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" , "sk_lock-AF_ALG" ,
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" , "slock-AF_ALG" ,
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" , "clock-AF_ALG" ,
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 ancillary 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 sk->sk_type == SOCK_STREAM)
599 tcp_set_keepalive(sk, valbool);
601 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
605 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
609 sk->sk_no_check = valbool;
613 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
614 sk->sk_priority = val;
620 if (optlen < sizeof(ling)) {
621 ret = -EINVAL; /* 1003.1g */
624 if (copy_from_user(&ling, optval, sizeof(ling))) {
629 sock_reset_flag(sk, SOCK_LINGER);
631 #if (BITS_PER_LONG == 32)
632 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
633 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
636 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
637 sock_set_flag(sk, SOCK_LINGER);
642 sock_warn_obsolete_bsdism("setsockopt");
647 set_bit(SOCK_PASSCRED, &sock->flags);
649 clear_bit(SOCK_PASSCRED, &sock->flags);
655 if (optname == SO_TIMESTAMP)
656 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
658 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
659 sock_set_flag(sk, SOCK_RCVTSTAMP);
660 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
662 sock_reset_flag(sk, SOCK_RCVTSTAMP);
663 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
667 case SO_TIMESTAMPING:
668 if (val & ~SOF_TIMESTAMPING_MASK) {
672 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
673 val & SOF_TIMESTAMPING_TX_HARDWARE);
674 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
675 val & SOF_TIMESTAMPING_TX_SOFTWARE);
676 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
677 val & SOF_TIMESTAMPING_RX_HARDWARE);
678 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
679 sock_enable_timestamp(sk,
680 SOCK_TIMESTAMPING_RX_SOFTWARE);
682 sock_disable_timestamp(sk,
683 SOCK_TIMESTAMPING_RX_SOFTWARE);
684 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
685 val & SOF_TIMESTAMPING_SOFTWARE);
686 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
687 val & SOF_TIMESTAMPING_SYS_HARDWARE);
688 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
689 val & SOF_TIMESTAMPING_RAW_HARDWARE);
695 sk->sk_rcvlowat = val ? : 1;
699 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
703 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
706 case SO_ATTACH_FILTER:
708 if (optlen == sizeof(struct sock_fprog)) {
709 struct sock_fprog fprog;
712 if (copy_from_user(&fprog, optval, sizeof(fprog)))
715 ret = sk_attach_filter(&fprog, sk);
719 case SO_DETACH_FILTER:
720 ret = sk_detach_filter(sk);
725 set_bit(SOCK_PASSSEC, &sock->flags);
727 clear_bit(SOCK_PASSSEC, &sock->flags);
730 if (!capable(CAP_NET_ADMIN))
736 /* We implement the SO_SNDLOWAT etc to
737 not be settable (1003.1g 5.3) */
740 sock_set_flag(sk, SOCK_RXQ_OVFL);
742 sock_reset_flag(sk, SOCK_RXQ_OVFL);
751 EXPORT_SYMBOL(sock_setsockopt);
754 void cred_to_ucred(struct pid *pid, const struct cred *cred,
755 struct ucred *ucred, bool use_effective)
757 ucred->pid = pid_vnr(pid);
758 ucred->uid = ucred->gid = -1;
760 struct user_namespace *current_ns = current_user_ns();
763 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
764 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
766 ucred->uid = user_ns_map_uid(current_ns, cred, cred->uid);
767 ucred->gid = user_ns_map_gid(current_ns, cred, cred->gid);
771 EXPORT_SYMBOL_GPL(cred_to_ucred);
773 int sock_getsockopt(struct socket *sock, int level, int optname,
774 char __user *optval, int __user *optlen)
776 struct sock *sk = sock->sk;
784 int lv = sizeof(int);
787 if (get_user(len, optlen))
792 memset(&v, 0, sizeof(v));
796 v.val = sock_flag(sk, SOCK_DBG);
800 v.val = sock_flag(sk, SOCK_LOCALROUTE);
804 v.val = !!sock_flag(sk, SOCK_BROADCAST);
808 v.val = sk->sk_sndbuf;
812 v.val = sk->sk_rcvbuf;
816 v.val = sk->sk_reuse;
820 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
828 v.val = sk->sk_protocol;
832 v.val = sk->sk_family;
836 v.val = -sock_error(sk);
838 v.val = xchg(&sk->sk_err_soft, 0);
842 v.val = !!sock_flag(sk, SOCK_URGINLINE);
846 v.val = sk->sk_no_check;
850 v.val = sk->sk_priority;
855 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
856 v.ling.l_linger = sk->sk_lingertime / HZ;
860 sock_warn_obsolete_bsdism("getsockopt");
864 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
865 !sock_flag(sk, SOCK_RCVTSTAMPNS);
869 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
872 case SO_TIMESTAMPING:
874 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
875 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
876 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
877 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
878 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
879 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
880 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
881 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
882 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
883 v.val |= SOF_TIMESTAMPING_SOFTWARE;
884 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
885 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
886 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
887 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
891 lv = sizeof(struct timeval);
892 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
896 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
897 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
902 lv = sizeof(struct timeval);
903 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
907 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
908 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
913 v.val = sk->sk_rcvlowat;
921 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
926 struct ucred peercred;
927 if (len > sizeof(peercred))
928 len = sizeof(peercred);
929 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred,
931 if (copy_to_user(optval, &peercred, len))
940 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
944 if (copy_to_user(optval, address, len))
949 /* Dubious BSD thing... Probably nobody even uses it, but
950 * the UNIX standard wants it for whatever reason... -DaveM
953 v.val = sk->sk_state == TCP_LISTEN;
957 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
961 return security_socket_getpeersec_stream(sock, optval, optlen, len);
968 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
977 if (copy_to_user(optval, &v, len))
980 if (put_user(len, optlen))
986 * Initialize an sk_lock.
988 * (We also register the sk_lock with the lock validator.)
990 static inline void sock_lock_init(struct sock *sk)
992 sock_lock_init_class_and_name(sk,
993 af_family_slock_key_strings[sk->sk_family],
994 af_family_slock_keys + sk->sk_family,
995 af_family_key_strings[sk->sk_family],
996 af_family_keys + sk->sk_family);
1000 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1001 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1002 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1004 static void sock_copy(struct sock *nsk, const struct sock *osk)
1006 #ifdef CONFIG_SECURITY_NETWORK
1007 void *sptr = nsk->sk_security;
1009 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1011 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1012 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1014 #ifdef CONFIG_SECURITY_NETWORK
1015 nsk->sk_security = sptr;
1016 security_sk_clone(osk, nsk);
1021 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1022 * un-modified. Special care is taken when initializing object to zero.
1024 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1026 if (offsetof(struct sock, sk_node.next) != 0)
1027 memset(sk, 0, offsetof(struct sock, sk_node.next));
1028 memset(&sk->sk_node.pprev, 0,
1029 size - offsetof(struct sock, sk_node.pprev));
1032 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1034 unsigned long nulls1, nulls2;
1036 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1037 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1038 if (nulls1 > nulls2)
1039 swap(nulls1, nulls2);
1042 memset((char *)sk, 0, nulls1);
1043 memset((char *)sk + nulls1 + sizeof(void *), 0,
1044 nulls2 - nulls1 - sizeof(void *));
1045 memset((char *)sk + nulls2 + sizeof(void *), 0,
1046 size - nulls2 - sizeof(void *));
1048 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1050 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1054 struct kmem_cache *slab;
1058 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1061 if (priority & __GFP_ZERO) {
1063 prot->clear_sk(sk, prot->obj_size);
1065 sk_prot_clear_nulls(sk, prot->obj_size);
1068 sk = kmalloc(prot->obj_size, priority);
1071 kmemcheck_annotate_bitfield(sk, flags);
1073 if (security_sk_alloc(sk, family, priority))
1076 if (!try_module_get(prot->owner))
1078 sk_tx_queue_clear(sk);
1084 security_sk_free(sk);
1087 kmem_cache_free(slab, sk);
1093 static void sk_prot_free(struct proto *prot, struct sock *sk)
1095 struct kmem_cache *slab;
1096 struct module *owner;
1098 owner = prot->owner;
1101 security_sk_free(sk);
1103 kmem_cache_free(slab, sk);
1109 #ifdef CONFIG_CGROUPS
1110 void sock_update_classid(struct sock *sk)
1114 rcu_read_lock(); /* doing current task, which cannot vanish. */
1115 classid = task_cls_classid(current);
1117 if (classid && classid != sk->sk_classid)
1118 sk->sk_classid = classid;
1120 EXPORT_SYMBOL(sock_update_classid);
1124 * sk_alloc - All socket objects are allocated here
1125 * @net: the applicable net namespace
1126 * @family: protocol family
1127 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1128 * @prot: struct proto associated with this new sock instance
1130 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1135 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1137 sk->sk_family = family;
1139 * See comment in struct sock definition to understand
1140 * why we need sk_prot_creator -acme
1142 sk->sk_prot = sk->sk_prot_creator = prot;
1144 sock_net_set(sk, get_net(net));
1145 atomic_set(&sk->sk_wmem_alloc, 1);
1147 sock_update_classid(sk);
1152 EXPORT_SYMBOL(sk_alloc);
1154 static void __sk_free(struct sock *sk)
1156 struct sk_filter *filter;
1158 if (sk->sk_destruct)
1159 sk->sk_destruct(sk);
1161 filter = rcu_dereference_check(sk->sk_filter,
1162 atomic_read(&sk->sk_wmem_alloc) == 0);
1164 sk_filter_uncharge(sk, filter);
1165 rcu_assign_pointer(sk->sk_filter, NULL);
1168 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1169 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1171 if (atomic_read(&sk->sk_omem_alloc))
1172 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1173 __func__, atomic_read(&sk->sk_omem_alloc));
1175 if (sk->sk_peer_cred)
1176 put_cred(sk->sk_peer_cred);
1177 put_pid(sk->sk_peer_pid);
1178 put_net(sock_net(sk));
1179 sk_prot_free(sk->sk_prot_creator, sk);
1182 void sk_free(struct sock *sk)
1185 * We subtract one from sk_wmem_alloc and can know if
1186 * some packets are still in some tx queue.
1187 * If not null, sock_wfree() will call __sk_free(sk) later
1189 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1192 EXPORT_SYMBOL(sk_free);
1195 * Last sock_put should drop reference to sk->sk_net. It has already
1196 * been dropped in sk_change_net. Taking reference to stopping namespace
1198 * Take reference to a socket to remove it from hash _alive_ and after that
1199 * destroy it in the context of init_net.
1201 void sk_release_kernel(struct sock *sk)
1203 if (sk == NULL || sk->sk_socket == NULL)
1207 sock_release(sk->sk_socket);
1208 release_net(sock_net(sk));
1209 sock_net_set(sk, get_net(&init_net));
1212 EXPORT_SYMBOL(sk_release_kernel);
1214 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1218 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1219 if (newsk != NULL) {
1220 struct sk_filter *filter;
1222 sock_copy(newsk, sk);
1225 get_net(sock_net(newsk));
1226 sk_node_init(&newsk->sk_node);
1227 sock_lock_init(newsk);
1228 bh_lock_sock(newsk);
1229 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1230 newsk->sk_backlog.len = 0;
1232 atomic_set(&newsk->sk_rmem_alloc, 0);
1234 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1236 atomic_set(&newsk->sk_wmem_alloc, 1);
1237 atomic_set(&newsk->sk_omem_alloc, 0);
1238 skb_queue_head_init(&newsk->sk_receive_queue);
1239 skb_queue_head_init(&newsk->sk_write_queue);
1240 #ifdef CONFIG_NET_DMA
1241 skb_queue_head_init(&newsk->sk_async_wait_queue);
1244 spin_lock_init(&newsk->sk_dst_lock);
1245 rwlock_init(&newsk->sk_callback_lock);
1246 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1247 af_callback_keys + newsk->sk_family,
1248 af_family_clock_key_strings[newsk->sk_family]);
1250 newsk->sk_dst_cache = NULL;
1251 newsk->sk_wmem_queued = 0;
1252 newsk->sk_forward_alloc = 0;
1253 newsk->sk_send_head = NULL;
1254 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1256 sock_reset_flag(newsk, SOCK_DONE);
1257 skb_queue_head_init(&newsk->sk_error_queue);
1259 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1261 sk_filter_charge(newsk, filter);
1263 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1264 /* It is still raw copy of parent, so invalidate
1265 * destructor and make plain sk_free() */
1266 newsk->sk_destruct = NULL;
1267 bh_unlock_sock(newsk);
1274 newsk->sk_priority = 0;
1276 * Before updating sk_refcnt, we must commit prior changes to memory
1277 * (Documentation/RCU/rculist_nulls.txt for details)
1280 atomic_set(&newsk->sk_refcnt, 2);
1283 * Increment the counter in the same struct proto as the master
1284 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1285 * is the same as sk->sk_prot->socks, as this field was copied
1288 * This _changes_ the previous behaviour, where
1289 * tcp_create_openreq_child always was incrementing the
1290 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1291 * to be taken into account in all callers. -acme
1293 sk_refcnt_debug_inc(newsk);
1294 sk_set_socket(newsk, NULL);
1295 newsk->sk_wq = NULL;
1297 if (newsk->sk_prot->sockets_allocated)
1298 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1300 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1301 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1302 net_enable_timestamp();
1307 EXPORT_SYMBOL_GPL(sk_clone);
1309 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1311 __sk_dst_set(sk, dst);
1312 sk->sk_route_caps = dst->dev->features;
1313 if (sk->sk_route_caps & NETIF_F_GSO)
1314 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1315 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1316 if (sk_can_gso(sk)) {
1317 if (dst->header_len) {
1318 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1320 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1321 sk->sk_gso_max_size = dst->dev->gso_max_size;
1322 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1326 EXPORT_SYMBOL_GPL(sk_setup_caps);
1328 void __init sk_init(void)
1330 if (totalram_pages <= 4096) {
1331 sysctl_wmem_max = 32767;
1332 sysctl_rmem_max = 32767;
1333 sysctl_wmem_default = 32767;
1334 sysctl_rmem_default = 32767;
1335 } else if (totalram_pages >= 131072) {
1336 sysctl_wmem_max = 131071;
1337 sysctl_rmem_max = 131071;
1342 * Simple resource managers for sockets.
1347 * Write buffer destructor automatically called from kfree_skb.
1349 void sock_wfree(struct sk_buff *skb)
1351 struct sock *sk = skb->sk;
1352 unsigned int len = skb->truesize;
1354 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1356 * Keep a reference on sk_wmem_alloc, this will be released
1357 * after sk_write_space() call
1359 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1360 sk->sk_write_space(sk);
1364 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1365 * could not do because of in-flight packets
1367 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1370 EXPORT_SYMBOL(sock_wfree);
1373 * Read buffer destructor automatically called from kfree_skb.
1375 void sock_rfree(struct sk_buff *skb)
1377 struct sock *sk = skb->sk;
1378 unsigned int len = skb->truesize;
1380 atomic_sub(len, &sk->sk_rmem_alloc);
1381 sk_mem_uncharge(sk, len);
1383 EXPORT_SYMBOL(sock_rfree);
1386 int sock_i_uid(struct sock *sk)
1390 read_lock_bh(&sk->sk_callback_lock);
1391 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1392 read_unlock_bh(&sk->sk_callback_lock);
1395 EXPORT_SYMBOL(sock_i_uid);
1397 unsigned long sock_i_ino(struct sock *sk)
1401 read_lock_bh(&sk->sk_callback_lock);
1402 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1403 read_unlock_bh(&sk->sk_callback_lock);
1406 EXPORT_SYMBOL(sock_i_ino);
1409 * Allocate a skb from the socket's send buffer.
1411 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1414 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1415 struct sk_buff *skb = alloc_skb(size, priority);
1417 skb_set_owner_w(skb, sk);
1423 EXPORT_SYMBOL(sock_wmalloc);
1426 * Allocate a skb from the socket's receive buffer.
1428 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1431 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1432 struct sk_buff *skb = alloc_skb(size, priority);
1434 skb_set_owner_r(skb, sk);
1442 * Allocate a memory block from the socket's option memory buffer.
1444 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1446 if ((unsigned)size <= sysctl_optmem_max &&
1447 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1449 /* First do the add, to avoid the race if kmalloc
1452 atomic_add(size, &sk->sk_omem_alloc);
1453 mem = kmalloc(size, priority);
1456 atomic_sub(size, &sk->sk_omem_alloc);
1460 EXPORT_SYMBOL(sock_kmalloc);
1463 * Free an option memory block.
1465 void sock_kfree_s(struct sock *sk, void *mem, int size)
1468 atomic_sub(size, &sk->sk_omem_alloc);
1470 EXPORT_SYMBOL(sock_kfree_s);
1472 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1473 I think, these locks should be removed for datagram sockets.
1475 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1479 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1483 if (signal_pending(current))
1485 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1486 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1487 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1489 if (sk->sk_shutdown & SEND_SHUTDOWN)
1493 timeo = schedule_timeout(timeo);
1495 finish_wait(sk_sleep(sk), &wait);
1501 * Generic send/receive buffer handlers
1504 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1505 unsigned long data_len, int noblock,
1508 struct sk_buff *skb;
1512 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1515 if (npages > MAX_SKB_FRAGS)
1518 gfp_mask = sk->sk_allocation;
1519 if (gfp_mask & __GFP_WAIT)
1520 gfp_mask |= __GFP_REPEAT;
1522 timeo = sock_sndtimeo(sk, noblock);
1524 err = sock_error(sk);
1529 if (sk->sk_shutdown & SEND_SHUTDOWN)
1532 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1533 skb = alloc_skb(header_len, gfp_mask);
1537 /* No pages, we're done... */
1541 skb->truesize += data_len;
1542 skb_shinfo(skb)->nr_frags = npages;
1543 for (i = 0; i < npages; i++) {
1547 page = alloc_pages(sk->sk_allocation, 0);
1550 skb_shinfo(skb)->nr_frags = i;
1555 frag = &skb_shinfo(skb)->frags[i];
1557 frag->page_offset = 0;
1558 frag->size = (data_len >= PAGE_SIZE ?
1561 data_len -= PAGE_SIZE;
1564 /* Full success... */
1570 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1571 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1575 if (signal_pending(current))
1577 timeo = sock_wait_for_wmem(sk, timeo);
1580 skb_set_owner_w(skb, sk);
1584 err = sock_intr_errno(timeo);
1589 EXPORT_SYMBOL(sock_alloc_send_pskb);
1591 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1592 int noblock, int *errcode)
1594 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1596 EXPORT_SYMBOL(sock_alloc_send_skb);
1598 static void __lock_sock(struct sock *sk)
1599 __releases(&sk->sk_lock.slock)
1600 __acquires(&sk->sk_lock.slock)
1605 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1606 TASK_UNINTERRUPTIBLE);
1607 spin_unlock_bh(&sk->sk_lock.slock);
1609 spin_lock_bh(&sk->sk_lock.slock);
1610 if (!sock_owned_by_user(sk))
1613 finish_wait(&sk->sk_lock.wq, &wait);
1616 static void __release_sock(struct sock *sk)
1617 __releases(&sk->sk_lock.slock)
1618 __acquires(&sk->sk_lock.slock)
1620 struct sk_buff *skb = sk->sk_backlog.head;
1623 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1627 struct sk_buff *next = skb->next;
1629 WARN_ON_ONCE(skb_dst_is_noref(skb));
1631 sk_backlog_rcv(sk, skb);
1634 * We are in process context here with softirqs
1635 * disabled, use cond_resched_softirq() to preempt.
1636 * This is safe to do because we've taken the backlog
1639 cond_resched_softirq();
1642 } while (skb != NULL);
1645 } while ((skb = sk->sk_backlog.head) != NULL);
1648 * Doing the zeroing here guarantee we can not loop forever
1649 * while a wild producer attempts to flood us.
1651 sk->sk_backlog.len = 0;
1655 * sk_wait_data - wait for data to arrive at sk_receive_queue
1656 * @sk: sock to wait on
1657 * @timeo: for how long
1659 * Now socket state including sk->sk_err is changed only under lock,
1660 * hence we may omit checks after joining wait queue.
1661 * We check receive queue before schedule() only as optimization;
1662 * it is very likely that release_sock() added new data.
1664 int sk_wait_data(struct sock *sk, long *timeo)
1669 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1670 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1671 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1672 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1673 finish_wait(sk_sleep(sk), &wait);
1676 EXPORT_SYMBOL(sk_wait_data);
1679 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1681 * @size: memory size to allocate
1682 * @kind: allocation type
1684 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1685 * rmem allocation. This function assumes that protocols which have
1686 * memory_pressure use sk_wmem_queued as write buffer accounting.
1688 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1690 struct proto *prot = sk->sk_prot;
1691 int amt = sk_mem_pages(size);
1694 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1695 allocated = atomic_long_add_return(amt, prot->memory_allocated);
1698 if (allocated <= prot->sysctl_mem[0]) {
1699 if (prot->memory_pressure && *prot->memory_pressure)
1700 *prot->memory_pressure = 0;
1704 /* Under pressure. */
1705 if (allocated > prot->sysctl_mem[1])
1706 if (prot->enter_memory_pressure)
1707 prot->enter_memory_pressure(sk);
1709 /* Over hard limit. */
1710 if (allocated > prot->sysctl_mem[2])
1711 goto suppress_allocation;
1713 /* guarantee minimum buffer size under pressure */
1714 if (kind == SK_MEM_RECV) {
1715 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1717 } else { /* SK_MEM_SEND */
1718 if (sk->sk_type == SOCK_STREAM) {
1719 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1721 } else if (atomic_read(&sk->sk_wmem_alloc) <
1722 prot->sysctl_wmem[0])
1726 if (prot->memory_pressure) {
1729 if (!*prot->memory_pressure)
1731 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1732 if (prot->sysctl_mem[2] > alloc *
1733 sk_mem_pages(sk->sk_wmem_queued +
1734 atomic_read(&sk->sk_rmem_alloc) +
1735 sk->sk_forward_alloc))
1739 suppress_allocation:
1741 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1742 sk_stream_moderate_sndbuf(sk);
1744 /* Fail only if socket is _under_ its sndbuf.
1745 * In this case we cannot block, so that we have to fail.
1747 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1751 /* Alas. Undo changes. */
1752 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1753 atomic_long_sub(amt, prot->memory_allocated);
1756 EXPORT_SYMBOL(__sk_mem_schedule);
1759 * __sk_reclaim - reclaim memory_allocated
1762 void __sk_mem_reclaim(struct sock *sk)
1764 struct proto *prot = sk->sk_prot;
1766 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1767 prot->memory_allocated);
1768 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1770 if (prot->memory_pressure && *prot->memory_pressure &&
1771 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1772 *prot->memory_pressure = 0;
1774 EXPORT_SYMBOL(__sk_mem_reclaim);
1778 * Set of default routines for initialising struct proto_ops when
1779 * the protocol does not support a particular function. In certain
1780 * cases where it makes no sense for a protocol to have a "do nothing"
1781 * function, some default processing is provided.
1784 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1788 EXPORT_SYMBOL(sock_no_bind);
1790 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1795 EXPORT_SYMBOL(sock_no_connect);
1797 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1801 EXPORT_SYMBOL(sock_no_socketpair);
1803 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1807 EXPORT_SYMBOL(sock_no_accept);
1809 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1814 EXPORT_SYMBOL(sock_no_getname);
1816 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1820 EXPORT_SYMBOL(sock_no_poll);
1822 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1826 EXPORT_SYMBOL(sock_no_ioctl);
1828 int sock_no_listen(struct socket *sock, int backlog)
1832 EXPORT_SYMBOL(sock_no_listen);
1834 int sock_no_shutdown(struct socket *sock, int how)
1838 EXPORT_SYMBOL(sock_no_shutdown);
1840 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1841 char __user *optval, unsigned int optlen)
1845 EXPORT_SYMBOL(sock_no_setsockopt);
1847 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1848 char __user *optval, int __user *optlen)
1852 EXPORT_SYMBOL(sock_no_getsockopt);
1854 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1859 EXPORT_SYMBOL(sock_no_sendmsg);
1861 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1862 size_t len, int flags)
1866 EXPORT_SYMBOL(sock_no_recvmsg);
1868 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1870 /* Mirror missing mmap method error code */
1873 EXPORT_SYMBOL(sock_no_mmap);
1875 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1878 struct msghdr msg = {.msg_flags = flags};
1880 char *kaddr = kmap(page);
1881 iov.iov_base = kaddr + offset;
1883 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1887 EXPORT_SYMBOL(sock_no_sendpage);
1890 * Default Socket Callbacks
1893 static void sock_def_wakeup(struct sock *sk)
1895 struct socket_wq *wq;
1898 wq = rcu_dereference(sk->sk_wq);
1899 if (wq_has_sleeper(wq))
1900 wake_up_interruptible_all(&wq->wait);
1904 static void sock_def_error_report(struct sock *sk)
1906 struct socket_wq *wq;
1909 wq = rcu_dereference(sk->sk_wq);
1910 if (wq_has_sleeper(wq))
1911 wake_up_interruptible_poll(&wq->wait, POLLERR);
1912 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1916 static void sock_def_readable(struct sock *sk, int len)
1918 struct socket_wq *wq;
1921 wq = rcu_dereference(sk->sk_wq);
1922 if (wq_has_sleeper(wq))
1923 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1924 POLLRDNORM | POLLRDBAND);
1925 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1929 static void sock_def_write_space(struct sock *sk)
1931 struct socket_wq *wq;
1935 /* Do not wake up a writer until he can make "significant"
1938 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1939 wq = rcu_dereference(sk->sk_wq);
1940 if (wq_has_sleeper(wq))
1941 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1942 POLLWRNORM | POLLWRBAND);
1944 /* Should agree with poll, otherwise some programs break */
1945 if (sock_writeable(sk))
1946 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1952 static void sock_def_destruct(struct sock *sk)
1954 kfree(sk->sk_protinfo);
1957 void sk_send_sigurg(struct sock *sk)
1959 if (sk->sk_socket && sk->sk_socket->file)
1960 if (send_sigurg(&sk->sk_socket->file->f_owner))
1961 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1963 EXPORT_SYMBOL(sk_send_sigurg);
1965 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1966 unsigned long expires)
1968 if (!mod_timer(timer, expires))
1971 EXPORT_SYMBOL(sk_reset_timer);
1973 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1975 if (timer_pending(timer) && del_timer(timer))
1978 EXPORT_SYMBOL(sk_stop_timer);
1980 void sock_init_data(struct socket *sock, struct sock *sk)
1982 skb_queue_head_init(&sk->sk_receive_queue);
1983 skb_queue_head_init(&sk->sk_write_queue);
1984 skb_queue_head_init(&sk->sk_error_queue);
1985 #ifdef CONFIG_NET_DMA
1986 skb_queue_head_init(&sk->sk_async_wait_queue);
1989 sk->sk_send_head = NULL;
1991 init_timer(&sk->sk_timer);
1993 sk->sk_allocation = GFP_KERNEL;
1994 sk->sk_rcvbuf = sysctl_rmem_default;
1995 sk->sk_sndbuf = sysctl_wmem_default;
1996 sk->sk_state = TCP_CLOSE;
1997 sk_set_socket(sk, sock);
1999 sock_set_flag(sk, SOCK_ZAPPED);
2002 sk->sk_type = sock->type;
2003 sk->sk_wq = sock->wq;
2008 spin_lock_init(&sk->sk_dst_lock);
2009 rwlock_init(&sk->sk_callback_lock);
2010 lockdep_set_class_and_name(&sk->sk_callback_lock,
2011 af_callback_keys + sk->sk_family,
2012 af_family_clock_key_strings[sk->sk_family]);
2014 sk->sk_state_change = sock_def_wakeup;
2015 sk->sk_data_ready = sock_def_readable;
2016 sk->sk_write_space = sock_def_write_space;
2017 sk->sk_error_report = sock_def_error_report;
2018 sk->sk_destruct = sock_def_destruct;
2020 sk->sk_sndmsg_page = NULL;
2021 sk->sk_sndmsg_off = 0;
2023 sk->sk_peer_pid = NULL;
2024 sk->sk_peer_cred = NULL;
2025 sk->sk_write_pending = 0;
2026 sk->sk_rcvlowat = 1;
2027 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2028 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2030 sk->sk_stamp = ktime_set(-1L, 0);
2033 * Before updating sk_refcnt, we must commit prior changes to memory
2034 * (Documentation/RCU/rculist_nulls.txt for details)
2037 atomic_set(&sk->sk_refcnt, 1);
2038 atomic_set(&sk->sk_drops, 0);
2040 EXPORT_SYMBOL(sock_init_data);
2042 void lock_sock_nested(struct sock *sk, int subclass)
2045 spin_lock_bh(&sk->sk_lock.slock);
2046 if (sk->sk_lock.owned)
2048 sk->sk_lock.owned = 1;
2049 spin_unlock(&sk->sk_lock.slock);
2051 * The sk_lock has mutex_lock() semantics here:
2053 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2056 EXPORT_SYMBOL(lock_sock_nested);
2058 void release_sock(struct sock *sk)
2061 * The sk_lock has mutex_unlock() semantics:
2063 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2065 spin_lock_bh(&sk->sk_lock.slock);
2066 if (sk->sk_backlog.tail)
2068 sk->sk_lock.owned = 0;
2069 if (waitqueue_active(&sk->sk_lock.wq))
2070 wake_up(&sk->sk_lock.wq);
2071 spin_unlock_bh(&sk->sk_lock.slock);
2073 EXPORT_SYMBOL(release_sock);
2076 * lock_sock_fast - fast version of lock_sock
2079 * This version should be used for very small section, where process wont block
2080 * return false if fast path is taken
2081 * sk_lock.slock locked, owned = 0, BH disabled
2082 * return true if slow path is taken
2083 * sk_lock.slock unlocked, owned = 1, BH enabled
2085 bool lock_sock_fast(struct sock *sk)
2088 spin_lock_bh(&sk->sk_lock.slock);
2090 if (!sk->sk_lock.owned)
2092 * Note : We must disable BH
2097 sk->sk_lock.owned = 1;
2098 spin_unlock(&sk->sk_lock.slock);
2100 * The sk_lock has mutex_lock() semantics here:
2102 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2106 EXPORT_SYMBOL(lock_sock_fast);
2108 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2111 if (!sock_flag(sk, SOCK_TIMESTAMP))
2112 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2113 tv = ktime_to_timeval(sk->sk_stamp);
2114 if (tv.tv_sec == -1)
2116 if (tv.tv_sec == 0) {
2117 sk->sk_stamp = ktime_get_real();
2118 tv = ktime_to_timeval(sk->sk_stamp);
2120 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2122 EXPORT_SYMBOL(sock_get_timestamp);
2124 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2127 if (!sock_flag(sk, SOCK_TIMESTAMP))
2128 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2129 ts = ktime_to_timespec(sk->sk_stamp);
2130 if (ts.tv_sec == -1)
2132 if (ts.tv_sec == 0) {
2133 sk->sk_stamp = ktime_get_real();
2134 ts = ktime_to_timespec(sk->sk_stamp);
2136 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2138 EXPORT_SYMBOL(sock_get_timestampns);
2140 void sock_enable_timestamp(struct sock *sk, int flag)
2142 if (!sock_flag(sk, flag)) {
2143 sock_set_flag(sk, flag);
2145 * we just set one of the two flags which require net
2146 * time stamping, but time stamping might have been on
2147 * already because of the other one
2150 flag == SOCK_TIMESTAMP ?
2151 SOCK_TIMESTAMPING_RX_SOFTWARE :
2153 net_enable_timestamp();
2158 * Get a socket option on an socket.
2160 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2161 * asynchronous errors should be reported by getsockopt. We assume
2162 * this means if you specify SO_ERROR (otherwise whats the point of it).
2164 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2165 char __user *optval, int __user *optlen)
2167 struct sock *sk = sock->sk;
2169 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2171 EXPORT_SYMBOL(sock_common_getsockopt);
2173 #ifdef CONFIG_COMPAT
2174 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2175 char __user *optval, int __user *optlen)
2177 struct sock *sk = sock->sk;
2179 if (sk->sk_prot->compat_getsockopt != NULL)
2180 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2182 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2184 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2187 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2188 struct msghdr *msg, size_t size, int flags)
2190 struct sock *sk = sock->sk;
2194 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2195 flags & ~MSG_DONTWAIT, &addr_len);
2197 msg->msg_namelen = addr_len;
2200 EXPORT_SYMBOL(sock_common_recvmsg);
2203 * Set socket options on an inet socket.
2205 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2206 char __user *optval, unsigned int optlen)
2208 struct sock *sk = sock->sk;
2210 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2212 EXPORT_SYMBOL(sock_common_setsockopt);
2214 #ifdef CONFIG_COMPAT
2215 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2216 char __user *optval, unsigned int optlen)
2218 struct sock *sk = sock->sk;
2220 if (sk->sk_prot->compat_setsockopt != NULL)
2221 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2223 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2225 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2228 void sk_common_release(struct sock *sk)
2230 if (sk->sk_prot->destroy)
2231 sk->sk_prot->destroy(sk);
2234 * Observation: when sock_common_release is called, processes have
2235 * no access to socket. But net still has.
2236 * Step one, detach it from networking:
2238 * A. Remove from hash tables.
2241 sk->sk_prot->unhash(sk);
2244 * In this point socket cannot receive new packets, but it is possible
2245 * that some packets are in flight because some CPU runs receiver and
2246 * did hash table lookup before we unhashed socket. They will achieve
2247 * receive queue and will be purged by socket destructor.
2249 * Also we still have packets pending on receive queue and probably,
2250 * our own packets waiting in device queues. sock_destroy will drain
2251 * receive queue, but transmitted packets will delay socket destruction
2252 * until the last reference will be released.
2257 xfrm_sk_free_policy(sk);
2259 sk_refcnt_debug_release(sk);
2262 EXPORT_SYMBOL(sk_common_release);
2264 static DEFINE_RWLOCK(proto_list_lock);
2265 static LIST_HEAD(proto_list);
2267 #ifdef CONFIG_PROC_FS
2268 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2270 int val[PROTO_INUSE_NR];
2273 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2275 #ifdef CONFIG_NET_NS
2276 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2278 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2280 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2282 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2284 int cpu, idx = prot->inuse_idx;
2287 for_each_possible_cpu(cpu)
2288 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2290 return res >= 0 ? res : 0;
2292 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2294 static int __net_init sock_inuse_init_net(struct net *net)
2296 net->core.inuse = alloc_percpu(struct prot_inuse);
2297 return net->core.inuse ? 0 : -ENOMEM;
2300 static void __net_exit sock_inuse_exit_net(struct net *net)
2302 free_percpu(net->core.inuse);
2305 static struct pernet_operations net_inuse_ops = {
2306 .init = sock_inuse_init_net,
2307 .exit = sock_inuse_exit_net,
2310 static __init int net_inuse_init(void)
2312 if (register_pernet_subsys(&net_inuse_ops))
2313 panic("Cannot initialize net inuse counters");
2318 core_initcall(net_inuse_init);
2320 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2322 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2324 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2326 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2328 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2330 int cpu, idx = prot->inuse_idx;
2333 for_each_possible_cpu(cpu)
2334 res += per_cpu(prot_inuse, cpu).val[idx];
2336 return res >= 0 ? res : 0;
2338 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2341 static void assign_proto_idx(struct proto *prot)
2343 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2345 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2346 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2350 set_bit(prot->inuse_idx, proto_inuse_idx);
2353 static void release_proto_idx(struct proto *prot)
2355 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2356 clear_bit(prot->inuse_idx, proto_inuse_idx);
2359 static inline void assign_proto_idx(struct proto *prot)
2363 static inline void release_proto_idx(struct proto *prot)
2368 int proto_register(struct proto *prot, int alloc_slab)
2371 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2372 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2375 if (prot->slab == NULL) {
2376 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2381 if (prot->rsk_prot != NULL) {
2382 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2383 if (prot->rsk_prot->slab_name == NULL)
2384 goto out_free_sock_slab;
2386 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2387 prot->rsk_prot->obj_size, 0,
2388 SLAB_HWCACHE_ALIGN, NULL);
2390 if (prot->rsk_prot->slab == NULL) {
2391 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2393 goto out_free_request_sock_slab_name;
2397 if (prot->twsk_prot != NULL) {
2398 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2400 if (prot->twsk_prot->twsk_slab_name == NULL)
2401 goto out_free_request_sock_slab;
2403 prot->twsk_prot->twsk_slab =
2404 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2405 prot->twsk_prot->twsk_obj_size,
2407 SLAB_HWCACHE_ALIGN |
2410 if (prot->twsk_prot->twsk_slab == NULL)
2411 goto out_free_timewait_sock_slab_name;
2415 write_lock(&proto_list_lock);
2416 list_add(&prot->node, &proto_list);
2417 assign_proto_idx(prot);
2418 write_unlock(&proto_list_lock);
2421 out_free_timewait_sock_slab_name:
2422 kfree(prot->twsk_prot->twsk_slab_name);
2423 out_free_request_sock_slab:
2424 if (prot->rsk_prot && prot->rsk_prot->slab) {
2425 kmem_cache_destroy(prot->rsk_prot->slab);
2426 prot->rsk_prot->slab = NULL;
2428 out_free_request_sock_slab_name:
2430 kfree(prot->rsk_prot->slab_name);
2432 kmem_cache_destroy(prot->slab);
2437 EXPORT_SYMBOL(proto_register);
2439 void proto_unregister(struct proto *prot)
2441 write_lock(&proto_list_lock);
2442 release_proto_idx(prot);
2443 list_del(&prot->node);
2444 write_unlock(&proto_list_lock);
2446 if (prot->slab != NULL) {
2447 kmem_cache_destroy(prot->slab);
2451 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2452 kmem_cache_destroy(prot->rsk_prot->slab);
2453 kfree(prot->rsk_prot->slab_name);
2454 prot->rsk_prot->slab = NULL;
2457 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2458 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2459 kfree(prot->twsk_prot->twsk_slab_name);
2460 prot->twsk_prot->twsk_slab = NULL;
2463 EXPORT_SYMBOL(proto_unregister);
2465 #ifdef CONFIG_PROC_FS
2466 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2467 __acquires(proto_list_lock)
2469 read_lock(&proto_list_lock);
2470 return seq_list_start_head(&proto_list, *pos);
2473 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2475 return seq_list_next(v, &proto_list, pos);
2478 static void proto_seq_stop(struct seq_file *seq, void *v)
2479 __releases(proto_list_lock)
2481 read_unlock(&proto_list_lock);
2484 static char proto_method_implemented(const void *method)
2486 return method == NULL ? 'n' : 'y';
2489 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2491 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2492 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2495 sock_prot_inuse_get(seq_file_net(seq), proto),
2496 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2497 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2499 proto->slab == NULL ? "no" : "yes",
2500 module_name(proto->owner),
2501 proto_method_implemented(proto->close),
2502 proto_method_implemented(proto->connect),
2503 proto_method_implemented(proto->disconnect),
2504 proto_method_implemented(proto->accept),
2505 proto_method_implemented(proto->ioctl),
2506 proto_method_implemented(proto->init),
2507 proto_method_implemented(proto->destroy),
2508 proto_method_implemented(proto->shutdown),
2509 proto_method_implemented(proto->setsockopt),
2510 proto_method_implemented(proto->getsockopt),
2511 proto_method_implemented(proto->sendmsg),
2512 proto_method_implemented(proto->recvmsg),
2513 proto_method_implemented(proto->sendpage),
2514 proto_method_implemented(proto->bind),
2515 proto_method_implemented(proto->backlog_rcv),
2516 proto_method_implemented(proto->hash),
2517 proto_method_implemented(proto->unhash),
2518 proto_method_implemented(proto->get_port),
2519 proto_method_implemented(proto->enter_memory_pressure));
2522 static int proto_seq_show(struct seq_file *seq, void *v)
2524 if (v == &proto_list)
2525 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2534 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2536 proto_seq_printf(seq, list_entry(v, struct proto, node));
2540 static const struct seq_operations proto_seq_ops = {
2541 .start = proto_seq_start,
2542 .next = proto_seq_next,
2543 .stop = proto_seq_stop,
2544 .show = proto_seq_show,
2547 static int proto_seq_open(struct inode *inode, struct file *file)
2549 return seq_open_net(inode, file, &proto_seq_ops,
2550 sizeof(struct seq_net_private));
2553 static const struct file_operations proto_seq_fops = {
2554 .owner = THIS_MODULE,
2555 .open = proto_seq_open,
2557 .llseek = seq_lseek,
2558 .release = seq_release_net,
2561 static __net_init int proto_init_net(struct net *net)
2563 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2569 static __net_exit void proto_exit_net(struct net *net)
2571 proc_net_remove(net, "protocols");
2575 static __net_initdata struct pernet_operations proto_net_ops = {
2576 .init = proto_init_net,
2577 .exit = proto_exit_net,
2580 static int __init proto_init(void)
2582 return register_pernet_subsys(&proto_net_ops);
2585 subsys_initcall(proto_init);
2587 #endif /* PROC_FS */