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);
797 if (sock->ops->set_peek_off)
798 sock->ops->set_peek_off(sk, val);
809 EXPORT_SYMBOL(sock_setsockopt);
812 void cred_to_ucred(struct pid *pid, const struct cred *cred,
815 ucred->pid = pid_vnr(pid);
816 ucred->uid = ucred->gid = -1;
818 struct user_namespace *current_ns = current_user_ns();
820 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
821 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
824 EXPORT_SYMBOL_GPL(cred_to_ucred);
826 int sock_getsockopt(struct socket *sock, int level, int optname,
827 char __user *optval, int __user *optlen)
829 struct sock *sk = sock->sk;
837 int lv = sizeof(int);
840 if (get_user(len, optlen))
845 memset(&v, 0, sizeof(v));
849 v.val = sock_flag(sk, SOCK_DBG);
853 v.val = sock_flag(sk, SOCK_LOCALROUTE);
857 v.val = !!sock_flag(sk, SOCK_BROADCAST);
861 v.val = sk->sk_sndbuf;
865 v.val = sk->sk_rcvbuf;
869 v.val = sk->sk_reuse;
873 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
881 v.val = sk->sk_protocol;
885 v.val = sk->sk_family;
889 v.val = -sock_error(sk);
891 v.val = xchg(&sk->sk_err_soft, 0);
895 v.val = !!sock_flag(sk, SOCK_URGINLINE);
899 v.val = sk->sk_no_check;
903 v.val = sk->sk_priority;
908 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
909 v.ling.l_linger = sk->sk_lingertime / HZ;
913 sock_warn_obsolete_bsdism("getsockopt");
917 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
918 !sock_flag(sk, SOCK_RCVTSTAMPNS);
922 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
925 case SO_TIMESTAMPING:
927 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
928 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
929 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
930 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
931 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
932 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
933 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
934 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
935 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
936 v.val |= SOF_TIMESTAMPING_SOFTWARE;
937 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
938 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
939 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
940 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
944 lv = sizeof(struct timeval);
945 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
949 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
950 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
955 lv = sizeof(struct timeval);
956 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
960 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
961 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
966 v.val = sk->sk_rcvlowat;
974 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
979 struct ucred peercred;
980 if (len > sizeof(peercred))
981 len = sizeof(peercred);
982 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
983 if (copy_to_user(optval, &peercred, len))
992 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
996 if (copy_to_user(optval, address, len))
1001 /* Dubious BSD thing... Probably nobody even uses it, but
1002 * the UNIX standard wants it for whatever reason... -DaveM
1005 v.val = sk->sk_state == TCP_LISTEN;
1009 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
1013 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1016 v.val = sk->sk_mark;
1020 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
1023 case SO_WIFI_STATUS:
1024 v.val = !!sock_flag(sk, SOCK_WIFI_STATUS);
1028 if (!sock->ops->set_peek_off)
1031 v.val = sk->sk_peek_off;
1034 return -ENOPROTOOPT;
1039 if (copy_to_user(optval, &v, len))
1042 if (put_user(len, optlen))
1048 * Initialize an sk_lock.
1050 * (We also register the sk_lock with the lock validator.)
1052 static inline void sock_lock_init(struct sock *sk)
1054 sock_lock_init_class_and_name(sk,
1055 af_family_slock_key_strings[sk->sk_family],
1056 af_family_slock_keys + sk->sk_family,
1057 af_family_key_strings[sk->sk_family],
1058 af_family_keys + sk->sk_family);
1062 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1063 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1064 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1066 static void sock_copy(struct sock *nsk, const struct sock *osk)
1068 #ifdef CONFIG_SECURITY_NETWORK
1069 void *sptr = nsk->sk_security;
1071 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1073 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1074 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1076 #ifdef CONFIG_SECURITY_NETWORK
1077 nsk->sk_security = sptr;
1078 security_sk_clone(osk, nsk);
1083 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1084 * un-modified. Special care is taken when initializing object to zero.
1086 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1088 if (offsetof(struct sock, sk_node.next) != 0)
1089 memset(sk, 0, offsetof(struct sock, sk_node.next));
1090 memset(&sk->sk_node.pprev, 0,
1091 size - offsetof(struct sock, sk_node.pprev));
1094 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1096 unsigned long nulls1, nulls2;
1098 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1099 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1100 if (nulls1 > nulls2)
1101 swap(nulls1, nulls2);
1104 memset((char *)sk, 0, nulls1);
1105 memset((char *)sk + nulls1 + sizeof(void *), 0,
1106 nulls2 - nulls1 - sizeof(void *));
1107 memset((char *)sk + nulls2 + sizeof(void *), 0,
1108 size - nulls2 - sizeof(void *));
1110 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1112 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1116 struct kmem_cache *slab;
1120 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1123 if (priority & __GFP_ZERO) {
1125 prot->clear_sk(sk, prot->obj_size);
1127 sk_prot_clear_nulls(sk, prot->obj_size);
1130 sk = kmalloc(prot->obj_size, priority);
1133 kmemcheck_annotate_bitfield(sk, flags);
1135 if (security_sk_alloc(sk, family, priority))
1138 if (!try_module_get(prot->owner))
1140 sk_tx_queue_clear(sk);
1146 security_sk_free(sk);
1149 kmem_cache_free(slab, sk);
1155 static void sk_prot_free(struct proto *prot, struct sock *sk)
1157 struct kmem_cache *slab;
1158 struct module *owner;
1160 owner = prot->owner;
1163 security_sk_free(sk);
1165 kmem_cache_free(slab, sk);
1171 #ifdef CONFIG_CGROUPS
1172 void sock_update_classid(struct sock *sk)
1176 rcu_read_lock(); /* doing current task, which cannot vanish. */
1177 classid = task_cls_classid(current);
1179 if (classid && classid != sk->sk_classid)
1180 sk->sk_classid = classid;
1182 EXPORT_SYMBOL(sock_update_classid);
1184 void sock_update_netprioidx(struct sock *sk)
1189 sk->sk_cgrp_prioidx = task_netprioidx(current);
1191 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1195 * sk_alloc - All socket objects are allocated here
1196 * @net: the applicable net namespace
1197 * @family: protocol family
1198 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1199 * @prot: struct proto associated with this new sock instance
1201 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1206 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1208 sk->sk_family = family;
1210 * See comment in struct sock definition to understand
1211 * why we need sk_prot_creator -acme
1213 sk->sk_prot = sk->sk_prot_creator = prot;
1215 sock_net_set(sk, get_net(net));
1216 atomic_set(&sk->sk_wmem_alloc, 1);
1218 sock_update_classid(sk);
1219 sock_update_netprioidx(sk);
1224 EXPORT_SYMBOL(sk_alloc);
1226 static void __sk_free(struct sock *sk)
1228 struct sk_filter *filter;
1230 if (sk->sk_destruct)
1231 sk->sk_destruct(sk);
1233 filter = rcu_dereference_check(sk->sk_filter,
1234 atomic_read(&sk->sk_wmem_alloc) == 0);
1236 sk_filter_uncharge(sk, filter);
1237 RCU_INIT_POINTER(sk->sk_filter, NULL);
1240 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1242 if (atomic_read(&sk->sk_omem_alloc))
1243 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1244 __func__, atomic_read(&sk->sk_omem_alloc));
1246 if (sk->sk_peer_cred)
1247 put_cred(sk->sk_peer_cred);
1248 put_pid(sk->sk_peer_pid);
1249 put_net(sock_net(sk));
1250 sk_prot_free(sk->sk_prot_creator, sk);
1253 void sk_free(struct sock *sk)
1256 * We subtract one from sk_wmem_alloc and can know if
1257 * some packets are still in some tx queue.
1258 * If not null, sock_wfree() will call __sk_free(sk) later
1260 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1263 EXPORT_SYMBOL(sk_free);
1266 * Last sock_put should drop reference to sk->sk_net. It has already
1267 * been dropped in sk_change_net. Taking reference to stopping namespace
1269 * Take reference to a socket to remove it from hash _alive_ and after that
1270 * destroy it in the context of init_net.
1272 void sk_release_kernel(struct sock *sk)
1274 if (sk == NULL || sk->sk_socket == NULL)
1278 sock_release(sk->sk_socket);
1279 release_net(sock_net(sk));
1280 sock_net_set(sk, get_net(&init_net));
1283 EXPORT_SYMBOL(sk_release_kernel);
1285 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1287 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1288 sock_update_memcg(newsk);
1292 * sk_clone_lock - clone a socket, and lock its clone
1293 * @sk: the socket to clone
1294 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1296 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1298 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1302 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1303 if (newsk != NULL) {
1304 struct sk_filter *filter;
1306 sock_copy(newsk, sk);
1309 get_net(sock_net(newsk));
1310 sk_node_init(&newsk->sk_node);
1311 sock_lock_init(newsk);
1312 bh_lock_sock(newsk);
1313 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1314 newsk->sk_backlog.len = 0;
1316 atomic_set(&newsk->sk_rmem_alloc, 0);
1318 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1320 atomic_set(&newsk->sk_wmem_alloc, 1);
1321 atomic_set(&newsk->sk_omem_alloc, 0);
1322 skb_queue_head_init(&newsk->sk_receive_queue);
1323 skb_queue_head_init(&newsk->sk_write_queue);
1324 #ifdef CONFIG_NET_DMA
1325 skb_queue_head_init(&newsk->sk_async_wait_queue);
1328 spin_lock_init(&newsk->sk_dst_lock);
1329 rwlock_init(&newsk->sk_callback_lock);
1330 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1331 af_callback_keys + newsk->sk_family,
1332 af_family_clock_key_strings[newsk->sk_family]);
1334 newsk->sk_dst_cache = NULL;
1335 newsk->sk_wmem_queued = 0;
1336 newsk->sk_forward_alloc = 0;
1337 newsk->sk_send_head = NULL;
1338 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1340 sock_reset_flag(newsk, SOCK_DONE);
1341 skb_queue_head_init(&newsk->sk_error_queue);
1343 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1345 sk_filter_charge(newsk, filter);
1347 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1348 /* It is still raw copy of parent, so invalidate
1349 * destructor and make plain sk_free() */
1350 newsk->sk_destruct = NULL;
1351 bh_unlock_sock(newsk);
1358 newsk->sk_priority = 0;
1360 * Before updating sk_refcnt, we must commit prior changes to memory
1361 * (Documentation/RCU/rculist_nulls.txt for details)
1364 atomic_set(&newsk->sk_refcnt, 2);
1367 * Increment the counter in the same struct proto as the master
1368 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1369 * is the same as sk->sk_prot->socks, as this field was copied
1372 * This _changes_ the previous behaviour, where
1373 * tcp_create_openreq_child always was incrementing the
1374 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1375 * to be taken into account in all callers. -acme
1377 sk_refcnt_debug_inc(newsk);
1378 sk_set_socket(newsk, NULL);
1379 newsk->sk_wq = NULL;
1381 sk_update_clone(sk, newsk);
1383 if (newsk->sk_prot->sockets_allocated)
1384 sk_sockets_allocated_inc(newsk);
1386 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1387 net_enable_timestamp();
1392 EXPORT_SYMBOL_GPL(sk_clone_lock);
1394 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1396 __sk_dst_set(sk, dst);
1397 sk->sk_route_caps = dst->dev->features;
1398 if (sk->sk_route_caps & NETIF_F_GSO)
1399 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1400 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1401 if (sk_can_gso(sk)) {
1402 if (dst->header_len) {
1403 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1405 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1406 sk->sk_gso_max_size = dst->dev->gso_max_size;
1410 EXPORT_SYMBOL_GPL(sk_setup_caps);
1412 void __init sk_init(void)
1414 if (totalram_pages <= 4096) {
1415 sysctl_wmem_max = 32767;
1416 sysctl_rmem_max = 32767;
1417 sysctl_wmem_default = 32767;
1418 sysctl_rmem_default = 32767;
1419 } else if (totalram_pages >= 131072) {
1420 sysctl_wmem_max = 131071;
1421 sysctl_rmem_max = 131071;
1426 * Simple resource managers for sockets.
1431 * Write buffer destructor automatically called from kfree_skb.
1433 void sock_wfree(struct sk_buff *skb)
1435 struct sock *sk = skb->sk;
1436 unsigned int len = skb->truesize;
1438 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1440 * Keep a reference on sk_wmem_alloc, this will be released
1441 * after sk_write_space() call
1443 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1444 sk->sk_write_space(sk);
1448 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1449 * could not do because of in-flight packets
1451 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1454 EXPORT_SYMBOL(sock_wfree);
1457 * Read buffer destructor automatically called from kfree_skb.
1459 void sock_rfree(struct sk_buff *skb)
1461 struct sock *sk = skb->sk;
1462 unsigned int len = skb->truesize;
1464 atomic_sub(len, &sk->sk_rmem_alloc);
1465 sk_mem_uncharge(sk, len);
1467 EXPORT_SYMBOL(sock_rfree);
1470 int sock_i_uid(struct sock *sk)
1474 read_lock_bh(&sk->sk_callback_lock);
1475 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1476 read_unlock_bh(&sk->sk_callback_lock);
1479 EXPORT_SYMBOL(sock_i_uid);
1481 unsigned long sock_i_ino(struct sock *sk)
1485 read_lock_bh(&sk->sk_callback_lock);
1486 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1487 read_unlock_bh(&sk->sk_callback_lock);
1490 EXPORT_SYMBOL(sock_i_ino);
1493 * Allocate a skb from the socket's send buffer.
1495 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1498 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1499 struct sk_buff *skb = alloc_skb(size, priority);
1501 skb_set_owner_w(skb, sk);
1507 EXPORT_SYMBOL(sock_wmalloc);
1510 * Allocate a skb from the socket's receive buffer.
1512 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1515 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1516 struct sk_buff *skb = alloc_skb(size, priority);
1518 skb_set_owner_r(skb, sk);
1526 * Allocate a memory block from the socket's option memory buffer.
1528 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1530 if ((unsigned)size <= sysctl_optmem_max &&
1531 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1533 /* First do the add, to avoid the race if kmalloc
1536 atomic_add(size, &sk->sk_omem_alloc);
1537 mem = kmalloc(size, priority);
1540 atomic_sub(size, &sk->sk_omem_alloc);
1544 EXPORT_SYMBOL(sock_kmalloc);
1547 * Free an option memory block.
1549 void sock_kfree_s(struct sock *sk, void *mem, int size)
1552 atomic_sub(size, &sk->sk_omem_alloc);
1554 EXPORT_SYMBOL(sock_kfree_s);
1556 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1557 I think, these locks should be removed for datagram sockets.
1559 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1563 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1567 if (signal_pending(current))
1569 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1570 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1571 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1573 if (sk->sk_shutdown & SEND_SHUTDOWN)
1577 timeo = schedule_timeout(timeo);
1579 finish_wait(sk_sleep(sk), &wait);
1585 * Generic send/receive buffer handlers
1588 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1589 unsigned long data_len, int noblock,
1592 struct sk_buff *skb;
1597 gfp_mask = sk->sk_allocation;
1598 if (gfp_mask & __GFP_WAIT)
1599 gfp_mask |= __GFP_REPEAT;
1601 timeo = sock_sndtimeo(sk, noblock);
1603 err = sock_error(sk);
1608 if (sk->sk_shutdown & SEND_SHUTDOWN)
1611 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1612 skb = alloc_skb(header_len, gfp_mask);
1617 /* No pages, we're done... */
1621 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1622 skb->truesize += data_len;
1623 skb_shinfo(skb)->nr_frags = npages;
1624 for (i = 0; i < npages; i++) {
1627 page = alloc_pages(sk->sk_allocation, 0);
1630 skb_shinfo(skb)->nr_frags = i;
1635 __skb_fill_page_desc(skb, i,
1637 (data_len >= PAGE_SIZE ?
1640 data_len -= PAGE_SIZE;
1643 /* Full success... */
1649 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1650 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1654 if (signal_pending(current))
1656 timeo = sock_wait_for_wmem(sk, timeo);
1659 skb_set_owner_w(skb, sk);
1663 err = sock_intr_errno(timeo);
1668 EXPORT_SYMBOL(sock_alloc_send_pskb);
1670 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1671 int noblock, int *errcode)
1673 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1675 EXPORT_SYMBOL(sock_alloc_send_skb);
1677 static void __lock_sock(struct sock *sk)
1678 __releases(&sk->sk_lock.slock)
1679 __acquires(&sk->sk_lock.slock)
1684 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1685 TASK_UNINTERRUPTIBLE);
1686 spin_unlock_bh(&sk->sk_lock.slock);
1688 spin_lock_bh(&sk->sk_lock.slock);
1689 if (!sock_owned_by_user(sk))
1692 finish_wait(&sk->sk_lock.wq, &wait);
1695 static void __release_sock(struct sock *sk)
1696 __releases(&sk->sk_lock.slock)
1697 __acquires(&sk->sk_lock.slock)
1699 struct sk_buff *skb = sk->sk_backlog.head;
1702 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1706 struct sk_buff *next = skb->next;
1708 WARN_ON_ONCE(skb_dst_is_noref(skb));
1710 sk_backlog_rcv(sk, skb);
1713 * We are in process context here with softirqs
1714 * disabled, use cond_resched_softirq() to preempt.
1715 * This is safe to do because we've taken the backlog
1718 cond_resched_softirq();
1721 } while (skb != NULL);
1724 } while ((skb = sk->sk_backlog.head) != NULL);
1727 * Doing the zeroing here guarantee we can not loop forever
1728 * while a wild producer attempts to flood us.
1730 sk->sk_backlog.len = 0;
1734 * sk_wait_data - wait for data to arrive at sk_receive_queue
1735 * @sk: sock to wait on
1736 * @timeo: for how long
1738 * Now socket state including sk->sk_err is changed only under lock,
1739 * hence we may omit checks after joining wait queue.
1740 * We check receive queue before schedule() only as optimization;
1741 * it is very likely that release_sock() added new data.
1743 int sk_wait_data(struct sock *sk, long *timeo)
1748 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1749 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1750 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1751 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1752 finish_wait(sk_sleep(sk), &wait);
1755 EXPORT_SYMBOL(sk_wait_data);
1758 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1760 * @size: memory size to allocate
1761 * @kind: allocation type
1763 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1764 * rmem allocation. This function assumes that protocols which have
1765 * memory_pressure use sk_wmem_queued as write buffer accounting.
1767 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1769 struct proto *prot = sk->sk_prot;
1770 int amt = sk_mem_pages(size);
1772 int parent_status = UNDER_LIMIT;
1774 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1776 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1779 if (parent_status == UNDER_LIMIT &&
1780 allocated <= sk_prot_mem_limits(sk, 0)) {
1781 sk_leave_memory_pressure(sk);
1785 /* Under pressure. (we or our parents) */
1786 if ((parent_status > SOFT_LIMIT) ||
1787 allocated > sk_prot_mem_limits(sk, 1))
1788 sk_enter_memory_pressure(sk);
1790 /* Over hard limit (we or our parents) */
1791 if ((parent_status == OVER_LIMIT) ||
1792 (allocated > sk_prot_mem_limits(sk, 2)))
1793 goto suppress_allocation;
1795 /* guarantee minimum buffer size under pressure */
1796 if (kind == SK_MEM_RECV) {
1797 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1800 } else { /* SK_MEM_SEND */
1801 if (sk->sk_type == SOCK_STREAM) {
1802 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1804 } else if (atomic_read(&sk->sk_wmem_alloc) <
1805 prot->sysctl_wmem[0])
1809 if (sk_has_memory_pressure(sk)) {
1812 if (!sk_under_memory_pressure(sk))
1814 alloc = sk_sockets_allocated_read_positive(sk);
1815 if (sk_prot_mem_limits(sk, 2) > alloc *
1816 sk_mem_pages(sk->sk_wmem_queued +
1817 atomic_read(&sk->sk_rmem_alloc) +
1818 sk->sk_forward_alloc))
1822 suppress_allocation:
1824 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1825 sk_stream_moderate_sndbuf(sk);
1827 /* Fail only if socket is _under_ its sndbuf.
1828 * In this case we cannot block, so that we have to fail.
1830 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1834 trace_sock_exceed_buf_limit(sk, prot, allocated);
1836 /* Alas. Undo changes. */
1837 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1839 sk_memory_allocated_sub(sk, amt);
1843 EXPORT_SYMBOL(__sk_mem_schedule);
1846 * __sk_reclaim - reclaim memory_allocated
1849 void __sk_mem_reclaim(struct sock *sk)
1851 sk_memory_allocated_sub(sk,
1852 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
1853 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1855 if (sk_under_memory_pressure(sk) &&
1856 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1857 sk_leave_memory_pressure(sk);
1859 EXPORT_SYMBOL(__sk_mem_reclaim);
1863 * Set of default routines for initialising struct proto_ops when
1864 * the protocol does not support a particular function. In certain
1865 * cases where it makes no sense for a protocol to have a "do nothing"
1866 * function, some default processing is provided.
1869 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1873 EXPORT_SYMBOL(sock_no_bind);
1875 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1880 EXPORT_SYMBOL(sock_no_connect);
1882 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1886 EXPORT_SYMBOL(sock_no_socketpair);
1888 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1892 EXPORT_SYMBOL(sock_no_accept);
1894 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1899 EXPORT_SYMBOL(sock_no_getname);
1901 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1905 EXPORT_SYMBOL(sock_no_poll);
1907 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1911 EXPORT_SYMBOL(sock_no_ioctl);
1913 int sock_no_listen(struct socket *sock, int backlog)
1917 EXPORT_SYMBOL(sock_no_listen);
1919 int sock_no_shutdown(struct socket *sock, int how)
1923 EXPORT_SYMBOL(sock_no_shutdown);
1925 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1926 char __user *optval, unsigned int optlen)
1930 EXPORT_SYMBOL(sock_no_setsockopt);
1932 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1933 char __user *optval, int __user *optlen)
1937 EXPORT_SYMBOL(sock_no_getsockopt);
1939 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1944 EXPORT_SYMBOL(sock_no_sendmsg);
1946 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1947 size_t len, int flags)
1951 EXPORT_SYMBOL(sock_no_recvmsg);
1953 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1955 /* Mirror missing mmap method error code */
1958 EXPORT_SYMBOL(sock_no_mmap);
1960 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1963 struct msghdr msg = {.msg_flags = flags};
1965 char *kaddr = kmap(page);
1966 iov.iov_base = kaddr + offset;
1968 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1972 EXPORT_SYMBOL(sock_no_sendpage);
1975 * Default Socket Callbacks
1978 static void sock_def_wakeup(struct sock *sk)
1980 struct socket_wq *wq;
1983 wq = rcu_dereference(sk->sk_wq);
1984 if (wq_has_sleeper(wq))
1985 wake_up_interruptible_all(&wq->wait);
1989 static void sock_def_error_report(struct sock *sk)
1991 struct socket_wq *wq;
1994 wq = rcu_dereference(sk->sk_wq);
1995 if (wq_has_sleeper(wq))
1996 wake_up_interruptible_poll(&wq->wait, POLLERR);
1997 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2001 static void sock_def_readable(struct sock *sk, int len)
2003 struct socket_wq *wq;
2006 wq = rcu_dereference(sk->sk_wq);
2007 if (wq_has_sleeper(wq))
2008 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2009 POLLRDNORM | POLLRDBAND);
2010 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2014 static void sock_def_write_space(struct sock *sk)
2016 struct socket_wq *wq;
2020 /* Do not wake up a writer until he can make "significant"
2023 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2024 wq = rcu_dereference(sk->sk_wq);
2025 if (wq_has_sleeper(wq))
2026 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2027 POLLWRNORM | POLLWRBAND);
2029 /* Should agree with poll, otherwise some programs break */
2030 if (sock_writeable(sk))
2031 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2037 static void sock_def_destruct(struct sock *sk)
2039 kfree(sk->sk_protinfo);
2042 void sk_send_sigurg(struct sock *sk)
2044 if (sk->sk_socket && sk->sk_socket->file)
2045 if (send_sigurg(&sk->sk_socket->file->f_owner))
2046 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2048 EXPORT_SYMBOL(sk_send_sigurg);
2050 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2051 unsigned long expires)
2053 if (!mod_timer(timer, expires))
2056 EXPORT_SYMBOL(sk_reset_timer);
2058 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2060 if (timer_pending(timer) && del_timer(timer))
2063 EXPORT_SYMBOL(sk_stop_timer);
2065 void sock_init_data(struct socket *sock, struct sock *sk)
2067 skb_queue_head_init(&sk->sk_receive_queue);
2068 skb_queue_head_init(&sk->sk_write_queue);
2069 skb_queue_head_init(&sk->sk_error_queue);
2070 #ifdef CONFIG_NET_DMA
2071 skb_queue_head_init(&sk->sk_async_wait_queue);
2074 sk->sk_send_head = NULL;
2076 init_timer(&sk->sk_timer);
2078 sk->sk_allocation = GFP_KERNEL;
2079 sk->sk_rcvbuf = sysctl_rmem_default;
2080 sk->sk_sndbuf = sysctl_wmem_default;
2081 sk->sk_state = TCP_CLOSE;
2082 sk_set_socket(sk, sock);
2084 sock_set_flag(sk, SOCK_ZAPPED);
2087 sk->sk_type = sock->type;
2088 sk->sk_wq = sock->wq;
2093 spin_lock_init(&sk->sk_dst_lock);
2094 rwlock_init(&sk->sk_callback_lock);
2095 lockdep_set_class_and_name(&sk->sk_callback_lock,
2096 af_callback_keys + sk->sk_family,
2097 af_family_clock_key_strings[sk->sk_family]);
2099 sk->sk_state_change = sock_def_wakeup;
2100 sk->sk_data_ready = sock_def_readable;
2101 sk->sk_write_space = sock_def_write_space;
2102 sk->sk_error_report = sock_def_error_report;
2103 sk->sk_destruct = sock_def_destruct;
2105 sk->sk_sndmsg_page = NULL;
2106 sk->sk_sndmsg_off = 0;
2107 sk->sk_peek_off = -1;
2109 sk->sk_peer_pid = NULL;
2110 sk->sk_peer_cred = NULL;
2111 sk->sk_write_pending = 0;
2112 sk->sk_rcvlowat = 1;
2113 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2114 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2116 sk->sk_stamp = ktime_set(-1L, 0);
2119 * Before updating sk_refcnt, we must commit prior changes to memory
2120 * (Documentation/RCU/rculist_nulls.txt for details)
2123 atomic_set(&sk->sk_refcnt, 1);
2124 atomic_set(&sk->sk_drops, 0);
2126 EXPORT_SYMBOL(sock_init_data);
2128 void lock_sock_nested(struct sock *sk, int subclass)
2131 spin_lock_bh(&sk->sk_lock.slock);
2132 if (sk->sk_lock.owned)
2134 sk->sk_lock.owned = 1;
2135 spin_unlock(&sk->sk_lock.slock);
2137 * The sk_lock has mutex_lock() semantics here:
2139 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2142 EXPORT_SYMBOL(lock_sock_nested);
2144 void release_sock(struct sock *sk)
2147 * The sk_lock has mutex_unlock() semantics:
2149 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2151 spin_lock_bh(&sk->sk_lock.slock);
2152 if (sk->sk_backlog.tail)
2154 sk->sk_lock.owned = 0;
2155 if (waitqueue_active(&sk->sk_lock.wq))
2156 wake_up(&sk->sk_lock.wq);
2157 spin_unlock_bh(&sk->sk_lock.slock);
2159 EXPORT_SYMBOL(release_sock);
2162 * lock_sock_fast - fast version of lock_sock
2165 * This version should be used for very small section, where process wont block
2166 * return false if fast path is taken
2167 * sk_lock.slock locked, owned = 0, BH disabled
2168 * return true if slow path is taken
2169 * sk_lock.slock unlocked, owned = 1, BH enabled
2171 bool lock_sock_fast(struct sock *sk)
2174 spin_lock_bh(&sk->sk_lock.slock);
2176 if (!sk->sk_lock.owned)
2178 * Note : We must disable BH
2183 sk->sk_lock.owned = 1;
2184 spin_unlock(&sk->sk_lock.slock);
2186 * The sk_lock has mutex_lock() semantics here:
2188 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2192 EXPORT_SYMBOL(lock_sock_fast);
2194 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2197 if (!sock_flag(sk, SOCK_TIMESTAMP))
2198 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2199 tv = ktime_to_timeval(sk->sk_stamp);
2200 if (tv.tv_sec == -1)
2202 if (tv.tv_sec == 0) {
2203 sk->sk_stamp = ktime_get_real();
2204 tv = ktime_to_timeval(sk->sk_stamp);
2206 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2208 EXPORT_SYMBOL(sock_get_timestamp);
2210 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2213 if (!sock_flag(sk, SOCK_TIMESTAMP))
2214 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2215 ts = ktime_to_timespec(sk->sk_stamp);
2216 if (ts.tv_sec == -1)
2218 if (ts.tv_sec == 0) {
2219 sk->sk_stamp = ktime_get_real();
2220 ts = ktime_to_timespec(sk->sk_stamp);
2222 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2224 EXPORT_SYMBOL(sock_get_timestampns);
2226 void sock_enable_timestamp(struct sock *sk, int flag)
2228 if (!sock_flag(sk, flag)) {
2229 unsigned long previous_flags = sk->sk_flags;
2231 sock_set_flag(sk, flag);
2233 * we just set one of the two flags which require net
2234 * time stamping, but time stamping might have been on
2235 * already because of the other one
2237 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2238 net_enable_timestamp();
2243 * Get a socket option on an socket.
2245 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2246 * asynchronous errors should be reported by getsockopt. We assume
2247 * this means if you specify SO_ERROR (otherwise whats the point of it).
2249 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2250 char __user *optval, int __user *optlen)
2252 struct sock *sk = sock->sk;
2254 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2256 EXPORT_SYMBOL(sock_common_getsockopt);
2258 #ifdef CONFIG_COMPAT
2259 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2260 char __user *optval, int __user *optlen)
2262 struct sock *sk = sock->sk;
2264 if (sk->sk_prot->compat_getsockopt != NULL)
2265 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2267 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2269 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2272 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2273 struct msghdr *msg, size_t size, int flags)
2275 struct sock *sk = sock->sk;
2279 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2280 flags & ~MSG_DONTWAIT, &addr_len);
2282 msg->msg_namelen = addr_len;
2285 EXPORT_SYMBOL(sock_common_recvmsg);
2288 * Set socket options on an inet socket.
2290 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2291 char __user *optval, unsigned int optlen)
2293 struct sock *sk = sock->sk;
2295 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2297 EXPORT_SYMBOL(sock_common_setsockopt);
2299 #ifdef CONFIG_COMPAT
2300 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2301 char __user *optval, unsigned int optlen)
2303 struct sock *sk = sock->sk;
2305 if (sk->sk_prot->compat_setsockopt != NULL)
2306 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2308 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2310 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2313 void sk_common_release(struct sock *sk)
2315 if (sk->sk_prot->destroy)
2316 sk->sk_prot->destroy(sk);
2319 * Observation: when sock_common_release is called, processes have
2320 * no access to socket. But net still has.
2321 * Step one, detach it from networking:
2323 * A. Remove from hash tables.
2326 sk->sk_prot->unhash(sk);
2329 * In this point socket cannot receive new packets, but it is possible
2330 * that some packets are in flight because some CPU runs receiver and
2331 * did hash table lookup before we unhashed socket. They will achieve
2332 * receive queue and will be purged by socket destructor.
2334 * Also we still have packets pending on receive queue and probably,
2335 * our own packets waiting in device queues. sock_destroy will drain
2336 * receive queue, but transmitted packets will delay socket destruction
2337 * until the last reference will be released.
2342 xfrm_sk_free_policy(sk);
2344 sk_refcnt_debug_release(sk);
2347 EXPORT_SYMBOL(sk_common_release);
2349 #ifdef CONFIG_PROC_FS
2350 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2352 int val[PROTO_INUSE_NR];
2355 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2357 #ifdef CONFIG_NET_NS
2358 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2360 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2362 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2364 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2366 int cpu, idx = prot->inuse_idx;
2369 for_each_possible_cpu(cpu)
2370 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2372 return res >= 0 ? res : 0;
2374 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2376 static int __net_init sock_inuse_init_net(struct net *net)
2378 net->core.inuse = alloc_percpu(struct prot_inuse);
2379 return net->core.inuse ? 0 : -ENOMEM;
2382 static void __net_exit sock_inuse_exit_net(struct net *net)
2384 free_percpu(net->core.inuse);
2387 static struct pernet_operations net_inuse_ops = {
2388 .init = sock_inuse_init_net,
2389 .exit = sock_inuse_exit_net,
2392 static __init int net_inuse_init(void)
2394 if (register_pernet_subsys(&net_inuse_ops))
2395 panic("Cannot initialize net inuse counters");
2400 core_initcall(net_inuse_init);
2402 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2404 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2406 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2408 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2410 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2412 int cpu, idx = prot->inuse_idx;
2415 for_each_possible_cpu(cpu)
2416 res += per_cpu(prot_inuse, cpu).val[idx];
2418 return res >= 0 ? res : 0;
2420 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2423 static void assign_proto_idx(struct proto *prot)
2425 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2427 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2428 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2432 set_bit(prot->inuse_idx, proto_inuse_idx);
2435 static void release_proto_idx(struct proto *prot)
2437 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2438 clear_bit(prot->inuse_idx, proto_inuse_idx);
2441 static inline void assign_proto_idx(struct proto *prot)
2445 static inline void release_proto_idx(struct proto *prot)
2450 int proto_register(struct proto *prot, int alloc_slab)
2453 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2454 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2457 if (prot->slab == NULL) {
2458 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2463 if (prot->rsk_prot != NULL) {
2464 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2465 if (prot->rsk_prot->slab_name == NULL)
2466 goto out_free_sock_slab;
2468 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2469 prot->rsk_prot->obj_size, 0,
2470 SLAB_HWCACHE_ALIGN, NULL);
2472 if (prot->rsk_prot->slab == NULL) {
2473 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2475 goto out_free_request_sock_slab_name;
2479 if (prot->twsk_prot != NULL) {
2480 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2482 if (prot->twsk_prot->twsk_slab_name == NULL)
2483 goto out_free_request_sock_slab;
2485 prot->twsk_prot->twsk_slab =
2486 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2487 prot->twsk_prot->twsk_obj_size,
2489 SLAB_HWCACHE_ALIGN |
2492 if (prot->twsk_prot->twsk_slab == NULL)
2493 goto out_free_timewait_sock_slab_name;
2497 mutex_lock(&proto_list_mutex);
2498 list_add(&prot->node, &proto_list);
2499 assign_proto_idx(prot);
2500 mutex_unlock(&proto_list_mutex);
2503 out_free_timewait_sock_slab_name:
2504 kfree(prot->twsk_prot->twsk_slab_name);
2505 out_free_request_sock_slab:
2506 if (prot->rsk_prot && prot->rsk_prot->slab) {
2507 kmem_cache_destroy(prot->rsk_prot->slab);
2508 prot->rsk_prot->slab = NULL;
2510 out_free_request_sock_slab_name:
2512 kfree(prot->rsk_prot->slab_name);
2514 kmem_cache_destroy(prot->slab);
2519 EXPORT_SYMBOL(proto_register);
2521 void proto_unregister(struct proto *prot)
2523 mutex_lock(&proto_list_mutex);
2524 release_proto_idx(prot);
2525 list_del(&prot->node);
2526 mutex_unlock(&proto_list_mutex);
2528 if (prot->slab != NULL) {
2529 kmem_cache_destroy(prot->slab);
2533 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2534 kmem_cache_destroy(prot->rsk_prot->slab);
2535 kfree(prot->rsk_prot->slab_name);
2536 prot->rsk_prot->slab = NULL;
2539 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2540 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2541 kfree(prot->twsk_prot->twsk_slab_name);
2542 prot->twsk_prot->twsk_slab = NULL;
2545 EXPORT_SYMBOL(proto_unregister);
2547 #ifdef CONFIG_PROC_FS
2548 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2549 __acquires(proto_list_mutex)
2551 mutex_lock(&proto_list_mutex);
2552 return seq_list_start_head(&proto_list, *pos);
2555 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2557 return seq_list_next(v, &proto_list, pos);
2560 static void proto_seq_stop(struct seq_file *seq, void *v)
2561 __releases(proto_list_mutex)
2563 mutex_unlock(&proto_list_mutex);
2566 static char proto_method_implemented(const void *method)
2568 return method == NULL ? 'n' : 'y';
2570 static long sock_prot_memory_allocated(struct proto *proto)
2572 return proto->memory_allocated != NULL ? proto_memory_allocated(proto): -1L;
2575 static char *sock_prot_memory_pressure(struct proto *proto)
2577 return proto->memory_pressure != NULL ?
2578 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2581 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2584 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2585 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2588 sock_prot_inuse_get(seq_file_net(seq), proto),
2589 sock_prot_memory_allocated(proto),
2590 sock_prot_memory_pressure(proto),
2592 proto->slab == NULL ? "no" : "yes",
2593 module_name(proto->owner),
2594 proto_method_implemented(proto->close),
2595 proto_method_implemented(proto->connect),
2596 proto_method_implemented(proto->disconnect),
2597 proto_method_implemented(proto->accept),
2598 proto_method_implemented(proto->ioctl),
2599 proto_method_implemented(proto->init),
2600 proto_method_implemented(proto->destroy),
2601 proto_method_implemented(proto->shutdown),
2602 proto_method_implemented(proto->setsockopt),
2603 proto_method_implemented(proto->getsockopt),
2604 proto_method_implemented(proto->sendmsg),
2605 proto_method_implemented(proto->recvmsg),
2606 proto_method_implemented(proto->sendpage),
2607 proto_method_implemented(proto->bind),
2608 proto_method_implemented(proto->backlog_rcv),
2609 proto_method_implemented(proto->hash),
2610 proto_method_implemented(proto->unhash),
2611 proto_method_implemented(proto->get_port),
2612 proto_method_implemented(proto->enter_memory_pressure));
2615 static int proto_seq_show(struct seq_file *seq, void *v)
2617 if (v == &proto_list)
2618 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2627 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2629 proto_seq_printf(seq, list_entry(v, struct proto, node));
2633 static const struct seq_operations proto_seq_ops = {
2634 .start = proto_seq_start,
2635 .next = proto_seq_next,
2636 .stop = proto_seq_stop,
2637 .show = proto_seq_show,
2640 static int proto_seq_open(struct inode *inode, struct file *file)
2642 return seq_open_net(inode, file, &proto_seq_ops,
2643 sizeof(struct seq_net_private));
2646 static const struct file_operations proto_seq_fops = {
2647 .owner = THIS_MODULE,
2648 .open = proto_seq_open,
2650 .llseek = seq_lseek,
2651 .release = seq_release_net,
2654 static __net_init int proto_init_net(struct net *net)
2656 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2662 static __net_exit void proto_exit_net(struct net *net)
2664 proc_net_remove(net, "protocols");
2668 static __net_initdata struct pernet_operations proto_net_ops = {
2669 .init = proto_init_net,
2670 .exit = proto_exit_net,
2673 static int __init proto_init(void)
2675 return register_pernet_subsys(&proto_net_ops);
2678 subsys_initcall(proto_init);
2680 #endif /* PROC_FS */