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>
114 #include <linux/jump_label.h>
115 #include <linux/memcontrol.h>
117 #include <asm/uaccess.h>
118 #include <asm/system.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
132 #include <linux/filter.h>
134 #include <trace/events/sock.h>
140 static DEFINE_MUTEX(proto_list_mutex);
141 static LIST_HEAD(proto_list);
143 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
144 int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss)
149 mutex_lock(&proto_list_mutex);
150 list_for_each_entry(proto, &proto_list, node) {
151 if (proto->init_cgroup) {
152 ret = proto->init_cgroup(cgrp, ss);
158 mutex_unlock(&proto_list_mutex);
161 list_for_each_entry_continue_reverse(proto, &proto_list, node)
162 if (proto->destroy_cgroup)
163 proto->destroy_cgroup(cgrp, ss);
164 mutex_unlock(&proto_list_mutex);
168 void mem_cgroup_sockets_destroy(struct cgroup *cgrp, struct cgroup_subsys *ss)
172 mutex_lock(&proto_list_mutex);
173 list_for_each_entry_reverse(proto, &proto_list, node)
174 if (proto->destroy_cgroup)
175 proto->destroy_cgroup(cgrp, ss);
176 mutex_unlock(&proto_list_mutex);
181 * Each address family might have different locking rules, so we have
182 * one slock key per address family:
184 static struct lock_class_key af_family_keys[AF_MAX];
185 static struct lock_class_key af_family_slock_keys[AF_MAX];
187 struct jump_label_key memcg_socket_limit_enabled;
188 EXPORT_SYMBOL(memcg_socket_limit_enabled);
191 * Make lock validator output more readable. (we pre-construct these
192 * strings build-time, so that runtime initialization of socket
195 static const char *const af_family_key_strings[AF_MAX+1] = {
196 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
197 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
198 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
199 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
200 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
201 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
202 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
203 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
204 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
205 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
206 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
207 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
208 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
209 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
211 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
212 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
213 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
214 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
215 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
216 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
217 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
218 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
219 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
220 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
221 "slock-27" , "slock-28" , "slock-AF_CAN" ,
222 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
223 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
224 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
225 "slock-AF_NFC" , "slock-AF_MAX"
227 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
228 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
229 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
230 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
231 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
232 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
233 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
234 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
235 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
236 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
237 "clock-27" , "clock-28" , "clock-AF_CAN" ,
238 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
239 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
240 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
241 "clock-AF_NFC" , "clock-AF_MAX"
245 * sk_callback_lock locking rules are per-address-family,
246 * so split the lock classes by using a per-AF key:
248 static struct lock_class_key af_callback_keys[AF_MAX];
250 /* Take into consideration the size of the struct sk_buff overhead in the
251 * determination of these values, since that is non-constant across
252 * platforms. This makes socket queueing behavior and performance
253 * not depend upon such differences.
255 #define _SK_MEM_PACKETS 256
256 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
257 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
258 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260 /* Run time adjustable parameters. */
261 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
262 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
263 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
264 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
266 /* Maximal space eaten by iovec or ancillary data plus some space */
267 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
268 EXPORT_SYMBOL(sysctl_optmem_max);
270 #if defined(CONFIG_CGROUPS)
271 #if !defined(CONFIG_NET_CLS_CGROUP)
272 int net_cls_subsys_id = -1;
273 EXPORT_SYMBOL_GPL(net_cls_subsys_id);
275 #if !defined(CONFIG_NETPRIO_CGROUP)
276 int net_prio_subsys_id = -1;
277 EXPORT_SYMBOL_GPL(net_prio_subsys_id);
281 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
285 if (optlen < sizeof(tv))
287 if (copy_from_user(&tv, optval, sizeof(tv)))
289 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
293 static int warned __read_mostly;
296 if (warned < 10 && net_ratelimit()) {
298 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
299 "tries to set negative timeout\n",
300 current->comm, task_pid_nr(current));
304 *timeo_p = MAX_SCHEDULE_TIMEOUT;
305 if (tv.tv_sec == 0 && tv.tv_usec == 0)
307 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
308 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
312 static void sock_warn_obsolete_bsdism(const char *name)
315 static char warncomm[TASK_COMM_LEN];
316 if (strcmp(warncomm, current->comm) && warned < 5) {
317 strcpy(warncomm, current->comm);
318 printk(KERN_WARNING "process `%s' is using obsolete "
319 "%s SO_BSDCOMPAT\n", warncomm, name);
324 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
326 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
328 if (sk->sk_flags & flags) {
329 sk->sk_flags &= ~flags;
330 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
331 net_disable_timestamp();
336 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
341 struct sk_buff_head *list = &sk->sk_receive_queue;
343 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
344 atomic_inc(&sk->sk_drops);
345 trace_sock_rcvqueue_full(sk, skb);
349 err = sk_filter(sk, skb);
353 if (!sk_rmem_schedule(sk, skb->truesize)) {
354 atomic_inc(&sk->sk_drops);
359 skb_set_owner_r(skb, sk);
361 /* Cache the SKB length before we tack it onto the receive
362 * queue. Once it is added it no longer belongs to us and
363 * may be freed by other threads of control pulling packets
368 /* we escape from rcu protected region, make sure we dont leak
373 spin_lock_irqsave(&list->lock, flags);
374 skb->dropcount = atomic_read(&sk->sk_drops);
375 __skb_queue_tail(list, skb);
376 spin_unlock_irqrestore(&list->lock, flags);
378 if (!sock_flag(sk, SOCK_DEAD))
379 sk->sk_data_ready(sk, skb_len);
382 EXPORT_SYMBOL(sock_queue_rcv_skb);
384 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
386 int rc = NET_RX_SUCCESS;
388 if (sk_filter(sk, skb))
389 goto discard_and_relse;
393 if (sk_rcvqueues_full(sk, skb)) {
394 atomic_inc(&sk->sk_drops);
395 goto discard_and_relse;
398 bh_lock_sock_nested(sk);
401 if (!sock_owned_by_user(sk)) {
403 * trylock + unlock semantics:
405 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
407 rc = sk_backlog_rcv(sk, skb);
409 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
410 } else if (sk_add_backlog(sk, skb)) {
412 atomic_inc(&sk->sk_drops);
413 goto discard_and_relse;
424 EXPORT_SYMBOL(sk_receive_skb);
426 void sk_reset_txq(struct sock *sk)
428 sk_tx_queue_clear(sk);
430 EXPORT_SYMBOL(sk_reset_txq);
432 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
434 struct dst_entry *dst = __sk_dst_get(sk);
436 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
437 sk_tx_queue_clear(sk);
438 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
445 EXPORT_SYMBOL(__sk_dst_check);
447 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
449 struct dst_entry *dst = sk_dst_get(sk);
451 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
459 EXPORT_SYMBOL(sk_dst_check);
461 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
463 int ret = -ENOPROTOOPT;
464 #ifdef CONFIG_NETDEVICES
465 struct net *net = sock_net(sk);
466 char devname[IFNAMSIZ];
471 if (!capable(CAP_NET_RAW))
478 /* Bind this socket to a particular device like "eth0",
479 * as specified in the passed interface name. If the
480 * name is "" or the option length is zero the socket
483 if (optlen > IFNAMSIZ - 1)
484 optlen = IFNAMSIZ - 1;
485 memset(devname, 0, sizeof(devname));
488 if (copy_from_user(devname, optval, optlen))
492 if (devname[0] != '\0') {
493 struct net_device *dev;
496 dev = dev_get_by_name_rcu(net, devname);
498 index = dev->ifindex;
506 sk->sk_bound_dev_if = index;
518 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
521 sock_set_flag(sk, bit);
523 sock_reset_flag(sk, bit);
527 * This is meant for all protocols to use and covers goings on
528 * at the socket level. Everything here is generic.
531 int sock_setsockopt(struct socket *sock, int level, int optname,
532 char __user *optval, unsigned int optlen)
534 struct sock *sk = sock->sk;
541 * Options without arguments
544 if (optname == SO_BINDTODEVICE)
545 return sock_bindtodevice(sk, optval, optlen);
547 if (optlen < sizeof(int))
550 if (get_user(val, (int __user *)optval))
553 valbool = val ? 1 : 0;
559 if (val && !capable(CAP_NET_ADMIN))
562 sock_valbool_flag(sk, SOCK_DBG, valbool);
565 sk->sk_reuse = valbool;
574 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
577 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
580 /* Don't error on this BSD doesn't and if you think
581 about it this is right. Otherwise apps have to
582 play 'guess the biggest size' games. RCVBUF/SNDBUF
583 are treated in BSD as hints */
585 if (val > sysctl_wmem_max)
586 val = sysctl_wmem_max;
588 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
589 if ((val * 2) < SOCK_MIN_SNDBUF)
590 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
592 sk->sk_sndbuf = val * 2;
595 * Wake up sending tasks if we
598 sk->sk_write_space(sk);
602 if (!capable(CAP_NET_ADMIN)) {
609 /* Don't error on this BSD doesn't and if you think
610 about it this is right. Otherwise apps have to
611 play 'guess the biggest size' games. RCVBUF/SNDBUF
612 are treated in BSD as hints */
614 if (val > sysctl_rmem_max)
615 val = sysctl_rmem_max;
617 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
619 * We double it on the way in to account for
620 * "struct sk_buff" etc. overhead. Applications
621 * assume that the SO_RCVBUF setting they make will
622 * allow that much actual data to be received on that
625 * Applications are unaware that "struct sk_buff" and
626 * other overheads allocate from the receive buffer
627 * during socket buffer allocation.
629 * And after considering the possible alternatives,
630 * returning the value we actually used in getsockopt
631 * is the most desirable behavior.
633 if ((val * 2) < SOCK_MIN_RCVBUF)
634 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
636 sk->sk_rcvbuf = val * 2;
640 if (!capable(CAP_NET_ADMIN)) {
648 if (sk->sk_protocol == IPPROTO_TCP)
649 tcp_set_keepalive(sk, valbool);
651 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
655 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
659 sk->sk_no_check = valbool;
663 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
664 sk->sk_priority = val;
670 if (optlen < sizeof(ling)) {
671 ret = -EINVAL; /* 1003.1g */
674 if (copy_from_user(&ling, optval, sizeof(ling))) {
679 sock_reset_flag(sk, SOCK_LINGER);
681 #if (BITS_PER_LONG == 32)
682 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
683 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
686 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
687 sock_set_flag(sk, SOCK_LINGER);
692 sock_warn_obsolete_bsdism("setsockopt");
697 set_bit(SOCK_PASSCRED, &sock->flags);
699 clear_bit(SOCK_PASSCRED, &sock->flags);
705 if (optname == SO_TIMESTAMP)
706 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
708 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
709 sock_set_flag(sk, SOCK_RCVTSTAMP);
710 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
712 sock_reset_flag(sk, SOCK_RCVTSTAMP);
713 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
717 case SO_TIMESTAMPING:
718 if (val & ~SOF_TIMESTAMPING_MASK) {
722 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
723 val & SOF_TIMESTAMPING_TX_HARDWARE);
724 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
725 val & SOF_TIMESTAMPING_TX_SOFTWARE);
726 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
727 val & SOF_TIMESTAMPING_RX_HARDWARE);
728 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
729 sock_enable_timestamp(sk,
730 SOCK_TIMESTAMPING_RX_SOFTWARE);
732 sock_disable_timestamp(sk,
733 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
734 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
735 val & SOF_TIMESTAMPING_SOFTWARE);
736 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
737 val & SOF_TIMESTAMPING_SYS_HARDWARE);
738 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
739 val & SOF_TIMESTAMPING_RAW_HARDWARE);
745 sk->sk_rcvlowat = val ? : 1;
749 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
753 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
756 case SO_ATTACH_FILTER:
758 if (optlen == sizeof(struct sock_fprog)) {
759 struct sock_fprog fprog;
762 if (copy_from_user(&fprog, optval, sizeof(fprog)))
765 ret = sk_attach_filter(&fprog, sk);
769 case SO_DETACH_FILTER:
770 ret = sk_detach_filter(sk);
775 set_bit(SOCK_PASSSEC, &sock->flags);
777 clear_bit(SOCK_PASSSEC, &sock->flags);
780 if (!capable(CAP_NET_ADMIN))
786 /* We implement the SO_SNDLOWAT etc to
787 not be settable (1003.1g 5.3) */
789 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
793 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
803 EXPORT_SYMBOL(sock_setsockopt);
806 void cred_to_ucred(struct pid *pid, const struct cred *cred,
809 ucred->pid = pid_vnr(pid);
810 ucred->uid = ucred->gid = -1;
812 struct user_namespace *current_ns = current_user_ns();
814 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
815 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
818 EXPORT_SYMBOL_GPL(cred_to_ucred);
820 int sock_getsockopt(struct socket *sock, int level, int optname,
821 char __user *optval, int __user *optlen)
823 struct sock *sk = sock->sk;
831 int lv = sizeof(int);
834 if (get_user(len, optlen))
839 memset(&v, 0, sizeof(v));
843 v.val = sock_flag(sk, SOCK_DBG);
847 v.val = sock_flag(sk, SOCK_LOCALROUTE);
851 v.val = !!sock_flag(sk, SOCK_BROADCAST);
855 v.val = sk->sk_sndbuf;
859 v.val = sk->sk_rcvbuf;
863 v.val = sk->sk_reuse;
867 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
875 v.val = sk->sk_protocol;
879 v.val = sk->sk_family;
883 v.val = -sock_error(sk);
885 v.val = xchg(&sk->sk_err_soft, 0);
889 v.val = !!sock_flag(sk, SOCK_URGINLINE);
893 v.val = sk->sk_no_check;
897 v.val = sk->sk_priority;
902 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
903 v.ling.l_linger = sk->sk_lingertime / HZ;
907 sock_warn_obsolete_bsdism("getsockopt");
911 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
912 !sock_flag(sk, SOCK_RCVTSTAMPNS);
916 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
919 case SO_TIMESTAMPING:
921 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
922 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
923 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
924 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
925 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
926 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
927 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
928 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
929 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
930 v.val |= SOF_TIMESTAMPING_SOFTWARE;
931 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
932 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
933 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
934 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
938 lv = sizeof(struct timeval);
939 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
943 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
944 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
949 lv = sizeof(struct timeval);
950 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
954 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
955 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
960 v.val = sk->sk_rcvlowat;
968 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
973 struct ucred peercred;
974 if (len > sizeof(peercred))
975 len = sizeof(peercred);
976 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
977 if (copy_to_user(optval, &peercred, len))
986 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
990 if (copy_to_user(optval, address, len))
995 /* Dubious BSD thing... Probably nobody even uses it, but
996 * the UNIX standard wants it for whatever reason... -DaveM
999 v.val = sk->sk_state == TCP_LISTEN;
1003 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
1007 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1010 v.val = sk->sk_mark;
1014 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
1017 case SO_WIFI_STATUS:
1018 v.val = !!sock_flag(sk, SOCK_WIFI_STATUS);
1022 return -ENOPROTOOPT;
1027 if (copy_to_user(optval, &v, len))
1030 if (put_user(len, optlen))
1036 * Initialize an sk_lock.
1038 * (We also register the sk_lock with the lock validator.)
1040 static inline void sock_lock_init(struct sock *sk)
1042 sock_lock_init_class_and_name(sk,
1043 af_family_slock_key_strings[sk->sk_family],
1044 af_family_slock_keys + sk->sk_family,
1045 af_family_key_strings[sk->sk_family],
1046 af_family_keys + sk->sk_family);
1050 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1051 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1052 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1054 static void sock_copy(struct sock *nsk, const struct sock *osk)
1056 #ifdef CONFIG_SECURITY_NETWORK
1057 void *sptr = nsk->sk_security;
1059 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1061 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1062 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1064 #ifdef CONFIG_SECURITY_NETWORK
1065 nsk->sk_security = sptr;
1066 security_sk_clone(osk, nsk);
1071 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1072 * un-modified. Special care is taken when initializing object to zero.
1074 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1076 if (offsetof(struct sock, sk_node.next) != 0)
1077 memset(sk, 0, offsetof(struct sock, sk_node.next));
1078 memset(&sk->sk_node.pprev, 0,
1079 size - offsetof(struct sock, sk_node.pprev));
1082 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1084 unsigned long nulls1, nulls2;
1086 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1087 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1088 if (nulls1 > nulls2)
1089 swap(nulls1, nulls2);
1092 memset((char *)sk, 0, nulls1);
1093 memset((char *)sk + nulls1 + sizeof(void *), 0,
1094 nulls2 - nulls1 - sizeof(void *));
1095 memset((char *)sk + nulls2 + sizeof(void *), 0,
1096 size - nulls2 - sizeof(void *));
1098 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1100 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1104 struct kmem_cache *slab;
1108 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1111 if (priority & __GFP_ZERO) {
1113 prot->clear_sk(sk, prot->obj_size);
1115 sk_prot_clear_nulls(sk, prot->obj_size);
1118 sk = kmalloc(prot->obj_size, priority);
1121 kmemcheck_annotate_bitfield(sk, flags);
1123 if (security_sk_alloc(sk, family, priority))
1126 if (!try_module_get(prot->owner))
1128 sk_tx_queue_clear(sk);
1134 security_sk_free(sk);
1137 kmem_cache_free(slab, sk);
1143 static void sk_prot_free(struct proto *prot, struct sock *sk)
1145 struct kmem_cache *slab;
1146 struct module *owner;
1148 owner = prot->owner;
1151 security_sk_free(sk);
1153 kmem_cache_free(slab, sk);
1159 #ifdef CONFIG_CGROUPS
1160 void sock_update_classid(struct sock *sk)
1164 rcu_read_lock(); /* doing current task, which cannot vanish. */
1165 classid = task_cls_classid(current);
1167 if (classid && classid != sk->sk_classid)
1168 sk->sk_classid = classid;
1170 EXPORT_SYMBOL(sock_update_classid);
1172 void sock_update_netprioidx(struct sock *sk)
1177 sk->sk_cgrp_prioidx = task_netprioidx(current);
1179 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1183 * sk_alloc - All socket objects are allocated here
1184 * @net: the applicable net namespace
1185 * @family: protocol family
1186 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1187 * @prot: struct proto associated with this new sock instance
1189 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1194 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1196 sk->sk_family = family;
1198 * See comment in struct sock definition to understand
1199 * why we need sk_prot_creator -acme
1201 sk->sk_prot = sk->sk_prot_creator = prot;
1203 sock_net_set(sk, get_net(net));
1204 atomic_set(&sk->sk_wmem_alloc, 1);
1206 sock_update_classid(sk);
1207 sock_update_netprioidx(sk);
1212 EXPORT_SYMBOL(sk_alloc);
1214 static void __sk_free(struct sock *sk)
1216 struct sk_filter *filter;
1218 if (sk->sk_destruct)
1219 sk->sk_destruct(sk);
1221 filter = rcu_dereference_check(sk->sk_filter,
1222 atomic_read(&sk->sk_wmem_alloc) == 0);
1224 sk_filter_uncharge(sk, filter);
1225 RCU_INIT_POINTER(sk->sk_filter, NULL);
1228 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1230 if (atomic_read(&sk->sk_omem_alloc))
1231 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1232 __func__, atomic_read(&sk->sk_omem_alloc));
1234 if (sk->sk_peer_cred)
1235 put_cred(sk->sk_peer_cred);
1236 put_pid(sk->sk_peer_pid);
1237 put_net(sock_net(sk));
1238 sk_prot_free(sk->sk_prot_creator, sk);
1241 void sk_free(struct sock *sk)
1244 * We subtract one from sk_wmem_alloc and can know if
1245 * some packets are still in some tx queue.
1246 * If not null, sock_wfree() will call __sk_free(sk) later
1248 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1251 EXPORT_SYMBOL(sk_free);
1254 * Last sock_put should drop reference to sk->sk_net. It has already
1255 * been dropped in sk_change_net. Taking reference to stopping namespace
1257 * Take reference to a socket to remove it from hash _alive_ and after that
1258 * destroy it in the context of init_net.
1260 void sk_release_kernel(struct sock *sk)
1262 if (sk == NULL || sk->sk_socket == NULL)
1266 sock_release(sk->sk_socket);
1267 release_net(sock_net(sk));
1268 sock_net_set(sk, get_net(&init_net));
1271 EXPORT_SYMBOL(sk_release_kernel);
1273 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1275 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1276 sock_update_memcg(newsk);
1280 * sk_clone_lock - clone a socket, and lock its clone
1281 * @sk: the socket to clone
1282 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1284 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1286 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1290 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1291 if (newsk != NULL) {
1292 struct sk_filter *filter;
1294 sock_copy(newsk, sk);
1297 get_net(sock_net(newsk));
1298 sk_node_init(&newsk->sk_node);
1299 sock_lock_init(newsk);
1300 bh_lock_sock(newsk);
1301 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1302 newsk->sk_backlog.len = 0;
1304 atomic_set(&newsk->sk_rmem_alloc, 0);
1306 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1308 atomic_set(&newsk->sk_wmem_alloc, 1);
1309 atomic_set(&newsk->sk_omem_alloc, 0);
1310 skb_queue_head_init(&newsk->sk_receive_queue);
1311 skb_queue_head_init(&newsk->sk_write_queue);
1312 #ifdef CONFIG_NET_DMA
1313 skb_queue_head_init(&newsk->sk_async_wait_queue);
1316 spin_lock_init(&newsk->sk_dst_lock);
1317 rwlock_init(&newsk->sk_callback_lock);
1318 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1319 af_callback_keys + newsk->sk_family,
1320 af_family_clock_key_strings[newsk->sk_family]);
1322 newsk->sk_dst_cache = NULL;
1323 newsk->sk_wmem_queued = 0;
1324 newsk->sk_forward_alloc = 0;
1325 newsk->sk_send_head = NULL;
1326 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1328 sock_reset_flag(newsk, SOCK_DONE);
1329 skb_queue_head_init(&newsk->sk_error_queue);
1331 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1333 sk_filter_charge(newsk, filter);
1335 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1336 /* It is still raw copy of parent, so invalidate
1337 * destructor and make plain sk_free() */
1338 newsk->sk_destruct = NULL;
1339 bh_unlock_sock(newsk);
1346 newsk->sk_priority = 0;
1348 * Before updating sk_refcnt, we must commit prior changes to memory
1349 * (Documentation/RCU/rculist_nulls.txt for details)
1352 atomic_set(&newsk->sk_refcnt, 2);
1355 * Increment the counter in the same struct proto as the master
1356 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1357 * is the same as sk->sk_prot->socks, as this field was copied
1360 * This _changes_ the previous behaviour, where
1361 * tcp_create_openreq_child always was incrementing the
1362 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1363 * to be taken into account in all callers. -acme
1365 sk_refcnt_debug_inc(newsk);
1366 sk_set_socket(newsk, NULL);
1367 newsk->sk_wq = NULL;
1369 sk_update_clone(sk, newsk);
1371 if (newsk->sk_prot->sockets_allocated)
1372 sk_sockets_allocated_inc(newsk);
1374 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1375 net_enable_timestamp();
1380 EXPORT_SYMBOL_GPL(sk_clone_lock);
1382 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1384 __sk_dst_set(sk, dst);
1385 sk->sk_route_caps = dst->dev->features;
1386 if (sk->sk_route_caps & NETIF_F_GSO)
1387 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1388 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1389 if (sk_can_gso(sk)) {
1390 if (dst->header_len) {
1391 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1393 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1394 sk->sk_gso_max_size = dst->dev->gso_max_size;
1398 EXPORT_SYMBOL_GPL(sk_setup_caps);
1400 void __init sk_init(void)
1402 if (totalram_pages <= 4096) {
1403 sysctl_wmem_max = 32767;
1404 sysctl_rmem_max = 32767;
1405 sysctl_wmem_default = 32767;
1406 sysctl_rmem_default = 32767;
1407 } else if (totalram_pages >= 131072) {
1408 sysctl_wmem_max = 131071;
1409 sysctl_rmem_max = 131071;
1414 * Simple resource managers for sockets.
1419 * Write buffer destructor automatically called from kfree_skb.
1421 void sock_wfree(struct sk_buff *skb)
1423 struct sock *sk = skb->sk;
1424 unsigned int len = skb->truesize;
1426 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1428 * Keep a reference on sk_wmem_alloc, this will be released
1429 * after sk_write_space() call
1431 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1432 sk->sk_write_space(sk);
1436 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1437 * could not do because of in-flight packets
1439 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1442 EXPORT_SYMBOL(sock_wfree);
1445 * Read buffer destructor automatically called from kfree_skb.
1447 void sock_rfree(struct sk_buff *skb)
1449 struct sock *sk = skb->sk;
1450 unsigned int len = skb->truesize;
1452 atomic_sub(len, &sk->sk_rmem_alloc);
1453 sk_mem_uncharge(sk, len);
1455 EXPORT_SYMBOL(sock_rfree);
1458 int sock_i_uid(struct sock *sk)
1462 read_lock_bh(&sk->sk_callback_lock);
1463 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1464 read_unlock_bh(&sk->sk_callback_lock);
1467 EXPORT_SYMBOL(sock_i_uid);
1469 unsigned long sock_i_ino(struct sock *sk)
1473 read_lock_bh(&sk->sk_callback_lock);
1474 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1475 read_unlock_bh(&sk->sk_callback_lock);
1478 EXPORT_SYMBOL(sock_i_ino);
1481 * Allocate a skb from the socket's send buffer.
1483 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1486 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1487 struct sk_buff *skb = alloc_skb(size, priority);
1489 skb_set_owner_w(skb, sk);
1495 EXPORT_SYMBOL(sock_wmalloc);
1498 * Allocate a skb from the socket's receive buffer.
1500 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1503 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1504 struct sk_buff *skb = alloc_skb(size, priority);
1506 skb_set_owner_r(skb, sk);
1514 * Allocate a memory block from the socket's option memory buffer.
1516 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1518 if ((unsigned)size <= sysctl_optmem_max &&
1519 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1521 /* First do the add, to avoid the race if kmalloc
1524 atomic_add(size, &sk->sk_omem_alloc);
1525 mem = kmalloc(size, priority);
1528 atomic_sub(size, &sk->sk_omem_alloc);
1532 EXPORT_SYMBOL(sock_kmalloc);
1535 * Free an option memory block.
1537 void sock_kfree_s(struct sock *sk, void *mem, int size)
1540 atomic_sub(size, &sk->sk_omem_alloc);
1542 EXPORT_SYMBOL(sock_kfree_s);
1544 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1545 I think, these locks should be removed for datagram sockets.
1547 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1551 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1555 if (signal_pending(current))
1557 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1558 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1559 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1561 if (sk->sk_shutdown & SEND_SHUTDOWN)
1565 timeo = schedule_timeout(timeo);
1567 finish_wait(sk_sleep(sk), &wait);
1573 * Generic send/receive buffer handlers
1576 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1577 unsigned long data_len, int noblock,
1580 struct sk_buff *skb;
1585 gfp_mask = sk->sk_allocation;
1586 if (gfp_mask & __GFP_WAIT)
1587 gfp_mask |= __GFP_REPEAT;
1589 timeo = sock_sndtimeo(sk, noblock);
1591 err = sock_error(sk);
1596 if (sk->sk_shutdown & SEND_SHUTDOWN)
1599 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1600 skb = alloc_skb(header_len, gfp_mask);
1605 /* No pages, we're done... */
1609 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1610 skb->truesize += data_len;
1611 skb_shinfo(skb)->nr_frags = npages;
1612 for (i = 0; i < npages; i++) {
1615 page = alloc_pages(sk->sk_allocation, 0);
1618 skb_shinfo(skb)->nr_frags = i;
1623 __skb_fill_page_desc(skb, i,
1625 (data_len >= PAGE_SIZE ?
1628 data_len -= PAGE_SIZE;
1631 /* Full success... */
1637 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1638 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1642 if (signal_pending(current))
1644 timeo = sock_wait_for_wmem(sk, timeo);
1647 skb_set_owner_w(skb, sk);
1651 err = sock_intr_errno(timeo);
1656 EXPORT_SYMBOL(sock_alloc_send_pskb);
1658 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1659 int noblock, int *errcode)
1661 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1663 EXPORT_SYMBOL(sock_alloc_send_skb);
1665 static void __lock_sock(struct sock *sk)
1666 __releases(&sk->sk_lock.slock)
1667 __acquires(&sk->sk_lock.slock)
1672 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1673 TASK_UNINTERRUPTIBLE);
1674 spin_unlock_bh(&sk->sk_lock.slock);
1676 spin_lock_bh(&sk->sk_lock.slock);
1677 if (!sock_owned_by_user(sk))
1680 finish_wait(&sk->sk_lock.wq, &wait);
1683 static void __release_sock(struct sock *sk)
1684 __releases(&sk->sk_lock.slock)
1685 __acquires(&sk->sk_lock.slock)
1687 struct sk_buff *skb = sk->sk_backlog.head;
1690 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1694 struct sk_buff *next = skb->next;
1696 WARN_ON_ONCE(skb_dst_is_noref(skb));
1698 sk_backlog_rcv(sk, skb);
1701 * We are in process context here with softirqs
1702 * disabled, use cond_resched_softirq() to preempt.
1703 * This is safe to do because we've taken the backlog
1706 cond_resched_softirq();
1709 } while (skb != NULL);
1712 } while ((skb = sk->sk_backlog.head) != NULL);
1715 * Doing the zeroing here guarantee we can not loop forever
1716 * while a wild producer attempts to flood us.
1718 sk->sk_backlog.len = 0;
1722 * sk_wait_data - wait for data to arrive at sk_receive_queue
1723 * @sk: sock to wait on
1724 * @timeo: for how long
1726 * Now socket state including sk->sk_err is changed only under lock,
1727 * hence we may omit checks after joining wait queue.
1728 * We check receive queue before schedule() only as optimization;
1729 * it is very likely that release_sock() added new data.
1731 int sk_wait_data(struct sock *sk, long *timeo)
1736 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1737 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1738 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1739 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1740 finish_wait(sk_sleep(sk), &wait);
1743 EXPORT_SYMBOL(sk_wait_data);
1746 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1748 * @size: memory size to allocate
1749 * @kind: allocation type
1751 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1752 * rmem allocation. This function assumes that protocols which have
1753 * memory_pressure use sk_wmem_queued as write buffer accounting.
1755 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1757 struct proto *prot = sk->sk_prot;
1758 int amt = sk_mem_pages(size);
1760 int parent_status = UNDER_LIMIT;
1762 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1764 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1767 if (parent_status == UNDER_LIMIT &&
1768 allocated <= sk_prot_mem_limits(sk, 0)) {
1769 sk_leave_memory_pressure(sk);
1773 /* Under pressure. (we or our parents) */
1774 if ((parent_status > SOFT_LIMIT) ||
1775 allocated > sk_prot_mem_limits(sk, 1))
1776 sk_enter_memory_pressure(sk);
1778 /* Over hard limit (we or our parents) */
1779 if ((parent_status == OVER_LIMIT) ||
1780 (allocated > sk_prot_mem_limits(sk, 2)))
1781 goto suppress_allocation;
1783 /* guarantee minimum buffer size under pressure */
1784 if (kind == SK_MEM_RECV) {
1785 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1788 } else { /* SK_MEM_SEND */
1789 if (sk->sk_type == SOCK_STREAM) {
1790 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1792 } else if (atomic_read(&sk->sk_wmem_alloc) <
1793 prot->sysctl_wmem[0])
1797 if (sk_has_memory_pressure(sk)) {
1800 if (!sk_under_memory_pressure(sk))
1802 alloc = sk_sockets_allocated_read_positive(sk);
1803 if (sk_prot_mem_limits(sk, 2) > alloc *
1804 sk_mem_pages(sk->sk_wmem_queued +
1805 atomic_read(&sk->sk_rmem_alloc) +
1806 sk->sk_forward_alloc))
1810 suppress_allocation:
1812 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1813 sk_stream_moderate_sndbuf(sk);
1815 /* Fail only if socket is _under_ its sndbuf.
1816 * In this case we cannot block, so that we have to fail.
1818 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1822 trace_sock_exceed_buf_limit(sk, prot, allocated);
1824 /* Alas. Undo changes. */
1825 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1827 sk_memory_allocated_sub(sk, amt);
1831 EXPORT_SYMBOL(__sk_mem_schedule);
1834 * __sk_reclaim - reclaim memory_allocated
1837 void __sk_mem_reclaim(struct sock *sk)
1839 sk_memory_allocated_sub(sk,
1840 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
1841 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1843 if (sk_under_memory_pressure(sk) &&
1844 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1845 sk_leave_memory_pressure(sk);
1847 EXPORT_SYMBOL(__sk_mem_reclaim);
1851 * Set of default routines for initialising struct proto_ops when
1852 * the protocol does not support a particular function. In certain
1853 * cases where it makes no sense for a protocol to have a "do nothing"
1854 * function, some default processing is provided.
1857 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1861 EXPORT_SYMBOL(sock_no_bind);
1863 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1868 EXPORT_SYMBOL(sock_no_connect);
1870 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1874 EXPORT_SYMBOL(sock_no_socketpair);
1876 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1880 EXPORT_SYMBOL(sock_no_accept);
1882 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1887 EXPORT_SYMBOL(sock_no_getname);
1889 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1893 EXPORT_SYMBOL(sock_no_poll);
1895 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1899 EXPORT_SYMBOL(sock_no_ioctl);
1901 int sock_no_listen(struct socket *sock, int backlog)
1905 EXPORT_SYMBOL(sock_no_listen);
1907 int sock_no_shutdown(struct socket *sock, int how)
1911 EXPORT_SYMBOL(sock_no_shutdown);
1913 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1914 char __user *optval, unsigned int optlen)
1918 EXPORT_SYMBOL(sock_no_setsockopt);
1920 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1921 char __user *optval, int __user *optlen)
1925 EXPORT_SYMBOL(sock_no_getsockopt);
1927 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1932 EXPORT_SYMBOL(sock_no_sendmsg);
1934 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1935 size_t len, int flags)
1939 EXPORT_SYMBOL(sock_no_recvmsg);
1941 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1943 /* Mirror missing mmap method error code */
1946 EXPORT_SYMBOL(sock_no_mmap);
1948 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1951 struct msghdr msg = {.msg_flags = flags};
1953 char *kaddr = kmap(page);
1954 iov.iov_base = kaddr + offset;
1956 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1960 EXPORT_SYMBOL(sock_no_sendpage);
1963 * Default Socket Callbacks
1966 static void sock_def_wakeup(struct sock *sk)
1968 struct socket_wq *wq;
1971 wq = rcu_dereference(sk->sk_wq);
1972 if (wq_has_sleeper(wq))
1973 wake_up_interruptible_all(&wq->wait);
1977 static void sock_def_error_report(struct sock *sk)
1979 struct socket_wq *wq;
1982 wq = rcu_dereference(sk->sk_wq);
1983 if (wq_has_sleeper(wq))
1984 wake_up_interruptible_poll(&wq->wait, POLLERR);
1985 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1989 static void sock_def_readable(struct sock *sk, int len)
1991 struct socket_wq *wq;
1994 wq = rcu_dereference(sk->sk_wq);
1995 if (wq_has_sleeper(wq))
1996 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1997 POLLRDNORM | POLLRDBAND);
1998 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2002 static void sock_def_write_space(struct sock *sk)
2004 struct socket_wq *wq;
2008 /* Do not wake up a writer until he can make "significant"
2011 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2012 wq = rcu_dereference(sk->sk_wq);
2013 if (wq_has_sleeper(wq))
2014 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2015 POLLWRNORM | POLLWRBAND);
2017 /* Should agree with poll, otherwise some programs break */
2018 if (sock_writeable(sk))
2019 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2025 static void sock_def_destruct(struct sock *sk)
2027 kfree(sk->sk_protinfo);
2030 void sk_send_sigurg(struct sock *sk)
2032 if (sk->sk_socket && sk->sk_socket->file)
2033 if (send_sigurg(&sk->sk_socket->file->f_owner))
2034 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2036 EXPORT_SYMBOL(sk_send_sigurg);
2038 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2039 unsigned long expires)
2041 if (!mod_timer(timer, expires))
2044 EXPORT_SYMBOL(sk_reset_timer);
2046 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2048 if (timer_pending(timer) && del_timer(timer))
2051 EXPORT_SYMBOL(sk_stop_timer);
2053 void sock_init_data(struct socket *sock, struct sock *sk)
2055 skb_queue_head_init(&sk->sk_receive_queue);
2056 skb_queue_head_init(&sk->sk_write_queue);
2057 skb_queue_head_init(&sk->sk_error_queue);
2058 #ifdef CONFIG_NET_DMA
2059 skb_queue_head_init(&sk->sk_async_wait_queue);
2062 sk->sk_send_head = NULL;
2064 init_timer(&sk->sk_timer);
2066 sk->sk_allocation = GFP_KERNEL;
2067 sk->sk_rcvbuf = sysctl_rmem_default;
2068 sk->sk_sndbuf = sysctl_wmem_default;
2069 sk->sk_state = TCP_CLOSE;
2070 sk_set_socket(sk, sock);
2072 sock_set_flag(sk, SOCK_ZAPPED);
2075 sk->sk_type = sock->type;
2076 sk->sk_wq = sock->wq;
2081 spin_lock_init(&sk->sk_dst_lock);
2082 rwlock_init(&sk->sk_callback_lock);
2083 lockdep_set_class_and_name(&sk->sk_callback_lock,
2084 af_callback_keys + sk->sk_family,
2085 af_family_clock_key_strings[sk->sk_family]);
2087 sk->sk_state_change = sock_def_wakeup;
2088 sk->sk_data_ready = sock_def_readable;
2089 sk->sk_write_space = sock_def_write_space;
2090 sk->sk_error_report = sock_def_error_report;
2091 sk->sk_destruct = sock_def_destruct;
2093 sk->sk_sndmsg_page = NULL;
2094 sk->sk_sndmsg_off = 0;
2096 sk->sk_peer_pid = NULL;
2097 sk->sk_peer_cred = NULL;
2098 sk->sk_write_pending = 0;
2099 sk->sk_rcvlowat = 1;
2100 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2101 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2103 sk->sk_stamp = ktime_set(-1L, 0);
2106 * Before updating sk_refcnt, we must commit prior changes to memory
2107 * (Documentation/RCU/rculist_nulls.txt for details)
2110 atomic_set(&sk->sk_refcnt, 1);
2111 atomic_set(&sk->sk_drops, 0);
2113 EXPORT_SYMBOL(sock_init_data);
2115 void lock_sock_nested(struct sock *sk, int subclass)
2118 spin_lock_bh(&sk->sk_lock.slock);
2119 if (sk->sk_lock.owned)
2121 sk->sk_lock.owned = 1;
2122 spin_unlock(&sk->sk_lock.slock);
2124 * The sk_lock has mutex_lock() semantics here:
2126 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2129 EXPORT_SYMBOL(lock_sock_nested);
2131 void release_sock(struct sock *sk)
2134 * The sk_lock has mutex_unlock() semantics:
2136 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2138 spin_lock_bh(&sk->sk_lock.slock);
2139 if (sk->sk_backlog.tail)
2141 sk->sk_lock.owned = 0;
2142 if (waitqueue_active(&sk->sk_lock.wq))
2143 wake_up(&sk->sk_lock.wq);
2144 spin_unlock_bh(&sk->sk_lock.slock);
2146 EXPORT_SYMBOL(release_sock);
2149 * lock_sock_fast - fast version of lock_sock
2152 * This version should be used for very small section, where process wont block
2153 * return false if fast path is taken
2154 * sk_lock.slock locked, owned = 0, BH disabled
2155 * return true if slow path is taken
2156 * sk_lock.slock unlocked, owned = 1, BH enabled
2158 bool lock_sock_fast(struct sock *sk)
2161 spin_lock_bh(&sk->sk_lock.slock);
2163 if (!sk->sk_lock.owned)
2165 * Note : We must disable BH
2170 sk->sk_lock.owned = 1;
2171 spin_unlock(&sk->sk_lock.slock);
2173 * The sk_lock has mutex_lock() semantics here:
2175 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2179 EXPORT_SYMBOL(lock_sock_fast);
2181 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2184 if (!sock_flag(sk, SOCK_TIMESTAMP))
2185 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2186 tv = ktime_to_timeval(sk->sk_stamp);
2187 if (tv.tv_sec == -1)
2189 if (tv.tv_sec == 0) {
2190 sk->sk_stamp = ktime_get_real();
2191 tv = ktime_to_timeval(sk->sk_stamp);
2193 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2195 EXPORT_SYMBOL(sock_get_timestamp);
2197 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2200 if (!sock_flag(sk, SOCK_TIMESTAMP))
2201 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2202 ts = ktime_to_timespec(sk->sk_stamp);
2203 if (ts.tv_sec == -1)
2205 if (ts.tv_sec == 0) {
2206 sk->sk_stamp = ktime_get_real();
2207 ts = ktime_to_timespec(sk->sk_stamp);
2209 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2211 EXPORT_SYMBOL(sock_get_timestampns);
2213 void sock_enable_timestamp(struct sock *sk, int flag)
2215 if (!sock_flag(sk, flag)) {
2216 unsigned long previous_flags = sk->sk_flags;
2218 sock_set_flag(sk, flag);
2220 * we just set one of the two flags which require net
2221 * time stamping, but time stamping might have been on
2222 * already because of the other one
2224 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2225 net_enable_timestamp();
2230 * Get a socket option on an socket.
2232 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2233 * asynchronous errors should be reported by getsockopt. We assume
2234 * this means if you specify SO_ERROR (otherwise whats the point of it).
2236 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2237 char __user *optval, int __user *optlen)
2239 struct sock *sk = sock->sk;
2241 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2243 EXPORT_SYMBOL(sock_common_getsockopt);
2245 #ifdef CONFIG_COMPAT
2246 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2247 char __user *optval, int __user *optlen)
2249 struct sock *sk = sock->sk;
2251 if (sk->sk_prot->compat_getsockopt != NULL)
2252 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2254 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2256 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2259 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2260 struct msghdr *msg, size_t size, int flags)
2262 struct sock *sk = sock->sk;
2266 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2267 flags & ~MSG_DONTWAIT, &addr_len);
2269 msg->msg_namelen = addr_len;
2272 EXPORT_SYMBOL(sock_common_recvmsg);
2275 * Set socket options on an inet socket.
2277 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2278 char __user *optval, unsigned int optlen)
2280 struct sock *sk = sock->sk;
2282 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2284 EXPORT_SYMBOL(sock_common_setsockopt);
2286 #ifdef CONFIG_COMPAT
2287 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2288 char __user *optval, unsigned int optlen)
2290 struct sock *sk = sock->sk;
2292 if (sk->sk_prot->compat_setsockopt != NULL)
2293 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2295 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2297 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2300 void sk_common_release(struct sock *sk)
2302 if (sk->sk_prot->destroy)
2303 sk->sk_prot->destroy(sk);
2306 * Observation: when sock_common_release is called, processes have
2307 * no access to socket. But net still has.
2308 * Step one, detach it from networking:
2310 * A. Remove from hash tables.
2313 sk->sk_prot->unhash(sk);
2316 * In this point socket cannot receive new packets, but it is possible
2317 * that some packets are in flight because some CPU runs receiver and
2318 * did hash table lookup before we unhashed socket. They will achieve
2319 * receive queue and will be purged by socket destructor.
2321 * Also we still have packets pending on receive queue and probably,
2322 * our own packets waiting in device queues. sock_destroy will drain
2323 * receive queue, but transmitted packets will delay socket destruction
2324 * until the last reference will be released.
2329 xfrm_sk_free_policy(sk);
2331 sk_refcnt_debug_release(sk);
2334 EXPORT_SYMBOL(sk_common_release);
2336 #ifdef CONFIG_PROC_FS
2337 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2339 int val[PROTO_INUSE_NR];
2342 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2344 #ifdef CONFIG_NET_NS
2345 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2347 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2349 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2351 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2353 int cpu, idx = prot->inuse_idx;
2356 for_each_possible_cpu(cpu)
2357 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2359 return res >= 0 ? res : 0;
2361 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2363 static int __net_init sock_inuse_init_net(struct net *net)
2365 net->core.inuse = alloc_percpu(struct prot_inuse);
2366 return net->core.inuse ? 0 : -ENOMEM;
2369 static void __net_exit sock_inuse_exit_net(struct net *net)
2371 free_percpu(net->core.inuse);
2374 static struct pernet_operations net_inuse_ops = {
2375 .init = sock_inuse_init_net,
2376 .exit = sock_inuse_exit_net,
2379 static __init int net_inuse_init(void)
2381 if (register_pernet_subsys(&net_inuse_ops))
2382 panic("Cannot initialize net inuse counters");
2387 core_initcall(net_inuse_init);
2389 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2391 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2393 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2395 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2397 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2399 int cpu, idx = prot->inuse_idx;
2402 for_each_possible_cpu(cpu)
2403 res += per_cpu(prot_inuse, cpu).val[idx];
2405 return res >= 0 ? res : 0;
2407 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2410 static void assign_proto_idx(struct proto *prot)
2412 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2414 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2415 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2419 set_bit(prot->inuse_idx, proto_inuse_idx);
2422 static void release_proto_idx(struct proto *prot)
2424 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2425 clear_bit(prot->inuse_idx, proto_inuse_idx);
2428 static inline void assign_proto_idx(struct proto *prot)
2432 static inline void release_proto_idx(struct proto *prot)
2437 int proto_register(struct proto *prot, int alloc_slab)
2440 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2441 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2444 if (prot->slab == NULL) {
2445 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2450 if (prot->rsk_prot != NULL) {
2451 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2452 if (prot->rsk_prot->slab_name == NULL)
2453 goto out_free_sock_slab;
2455 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2456 prot->rsk_prot->obj_size, 0,
2457 SLAB_HWCACHE_ALIGN, NULL);
2459 if (prot->rsk_prot->slab == NULL) {
2460 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2462 goto out_free_request_sock_slab_name;
2466 if (prot->twsk_prot != NULL) {
2467 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2469 if (prot->twsk_prot->twsk_slab_name == NULL)
2470 goto out_free_request_sock_slab;
2472 prot->twsk_prot->twsk_slab =
2473 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2474 prot->twsk_prot->twsk_obj_size,
2476 SLAB_HWCACHE_ALIGN |
2479 if (prot->twsk_prot->twsk_slab == NULL)
2480 goto out_free_timewait_sock_slab_name;
2484 mutex_lock(&proto_list_mutex);
2485 list_add(&prot->node, &proto_list);
2486 assign_proto_idx(prot);
2487 mutex_unlock(&proto_list_mutex);
2490 out_free_timewait_sock_slab_name:
2491 kfree(prot->twsk_prot->twsk_slab_name);
2492 out_free_request_sock_slab:
2493 if (prot->rsk_prot && prot->rsk_prot->slab) {
2494 kmem_cache_destroy(prot->rsk_prot->slab);
2495 prot->rsk_prot->slab = NULL;
2497 out_free_request_sock_slab_name:
2499 kfree(prot->rsk_prot->slab_name);
2501 kmem_cache_destroy(prot->slab);
2506 EXPORT_SYMBOL(proto_register);
2508 void proto_unregister(struct proto *prot)
2510 mutex_lock(&proto_list_mutex);
2511 release_proto_idx(prot);
2512 list_del(&prot->node);
2513 mutex_unlock(&proto_list_mutex);
2515 if (prot->slab != NULL) {
2516 kmem_cache_destroy(prot->slab);
2520 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2521 kmem_cache_destroy(prot->rsk_prot->slab);
2522 kfree(prot->rsk_prot->slab_name);
2523 prot->rsk_prot->slab = NULL;
2526 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2527 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2528 kfree(prot->twsk_prot->twsk_slab_name);
2529 prot->twsk_prot->twsk_slab = NULL;
2532 EXPORT_SYMBOL(proto_unregister);
2534 #ifdef CONFIG_PROC_FS
2535 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2536 __acquires(proto_list_mutex)
2538 mutex_lock(&proto_list_mutex);
2539 return seq_list_start_head(&proto_list, *pos);
2542 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2544 return seq_list_next(v, &proto_list, pos);
2547 static void proto_seq_stop(struct seq_file *seq, void *v)
2548 __releases(proto_list_mutex)
2550 mutex_unlock(&proto_list_mutex);
2553 static char proto_method_implemented(const void *method)
2555 return method == NULL ? 'n' : 'y';
2557 static long sock_prot_memory_allocated(struct proto *proto)
2559 return proto->memory_allocated != NULL ? proto_memory_allocated(proto): -1L;
2562 static char *sock_prot_memory_pressure(struct proto *proto)
2564 return proto->memory_pressure != NULL ?
2565 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2568 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2571 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2572 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2575 sock_prot_inuse_get(seq_file_net(seq), proto),
2576 sock_prot_memory_allocated(proto),
2577 sock_prot_memory_pressure(proto),
2579 proto->slab == NULL ? "no" : "yes",
2580 module_name(proto->owner),
2581 proto_method_implemented(proto->close),
2582 proto_method_implemented(proto->connect),
2583 proto_method_implemented(proto->disconnect),
2584 proto_method_implemented(proto->accept),
2585 proto_method_implemented(proto->ioctl),
2586 proto_method_implemented(proto->init),
2587 proto_method_implemented(proto->destroy),
2588 proto_method_implemented(proto->shutdown),
2589 proto_method_implemented(proto->setsockopt),
2590 proto_method_implemented(proto->getsockopt),
2591 proto_method_implemented(proto->sendmsg),
2592 proto_method_implemented(proto->recvmsg),
2593 proto_method_implemented(proto->sendpage),
2594 proto_method_implemented(proto->bind),
2595 proto_method_implemented(proto->backlog_rcv),
2596 proto_method_implemented(proto->hash),
2597 proto_method_implemented(proto->unhash),
2598 proto_method_implemented(proto->get_port),
2599 proto_method_implemented(proto->enter_memory_pressure));
2602 static int proto_seq_show(struct seq_file *seq, void *v)
2604 if (v == &proto_list)
2605 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2614 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2616 proto_seq_printf(seq, list_entry(v, struct proto, node));
2620 static const struct seq_operations proto_seq_ops = {
2621 .start = proto_seq_start,
2622 .next = proto_seq_next,
2623 .stop = proto_seq_stop,
2624 .show = proto_seq_show,
2627 static int proto_seq_open(struct inode *inode, struct file *file)
2629 return seq_open_net(inode, file, &proto_seq_ops,
2630 sizeof(struct seq_net_private));
2633 static const struct file_operations proto_seq_fops = {
2634 .owner = THIS_MODULE,
2635 .open = proto_seq_open,
2637 .llseek = seq_lseek,
2638 .release = seq_release_net,
2641 static __net_init int proto_init_net(struct net *net)
2643 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2649 static __net_exit void proto_exit_net(struct net *net)
2651 proc_net_remove(net, "protocols");
2655 static __net_initdata struct pernet_operations proto_net_ops = {
2656 .init = proto_init_net,
2657 .exit = proto_exit_net,
2660 static int __init proto_init(void)
2662 return register_pernet_subsys(&proto_net_ops);
2665 subsys_initcall(proto_init);
2667 #endif /* PROC_FS */