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 #include <linux/capability.h>
93 #include <linux/errno.h>
94 #include <linux/types.h>
95 #include <linux/socket.h>
97 #include <linux/kernel.h>
98 #include <linux/module.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/mm.h>
107 #include <linux/slab.h>
108 #include <linux/interrupt.h>
109 #include <linux/poll.h>
110 #include <linux/tcp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
115 #include <asm/uaccess.h>
116 #include <asm/system.h>
118 #include <linux/netdevice.h>
119 #include <net/protocol.h>
120 #include <linux/skbuff.h>
121 #include <net/net_namespace.h>
122 #include <net/request_sock.h>
123 #include <net/sock.h>
124 #include <linux/net_tstamp.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
127 #include <net/cls_cgroup.h>
129 #include <linux/filter.h>
136 * Each address family might have different locking rules, so we have
137 * one slock key per address family:
139 static struct lock_class_key af_family_keys[AF_MAX];
140 static struct lock_class_key af_family_slock_keys[AF_MAX];
143 * Make lock validator output more readable. (we pre-construct these
144 * strings build-time, so that runtime initialization of socket
147 static const char *const af_family_key_strings[AF_MAX+1] = {
148 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
149 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
150 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
151 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
152 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
153 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
154 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
155 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
157 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
160 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" ,
163 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
164 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
165 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
166 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
167 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
168 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
169 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
170 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
171 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
172 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
173 "slock-27" , "slock-28" , "slock-AF_CAN" ,
174 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
175 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
176 "slock-AF_IEEE802154", "slock-AF_CAIF" ,
179 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
180 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
181 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
182 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
183 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
184 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
185 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
186 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
187 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
188 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
189 "clock-27" , "clock-28" , "clock-AF_CAN" ,
190 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
191 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
192 "clock-AF_IEEE802154", "clock-AF_CAIF" ,
197 * sk_callback_lock locking rules are per-address-family,
198 * so split the lock classes by using a per-AF key:
200 static struct lock_class_key af_callback_keys[AF_MAX];
202 /* Take into consideration the size of the struct sk_buff overhead in the
203 * determination of these values, since that is non-constant across
204 * platforms. This makes socket queueing behavior and performance
205 * not depend upon such differences.
207 #define _SK_MEM_PACKETS 256
208 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
209 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
210 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
212 /* Run time adjustable parameters. */
213 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
214 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
215 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
216 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
218 /* Maximal space eaten by iovec or ancilliary data plus some space */
219 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
220 EXPORT_SYMBOL(sysctl_optmem_max);
222 #if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
223 int net_cls_subsys_id = -1;
224 EXPORT_SYMBOL_GPL(net_cls_subsys_id);
227 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
231 if (optlen < sizeof(tv))
233 if (copy_from_user(&tv, optval, sizeof(tv)))
235 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
239 static int warned __read_mostly;
242 if (warned < 10 && net_ratelimit()) {
244 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
245 "tries to set negative timeout\n",
246 current->comm, task_pid_nr(current));
250 *timeo_p = MAX_SCHEDULE_TIMEOUT;
251 if (tv.tv_sec == 0 && tv.tv_usec == 0)
253 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
254 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
258 static void sock_warn_obsolete_bsdism(const char *name)
261 static char warncomm[TASK_COMM_LEN];
262 if (strcmp(warncomm, current->comm) && warned < 5) {
263 strcpy(warncomm, current->comm);
264 printk(KERN_WARNING "process `%s' is using obsolete "
265 "%s SO_BSDCOMPAT\n", warncomm, name);
270 static void sock_disable_timestamp(struct sock *sk, int flag)
272 if (sock_flag(sk, flag)) {
273 sock_reset_flag(sk, flag);
274 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
275 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
276 net_disable_timestamp();
282 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
287 struct sk_buff_head *list = &sk->sk_receive_queue;
289 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
290 number of warnings when compiling with -W --ANK
292 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
293 (unsigned)sk->sk_rcvbuf) {
294 atomic_inc(&sk->sk_drops);
298 err = sk_filter(sk, skb);
302 if (!sk_rmem_schedule(sk, skb->truesize)) {
303 atomic_inc(&sk->sk_drops);
308 skb_set_owner_r(skb, sk);
310 /* Cache the SKB length before we tack it onto the receive
311 * queue. Once it is added it no longer belongs to us and
312 * may be freed by other threads of control pulling packets
317 /* we escape from rcu protected region, make sure we dont leak
322 spin_lock_irqsave(&list->lock, flags);
323 skb->dropcount = atomic_read(&sk->sk_drops);
324 __skb_queue_tail(list, skb);
325 spin_unlock_irqrestore(&list->lock, flags);
327 if (!sock_flag(sk, SOCK_DEAD))
328 sk->sk_data_ready(sk, skb_len);
331 EXPORT_SYMBOL(sock_queue_rcv_skb);
333 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
335 int rc = NET_RX_SUCCESS;
337 if (sk_filter(sk, skb))
338 goto discard_and_relse;
342 if (sk_rcvqueues_full(sk, skb)) {
343 atomic_inc(&sk->sk_drops);
344 goto discard_and_relse;
347 bh_lock_sock_nested(sk);
350 if (!sock_owned_by_user(sk)) {
352 * trylock + unlock semantics:
354 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
356 rc = sk_backlog_rcv(sk, skb);
358 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
359 } else if (sk_add_backlog(sk, skb)) {
361 atomic_inc(&sk->sk_drops);
362 goto discard_and_relse;
373 EXPORT_SYMBOL(sk_receive_skb);
375 void sk_reset_txq(struct sock *sk)
377 sk_tx_queue_clear(sk);
379 EXPORT_SYMBOL(sk_reset_txq);
381 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
383 struct dst_entry *dst = __sk_dst_get(sk);
385 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
386 sk_tx_queue_clear(sk);
387 rcu_assign_pointer(sk->sk_dst_cache, NULL);
394 EXPORT_SYMBOL(__sk_dst_check);
396 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
398 struct dst_entry *dst = sk_dst_get(sk);
400 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
408 EXPORT_SYMBOL(sk_dst_check);
410 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
412 int ret = -ENOPROTOOPT;
413 #ifdef CONFIG_NETDEVICES
414 struct net *net = sock_net(sk);
415 char devname[IFNAMSIZ];
420 if (!capable(CAP_NET_RAW))
427 /* Bind this socket to a particular device like "eth0",
428 * as specified in the passed interface name. If the
429 * name is "" or the option length is zero the socket
432 if (optlen > IFNAMSIZ - 1)
433 optlen = IFNAMSIZ - 1;
434 memset(devname, 0, sizeof(devname));
437 if (copy_from_user(devname, optval, optlen))
441 if (devname[0] != '\0') {
442 struct net_device *dev;
445 dev = dev_get_by_name_rcu(net, devname);
447 index = dev->ifindex;
455 sk->sk_bound_dev_if = index;
467 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
470 sock_set_flag(sk, bit);
472 sock_reset_flag(sk, bit);
476 * This is meant for all protocols to use and covers goings on
477 * at the socket level. Everything here is generic.
480 int sock_setsockopt(struct socket *sock, int level, int optname,
481 char __user *optval, unsigned int optlen)
483 struct sock *sk = sock->sk;
490 * Options without arguments
493 if (optname == SO_BINDTODEVICE)
494 return sock_bindtodevice(sk, optval, optlen);
496 if (optlen < sizeof(int))
499 if (get_user(val, (int __user *)optval))
502 valbool = val ? 1 : 0;
508 if (val && !capable(CAP_NET_ADMIN))
511 sock_valbool_flag(sk, SOCK_DBG, valbool);
514 sk->sk_reuse = valbool;
523 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
526 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
529 /* Don't error on this BSD doesn't and if you think
530 about it this is right. Otherwise apps have to
531 play 'guess the biggest size' games. RCVBUF/SNDBUF
532 are treated in BSD as hints */
534 if (val > sysctl_wmem_max)
535 val = sysctl_wmem_max;
537 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
538 if ((val * 2) < SOCK_MIN_SNDBUF)
539 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
541 sk->sk_sndbuf = val * 2;
544 * Wake up sending tasks if we
547 sk->sk_write_space(sk);
551 if (!capable(CAP_NET_ADMIN)) {
558 /* Don't error on this BSD doesn't and if you think
559 about it this is right. Otherwise apps have to
560 play 'guess the biggest size' games. RCVBUF/SNDBUF
561 are treated in BSD as hints */
563 if (val > sysctl_rmem_max)
564 val = sysctl_rmem_max;
566 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
568 * We double it on the way in to account for
569 * "struct sk_buff" etc. overhead. Applications
570 * assume that the SO_RCVBUF setting they make will
571 * allow that much actual data to be received on that
574 * Applications are unaware that "struct sk_buff" and
575 * other overheads allocate from the receive buffer
576 * during socket buffer allocation.
578 * And after considering the possible alternatives,
579 * returning the value we actually used in getsockopt
580 * is the most desirable behavior.
582 if ((val * 2) < SOCK_MIN_RCVBUF)
583 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
585 sk->sk_rcvbuf = val * 2;
589 if (!capable(CAP_NET_ADMIN)) {
597 if (sk->sk_protocol == IPPROTO_TCP)
598 tcp_set_keepalive(sk, valbool);
600 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
604 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
608 sk->sk_no_check = valbool;
612 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
613 sk->sk_priority = val;
619 if (optlen < sizeof(ling)) {
620 ret = -EINVAL; /* 1003.1g */
623 if (copy_from_user(&ling, optval, sizeof(ling))) {
628 sock_reset_flag(sk, SOCK_LINGER);
630 #if (BITS_PER_LONG == 32)
631 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
632 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
635 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
636 sock_set_flag(sk, SOCK_LINGER);
641 sock_warn_obsolete_bsdism("setsockopt");
646 set_bit(SOCK_PASSCRED, &sock->flags);
648 clear_bit(SOCK_PASSCRED, &sock->flags);
654 if (optname == SO_TIMESTAMP)
655 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
657 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
658 sock_set_flag(sk, SOCK_RCVTSTAMP);
659 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
661 sock_reset_flag(sk, SOCK_RCVTSTAMP);
662 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
666 case SO_TIMESTAMPING:
667 if (val & ~SOF_TIMESTAMPING_MASK) {
671 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
672 val & SOF_TIMESTAMPING_TX_HARDWARE);
673 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
674 val & SOF_TIMESTAMPING_TX_SOFTWARE);
675 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
676 val & SOF_TIMESTAMPING_RX_HARDWARE);
677 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
678 sock_enable_timestamp(sk,
679 SOCK_TIMESTAMPING_RX_SOFTWARE);
681 sock_disable_timestamp(sk,
682 SOCK_TIMESTAMPING_RX_SOFTWARE);
683 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
684 val & SOF_TIMESTAMPING_SOFTWARE);
685 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
686 val & SOF_TIMESTAMPING_SYS_HARDWARE);
687 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
688 val & SOF_TIMESTAMPING_RAW_HARDWARE);
694 sk->sk_rcvlowat = val ? : 1;
698 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
702 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
705 case SO_ATTACH_FILTER:
707 if (optlen == sizeof(struct sock_fprog)) {
708 struct sock_fprog fprog;
711 if (copy_from_user(&fprog, optval, sizeof(fprog)))
714 ret = sk_attach_filter(&fprog, sk);
718 case SO_DETACH_FILTER:
719 ret = sk_detach_filter(sk);
724 set_bit(SOCK_PASSSEC, &sock->flags);
726 clear_bit(SOCK_PASSSEC, &sock->flags);
729 if (!capable(CAP_NET_ADMIN))
735 /* We implement the SO_SNDLOWAT etc to
736 not be settable (1003.1g 5.3) */
739 sock_set_flag(sk, SOCK_RXQ_OVFL);
741 sock_reset_flag(sk, SOCK_RXQ_OVFL);
750 EXPORT_SYMBOL(sock_setsockopt);
753 void cred_to_ucred(struct pid *pid, const struct cred *cred,
756 ucred->pid = pid_vnr(pid);
757 ucred->uid = ucred->gid = -1;
759 struct user_namespace *current_ns = current_user_ns();
761 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
762 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
766 int sock_getsockopt(struct socket *sock, int level, int optname,
767 char __user *optval, int __user *optlen)
769 struct sock *sk = sock->sk;
777 int lv = sizeof(int);
780 if (get_user(len, optlen))
785 memset(&v, 0, sizeof(v));
789 v.val = sock_flag(sk, SOCK_DBG);
793 v.val = sock_flag(sk, SOCK_LOCALROUTE);
797 v.val = !!sock_flag(sk, SOCK_BROADCAST);
801 v.val = sk->sk_sndbuf;
805 v.val = sk->sk_rcvbuf;
809 v.val = sk->sk_reuse;
813 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
821 v.val = sk->sk_protocol;
825 v.val = sk->sk_family;
829 v.val = -sock_error(sk);
831 v.val = xchg(&sk->sk_err_soft, 0);
835 v.val = !!sock_flag(sk, SOCK_URGINLINE);
839 v.val = sk->sk_no_check;
843 v.val = sk->sk_priority;
848 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
849 v.ling.l_linger = sk->sk_lingertime / HZ;
853 sock_warn_obsolete_bsdism("getsockopt");
857 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
858 !sock_flag(sk, SOCK_RCVTSTAMPNS);
862 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
865 case SO_TIMESTAMPING:
867 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
868 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
869 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
870 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
871 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
872 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
873 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
874 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
875 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
876 v.val |= SOF_TIMESTAMPING_SOFTWARE;
877 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
878 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
879 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
880 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
884 lv = sizeof(struct timeval);
885 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
889 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
890 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
895 lv = sizeof(struct timeval);
896 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
900 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
901 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
906 v.val = sk->sk_rcvlowat;
914 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
919 struct ucred peercred;
920 if (len > sizeof(peercred))
921 len = sizeof(peercred);
922 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
923 if (copy_to_user(optval, &peercred, len))
932 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
936 if (copy_to_user(optval, address, len))
941 /* Dubious BSD thing... Probably nobody even uses it, but
942 * the UNIX standard wants it for whatever reason... -DaveM
945 v.val = sk->sk_state == TCP_LISTEN;
949 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
953 return security_socket_getpeersec_stream(sock, optval, optlen, len);
960 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
969 if (copy_to_user(optval, &v, len))
972 if (put_user(len, optlen))
978 * Initialize an sk_lock.
980 * (We also register the sk_lock with the lock validator.)
982 static inline void sock_lock_init(struct sock *sk)
984 sock_lock_init_class_and_name(sk,
985 af_family_slock_key_strings[sk->sk_family],
986 af_family_slock_keys + sk->sk_family,
987 af_family_key_strings[sk->sk_family],
988 af_family_keys + sk->sk_family);
992 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
993 * even temporarly, because of RCU lookups. sk_node should also be left as is.
995 static void sock_copy(struct sock *nsk, const struct sock *osk)
997 #ifdef CONFIG_SECURITY_NETWORK
998 void *sptr = nsk->sk_security;
1000 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
1001 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
1002 sizeof(osk->sk_tx_queue_mapping));
1003 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
1004 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
1005 #ifdef CONFIG_SECURITY_NETWORK
1006 nsk->sk_security = sptr;
1007 security_sk_clone(osk, nsk);
1011 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1015 struct kmem_cache *slab;
1019 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1022 if (priority & __GFP_ZERO) {
1024 * caches using SLAB_DESTROY_BY_RCU should let
1025 * sk_node.next un-modified. Special care is taken
1026 * when initializing object to zero.
1028 if (offsetof(struct sock, sk_node.next) != 0)
1029 memset(sk, 0, offsetof(struct sock, sk_node.next));
1030 memset(&sk->sk_node.pprev, 0,
1031 prot->obj_size - offsetof(struct sock,
1036 sk = kmalloc(prot->obj_size, priority);
1039 kmemcheck_annotate_bitfield(sk, flags);
1041 if (security_sk_alloc(sk, family, priority))
1044 if (!try_module_get(prot->owner))
1046 sk_tx_queue_clear(sk);
1052 security_sk_free(sk);
1055 kmem_cache_free(slab, sk);
1061 static void sk_prot_free(struct proto *prot, struct sock *sk)
1063 struct kmem_cache *slab;
1064 struct module *owner;
1066 owner = prot->owner;
1069 security_sk_free(sk);
1071 kmem_cache_free(slab, sk);
1077 #ifdef CONFIG_CGROUPS
1078 void sock_update_classid(struct sock *sk)
1080 u32 classid = task_cls_classid(current);
1082 if (classid && classid != sk->sk_classid)
1083 sk->sk_classid = classid;
1085 EXPORT_SYMBOL(sock_update_classid);
1089 * sk_alloc - All socket objects are allocated here
1090 * @net: the applicable net namespace
1091 * @family: protocol family
1092 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1093 * @prot: struct proto associated with this new sock instance
1095 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1100 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1102 sk->sk_family = family;
1104 * See comment in struct sock definition to understand
1105 * why we need sk_prot_creator -acme
1107 sk->sk_prot = sk->sk_prot_creator = prot;
1109 sock_net_set(sk, get_net(net));
1110 atomic_set(&sk->sk_wmem_alloc, 1);
1112 sock_update_classid(sk);
1117 EXPORT_SYMBOL(sk_alloc);
1119 static void __sk_free(struct sock *sk)
1121 struct sk_filter *filter;
1123 if (sk->sk_destruct)
1124 sk->sk_destruct(sk);
1126 filter = rcu_dereference_check(sk->sk_filter,
1127 atomic_read(&sk->sk_wmem_alloc) == 0);
1129 sk_filter_uncharge(sk, filter);
1130 rcu_assign_pointer(sk->sk_filter, NULL);
1133 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1134 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1136 if (atomic_read(&sk->sk_omem_alloc))
1137 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1138 __func__, atomic_read(&sk->sk_omem_alloc));
1140 if (sk->sk_peer_cred)
1141 put_cred(sk->sk_peer_cred);
1142 put_pid(sk->sk_peer_pid);
1143 put_net(sock_net(sk));
1144 sk_prot_free(sk->sk_prot_creator, sk);
1147 void sk_free(struct sock *sk)
1150 * We substract one from sk_wmem_alloc and can know if
1151 * some packets are still in some tx queue.
1152 * If not null, sock_wfree() will call __sk_free(sk) later
1154 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1157 EXPORT_SYMBOL(sk_free);
1160 * Last sock_put should drop referrence to sk->sk_net. It has already
1161 * been dropped in sk_change_net. Taking referrence to stopping namespace
1163 * Take referrence to a socket to remove it from hash _alive_ and after that
1164 * destroy it in the context of init_net.
1166 void sk_release_kernel(struct sock *sk)
1168 if (sk == NULL || sk->sk_socket == NULL)
1172 sock_release(sk->sk_socket);
1173 release_net(sock_net(sk));
1174 sock_net_set(sk, get_net(&init_net));
1177 EXPORT_SYMBOL(sk_release_kernel);
1179 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1183 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1184 if (newsk != NULL) {
1185 struct sk_filter *filter;
1187 sock_copy(newsk, sk);
1190 get_net(sock_net(newsk));
1191 sk_node_init(&newsk->sk_node);
1192 sock_lock_init(newsk);
1193 bh_lock_sock(newsk);
1194 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1195 newsk->sk_backlog.len = 0;
1197 atomic_set(&newsk->sk_rmem_alloc, 0);
1199 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1201 atomic_set(&newsk->sk_wmem_alloc, 1);
1202 atomic_set(&newsk->sk_omem_alloc, 0);
1203 skb_queue_head_init(&newsk->sk_receive_queue);
1204 skb_queue_head_init(&newsk->sk_write_queue);
1205 #ifdef CONFIG_NET_DMA
1206 skb_queue_head_init(&newsk->sk_async_wait_queue);
1209 spin_lock_init(&newsk->sk_dst_lock);
1210 rwlock_init(&newsk->sk_callback_lock);
1211 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1212 af_callback_keys + newsk->sk_family,
1213 af_family_clock_key_strings[newsk->sk_family]);
1215 newsk->sk_dst_cache = NULL;
1216 newsk->sk_wmem_queued = 0;
1217 newsk->sk_forward_alloc = 0;
1218 newsk->sk_send_head = NULL;
1219 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1221 sock_reset_flag(newsk, SOCK_DONE);
1222 skb_queue_head_init(&newsk->sk_error_queue);
1224 filter = newsk->sk_filter;
1226 sk_filter_charge(newsk, filter);
1228 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1229 /* It is still raw copy of parent, so invalidate
1230 * destructor and make plain sk_free() */
1231 newsk->sk_destruct = NULL;
1238 newsk->sk_priority = 0;
1240 * Before updating sk_refcnt, we must commit prior changes to memory
1241 * (Documentation/RCU/rculist_nulls.txt for details)
1244 atomic_set(&newsk->sk_refcnt, 2);
1247 * Increment the counter in the same struct proto as the master
1248 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1249 * is the same as sk->sk_prot->socks, as this field was copied
1252 * This _changes_ the previous behaviour, where
1253 * tcp_create_openreq_child always was incrementing the
1254 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1255 * to be taken into account in all callers. -acme
1257 sk_refcnt_debug_inc(newsk);
1258 sk_set_socket(newsk, NULL);
1259 newsk->sk_wq = NULL;
1261 if (newsk->sk_prot->sockets_allocated)
1262 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1264 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1265 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1266 net_enable_timestamp();
1271 EXPORT_SYMBOL_GPL(sk_clone);
1273 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1275 __sk_dst_set(sk, dst);
1276 sk->sk_route_caps = dst->dev->features;
1277 if (sk->sk_route_caps & NETIF_F_GSO)
1278 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1279 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1280 if (sk_can_gso(sk)) {
1281 if (dst->header_len) {
1282 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1284 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1285 sk->sk_gso_max_size = dst->dev->gso_max_size;
1289 EXPORT_SYMBOL_GPL(sk_setup_caps);
1291 void __init sk_init(void)
1293 if (totalram_pages <= 4096) {
1294 sysctl_wmem_max = 32767;
1295 sysctl_rmem_max = 32767;
1296 sysctl_wmem_default = 32767;
1297 sysctl_rmem_default = 32767;
1298 } else if (totalram_pages >= 131072) {
1299 sysctl_wmem_max = 131071;
1300 sysctl_rmem_max = 131071;
1305 * Simple resource managers for sockets.
1310 * Write buffer destructor automatically called from kfree_skb.
1312 void sock_wfree(struct sk_buff *skb)
1314 struct sock *sk = skb->sk;
1315 unsigned int len = skb->truesize;
1317 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1319 * Keep a reference on sk_wmem_alloc, this will be released
1320 * after sk_write_space() call
1322 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1323 sk->sk_write_space(sk);
1327 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1328 * could not do because of in-flight packets
1330 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1333 EXPORT_SYMBOL(sock_wfree);
1336 * Read buffer destructor automatically called from kfree_skb.
1338 void sock_rfree(struct sk_buff *skb)
1340 struct sock *sk = skb->sk;
1342 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1343 sk_mem_uncharge(skb->sk, skb->truesize);
1345 EXPORT_SYMBOL(sock_rfree);
1348 int sock_i_uid(struct sock *sk)
1352 read_lock(&sk->sk_callback_lock);
1353 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1354 read_unlock(&sk->sk_callback_lock);
1357 EXPORT_SYMBOL(sock_i_uid);
1359 unsigned long sock_i_ino(struct sock *sk)
1363 read_lock(&sk->sk_callback_lock);
1364 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1365 read_unlock(&sk->sk_callback_lock);
1368 EXPORT_SYMBOL(sock_i_ino);
1371 * Allocate a skb from the socket's send buffer.
1373 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1376 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1377 struct sk_buff *skb = alloc_skb(size, priority);
1379 skb_set_owner_w(skb, sk);
1385 EXPORT_SYMBOL(sock_wmalloc);
1388 * Allocate a skb from the socket's receive buffer.
1390 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1393 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1394 struct sk_buff *skb = alloc_skb(size, priority);
1396 skb_set_owner_r(skb, sk);
1404 * Allocate a memory block from the socket's option memory buffer.
1406 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1408 if ((unsigned)size <= sysctl_optmem_max &&
1409 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1411 /* First do the add, to avoid the race if kmalloc
1414 atomic_add(size, &sk->sk_omem_alloc);
1415 mem = kmalloc(size, priority);
1418 atomic_sub(size, &sk->sk_omem_alloc);
1422 EXPORT_SYMBOL(sock_kmalloc);
1425 * Free an option memory block.
1427 void sock_kfree_s(struct sock *sk, void *mem, int size)
1430 atomic_sub(size, &sk->sk_omem_alloc);
1432 EXPORT_SYMBOL(sock_kfree_s);
1434 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1435 I think, these locks should be removed for datagram sockets.
1437 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1441 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1445 if (signal_pending(current))
1447 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1448 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1449 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1451 if (sk->sk_shutdown & SEND_SHUTDOWN)
1455 timeo = schedule_timeout(timeo);
1457 finish_wait(sk_sleep(sk), &wait);
1463 * Generic send/receive buffer handlers
1466 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1467 unsigned long data_len, int noblock,
1470 struct sk_buff *skb;
1475 gfp_mask = sk->sk_allocation;
1476 if (gfp_mask & __GFP_WAIT)
1477 gfp_mask |= __GFP_REPEAT;
1479 timeo = sock_sndtimeo(sk, noblock);
1481 err = sock_error(sk);
1486 if (sk->sk_shutdown & SEND_SHUTDOWN)
1489 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1490 skb = alloc_skb(header_len, gfp_mask);
1495 /* No pages, we're done... */
1499 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1500 skb->truesize += data_len;
1501 skb_shinfo(skb)->nr_frags = npages;
1502 for (i = 0; i < npages; i++) {
1506 page = alloc_pages(sk->sk_allocation, 0);
1509 skb_shinfo(skb)->nr_frags = i;
1514 frag = &skb_shinfo(skb)->frags[i];
1516 frag->page_offset = 0;
1517 frag->size = (data_len >= PAGE_SIZE ?
1520 data_len -= PAGE_SIZE;
1523 /* Full success... */
1529 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1530 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1534 if (signal_pending(current))
1536 timeo = sock_wait_for_wmem(sk, timeo);
1539 skb_set_owner_w(skb, sk);
1543 err = sock_intr_errno(timeo);
1548 EXPORT_SYMBOL(sock_alloc_send_pskb);
1550 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1551 int noblock, int *errcode)
1553 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1555 EXPORT_SYMBOL(sock_alloc_send_skb);
1557 static void __lock_sock(struct sock *sk)
1562 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1563 TASK_UNINTERRUPTIBLE);
1564 spin_unlock_bh(&sk->sk_lock.slock);
1566 spin_lock_bh(&sk->sk_lock.slock);
1567 if (!sock_owned_by_user(sk))
1570 finish_wait(&sk->sk_lock.wq, &wait);
1573 static void __release_sock(struct sock *sk)
1575 struct sk_buff *skb = sk->sk_backlog.head;
1578 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1582 struct sk_buff *next = skb->next;
1584 WARN_ON_ONCE(skb_dst_is_noref(skb));
1586 sk_backlog_rcv(sk, skb);
1589 * We are in process context here with softirqs
1590 * disabled, use cond_resched_softirq() to preempt.
1591 * This is safe to do because we've taken the backlog
1594 cond_resched_softirq();
1597 } while (skb != NULL);
1600 } while ((skb = sk->sk_backlog.head) != NULL);
1603 * Doing the zeroing here guarantee we can not loop forever
1604 * while a wild producer attempts to flood us.
1606 sk->sk_backlog.len = 0;
1610 * sk_wait_data - wait for data to arrive at sk_receive_queue
1611 * @sk: sock to wait on
1612 * @timeo: for how long
1614 * Now socket state including sk->sk_err is changed only under lock,
1615 * hence we may omit checks after joining wait queue.
1616 * We check receive queue before schedule() only as optimization;
1617 * it is very likely that release_sock() added new data.
1619 int sk_wait_data(struct sock *sk, long *timeo)
1624 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1625 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1626 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1627 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1628 finish_wait(sk_sleep(sk), &wait);
1631 EXPORT_SYMBOL(sk_wait_data);
1634 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1636 * @size: memory size to allocate
1637 * @kind: allocation type
1639 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1640 * rmem allocation. This function assumes that protocols which have
1641 * memory_pressure use sk_wmem_queued as write buffer accounting.
1643 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1645 struct proto *prot = sk->sk_prot;
1646 int amt = sk_mem_pages(size);
1649 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1650 allocated = atomic_add_return(amt, prot->memory_allocated);
1653 if (allocated <= prot->sysctl_mem[0]) {
1654 if (prot->memory_pressure && *prot->memory_pressure)
1655 *prot->memory_pressure = 0;
1659 /* Under pressure. */
1660 if (allocated > prot->sysctl_mem[1])
1661 if (prot->enter_memory_pressure)
1662 prot->enter_memory_pressure(sk);
1664 /* Over hard limit. */
1665 if (allocated > prot->sysctl_mem[2])
1666 goto suppress_allocation;
1668 /* guarantee minimum buffer size under pressure */
1669 if (kind == SK_MEM_RECV) {
1670 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1672 } else { /* SK_MEM_SEND */
1673 if (sk->sk_type == SOCK_STREAM) {
1674 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1676 } else if (atomic_read(&sk->sk_wmem_alloc) <
1677 prot->sysctl_wmem[0])
1681 if (prot->memory_pressure) {
1684 if (!*prot->memory_pressure)
1686 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1687 if (prot->sysctl_mem[2] > alloc *
1688 sk_mem_pages(sk->sk_wmem_queued +
1689 atomic_read(&sk->sk_rmem_alloc) +
1690 sk->sk_forward_alloc))
1694 suppress_allocation:
1696 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1697 sk_stream_moderate_sndbuf(sk);
1699 /* Fail only if socket is _under_ its sndbuf.
1700 * In this case we cannot block, so that we have to fail.
1702 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1706 /* Alas. Undo changes. */
1707 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1708 atomic_sub(amt, prot->memory_allocated);
1711 EXPORT_SYMBOL(__sk_mem_schedule);
1714 * __sk_reclaim - reclaim memory_allocated
1717 void __sk_mem_reclaim(struct sock *sk)
1719 struct proto *prot = sk->sk_prot;
1721 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1722 prot->memory_allocated);
1723 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1725 if (prot->memory_pressure && *prot->memory_pressure &&
1726 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1727 *prot->memory_pressure = 0;
1729 EXPORT_SYMBOL(__sk_mem_reclaim);
1733 * Set of default routines for initialising struct proto_ops when
1734 * the protocol does not support a particular function. In certain
1735 * cases where it makes no sense for a protocol to have a "do nothing"
1736 * function, some default processing is provided.
1739 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1743 EXPORT_SYMBOL(sock_no_bind);
1745 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1750 EXPORT_SYMBOL(sock_no_connect);
1752 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1756 EXPORT_SYMBOL(sock_no_socketpair);
1758 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1762 EXPORT_SYMBOL(sock_no_accept);
1764 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1769 EXPORT_SYMBOL(sock_no_getname);
1771 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1775 EXPORT_SYMBOL(sock_no_poll);
1777 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1781 EXPORT_SYMBOL(sock_no_ioctl);
1783 int sock_no_listen(struct socket *sock, int backlog)
1787 EXPORT_SYMBOL(sock_no_listen);
1789 int sock_no_shutdown(struct socket *sock, int how)
1793 EXPORT_SYMBOL(sock_no_shutdown);
1795 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1796 char __user *optval, unsigned int optlen)
1800 EXPORT_SYMBOL(sock_no_setsockopt);
1802 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1803 char __user *optval, int __user *optlen)
1807 EXPORT_SYMBOL(sock_no_getsockopt);
1809 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1814 EXPORT_SYMBOL(sock_no_sendmsg);
1816 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1817 size_t len, int flags)
1821 EXPORT_SYMBOL(sock_no_recvmsg);
1823 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1825 /* Mirror missing mmap method error code */
1828 EXPORT_SYMBOL(sock_no_mmap);
1830 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1833 struct msghdr msg = {.msg_flags = flags};
1835 char *kaddr = kmap(page);
1836 iov.iov_base = kaddr + offset;
1838 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1842 EXPORT_SYMBOL(sock_no_sendpage);
1845 * Default Socket Callbacks
1848 static void sock_def_wakeup(struct sock *sk)
1850 struct socket_wq *wq;
1853 wq = rcu_dereference(sk->sk_wq);
1854 if (wq_has_sleeper(wq))
1855 wake_up_interruptible_all(&wq->wait);
1859 static void sock_def_error_report(struct sock *sk)
1861 struct socket_wq *wq;
1864 wq = rcu_dereference(sk->sk_wq);
1865 if (wq_has_sleeper(wq))
1866 wake_up_interruptible_poll(&wq->wait, POLLERR);
1867 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1871 static void sock_def_readable(struct sock *sk, int len)
1873 struct socket_wq *wq;
1876 wq = rcu_dereference(sk->sk_wq);
1877 if (wq_has_sleeper(wq))
1878 wake_up_interruptible_sync_poll(&wq->wait, POLLIN |
1879 POLLRDNORM | POLLRDBAND);
1880 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1884 static void sock_def_write_space(struct sock *sk)
1886 struct socket_wq *wq;
1890 /* Do not wake up a writer until he can make "significant"
1893 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1894 wq = rcu_dereference(sk->sk_wq);
1895 if (wq_has_sleeper(wq))
1896 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1897 POLLWRNORM | POLLWRBAND);
1899 /* Should agree with poll, otherwise some programs break */
1900 if (sock_writeable(sk))
1901 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1907 static void sock_def_destruct(struct sock *sk)
1909 kfree(sk->sk_protinfo);
1912 void sk_send_sigurg(struct sock *sk)
1914 if (sk->sk_socket && sk->sk_socket->file)
1915 if (send_sigurg(&sk->sk_socket->file->f_owner))
1916 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1918 EXPORT_SYMBOL(sk_send_sigurg);
1920 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1921 unsigned long expires)
1923 if (!mod_timer(timer, expires))
1926 EXPORT_SYMBOL(sk_reset_timer);
1928 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1930 if (timer_pending(timer) && del_timer(timer))
1933 EXPORT_SYMBOL(sk_stop_timer);
1935 void sock_init_data(struct socket *sock, struct sock *sk)
1937 skb_queue_head_init(&sk->sk_receive_queue);
1938 skb_queue_head_init(&sk->sk_write_queue);
1939 skb_queue_head_init(&sk->sk_error_queue);
1940 #ifdef CONFIG_NET_DMA
1941 skb_queue_head_init(&sk->sk_async_wait_queue);
1944 sk->sk_send_head = NULL;
1946 init_timer(&sk->sk_timer);
1948 sk->sk_allocation = GFP_KERNEL;
1949 sk->sk_rcvbuf = sysctl_rmem_default;
1950 sk->sk_sndbuf = sysctl_wmem_default;
1951 sk->sk_state = TCP_CLOSE;
1952 sk_set_socket(sk, sock);
1954 sock_set_flag(sk, SOCK_ZAPPED);
1957 sk->sk_type = sock->type;
1958 sk->sk_wq = sock->wq;
1963 spin_lock_init(&sk->sk_dst_lock);
1964 rwlock_init(&sk->sk_callback_lock);
1965 lockdep_set_class_and_name(&sk->sk_callback_lock,
1966 af_callback_keys + sk->sk_family,
1967 af_family_clock_key_strings[sk->sk_family]);
1969 sk->sk_state_change = sock_def_wakeup;
1970 sk->sk_data_ready = sock_def_readable;
1971 sk->sk_write_space = sock_def_write_space;
1972 sk->sk_error_report = sock_def_error_report;
1973 sk->sk_destruct = sock_def_destruct;
1975 sk->sk_sndmsg_page = NULL;
1976 sk->sk_sndmsg_off = 0;
1978 sk->sk_peer_pid = NULL;
1979 sk->sk_peer_cred = NULL;
1980 sk->sk_write_pending = 0;
1981 sk->sk_rcvlowat = 1;
1982 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1983 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1985 sk->sk_stamp = ktime_set(-1L, 0);
1988 * Before updating sk_refcnt, we must commit prior changes to memory
1989 * (Documentation/RCU/rculist_nulls.txt for details)
1992 atomic_set(&sk->sk_refcnt, 1);
1993 atomic_set(&sk->sk_drops, 0);
1995 EXPORT_SYMBOL(sock_init_data);
1997 void lock_sock_nested(struct sock *sk, int subclass)
2000 spin_lock_bh(&sk->sk_lock.slock);
2001 if (sk->sk_lock.owned)
2003 sk->sk_lock.owned = 1;
2004 spin_unlock(&sk->sk_lock.slock);
2006 * The sk_lock has mutex_lock() semantics here:
2008 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2011 EXPORT_SYMBOL(lock_sock_nested);
2013 void release_sock(struct sock *sk)
2016 * The sk_lock has mutex_unlock() semantics:
2018 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2020 spin_lock_bh(&sk->sk_lock.slock);
2021 if (sk->sk_backlog.tail)
2023 sk->sk_lock.owned = 0;
2024 if (waitqueue_active(&sk->sk_lock.wq))
2025 wake_up(&sk->sk_lock.wq);
2026 spin_unlock_bh(&sk->sk_lock.slock);
2028 EXPORT_SYMBOL(release_sock);
2031 * lock_sock_fast - fast version of lock_sock
2034 * This version should be used for very small section, where process wont block
2035 * return false if fast path is taken
2036 * sk_lock.slock locked, owned = 0, BH disabled
2037 * return true if slow path is taken
2038 * sk_lock.slock unlocked, owned = 1, BH enabled
2040 bool lock_sock_fast(struct sock *sk)
2043 spin_lock_bh(&sk->sk_lock.slock);
2045 if (!sk->sk_lock.owned)
2047 * Note : We must disable BH
2052 sk->sk_lock.owned = 1;
2053 spin_unlock(&sk->sk_lock.slock);
2055 * The sk_lock has mutex_lock() semantics here:
2057 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2061 EXPORT_SYMBOL(lock_sock_fast);
2063 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2066 if (!sock_flag(sk, SOCK_TIMESTAMP))
2067 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2068 tv = ktime_to_timeval(sk->sk_stamp);
2069 if (tv.tv_sec == -1)
2071 if (tv.tv_sec == 0) {
2072 sk->sk_stamp = ktime_get_real();
2073 tv = ktime_to_timeval(sk->sk_stamp);
2075 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2077 EXPORT_SYMBOL(sock_get_timestamp);
2079 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2082 if (!sock_flag(sk, SOCK_TIMESTAMP))
2083 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2084 ts = ktime_to_timespec(sk->sk_stamp);
2085 if (ts.tv_sec == -1)
2087 if (ts.tv_sec == 0) {
2088 sk->sk_stamp = ktime_get_real();
2089 ts = ktime_to_timespec(sk->sk_stamp);
2091 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2093 EXPORT_SYMBOL(sock_get_timestampns);
2095 void sock_enable_timestamp(struct sock *sk, int flag)
2097 if (!sock_flag(sk, flag)) {
2098 sock_set_flag(sk, flag);
2100 * we just set one of the two flags which require net
2101 * time stamping, but time stamping might have been on
2102 * already because of the other one
2105 flag == SOCK_TIMESTAMP ?
2106 SOCK_TIMESTAMPING_RX_SOFTWARE :
2108 net_enable_timestamp();
2113 * Get a socket option on an socket.
2115 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2116 * asynchronous errors should be reported by getsockopt. We assume
2117 * this means if you specify SO_ERROR (otherwise whats the point of it).
2119 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2120 char __user *optval, int __user *optlen)
2122 struct sock *sk = sock->sk;
2124 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2126 EXPORT_SYMBOL(sock_common_getsockopt);
2128 #ifdef CONFIG_COMPAT
2129 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2130 char __user *optval, int __user *optlen)
2132 struct sock *sk = sock->sk;
2134 if (sk->sk_prot->compat_getsockopt != NULL)
2135 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2137 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2139 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2142 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2143 struct msghdr *msg, size_t size, int flags)
2145 struct sock *sk = sock->sk;
2149 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2150 flags & ~MSG_DONTWAIT, &addr_len);
2152 msg->msg_namelen = addr_len;
2155 EXPORT_SYMBOL(sock_common_recvmsg);
2158 * Set socket options on an inet socket.
2160 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2161 char __user *optval, unsigned int optlen)
2163 struct sock *sk = sock->sk;
2165 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2167 EXPORT_SYMBOL(sock_common_setsockopt);
2169 #ifdef CONFIG_COMPAT
2170 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2171 char __user *optval, unsigned int optlen)
2173 struct sock *sk = sock->sk;
2175 if (sk->sk_prot->compat_setsockopt != NULL)
2176 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2178 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2180 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2183 void sk_common_release(struct sock *sk)
2185 if (sk->sk_prot->destroy)
2186 sk->sk_prot->destroy(sk);
2189 * Observation: when sock_common_release is called, processes have
2190 * no access to socket. But net still has.
2191 * Step one, detach it from networking:
2193 * A. Remove from hash tables.
2196 sk->sk_prot->unhash(sk);
2199 * In this point socket cannot receive new packets, but it is possible
2200 * that some packets are in flight because some CPU runs receiver and
2201 * did hash table lookup before we unhashed socket. They will achieve
2202 * receive queue and will be purged by socket destructor.
2204 * Also we still have packets pending on receive queue and probably,
2205 * our own packets waiting in device queues. sock_destroy will drain
2206 * receive queue, but transmitted packets will delay socket destruction
2207 * until the last reference will be released.
2212 xfrm_sk_free_policy(sk);
2214 sk_refcnt_debug_release(sk);
2217 EXPORT_SYMBOL(sk_common_release);
2219 static DEFINE_RWLOCK(proto_list_lock);
2220 static LIST_HEAD(proto_list);
2222 #ifdef CONFIG_PROC_FS
2223 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2225 int val[PROTO_INUSE_NR];
2228 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2230 #ifdef CONFIG_NET_NS
2231 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2233 int cpu = smp_processor_id();
2234 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2236 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2238 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2240 int cpu, idx = prot->inuse_idx;
2243 for_each_possible_cpu(cpu)
2244 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2246 return res >= 0 ? res : 0;
2248 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2250 static int __net_init sock_inuse_init_net(struct net *net)
2252 net->core.inuse = alloc_percpu(struct prot_inuse);
2253 return net->core.inuse ? 0 : -ENOMEM;
2256 static void __net_exit sock_inuse_exit_net(struct net *net)
2258 free_percpu(net->core.inuse);
2261 static struct pernet_operations net_inuse_ops = {
2262 .init = sock_inuse_init_net,
2263 .exit = sock_inuse_exit_net,
2266 static __init int net_inuse_init(void)
2268 if (register_pernet_subsys(&net_inuse_ops))
2269 panic("Cannot initialize net inuse counters");
2274 core_initcall(net_inuse_init);
2276 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2278 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2280 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2282 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2284 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2286 int cpu, idx = prot->inuse_idx;
2289 for_each_possible_cpu(cpu)
2290 res += per_cpu(prot_inuse, cpu).val[idx];
2292 return res >= 0 ? res : 0;
2294 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2297 static void assign_proto_idx(struct proto *prot)
2299 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2301 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2302 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2306 set_bit(prot->inuse_idx, proto_inuse_idx);
2309 static void release_proto_idx(struct proto *prot)
2311 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2312 clear_bit(prot->inuse_idx, proto_inuse_idx);
2315 static inline void assign_proto_idx(struct proto *prot)
2319 static inline void release_proto_idx(struct proto *prot)
2324 int proto_register(struct proto *prot, int alloc_slab)
2327 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2328 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2331 if (prot->slab == NULL) {
2332 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2337 if (prot->rsk_prot != NULL) {
2338 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2339 if (prot->rsk_prot->slab_name == NULL)
2340 goto out_free_sock_slab;
2342 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2343 prot->rsk_prot->obj_size, 0,
2344 SLAB_HWCACHE_ALIGN, NULL);
2346 if (prot->rsk_prot->slab == NULL) {
2347 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2349 goto out_free_request_sock_slab_name;
2353 if (prot->twsk_prot != NULL) {
2354 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2356 if (prot->twsk_prot->twsk_slab_name == NULL)
2357 goto out_free_request_sock_slab;
2359 prot->twsk_prot->twsk_slab =
2360 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2361 prot->twsk_prot->twsk_obj_size,
2363 SLAB_HWCACHE_ALIGN |
2366 if (prot->twsk_prot->twsk_slab == NULL)
2367 goto out_free_timewait_sock_slab_name;
2371 write_lock(&proto_list_lock);
2372 list_add(&prot->node, &proto_list);
2373 assign_proto_idx(prot);
2374 write_unlock(&proto_list_lock);
2377 out_free_timewait_sock_slab_name:
2378 kfree(prot->twsk_prot->twsk_slab_name);
2379 out_free_request_sock_slab:
2380 if (prot->rsk_prot && prot->rsk_prot->slab) {
2381 kmem_cache_destroy(prot->rsk_prot->slab);
2382 prot->rsk_prot->slab = NULL;
2384 out_free_request_sock_slab_name:
2386 kfree(prot->rsk_prot->slab_name);
2388 kmem_cache_destroy(prot->slab);
2393 EXPORT_SYMBOL(proto_register);
2395 void proto_unregister(struct proto *prot)
2397 write_lock(&proto_list_lock);
2398 release_proto_idx(prot);
2399 list_del(&prot->node);
2400 write_unlock(&proto_list_lock);
2402 if (prot->slab != NULL) {
2403 kmem_cache_destroy(prot->slab);
2407 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2408 kmem_cache_destroy(prot->rsk_prot->slab);
2409 kfree(prot->rsk_prot->slab_name);
2410 prot->rsk_prot->slab = NULL;
2413 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2414 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2415 kfree(prot->twsk_prot->twsk_slab_name);
2416 prot->twsk_prot->twsk_slab = NULL;
2419 EXPORT_SYMBOL(proto_unregister);
2421 #ifdef CONFIG_PROC_FS
2422 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2423 __acquires(proto_list_lock)
2425 read_lock(&proto_list_lock);
2426 return seq_list_start_head(&proto_list, *pos);
2429 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2431 return seq_list_next(v, &proto_list, pos);
2434 static void proto_seq_stop(struct seq_file *seq, void *v)
2435 __releases(proto_list_lock)
2437 read_unlock(&proto_list_lock);
2440 static char proto_method_implemented(const void *method)
2442 return method == NULL ? 'n' : 'y';
2445 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2447 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2448 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2451 sock_prot_inuse_get(seq_file_net(seq), proto),
2452 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2453 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2455 proto->slab == NULL ? "no" : "yes",
2456 module_name(proto->owner),
2457 proto_method_implemented(proto->close),
2458 proto_method_implemented(proto->connect),
2459 proto_method_implemented(proto->disconnect),
2460 proto_method_implemented(proto->accept),
2461 proto_method_implemented(proto->ioctl),
2462 proto_method_implemented(proto->init),
2463 proto_method_implemented(proto->destroy),
2464 proto_method_implemented(proto->shutdown),
2465 proto_method_implemented(proto->setsockopt),
2466 proto_method_implemented(proto->getsockopt),
2467 proto_method_implemented(proto->sendmsg),
2468 proto_method_implemented(proto->recvmsg),
2469 proto_method_implemented(proto->sendpage),
2470 proto_method_implemented(proto->bind),
2471 proto_method_implemented(proto->backlog_rcv),
2472 proto_method_implemented(proto->hash),
2473 proto_method_implemented(proto->unhash),
2474 proto_method_implemented(proto->get_port),
2475 proto_method_implemented(proto->enter_memory_pressure));
2478 static int proto_seq_show(struct seq_file *seq, void *v)
2480 if (v == &proto_list)
2481 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2490 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2492 proto_seq_printf(seq, list_entry(v, struct proto, node));
2496 static const struct seq_operations proto_seq_ops = {
2497 .start = proto_seq_start,
2498 .next = proto_seq_next,
2499 .stop = proto_seq_stop,
2500 .show = proto_seq_show,
2503 static int proto_seq_open(struct inode *inode, struct file *file)
2505 return seq_open_net(inode, file, &proto_seq_ops,
2506 sizeof(struct seq_net_private));
2509 static const struct file_operations proto_seq_fops = {
2510 .owner = THIS_MODULE,
2511 .open = proto_seq_open,
2513 .llseek = seq_lseek,
2514 .release = seq_release_net,
2517 static __net_init int proto_init_net(struct net *net)
2519 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2525 static __net_exit void proto_exit_net(struct net *net)
2527 proc_net_remove(net, "protocols");
2531 static __net_initdata struct pernet_operations proto_net_ops = {
2532 .init = proto_init_net,
2533 .exit = proto_exit_net,
2536 static int __init proto_init(void)
2538 return register_pernet_subsys(&proto_net_ops);
2541 subsys_initcall(proto_init);
2543 #endif /* PROC_FS */