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/types.h>
97 #include <linux/socket.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
115 #include <linux/user_namespace.h>
116 #include <linux/static_key.h>
117 #include <linux/memcontrol.h>
118 #include <linux/prefetch.h>
120 #include <asm/uaccess.h>
122 #include <linux/netdevice.h>
123 #include <net/protocol.h>
124 #include <linux/skbuff.h>
125 #include <net/net_namespace.h>
126 #include <net/request_sock.h>
127 #include <net/sock.h>
128 #include <linux/net_tstamp.h>
129 #include <net/xfrm.h>
130 #include <linux/ipsec.h>
131 #include <net/cls_cgroup.h>
132 #include <net/netprio_cgroup.h>
134 #include <linux/filter.h>
136 #include <trace/events/sock.h>
142 static DEFINE_MUTEX(proto_list_mutex);
143 static LIST_HEAD(proto_list);
145 #ifdef CONFIG_MEMCG_KMEM
146 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
151 mutex_lock(&proto_list_mutex);
152 list_for_each_entry(proto, &proto_list, node) {
153 if (proto->init_cgroup) {
154 ret = proto->init_cgroup(memcg, ss);
160 mutex_unlock(&proto_list_mutex);
163 list_for_each_entry_continue_reverse(proto, &proto_list, node)
164 if (proto->destroy_cgroup)
165 proto->destroy_cgroup(memcg);
166 mutex_unlock(&proto_list_mutex);
170 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
174 mutex_lock(&proto_list_mutex);
175 list_for_each_entry_reverse(proto, &proto_list, node)
176 if (proto->destroy_cgroup)
177 proto->destroy_cgroup(memcg);
178 mutex_unlock(&proto_list_mutex);
183 * Each address family might have different locking rules, so we have
184 * one slock key per address family:
186 static struct lock_class_key af_family_keys[AF_MAX];
187 static struct lock_class_key af_family_slock_keys[AF_MAX];
189 struct static_key memcg_socket_limit_enabled;
190 EXPORT_SYMBOL(memcg_socket_limit_enabled);
193 * Make lock validator output more readable. (we pre-construct these
194 * strings build-time, so that runtime initialization of socket
197 static const char *const af_family_key_strings[AF_MAX+1] = {
198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
213 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
223 "slock-27" , "slock-28" , "slock-AF_CAN" ,
224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
227 "slock-AF_NFC" , "slock-AF_MAX"
229 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
239 "clock-27" , "clock-28" , "clock-AF_CAN" ,
240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
243 "clock-AF_NFC" , "clock-AF_MAX"
247 * sk_callback_lock locking rules are per-address-family,
248 * so split the lock classes by using a per-AF key:
250 static struct lock_class_key af_callback_keys[AF_MAX];
252 /* Take into consideration the size of the struct sk_buff overhead in the
253 * determination of these values, since that is non-constant across
254 * platforms. This makes socket queueing behavior and performance
255 * not depend upon such differences.
257 #define _SK_MEM_PACKETS 256
258 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
259 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
262 /* Run time adjustable parameters. */
263 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
264 EXPORT_SYMBOL(sysctl_wmem_max);
265 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
266 EXPORT_SYMBOL(sysctl_rmem_max);
267 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
268 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
270 /* Maximal space eaten by iovec or ancillary data plus some space */
271 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
272 EXPORT_SYMBOL(sysctl_optmem_max);
275 * sk_set_memalloc - sets %SOCK_MEMALLOC
276 * @sk: socket to set it on
278 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
279 * It's the responsibility of the admin to adjust min_free_kbytes
280 * to meet the requirements
282 void sk_set_memalloc(struct sock *sk)
284 sock_set_flag(sk, SOCK_MEMALLOC);
285 sk->sk_allocation |= __GFP_MEMALLOC;
287 EXPORT_SYMBOL_GPL(sk_set_memalloc);
289 void sk_clear_memalloc(struct sock *sk)
291 sock_reset_flag(sk, SOCK_MEMALLOC);
292 sk->sk_allocation &= ~__GFP_MEMALLOC;
294 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
296 #if defined(CONFIG_CGROUPS)
297 #if !defined(CONFIG_NET_CLS_CGROUP)
298 int net_cls_subsys_id = -1;
299 EXPORT_SYMBOL_GPL(net_cls_subsys_id);
301 #if !defined(CONFIG_NETPRIO_CGROUP)
302 int net_prio_subsys_id = -1;
303 EXPORT_SYMBOL_GPL(net_prio_subsys_id);
307 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
311 if (optlen < sizeof(tv))
313 if (copy_from_user(&tv, optval, sizeof(tv)))
315 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
319 static int warned __read_mostly;
322 if (warned < 10 && net_ratelimit()) {
324 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
325 __func__, current->comm, task_pid_nr(current));
329 *timeo_p = MAX_SCHEDULE_TIMEOUT;
330 if (tv.tv_sec == 0 && tv.tv_usec == 0)
332 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
333 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
337 static void sock_warn_obsolete_bsdism(const char *name)
340 static char warncomm[TASK_COMM_LEN];
341 if (strcmp(warncomm, current->comm) && warned < 5) {
342 strcpy(warncomm, current->comm);
343 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
349 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
351 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
353 if (sk->sk_flags & flags) {
354 sk->sk_flags &= ~flags;
355 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
356 net_disable_timestamp();
361 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
366 struct sk_buff_head *list = &sk->sk_receive_queue;
368 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
369 atomic_inc(&sk->sk_drops);
370 trace_sock_rcvqueue_full(sk, skb);
374 err = sk_filter(sk, skb);
378 if (!sk_rmem_schedule(sk, skb->truesize)) {
379 atomic_inc(&sk->sk_drops);
384 skb_set_owner_r(skb, sk);
386 /* Cache the SKB length before we tack it onto the receive
387 * queue. Once it is added it no longer belongs to us and
388 * may be freed by other threads of control pulling packets
393 /* we escape from rcu protected region, make sure we dont leak
398 spin_lock_irqsave(&list->lock, flags);
399 skb->dropcount = atomic_read(&sk->sk_drops);
400 __skb_queue_tail(list, skb);
401 spin_unlock_irqrestore(&list->lock, flags);
403 if (!sock_flag(sk, SOCK_DEAD))
404 sk->sk_data_ready(sk, skb_len);
407 EXPORT_SYMBOL(sock_queue_rcv_skb);
409 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
411 int rc = NET_RX_SUCCESS;
413 if (sk_filter(sk, skb))
414 goto discard_and_relse;
418 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
419 atomic_inc(&sk->sk_drops);
420 goto discard_and_relse;
423 bh_lock_sock_nested(sk);
426 if (!sock_owned_by_user(sk)) {
428 * trylock + unlock semantics:
430 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
432 rc = sk_backlog_rcv(sk, skb);
434 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
435 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
437 atomic_inc(&sk->sk_drops);
438 goto discard_and_relse;
449 EXPORT_SYMBOL(sk_receive_skb);
451 void sk_reset_txq(struct sock *sk)
453 sk_tx_queue_clear(sk);
455 EXPORT_SYMBOL(sk_reset_txq);
457 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
459 struct dst_entry *dst = __sk_dst_get(sk);
461 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
462 sk_tx_queue_clear(sk);
463 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
470 EXPORT_SYMBOL(__sk_dst_check);
472 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
474 struct dst_entry *dst = sk_dst_get(sk);
476 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
484 EXPORT_SYMBOL(sk_dst_check);
486 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
488 int ret = -ENOPROTOOPT;
489 #ifdef CONFIG_NETDEVICES
490 struct net *net = sock_net(sk);
491 char devname[IFNAMSIZ];
496 if (!capable(CAP_NET_RAW))
503 /* Bind this socket to a particular device like "eth0",
504 * as specified in the passed interface name. If the
505 * name is "" or the option length is zero the socket
508 if (optlen > IFNAMSIZ - 1)
509 optlen = IFNAMSIZ - 1;
510 memset(devname, 0, sizeof(devname));
513 if (copy_from_user(devname, optval, optlen))
517 if (devname[0] != '\0') {
518 struct net_device *dev;
521 dev = dev_get_by_name_rcu(net, devname);
523 index = dev->ifindex;
531 sk->sk_bound_dev_if = index;
543 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
546 sock_set_flag(sk, bit);
548 sock_reset_flag(sk, bit);
552 * This is meant for all protocols to use and covers goings on
553 * at the socket level. Everything here is generic.
556 int sock_setsockopt(struct socket *sock, int level, int optname,
557 char __user *optval, unsigned int optlen)
559 struct sock *sk = sock->sk;
566 * Options without arguments
569 if (optname == SO_BINDTODEVICE)
570 return sock_bindtodevice(sk, optval, optlen);
572 if (optlen < sizeof(int))
575 if (get_user(val, (int __user *)optval))
578 valbool = val ? 1 : 0;
584 if (val && !capable(CAP_NET_ADMIN))
587 sock_valbool_flag(sk, SOCK_DBG, valbool);
590 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
599 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
602 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
605 /* Don't error on this BSD doesn't and if you think
606 * about it this is right. Otherwise apps have to
607 * play 'guess the biggest size' games. RCVBUF/SNDBUF
608 * are treated in BSD as hints
610 val = min_t(u32, val, sysctl_wmem_max);
612 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
613 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
614 /* Wake up sending tasks if we upped the value. */
615 sk->sk_write_space(sk);
619 if (!capable(CAP_NET_ADMIN)) {
626 /* Don't error on this BSD doesn't and if you think
627 * about it this is right. Otherwise apps have to
628 * play 'guess the biggest size' games. RCVBUF/SNDBUF
629 * are treated in BSD as hints
631 val = min_t(u32, val, sysctl_rmem_max);
633 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
635 * We double it on the way in to account for
636 * "struct sk_buff" etc. overhead. Applications
637 * assume that the SO_RCVBUF setting they make will
638 * allow that much actual data to be received on that
641 * Applications are unaware that "struct sk_buff" and
642 * other overheads allocate from the receive buffer
643 * during socket buffer allocation.
645 * And after considering the possible alternatives,
646 * returning the value we actually used in getsockopt
647 * is the most desirable behavior.
649 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
653 if (!capable(CAP_NET_ADMIN)) {
661 if (sk->sk_protocol == IPPROTO_TCP)
662 tcp_set_keepalive(sk, valbool);
664 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
668 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
672 sk->sk_no_check = valbool;
676 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
677 sk->sk_priority = val;
683 if (optlen < sizeof(ling)) {
684 ret = -EINVAL; /* 1003.1g */
687 if (copy_from_user(&ling, optval, sizeof(ling))) {
692 sock_reset_flag(sk, SOCK_LINGER);
694 #if (BITS_PER_LONG == 32)
695 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
696 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
699 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
700 sock_set_flag(sk, SOCK_LINGER);
705 sock_warn_obsolete_bsdism("setsockopt");
710 set_bit(SOCK_PASSCRED, &sock->flags);
712 clear_bit(SOCK_PASSCRED, &sock->flags);
718 if (optname == SO_TIMESTAMP)
719 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
721 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
722 sock_set_flag(sk, SOCK_RCVTSTAMP);
723 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
725 sock_reset_flag(sk, SOCK_RCVTSTAMP);
726 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
730 case SO_TIMESTAMPING:
731 if (val & ~SOF_TIMESTAMPING_MASK) {
735 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
736 val & SOF_TIMESTAMPING_TX_HARDWARE);
737 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
738 val & SOF_TIMESTAMPING_TX_SOFTWARE);
739 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
740 val & SOF_TIMESTAMPING_RX_HARDWARE);
741 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
742 sock_enable_timestamp(sk,
743 SOCK_TIMESTAMPING_RX_SOFTWARE);
745 sock_disable_timestamp(sk,
746 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
747 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
748 val & SOF_TIMESTAMPING_SOFTWARE);
749 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
750 val & SOF_TIMESTAMPING_SYS_HARDWARE);
751 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
752 val & SOF_TIMESTAMPING_RAW_HARDWARE);
758 sk->sk_rcvlowat = val ? : 1;
762 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
766 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
769 case SO_ATTACH_FILTER:
771 if (optlen == sizeof(struct sock_fprog)) {
772 struct sock_fprog fprog;
775 if (copy_from_user(&fprog, optval, sizeof(fprog)))
778 ret = sk_attach_filter(&fprog, sk);
782 case SO_DETACH_FILTER:
783 ret = sk_detach_filter(sk);
788 set_bit(SOCK_PASSSEC, &sock->flags);
790 clear_bit(SOCK_PASSSEC, &sock->flags);
793 if (!capable(CAP_NET_ADMIN))
799 /* We implement the SO_SNDLOWAT etc to
800 not be settable (1003.1g 5.3) */
802 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
806 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
810 if (sock->ops->set_peek_off)
811 sock->ops->set_peek_off(sk, val);
817 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
827 EXPORT_SYMBOL(sock_setsockopt);
830 void cred_to_ucred(struct pid *pid, const struct cred *cred,
833 ucred->pid = pid_vnr(pid);
834 ucred->uid = ucred->gid = -1;
836 struct user_namespace *current_ns = current_user_ns();
838 ucred->uid = from_kuid(current_ns, cred->euid);
839 ucred->gid = from_kgid(current_ns, cred->egid);
842 EXPORT_SYMBOL_GPL(cred_to_ucred);
844 int sock_getsockopt(struct socket *sock, int level, int optname,
845 char __user *optval, int __user *optlen)
847 struct sock *sk = sock->sk;
855 int lv = sizeof(int);
858 if (get_user(len, optlen))
863 memset(&v, 0, sizeof(v));
867 v.val = sock_flag(sk, SOCK_DBG);
871 v.val = sock_flag(sk, SOCK_LOCALROUTE);
875 v.val = sock_flag(sk, SOCK_BROADCAST);
879 v.val = sk->sk_sndbuf;
883 v.val = sk->sk_rcvbuf;
887 v.val = sk->sk_reuse;
891 v.val = sock_flag(sk, SOCK_KEEPOPEN);
899 v.val = sk->sk_protocol;
903 v.val = sk->sk_family;
907 v.val = -sock_error(sk);
909 v.val = xchg(&sk->sk_err_soft, 0);
913 v.val = sock_flag(sk, SOCK_URGINLINE);
917 v.val = sk->sk_no_check;
921 v.val = sk->sk_priority;
926 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
927 v.ling.l_linger = sk->sk_lingertime / HZ;
931 sock_warn_obsolete_bsdism("getsockopt");
935 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
936 !sock_flag(sk, SOCK_RCVTSTAMPNS);
940 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
943 case SO_TIMESTAMPING:
945 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
946 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
947 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
948 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
949 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
950 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
951 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
952 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
953 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
954 v.val |= SOF_TIMESTAMPING_SOFTWARE;
955 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
956 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
957 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
958 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
962 lv = sizeof(struct timeval);
963 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
967 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
968 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
973 lv = sizeof(struct timeval);
974 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
978 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
979 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
984 v.val = sk->sk_rcvlowat;
992 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
997 struct ucred peercred;
998 if (len > sizeof(peercred))
999 len = sizeof(peercred);
1000 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1001 if (copy_to_user(optval, &peercred, len))
1010 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1014 if (copy_to_user(optval, address, len))
1019 /* Dubious BSD thing... Probably nobody even uses it, but
1020 * the UNIX standard wants it for whatever reason... -DaveM
1023 v.val = sk->sk_state == TCP_LISTEN;
1027 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1031 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1034 v.val = sk->sk_mark;
1038 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1041 case SO_WIFI_STATUS:
1042 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1046 if (!sock->ops->set_peek_off)
1049 v.val = sk->sk_peek_off;
1052 v.val = sock_flag(sk, SOCK_NOFCS);
1055 return -ENOPROTOOPT;
1060 if (copy_to_user(optval, &v, len))
1063 if (put_user(len, optlen))
1069 * Initialize an sk_lock.
1071 * (We also register the sk_lock with the lock validator.)
1073 static inline void sock_lock_init(struct sock *sk)
1075 sock_lock_init_class_and_name(sk,
1076 af_family_slock_key_strings[sk->sk_family],
1077 af_family_slock_keys + sk->sk_family,
1078 af_family_key_strings[sk->sk_family],
1079 af_family_keys + sk->sk_family);
1083 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1084 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1085 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1087 static void sock_copy(struct sock *nsk, const struct sock *osk)
1089 #ifdef CONFIG_SECURITY_NETWORK
1090 void *sptr = nsk->sk_security;
1092 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1094 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1095 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1097 #ifdef CONFIG_SECURITY_NETWORK
1098 nsk->sk_security = sptr;
1099 security_sk_clone(osk, nsk);
1104 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1105 * un-modified. Special care is taken when initializing object to zero.
1107 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1109 if (offsetof(struct sock, sk_node.next) != 0)
1110 memset(sk, 0, offsetof(struct sock, sk_node.next));
1111 memset(&sk->sk_node.pprev, 0,
1112 size - offsetof(struct sock, sk_node.pprev));
1115 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1117 unsigned long nulls1, nulls2;
1119 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1120 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1121 if (nulls1 > nulls2)
1122 swap(nulls1, nulls2);
1125 memset((char *)sk, 0, nulls1);
1126 memset((char *)sk + nulls1 + sizeof(void *), 0,
1127 nulls2 - nulls1 - sizeof(void *));
1128 memset((char *)sk + nulls2 + sizeof(void *), 0,
1129 size - nulls2 - sizeof(void *));
1131 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1133 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1137 struct kmem_cache *slab;
1141 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1144 if (priority & __GFP_ZERO) {
1146 prot->clear_sk(sk, prot->obj_size);
1148 sk_prot_clear_nulls(sk, prot->obj_size);
1151 sk = kmalloc(prot->obj_size, priority);
1154 kmemcheck_annotate_bitfield(sk, flags);
1156 if (security_sk_alloc(sk, family, priority))
1159 if (!try_module_get(prot->owner))
1161 sk_tx_queue_clear(sk);
1167 security_sk_free(sk);
1170 kmem_cache_free(slab, sk);
1176 static void sk_prot_free(struct proto *prot, struct sock *sk)
1178 struct kmem_cache *slab;
1179 struct module *owner;
1181 owner = prot->owner;
1184 security_sk_free(sk);
1186 kmem_cache_free(slab, sk);
1192 #ifdef CONFIG_CGROUPS
1193 void sock_update_classid(struct sock *sk)
1197 rcu_read_lock(); /* doing current task, which cannot vanish. */
1198 classid = task_cls_classid(current);
1200 if (classid && classid != sk->sk_classid)
1201 sk->sk_classid = classid;
1203 EXPORT_SYMBOL(sock_update_classid);
1205 void sock_update_netprioidx(struct sock *sk, struct task_struct *task)
1210 sk->sk_cgrp_prioidx = task_netprioidx(task);
1212 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1216 * sk_alloc - All socket objects are allocated here
1217 * @net: the applicable net namespace
1218 * @family: protocol family
1219 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1220 * @prot: struct proto associated with this new sock instance
1222 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1227 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1229 sk->sk_family = family;
1231 * See comment in struct sock definition to understand
1232 * why we need sk_prot_creator -acme
1234 sk->sk_prot = sk->sk_prot_creator = prot;
1236 sock_net_set(sk, get_net(net));
1237 atomic_set(&sk->sk_wmem_alloc, 1);
1239 sock_update_classid(sk);
1240 sock_update_netprioidx(sk, current);
1245 EXPORT_SYMBOL(sk_alloc);
1247 static void __sk_free(struct sock *sk)
1249 struct sk_filter *filter;
1251 if (sk->sk_destruct)
1252 sk->sk_destruct(sk);
1254 filter = rcu_dereference_check(sk->sk_filter,
1255 atomic_read(&sk->sk_wmem_alloc) == 0);
1257 sk_filter_uncharge(sk, filter);
1258 RCU_INIT_POINTER(sk->sk_filter, NULL);
1261 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1263 if (atomic_read(&sk->sk_omem_alloc))
1264 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1265 __func__, atomic_read(&sk->sk_omem_alloc));
1267 if (sk->sk_peer_cred)
1268 put_cred(sk->sk_peer_cred);
1269 put_pid(sk->sk_peer_pid);
1270 put_net(sock_net(sk));
1271 sk_prot_free(sk->sk_prot_creator, sk);
1274 void sk_free(struct sock *sk)
1277 * We subtract one from sk_wmem_alloc and can know if
1278 * some packets are still in some tx queue.
1279 * If not null, sock_wfree() will call __sk_free(sk) later
1281 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1284 EXPORT_SYMBOL(sk_free);
1287 * Last sock_put should drop reference to sk->sk_net. It has already
1288 * been dropped in sk_change_net. Taking reference to stopping namespace
1290 * Take reference to a socket to remove it from hash _alive_ and after that
1291 * destroy it in the context of init_net.
1293 void sk_release_kernel(struct sock *sk)
1295 if (sk == NULL || sk->sk_socket == NULL)
1299 sock_release(sk->sk_socket);
1300 release_net(sock_net(sk));
1301 sock_net_set(sk, get_net(&init_net));
1304 EXPORT_SYMBOL(sk_release_kernel);
1306 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1308 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1309 sock_update_memcg(newsk);
1313 * sk_clone_lock - clone a socket, and lock its clone
1314 * @sk: the socket to clone
1315 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1317 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1319 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1323 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1324 if (newsk != NULL) {
1325 struct sk_filter *filter;
1327 sock_copy(newsk, sk);
1330 get_net(sock_net(newsk));
1331 sk_node_init(&newsk->sk_node);
1332 sock_lock_init(newsk);
1333 bh_lock_sock(newsk);
1334 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1335 newsk->sk_backlog.len = 0;
1337 atomic_set(&newsk->sk_rmem_alloc, 0);
1339 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1341 atomic_set(&newsk->sk_wmem_alloc, 1);
1342 atomic_set(&newsk->sk_omem_alloc, 0);
1343 skb_queue_head_init(&newsk->sk_receive_queue);
1344 skb_queue_head_init(&newsk->sk_write_queue);
1345 #ifdef CONFIG_NET_DMA
1346 skb_queue_head_init(&newsk->sk_async_wait_queue);
1349 spin_lock_init(&newsk->sk_dst_lock);
1350 rwlock_init(&newsk->sk_callback_lock);
1351 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1352 af_callback_keys + newsk->sk_family,
1353 af_family_clock_key_strings[newsk->sk_family]);
1355 newsk->sk_dst_cache = NULL;
1356 newsk->sk_wmem_queued = 0;
1357 newsk->sk_forward_alloc = 0;
1358 newsk->sk_send_head = NULL;
1359 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1361 sock_reset_flag(newsk, SOCK_DONE);
1362 skb_queue_head_init(&newsk->sk_error_queue);
1364 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1366 sk_filter_charge(newsk, filter);
1368 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1369 /* It is still raw copy of parent, so invalidate
1370 * destructor and make plain sk_free() */
1371 newsk->sk_destruct = NULL;
1372 bh_unlock_sock(newsk);
1379 newsk->sk_priority = 0;
1381 * Before updating sk_refcnt, we must commit prior changes to memory
1382 * (Documentation/RCU/rculist_nulls.txt for details)
1385 atomic_set(&newsk->sk_refcnt, 2);
1388 * Increment the counter in the same struct proto as the master
1389 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1390 * is the same as sk->sk_prot->socks, as this field was copied
1393 * This _changes_ the previous behaviour, where
1394 * tcp_create_openreq_child always was incrementing the
1395 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1396 * to be taken into account in all callers. -acme
1398 sk_refcnt_debug_inc(newsk);
1399 sk_set_socket(newsk, NULL);
1400 newsk->sk_wq = NULL;
1402 sk_update_clone(sk, newsk);
1404 if (newsk->sk_prot->sockets_allocated)
1405 sk_sockets_allocated_inc(newsk);
1407 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1408 net_enable_timestamp();
1413 EXPORT_SYMBOL_GPL(sk_clone_lock);
1415 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1417 __sk_dst_set(sk, dst);
1418 sk->sk_route_caps = dst->dev->features;
1419 if (sk->sk_route_caps & NETIF_F_GSO)
1420 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1421 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1422 if (sk_can_gso(sk)) {
1423 if (dst->header_len) {
1424 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1426 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1427 sk->sk_gso_max_size = dst->dev->gso_max_size;
1431 EXPORT_SYMBOL_GPL(sk_setup_caps);
1433 void __init sk_init(void)
1435 if (totalram_pages <= 4096) {
1436 sysctl_wmem_max = 32767;
1437 sysctl_rmem_max = 32767;
1438 sysctl_wmem_default = 32767;
1439 sysctl_rmem_default = 32767;
1440 } else if (totalram_pages >= 131072) {
1441 sysctl_wmem_max = 131071;
1442 sysctl_rmem_max = 131071;
1447 * Simple resource managers for sockets.
1452 * Write buffer destructor automatically called from kfree_skb.
1454 void sock_wfree(struct sk_buff *skb)
1456 struct sock *sk = skb->sk;
1457 unsigned int len = skb->truesize;
1459 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1461 * Keep a reference on sk_wmem_alloc, this will be released
1462 * after sk_write_space() call
1464 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1465 sk->sk_write_space(sk);
1469 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1470 * could not do because of in-flight packets
1472 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1475 EXPORT_SYMBOL(sock_wfree);
1478 * Read buffer destructor automatically called from kfree_skb.
1480 void sock_rfree(struct sk_buff *skb)
1482 struct sock *sk = skb->sk;
1483 unsigned int len = skb->truesize;
1485 atomic_sub(len, &sk->sk_rmem_alloc);
1486 sk_mem_uncharge(sk, len);
1488 EXPORT_SYMBOL(sock_rfree);
1490 void sock_edemux(struct sk_buff *skb)
1494 EXPORT_SYMBOL(sock_edemux);
1496 int sock_i_uid(struct sock *sk)
1500 read_lock_bh(&sk->sk_callback_lock);
1501 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1502 read_unlock_bh(&sk->sk_callback_lock);
1505 EXPORT_SYMBOL(sock_i_uid);
1507 unsigned long sock_i_ino(struct sock *sk)
1511 read_lock_bh(&sk->sk_callback_lock);
1512 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1513 read_unlock_bh(&sk->sk_callback_lock);
1516 EXPORT_SYMBOL(sock_i_ino);
1519 * Allocate a skb from the socket's send buffer.
1521 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1524 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1525 struct sk_buff *skb = alloc_skb(size, priority);
1527 skb_set_owner_w(skb, sk);
1533 EXPORT_SYMBOL(sock_wmalloc);
1536 * Allocate a skb from the socket's receive buffer.
1538 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1541 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1542 struct sk_buff *skb = alloc_skb(size, priority);
1544 skb_set_owner_r(skb, sk);
1552 * Allocate a memory block from the socket's option memory buffer.
1554 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1556 if ((unsigned int)size <= sysctl_optmem_max &&
1557 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1559 /* First do the add, to avoid the race if kmalloc
1562 atomic_add(size, &sk->sk_omem_alloc);
1563 mem = kmalloc(size, priority);
1566 atomic_sub(size, &sk->sk_omem_alloc);
1570 EXPORT_SYMBOL(sock_kmalloc);
1573 * Free an option memory block.
1575 void sock_kfree_s(struct sock *sk, void *mem, int size)
1578 atomic_sub(size, &sk->sk_omem_alloc);
1580 EXPORT_SYMBOL(sock_kfree_s);
1582 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1583 I think, these locks should be removed for datagram sockets.
1585 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1589 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1593 if (signal_pending(current))
1595 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1596 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1597 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1599 if (sk->sk_shutdown & SEND_SHUTDOWN)
1603 timeo = schedule_timeout(timeo);
1605 finish_wait(sk_sleep(sk), &wait);
1611 * Generic send/receive buffer handlers
1614 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1615 unsigned long data_len, int noblock,
1618 struct sk_buff *skb;
1622 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1625 if (npages > MAX_SKB_FRAGS)
1628 gfp_mask = sk->sk_allocation;
1629 if (gfp_mask & __GFP_WAIT)
1630 gfp_mask |= __GFP_REPEAT;
1632 timeo = sock_sndtimeo(sk, noblock);
1634 err = sock_error(sk);
1639 if (sk->sk_shutdown & SEND_SHUTDOWN)
1642 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1643 skb = alloc_skb(header_len, gfp_mask);
1647 /* No pages, we're done... */
1651 skb->truesize += data_len;
1652 skb_shinfo(skb)->nr_frags = npages;
1653 for (i = 0; i < npages; i++) {
1656 page = alloc_pages(sk->sk_allocation, 0);
1659 skb_shinfo(skb)->nr_frags = i;
1664 __skb_fill_page_desc(skb, i,
1666 (data_len >= PAGE_SIZE ?
1669 data_len -= PAGE_SIZE;
1672 /* Full success... */
1678 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1679 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1683 if (signal_pending(current))
1685 timeo = sock_wait_for_wmem(sk, timeo);
1688 skb_set_owner_w(skb, sk);
1692 err = sock_intr_errno(timeo);
1697 EXPORT_SYMBOL(sock_alloc_send_pskb);
1699 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1700 int noblock, int *errcode)
1702 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1704 EXPORT_SYMBOL(sock_alloc_send_skb);
1706 static void __lock_sock(struct sock *sk)
1707 __releases(&sk->sk_lock.slock)
1708 __acquires(&sk->sk_lock.slock)
1713 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1714 TASK_UNINTERRUPTIBLE);
1715 spin_unlock_bh(&sk->sk_lock.slock);
1717 spin_lock_bh(&sk->sk_lock.slock);
1718 if (!sock_owned_by_user(sk))
1721 finish_wait(&sk->sk_lock.wq, &wait);
1724 static void __release_sock(struct sock *sk)
1725 __releases(&sk->sk_lock.slock)
1726 __acquires(&sk->sk_lock.slock)
1728 struct sk_buff *skb = sk->sk_backlog.head;
1731 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1735 struct sk_buff *next = skb->next;
1738 WARN_ON_ONCE(skb_dst_is_noref(skb));
1740 sk_backlog_rcv(sk, skb);
1743 * We are in process context here with softirqs
1744 * disabled, use cond_resched_softirq() to preempt.
1745 * This is safe to do because we've taken the backlog
1748 cond_resched_softirq();
1751 } while (skb != NULL);
1754 } while ((skb = sk->sk_backlog.head) != NULL);
1757 * Doing the zeroing here guarantee we can not loop forever
1758 * while a wild producer attempts to flood us.
1760 sk->sk_backlog.len = 0;
1764 * sk_wait_data - wait for data to arrive at sk_receive_queue
1765 * @sk: sock to wait on
1766 * @timeo: for how long
1768 * Now socket state including sk->sk_err is changed only under lock,
1769 * hence we may omit checks after joining wait queue.
1770 * We check receive queue before schedule() only as optimization;
1771 * it is very likely that release_sock() added new data.
1773 int sk_wait_data(struct sock *sk, long *timeo)
1778 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1779 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1780 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1781 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1782 finish_wait(sk_sleep(sk), &wait);
1785 EXPORT_SYMBOL(sk_wait_data);
1788 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1790 * @size: memory size to allocate
1791 * @kind: allocation type
1793 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1794 * rmem allocation. This function assumes that protocols which have
1795 * memory_pressure use sk_wmem_queued as write buffer accounting.
1797 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1799 struct proto *prot = sk->sk_prot;
1800 int amt = sk_mem_pages(size);
1802 int parent_status = UNDER_LIMIT;
1804 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1806 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1809 if (parent_status == UNDER_LIMIT &&
1810 allocated <= sk_prot_mem_limits(sk, 0)) {
1811 sk_leave_memory_pressure(sk);
1815 /* Under pressure. (we or our parents) */
1816 if ((parent_status > SOFT_LIMIT) ||
1817 allocated > sk_prot_mem_limits(sk, 1))
1818 sk_enter_memory_pressure(sk);
1820 /* Over hard limit (we or our parents) */
1821 if ((parent_status == OVER_LIMIT) ||
1822 (allocated > sk_prot_mem_limits(sk, 2)))
1823 goto suppress_allocation;
1825 /* guarantee minimum buffer size under pressure */
1826 if (kind == SK_MEM_RECV) {
1827 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1830 } else { /* SK_MEM_SEND */
1831 if (sk->sk_type == SOCK_STREAM) {
1832 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1834 } else if (atomic_read(&sk->sk_wmem_alloc) <
1835 prot->sysctl_wmem[0])
1839 if (sk_has_memory_pressure(sk)) {
1842 if (!sk_under_memory_pressure(sk))
1844 alloc = sk_sockets_allocated_read_positive(sk);
1845 if (sk_prot_mem_limits(sk, 2) > alloc *
1846 sk_mem_pages(sk->sk_wmem_queued +
1847 atomic_read(&sk->sk_rmem_alloc) +
1848 sk->sk_forward_alloc))
1852 suppress_allocation:
1854 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1855 sk_stream_moderate_sndbuf(sk);
1857 /* Fail only if socket is _under_ its sndbuf.
1858 * In this case we cannot block, so that we have to fail.
1860 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1864 trace_sock_exceed_buf_limit(sk, prot, allocated);
1866 /* Alas. Undo changes. */
1867 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1869 sk_memory_allocated_sub(sk, amt);
1873 EXPORT_SYMBOL(__sk_mem_schedule);
1876 * __sk_reclaim - reclaim memory_allocated
1879 void __sk_mem_reclaim(struct sock *sk)
1881 sk_memory_allocated_sub(sk,
1882 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
1883 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1885 if (sk_under_memory_pressure(sk) &&
1886 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1887 sk_leave_memory_pressure(sk);
1889 EXPORT_SYMBOL(__sk_mem_reclaim);
1893 * Set of default routines for initialising struct proto_ops when
1894 * the protocol does not support a particular function. In certain
1895 * cases where it makes no sense for a protocol to have a "do nothing"
1896 * function, some default processing is provided.
1899 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1903 EXPORT_SYMBOL(sock_no_bind);
1905 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1910 EXPORT_SYMBOL(sock_no_connect);
1912 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1916 EXPORT_SYMBOL(sock_no_socketpair);
1918 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1922 EXPORT_SYMBOL(sock_no_accept);
1924 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1929 EXPORT_SYMBOL(sock_no_getname);
1931 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1935 EXPORT_SYMBOL(sock_no_poll);
1937 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1941 EXPORT_SYMBOL(sock_no_ioctl);
1943 int sock_no_listen(struct socket *sock, int backlog)
1947 EXPORT_SYMBOL(sock_no_listen);
1949 int sock_no_shutdown(struct socket *sock, int how)
1953 EXPORT_SYMBOL(sock_no_shutdown);
1955 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1956 char __user *optval, unsigned int optlen)
1960 EXPORT_SYMBOL(sock_no_setsockopt);
1962 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1963 char __user *optval, int __user *optlen)
1967 EXPORT_SYMBOL(sock_no_getsockopt);
1969 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1974 EXPORT_SYMBOL(sock_no_sendmsg);
1976 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1977 size_t len, int flags)
1981 EXPORT_SYMBOL(sock_no_recvmsg);
1983 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1985 /* Mirror missing mmap method error code */
1988 EXPORT_SYMBOL(sock_no_mmap);
1990 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1993 struct msghdr msg = {.msg_flags = flags};
1995 char *kaddr = kmap(page);
1996 iov.iov_base = kaddr + offset;
1998 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2002 EXPORT_SYMBOL(sock_no_sendpage);
2005 * Default Socket Callbacks
2008 static void sock_def_wakeup(struct sock *sk)
2010 struct socket_wq *wq;
2013 wq = rcu_dereference(sk->sk_wq);
2014 if (wq_has_sleeper(wq))
2015 wake_up_interruptible_all(&wq->wait);
2019 static void sock_def_error_report(struct sock *sk)
2021 struct socket_wq *wq;
2024 wq = rcu_dereference(sk->sk_wq);
2025 if (wq_has_sleeper(wq))
2026 wake_up_interruptible_poll(&wq->wait, POLLERR);
2027 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2031 static void sock_def_readable(struct sock *sk, int len)
2033 struct socket_wq *wq;
2036 wq = rcu_dereference(sk->sk_wq);
2037 if (wq_has_sleeper(wq))
2038 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2039 POLLRDNORM | POLLRDBAND);
2040 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2044 static void sock_def_write_space(struct sock *sk)
2046 struct socket_wq *wq;
2050 /* Do not wake up a writer until he can make "significant"
2053 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2054 wq = rcu_dereference(sk->sk_wq);
2055 if (wq_has_sleeper(wq))
2056 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2057 POLLWRNORM | POLLWRBAND);
2059 /* Should agree with poll, otherwise some programs break */
2060 if (sock_writeable(sk))
2061 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2067 static void sock_def_destruct(struct sock *sk)
2069 kfree(sk->sk_protinfo);
2072 void sk_send_sigurg(struct sock *sk)
2074 if (sk->sk_socket && sk->sk_socket->file)
2075 if (send_sigurg(&sk->sk_socket->file->f_owner))
2076 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2078 EXPORT_SYMBOL(sk_send_sigurg);
2080 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2081 unsigned long expires)
2083 if (!mod_timer(timer, expires))
2086 EXPORT_SYMBOL(sk_reset_timer);
2088 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2090 if (timer_pending(timer) && del_timer(timer))
2093 EXPORT_SYMBOL(sk_stop_timer);
2095 void sock_init_data(struct socket *sock, struct sock *sk)
2097 skb_queue_head_init(&sk->sk_receive_queue);
2098 skb_queue_head_init(&sk->sk_write_queue);
2099 skb_queue_head_init(&sk->sk_error_queue);
2100 #ifdef CONFIG_NET_DMA
2101 skb_queue_head_init(&sk->sk_async_wait_queue);
2104 sk->sk_send_head = NULL;
2106 init_timer(&sk->sk_timer);
2108 sk->sk_allocation = GFP_KERNEL;
2109 sk->sk_rcvbuf = sysctl_rmem_default;
2110 sk->sk_sndbuf = sysctl_wmem_default;
2111 sk->sk_state = TCP_CLOSE;
2112 sk_set_socket(sk, sock);
2114 sock_set_flag(sk, SOCK_ZAPPED);
2117 sk->sk_type = sock->type;
2118 sk->sk_wq = sock->wq;
2123 spin_lock_init(&sk->sk_dst_lock);
2124 rwlock_init(&sk->sk_callback_lock);
2125 lockdep_set_class_and_name(&sk->sk_callback_lock,
2126 af_callback_keys + sk->sk_family,
2127 af_family_clock_key_strings[sk->sk_family]);
2129 sk->sk_state_change = sock_def_wakeup;
2130 sk->sk_data_ready = sock_def_readable;
2131 sk->sk_write_space = sock_def_write_space;
2132 sk->sk_error_report = sock_def_error_report;
2133 sk->sk_destruct = sock_def_destruct;
2135 sk->sk_sndmsg_page = NULL;
2136 sk->sk_sndmsg_off = 0;
2137 sk->sk_peek_off = -1;
2139 sk->sk_peer_pid = NULL;
2140 sk->sk_peer_cred = NULL;
2141 sk->sk_write_pending = 0;
2142 sk->sk_rcvlowat = 1;
2143 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2144 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2146 sk->sk_stamp = ktime_set(-1L, 0);
2149 * Before updating sk_refcnt, we must commit prior changes to memory
2150 * (Documentation/RCU/rculist_nulls.txt for details)
2153 atomic_set(&sk->sk_refcnt, 1);
2154 atomic_set(&sk->sk_drops, 0);
2156 EXPORT_SYMBOL(sock_init_data);
2158 void lock_sock_nested(struct sock *sk, int subclass)
2161 spin_lock_bh(&sk->sk_lock.slock);
2162 if (sk->sk_lock.owned)
2164 sk->sk_lock.owned = 1;
2165 spin_unlock(&sk->sk_lock.slock);
2167 * The sk_lock has mutex_lock() semantics here:
2169 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2172 EXPORT_SYMBOL(lock_sock_nested);
2174 void release_sock(struct sock *sk)
2177 * The sk_lock has mutex_unlock() semantics:
2179 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2181 spin_lock_bh(&sk->sk_lock.slock);
2182 if (sk->sk_backlog.tail)
2185 if (sk->sk_prot->release_cb)
2186 sk->sk_prot->release_cb(sk);
2188 sk->sk_lock.owned = 0;
2189 if (waitqueue_active(&sk->sk_lock.wq))
2190 wake_up(&sk->sk_lock.wq);
2191 spin_unlock_bh(&sk->sk_lock.slock);
2193 EXPORT_SYMBOL(release_sock);
2196 * lock_sock_fast - fast version of lock_sock
2199 * This version should be used for very small section, where process wont block
2200 * return false if fast path is taken
2201 * sk_lock.slock locked, owned = 0, BH disabled
2202 * return true if slow path is taken
2203 * sk_lock.slock unlocked, owned = 1, BH enabled
2205 bool lock_sock_fast(struct sock *sk)
2208 spin_lock_bh(&sk->sk_lock.slock);
2210 if (!sk->sk_lock.owned)
2212 * Note : We must disable BH
2217 sk->sk_lock.owned = 1;
2218 spin_unlock(&sk->sk_lock.slock);
2220 * The sk_lock has mutex_lock() semantics here:
2222 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2226 EXPORT_SYMBOL(lock_sock_fast);
2228 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2231 if (!sock_flag(sk, SOCK_TIMESTAMP))
2232 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2233 tv = ktime_to_timeval(sk->sk_stamp);
2234 if (tv.tv_sec == -1)
2236 if (tv.tv_sec == 0) {
2237 sk->sk_stamp = ktime_get_real();
2238 tv = ktime_to_timeval(sk->sk_stamp);
2240 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2242 EXPORT_SYMBOL(sock_get_timestamp);
2244 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2247 if (!sock_flag(sk, SOCK_TIMESTAMP))
2248 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2249 ts = ktime_to_timespec(sk->sk_stamp);
2250 if (ts.tv_sec == -1)
2252 if (ts.tv_sec == 0) {
2253 sk->sk_stamp = ktime_get_real();
2254 ts = ktime_to_timespec(sk->sk_stamp);
2256 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2258 EXPORT_SYMBOL(sock_get_timestampns);
2260 void sock_enable_timestamp(struct sock *sk, int flag)
2262 if (!sock_flag(sk, flag)) {
2263 unsigned long previous_flags = sk->sk_flags;
2265 sock_set_flag(sk, flag);
2267 * we just set one of the two flags which require net
2268 * time stamping, but time stamping might have been on
2269 * already because of the other one
2271 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2272 net_enable_timestamp();
2277 * Get a socket option on an socket.
2279 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2280 * asynchronous errors should be reported by getsockopt. We assume
2281 * this means if you specify SO_ERROR (otherwise whats the point of it).
2283 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2284 char __user *optval, int __user *optlen)
2286 struct sock *sk = sock->sk;
2288 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2290 EXPORT_SYMBOL(sock_common_getsockopt);
2292 #ifdef CONFIG_COMPAT
2293 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2294 char __user *optval, int __user *optlen)
2296 struct sock *sk = sock->sk;
2298 if (sk->sk_prot->compat_getsockopt != NULL)
2299 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2301 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2303 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2306 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2307 struct msghdr *msg, size_t size, int flags)
2309 struct sock *sk = sock->sk;
2313 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2314 flags & ~MSG_DONTWAIT, &addr_len);
2316 msg->msg_namelen = addr_len;
2319 EXPORT_SYMBOL(sock_common_recvmsg);
2322 * Set socket options on an inet socket.
2324 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2325 char __user *optval, unsigned int optlen)
2327 struct sock *sk = sock->sk;
2329 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2331 EXPORT_SYMBOL(sock_common_setsockopt);
2333 #ifdef CONFIG_COMPAT
2334 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2335 char __user *optval, unsigned int optlen)
2337 struct sock *sk = sock->sk;
2339 if (sk->sk_prot->compat_setsockopt != NULL)
2340 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2342 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2344 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2347 void sk_common_release(struct sock *sk)
2349 if (sk->sk_prot->destroy)
2350 sk->sk_prot->destroy(sk);
2353 * Observation: when sock_common_release is called, processes have
2354 * no access to socket. But net still has.
2355 * Step one, detach it from networking:
2357 * A. Remove from hash tables.
2360 sk->sk_prot->unhash(sk);
2363 * In this point socket cannot receive new packets, but it is possible
2364 * that some packets are in flight because some CPU runs receiver and
2365 * did hash table lookup before we unhashed socket. They will achieve
2366 * receive queue and will be purged by socket destructor.
2368 * Also we still have packets pending on receive queue and probably,
2369 * our own packets waiting in device queues. sock_destroy will drain
2370 * receive queue, but transmitted packets will delay socket destruction
2371 * until the last reference will be released.
2376 xfrm_sk_free_policy(sk);
2378 sk_refcnt_debug_release(sk);
2381 EXPORT_SYMBOL(sk_common_release);
2383 #ifdef CONFIG_PROC_FS
2384 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2386 int val[PROTO_INUSE_NR];
2389 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2391 #ifdef CONFIG_NET_NS
2392 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2394 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2396 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2398 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2400 int cpu, idx = prot->inuse_idx;
2403 for_each_possible_cpu(cpu)
2404 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2406 return res >= 0 ? res : 0;
2408 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2410 static int __net_init sock_inuse_init_net(struct net *net)
2412 net->core.inuse = alloc_percpu(struct prot_inuse);
2413 return net->core.inuse ? 0 : -ENOMEM;
2416 static void __net_exit sock_inuse_exit_net(struct net *net)
2418 free_percpu(net->core.inuse);
2421 static struct pernet_operations net_inuse_ops = {
2422 .init = sock_inuse_init_net,
2423 .exit = sock_inuse_exit_net,
2426 static __init int net_inuse_init(void)
2428 if (register_pernet_subsys(&net_inuse_ops))
2429 panic("Cannot initialize net inuse counters");
2434 core_initcall(net_inuse_init);
2436 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2438 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2440 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2442 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2444 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2446 int cpu, idx = prot->inuse_idx;
2449 for_each_possible_cpu(cpu)
2450 res += per_cpu(prot_inuse, cpu).val[idx];
2452 return res >= 0 ? res : 0;
2454 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2457 static void assign_proto_idx(struct proto *prot)
2459 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2461 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2462 pr_err("PROTO_INUSE_NR exhausted\n");
2466 set_bit(prot->inuse_idx, proto_inuse_idx);
2469 static void release_proto_idx(struct proto *prot)
2471 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2472 clear_bit(prot->inuse_idx, proto_inuse_idx);
2475 static inline void assign_proto_idx(struct proto *prot)
2479 static inline void release_proto_idx(struct proto *prot)
2484 int proto_register(struct proto *prot, int alloc_slab)
2487 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2488 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2491 if (prot->slab == NULL) {
2492 pr_crit("%s: Can't create sock SLAB cache!\n",
2497 if (prot->rsk_prot != NULL) {
2498 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2499 if (prot->rsk_prot->slab_name == NULL)
2500 goto out_free_sock_slab;
2502 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2503 prot->rsk_prot->obj_size, 0,
2504 SLAB_HWCACHE_ALIGN, NULL);
2506 if (prot->rsk_prot->slab == NULL) {
2507 pr_crit("%s: Can't create request sock SLAB cache!\n",
2509 goto out_free_request_sock_slab_name;
2513 if (prot->twsk_prot != NULL) {
2514 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2516 if (prot->twsk_prot->twsk_slab_name == NULL)
2517 goto out_free_request_sock_slab;
2519 prot->twsk_prot->twsk_slab =
2520 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2521 prot->twsk_prot->twsk_obj_size,
2523 SLAB_HWCACHE_ALIGN |
2526 if (prot->twsk_prot->twsk_slab == NULL)
2527 goto out_free_timewait_sock_slab_name;
2531 mutex_lock(&proto_list_mutex);
2532 list_add(&prot->node, &proto_list);
2533 assign_proto_idx(prot);
2534 mutex_unlock(&proto_list_mutex);
2537 out_free_timewait_sock_slab_name:
2538 kfree(prot->twsk_prot->twsk_slab_name);
2539 out_free_request_sock_slab:
2540 if (prot->rsk_prot && prot->rsk_prot->slab) {
2541 kmem_cache_destroy(prot->rsk_prot->slab);
2542 prot->rsk_prot->slab = NULL;
2544 out_free_request_sock_slab_name:
2546 kfree(prot->rsk_prot->slab_name);
2548 kmem_cache_destroy(prot->slab);
2553 EXPORT_SYMBOL(proto_register);
2555 void proto_unregister(struct proto *prot)
2557 mutex_lock(&proto_list_mutex);
2558 release_proto_idx(prot);
2559 list_del(&prot->node);
2560 mutex_unlock(&proto_list_mutex);
2562 if (prot->slab != NULL) {
2563 kmem_cache_destroy(prot->slab);
2567 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2568 kmem_cache_destroy(prot->rsk_prot->slab);
2569 kfree(prot->rsk_prot->slab_name);
2570 prot->rsk_prot->slab = NULL;
2573 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2574 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2575 kfree(prot->twsk_prot->twsk_slab_name);
2576 prot->twsk_prot->twsk_slab = NULL;
2579 EXPORT_SYMBOL(proto_unregister);
2581 #ifdef CONFIG_PROC_FS
2582 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2583 __acquires(proto_list_mutex)
2585 mutex_lock(&proto_list_mutex);
2586 return seq_list_start_head(&proto_list, *pos);
2589 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2591 return seq_list_next(v, &proto_list, pos);
2594 static void proto_seq_stop(struct seq_file *seq, void *v)
2595 __releases(proto_list_mutex)
2597 mutex_unlock(&proto_list_mutex);
2600 static char proto_method_implemented(const void *method)
2602 return method == NULL ? 'n' : 'y';
2604 static long sock_prot_memory_allocated(struct proto *proto)
2606 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2609 static char *sock_prot_memory_pressure(struct proto *proto)
2611 return proto->memory_pressure != NULL ?
2612 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2615 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2618 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2619 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2622 sock_prot_inuse_get(seq_file_net(seq), proto),
2623 sock_prot_memory_allocated(proto),
2624 sock_prot_memory_pressure(proto),
2626 proto->slab == NULL ? "no" : "yes",
2627 module_name(proto->owner),
2628 proto_method_implemented(proto->close),
2629 proto_method_implemented(proto->connect),
2630 proto_method_implemented(proto->disconnect),
2631 proto_method_implemented(proto->accept),
2632 proto_method_implemented(proto->ioctl),
2633 proto_method_implemented(proto->init),
2634 proto_method_implemented(proto->destroy),
2635 proto_method_implemented(proto->shutdown),
2636 proto_method_implemented(proto->setsockopt),
2637 proto_method_implemented(proto->getsockopt),
2638 proto_method_implemented(proto->sendmsg),
2639 proto_method_implemented(proto->recvmsg),
2640 proto_method_implemented(proto->sendpage),
2641 proto_method_implemented(proto->bind),
2642 proto_method_implemented(proto->backlog_rcv),
2643 proto_method_implemented(proto->hash),
2644 proto_method_implemented(proto->unhash),
2645 proto_method_implemented(proto->get_port),
2646 proto_method_implemented(proto->enter_memory_pressure));
2649 static int proto_seq_show(struct seq_file *seq, void *v)
2651 if (v == &proto_list)
2652 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2661 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2663 proto_seq_printf(seq, list_entry(v, struct proto, node));
2667 static const struct seq_operations proto_seq_ops = {
2668 .start = proto_seq_start,
2669 .next = proto_seq_next,
2670 .stop = proto_seq_stop,
2671 .show = proto_seq_show,
2674 static int proto_seq_open(struct inode *inode, struct file *file)
2676 return seq_open_net(inode, file, &proto_seq_ops,
2677 sizeof(struct seq_net_private));
2680 static const struct file_operations proto_seq_fops = {
2681 .owner = THIS_MODULE,
2682 .open = proto_seq_open,
2684 .llseek = seq_lseek,
2685 .release = seq_release_net,
2688 static __net_init int proto_init_net(struct net *net)
2690 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2696 static __net_exit void proto_exit_net(struct net *net)
2698 proc_net_remove(net, "protocols");
2702 static __net_initdata struct pernet_operations proto_net_ops = {
2703 .init = proto_init_net,
2704 .exit = proto_exit_net,
2707 static int __init proto_init(void)
2709 return register_pernet_subsys(&proto_net_ops);
2712 subsys_initcall(proto_init);
2714 #endif /* PROC_FS */