcall(proto_init);
2976 #endif /* PROC_FS */