2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #include <linux/capability.h>
93 #include <linux/errno.h>
94 #include <linux/types.h>
95 #include <linux/socket.h>
97 #include <linux/kernel.h>
98 #include <linux/module.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/mm.h>
107 #include <linux/slab.h>
108 #include <linux/interrupt.h>
109 #include <linux/poll.h>
110 #include <linux/tcp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
115 #include <asm/uaccess.h>
116 #include <asm/system.h>
118 #include <linux/netdevice.h>
119 #include <net/protocol.h>
120 #include <linux/skbuff.h>
121 #include <net/net_namespace.h>
122 #include <net/request_sock.h>
123 #include <net/sock.h>
124 #include <linux/net_tstamp.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
127 #include <net/cls_cgroup.h>
128 #include <net/netprio_cgroup.h>
130 #include <linux/filter.h>
132 #include <trace/events/sock.h>
139 * Each address family might have different locking rules, so we have
140 * one slock key per address family:
142 static struct lock_class_key af_family_keys[AF_MAX];
143 static struct lock_class_key af_family_slock_keys[AF_MAX];
146 * Make lock validator output more readable. (we pre-construct these
147 * strings build-time, so that runtime initialization of socket
150 static const char *const af_family_key_strings[AF_MAX+1] = {
151 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
152 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
153 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
154 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
155 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
156 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
157 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
158 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
159 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
160 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
161 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
162 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
163 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
164 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
166 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
167 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
168 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
169 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
170 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
171 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
172 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
173 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
174 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
175 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
176 "slock-27" , "slock-28" , "slock-AF_CAN" ,
177 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
178 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
179 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
180 "slock-AF_NFC" , "slock-AF_MAX"
182 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
183 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
184 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
185 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
186 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
187 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
188 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
189 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
190 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
191 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
192 "clock-27" , "clock-28" , "clock-AF_CAN" ,
193 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
194 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
195 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
196 "clock-AF_NFC" , "clock-AF_MAX"
200 * sk_callback_lock locking rules are per-address-family,
201 * so split the lock classes by using a per-AF key:
203 static struct lock_class_key af_callback_keys[AF_MAX];
205 /* Take into consideration the size of the struct sk_buff overhead in the
206 * determination of these values, since that is non-constant across
207 * platforms. This makes socket queueing behavior and performance
208 * not depend upon such differences.
210 #define _SK_MEM_PACKETS 256
211 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
212 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
213 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
215 /* Run time adjustable parameters. */
216 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
217 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
218 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
219 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
221 /* Maximal space eaten by iovec or ancillary data plus some space */
222 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
223 EXPORT_SYMBOL(sysctl_optmem_max);
225 #if defined(CONFIG_CGROUPS)
226 #if !defined(CONFIG_NET_CLS_CGROUP)
227 int net_cls_subsys_id = -1;
228 EXPORT_SYMBOL_GPL(net_cls_subsys_id);
230 #if !defined(CONFIG_NETPRIO_CGROUP)
231 int net_prio_subsys_id = -1;
232 EXPORT_SYMBOL_GPL(net_prio_subsys_id);
236 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
240 if (optlen < sizeof(tv))
242 if (copy_from_user(&tv, optval, sizeof(tv)))
244 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
248 static int warned __read_mostly;
251 if (warned < 10 && net_ratelimit()) {
253 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
254 "tries to set negative timeout\n",
255 current->comm, task_pid_nr(current));
259 *timeo_p = MAX_SCHEDULE_TIMEOUT;
260 if (tv.tv_sec == 0 && tv.tv_usec == 0)
262 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
263 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
267 static void sock_warn_obsolete_bsdism(const char *name)
270 static char warncomm[TASK_COMM_LEN];
271 if (strcmp(warncomm, current->comm) && warned < 5) {
272 strcpy(warncomm, current->comm);
273 printk(KERN_WARNING "process `%s' is using obsolete "
274 "%s SO_BSDCOMPAT\n", warncomm, name);
279 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
281 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
283 if (sk->sk_flags & flags) {
284 sk->sk_flags &= ~flags;
285 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
286 net_disable_timestamp();
291 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
296 struct sk_buff_head *list = &sk->sk_receive_queue;
298 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
299 number of warnings when compiling with -W --ANK
301 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
302 (unsigned)sk->sk_rcvbuf) {
303 atomic_inc(&sk->sk_drops);
304 trace_sock_rcvqueue_full(sk, skb);
308 err = sk_filter(sk, skb);
312 if (!sk_rmem_schedule(sk, skb->truesize)) {
313 atomic_inc(&sk->sk_drops);
318 skb_set_owner_r(skb, sk);
320 /* Cache the SKB length before we tack it onto the receive
321 * queue. Once it is added it no longer belongs to us and
322 * may be freed by other threads of control pulling packets
327 /* we escape from rcu protected region, make sure we dont leak
332 spin_lock_irqsave(&list->lock, flags);
333 skb->dropcount = atomic_read(&sk->sk_drops);
334 __skb_queue_tail(list, skb);
335 spin_unlock_irqrestore(&list->lock, flags);
337 if (!sock_flag(sk, SOCK_DEAD))
338 sk->sk_data_ready(sk, skb_len);
341 EXPORT_SYMBOL(sock_queue_rcv_skb);
343 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
345 int rc = NET_RX_SUCCESS;
347 if (sk_filter(sk, skb))
348 goto discard_and_relse;
352 if (sk_rcvqueues_full(sk, skb)) {
353 atomic_inc(&sk->sk_drops);
354 goto discard_and_relse;
357 bh_lock_sock_nested(sk);
360 if (!sock_owned_by_user(sk)) {
362 * trylock + unlock semantics:
364 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
366 rc = sk_backlog_rcv(sk, skb);
368 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
369 } else if (sk_add_backlog(sk, skb)) {
371 atomic_inc(&sk->sk_drops);
372 goto discard_and_relse;
383 EXPORT_SYMBOL(sk_receive_skb);
385 void sk_reset_txq(struct sock *sk)
387 sk_tx_queue_clear(sk);
389 EXPORT_SYMBOL(sk_reset_txq);
391 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
393 struct dst_entry *dst = __sk_dst_get(sk);
395 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
396 sk_tx_queue_clear(sk);
397 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
404 EXPORT_SYMBOL(__sk_dst_check);
406 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
408 struct dst_entry *dst = sk_dst_get(sk);
410 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
418 EXPORT_SYMBOL(sk_dst_check);
420 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
422 int ret = -ENOPROTOOPT;
423 #ifdef CONFIG_NETDEVICES
424 struct net *net = sock_net(sk);
425 char devname[IFNAMSIZ];
430 if (!capable(CAP_NET_RAW))
437 /* Bind this socket to a particular device like "eth0",
438 * as specified in the passed interface name. If the
439 * name is "" or the option length is zero the socket
442 if (optlen > IFNAMSIZ - 1)
443 optlen = IFNAMSIZ - 1;
444 memset(devname, 0, sizeof(devname));
447 if (copy_from_user(devname, optval, optlen))
451 if (devname[0] != '\0') {
452 struct net_device *dev;
455 dev = dev_get_by_name_rcu(net, devname);
457 index = dev->ifindex;
465 sk->sk_bound_dev_if = index;
477 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
480 sock_set_flag(sk, bit);
482 sock_reset_flag(sk, bit);
486 * This is meant for all protocols to use and covers goings on
487 * at the socket level. Everything here is generic.
490 int sock_setsockopt(struct socket *sock, int level, int optname,
491 char __user *optval, unsigned int optlen)
493 struct sock *sk = sock->sk;
500 * Options without arguments
503 if (optname == SO_BINDTODEVICE)
504 return sock_bindtodevice(sk, optval, optlen);
506 if (optlen < sizeof(int))
509 if (get_user(val, (int __user *)optval))
512 valbool = val ? 1 : 0;
518 if (val && !capable(CAP_NET_ADMIN))
521 sock_valbool_flag(sk, SOCK_DBG, valbool);
524 sk->sk_reuse = valbool;
533 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
536 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
539 /* Don't error on this BSD doesn't and if you think
540 about it this is right. Otherwise apps have to
541 play 'guess the biggest size' games. RCVBUF/SNDBUF
542 are treated in BSD as hints */
544 if (val > sysctl_wmem_max)
545 val = sysctl_wmem_max;
547 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
548 if ((val * 2) < SOCK_MIN_SNDBUF)
549 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
551 sk->sk_sndbuf = val * 2;
554 * Wake up sending tasks if we
557 sk->sk_write_space(sk);
561 if (!capable(CAP_NET_ADMIN)) {
568 /* Don't error on this BSD doesn't and if you think
569 about it this is right. Otherwise apps have to
570 play 'guess the biggest size' games. RCVBUF/SNDBUF
571 are treated in BSD as hints */
573 if (val > sysctl_rmem_max)
574 val = sysctl_rmem_max;
576 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
578 * We double it on the way in to account for
579 * "struct sk_buff" etc. overhead. Applications
580 * assume that the SO_RCVBUF setting they make will
581 * allow that much actual data to be received on that
584 * Applications are unaware that "struct sk_buff" and
585 * other overheads allocate from the receive buffer
586 * during socket buffer allocation.
588 * And after considering the possible alternatives,
589 * returning the value we actually used in getsockopt
590 * is the most desirable behavior.
592 if ((val * 2) < SOCK_MIN_RCVBUF)
593 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
595 sk->sk_rcvbuf = val * 2;
599 if (!capable(CAP_NET_ADMIN)) {
607 if (sk->sk_protocol == IPPROTO_TCP)
608 tcp_set_keepalive(sk, valbool);
610 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
614 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
618 sk->sk_no_check = valbool;
622 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
623 sk->sk_priority = val;
629 if (optlen < sizeof(ling)) {
630 ret = -EINVAL; /* 1003.1g */
633 if (copy_from_user(&ling, optval, sizeof(ling))) {
638 sock_reset_flag(sk, SOCK_LINGER);
640 #if (BITS_PER_LONG == 32)
641 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
642 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
645 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
646 sock_set_flag(sk, SOCK_LINGER);
651 sock_warn_obsolete_bsdism("setsockopt");
656 set_bit(SOCK_PASSCRED, &sock->flags);
658 clear_bit(SOCK_PASSCRED, &sock->flags);
664 if (optname == SO_TIMESTAMP)
665 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
667 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
668 sock_set_flag(sk, SOCK_RCVTSTAMP);
669 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
671 sock_reset_flag(sk, SOCK_RCVTSTAMP);
672 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
676 case SO_TIMESTAMPING:
677 if (val & ~SOF_TIMESTAMPING_MASK) {
681 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
682 val & SOF_TIMESTAMPING_TX_HARDWARE);
683 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
684 val & SOF_TIMESTAMPING_TX_SOFTWARE);
685 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
686 val & SOF_TIMESTAMPING_RX_HARDWARE);
687 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
688 sock_enable_timestamp(sk,
689 SOCK_TIMESTAMPING_RX_SOFTWARE);
691 sock_disable_timestamp(sk,
692 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
693 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
694 val & SOF_TIMESTAMPING_SOFTWARE);
695 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
696 val & SOF_TIMESTAMPING_SYS_HARDWARE);
697 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
698 val & SOF_TIMESTAMPING_RAW_HARDWARE);
704 sk->sk_rcvlowat = val ? : 1;
708 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
712 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
715 case SO_ATTACH_FILTER:
717 if (optlen == sizeof(struct sock_fprog)) {
718 struct sock_fprog fprog;
721 if (copy_from_user(&fprog, optval, sizeof(fprog)))
724 ret = sk_attach_filter(&fprog, sk);
728 case SO_DETACH_FILTER:
729 ret = sk_detach_filter(sk);
734 set_bit(SOCK_PASSSEC, &sock->flags);
736 clear_bit(SOCK_PASSSEC, &sock->flags);
739 if (!capable(CAP_NET_ADMIN))
745 /* We implement the SO_SNDLOWAT etc to
746 not be settable (1003.1g 5.3) */
748 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
752 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
762 EXPORT_SYMBOL(sock_setsockopt);
765 void cred_to_ucred(struct pid *pid, const struct cred *cred,
768 ucred->pid = pid_vnr(pid);
769 ucred->uid = ucred->gid = -1;
771 struct user_namespace *current_ns = current_user_ns();
773 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
774 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
777 EXPORT_SYMBOL_GPL(cred_to_ucred);
779 int sock_getsockopt(struct socket *sock, int level, int optname,
780 char __user *optval, int __user *optlen)
782 struct sock *sk = sock->sk;
790 int lv = sizeof(int);
793 if (get_user(len, optlen))
798 memset(&v, 0, sizeof(v));
802 v.val = sock_flag(sk, SOCK_DBG);
806 v.val = sock_flag(sk, SOCK_LOCALROUTE);
810 v.val = !!sock_flag(sk, SOCK_BROADCAST);
814 v.val = sk->sk_sndbuf;
818 v.val = sk->sk_rcvbuf;
822 v.val = sk->sk_reuse;
826 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
834 v.val = sk->sk_protocol;
838 v.val = sk->sk_family;
842 v.val = -sock_error(sk);
844 v.val = xchg(&sk->sk_err_soft, 0);
848 v.val = !!sock_flag(sk, SOCK_URGINLINE);
852 v.val = sk->sk_no_check;
856 v.val = sk->sk_priority;
861 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
862 v.ling.l_linger = sk->sk_lingertime / HZ;
866 sock_warn_obsolete_bsdism("getsockopt");
870 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
871 !sock_flag(sk, SOCK_RCVTSTAMPNS);
875 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
878 case SO_TIMESTAMPING:
880 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
881 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
882 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
883 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
884 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
885 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
886 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
887 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
888 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
889 v.val |= SOF_TIMESTAMPING_SOFTWARE;
890 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
891 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
892 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
893 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
897 lv = sizeof(struct timeval);
898 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
902 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
903 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
908 lv = sizeof(struct timeval);
909 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
913 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
914 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
919 v.val = sk->sk_rcvlowat;
927 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
932 struct ucred peercred;
933 if (len > sizeof(peercred))
934 len = sizeof(peercred);
935 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
936 if (copy_to_user(optval, &peercred, len))
945 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
949 if (copy_to_user(optval, address, len))
954 /* Dubious BSD thing... Probably nobody even uses it, but
955 * the UNIX standard wants it for whatever reason... -DaveM
958 v.val = sk->sk_state == TCP_LISTEN;
962 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
966 return security_socket_getpeersec_stream(sock, optval, optlen, len);
973 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
977 v.val = !!sock_flag(sk, SOCK_WIFI_STATUS);
986 if (copy_to_user(optval, &v, len))
989 if (put_user(len, optlen))
995 * Initialize an sk_lock.
997 * (We also register the sk_lock with the lock validator.)
999 static inline void sock_lock_init(struct sock *sk)
1001 sock_lock_init_class_and_name(sk,
1002 af_family_slock_key_strings[sk->sk_family],
1003 af_family_slock_keys + sk->sk_family,
1004 af_family_key_strings[sk->sk_family],
1005 af_family_keys + sk->sk_family);
1009 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1010 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1011 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1013 static void sock_copy(struct sock *nsk, const struct sock *osk)
1015 #ifdef CONFIG_SECURITY_NETWORK
1016 void *sptr = nsk->sk_security;
1018 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1020 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1021 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1023 #ifdef CONFIG_SECURITY_NETWORK
1024 nsk->sk_security = sptr;
1025 security_sk_clone(osk, nsk);
1030 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1031 * un-modified. Special care is taken when initializing object to zero.
1033 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1035 if (offsetof(struct sock, sk_node.next) != 0)
1036 memset(sk, 0, offsetof(struct sock, sk_node.next));
1037 memset(&sk->sk_node.pprev, 0,
1038 size - offsetof(struct sock, sk_node.pprev));
1041 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1043 unsigned long nulls1, nulls2;
1045 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1046 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1047 if (nulls1 > nulls2)
1048 swap(nulls1, nulls2);
1051 memset((char *)sk, 0, nulls1);
1052 memset((char *)sk + nulls1 + sizeof(void *), 0,
1053 nulls2 - nulls1 - sizeof(void *));
1054 memset((char *)sk + nulls2 + sizeof(void *), 0,
1055 size - nulls2 - sizeof(void *));
1057 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1059 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1063 struct kmem_cache *slab;
1067 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1070 if (priority & __GFP_ZERO) {
1072 prot->clear_sk(sk, prot->obj_size);
1074 sk_prot_clear_nulls(sk, prot->obj_size);
1077 sk = kmalloc(prot->obj_size, priority);
1080 kmemcheck_annotate_bitfield(sk, flags);
1082 if (security_sk_alloc(sk, family, priority))
1085 if (!try_module_get(prot->owner))
1087 sk_tx_queue_clear(sk);
1093 security_sk_free(sk);
1096 kmem_cache_free(slab, sk);
1102 static void sk_prot_free(struct proto *prot, struct sock *sk)
1104 struct kmem_cache *slab;
1105 struct module *owner;
1107 owner = prot->owner;
1110 security_sk_free(sk);
1112 kmem_cache_free(slab, sk);
1118 #ifdef CONFIG_CGROUPS
1119 void sock_update_classid(struct sock *sk)
1123 rcu_read_lock(); /* doing current task, which cannot vanish. */
1124 classid = task_cls_classid(current);
1126 if (classid && classid != sk->sk_classid)
1127 sk->sk_classid = classid;
1129 EXPORT_SYMBOL(sock_update_classid);
1131 void sock_update_netprioidx(struct sock *sk)
1133 struct cgroup_netprio_state *state;
1137 state = task_netprio_state(current);
1138 sk->sk_cgrp_prioidx = state ? state->prioidx : 0;
1141 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1145 * sk_alloc - All socket objects are allocated here
1146 * @net: the applicable net namespace
1147 * @family: protocol family
1148 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1149 * @prot: struct proto associated with this new sock instance
1151 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1156 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1158 sk->sk_family = family;
1160 * See comment in struct sock definition to understand
1161 * why we need sk_prot_creator -acme
1163 sk->sk_prot = sk->sk_prot_creator = prot;
1165 sock_net_set(sk, get_net(net));
1166 atomic_set(&sk->sk_wmem_alloc, 1);
1168 sock_update_classid(sk);
1169 sock_update_netprioidx(sk);
1174 EXPORT_SYMBOL(sk_alloc);
1176 static void __sk_free(struct sock *sk)
1178 struct sk_filter *filter;
1180 if (sk->sk_destruct)
1181 sk->sk_destruct(sk);
1183 filter = rcu_dereference_check(sk->sk_filter,
1184 atomic_read(&sk->sk_wmem_alloc) == 0);
1186 sk_filter_uncharge(sk, filter);
1187 RCU_INIT_POINTER(sk->sk_filter, NULL);
1190 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1192 if (atomic_read(&sk->sk_omem_alloc))
1193 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1194 __func__, atomic_read(&sk->sk_omem_alloc));
1196 if (sk->sk_peer_cred)
1197 put_cred(sk->sk_peer_cred);
1198 put_pid(sk->sk_peer_pid);
1199 put_net(sock_net(sk));
1200 sk_prot_free(sk->sk_prot_creator, sk);
1203 void sk_free(struct sock *sk)
1206 * We subtract one from sk_wmem_alloc and can know if
1207 * some packets are still in some tx queue.
1208 * If not null, sock_wfree() will call __sk_free(sk) later
1210 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1213 EXPORT_SYMBOL(sk_free);
1216 * Last sock_put should drop reference to sk->sk_net. It has already
1217 * been dropped in sk_change_net. Taking reference to stopping namespace
1219 * Take reference to a socket to remove it from hash _alive_ and after that
1220 * destroy it in the context of init_net.
1222 void sk_release_kernel(struct sock *sk)
1224 if (sk == NULL || sk->sk_socket == NULL)
1228 sock_release(sk->sk_socket);
1229 release_net(sock_net(sk));
1230 sock_net_set(sk, get_net(&init_net));
1233 EXPORT_SYMBOL(sk_release_kernel);
1236 * sk_clone_lock - clone a socket, and lock its clone
1237 * @sk: the socket to clone
1238 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1240 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1242 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1246 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1247 if (newsk != NULL) {
1248 struct sk_filter *filter;
1250 sock_copy(newsk, sk);
1253 get_net(sock_net(newsk));
1254 sk_node_init(&newsk->sk_node);
1255 sock_lock_init(newsk);
1256 bh_lock_sock(newsk);
1257 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1258 newsk->sk_backlog.len = 0;
1260 atomic_set(&newsk->sk_rmem_alloc, 0);
1262 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1264 atomic_set(&newsk->sk_wmem_alloc, 1);
1265 atomic_set(&newsk->sk_omem_alloc, 0);
1266 skb_queue_head_init(&newsk->sk_receive_queue);
1267 skb_queue_head_init(&newsk->sk_write_queue);
1268 #ifdef CONFIG_NET_DMA
1269 skb_queue_head_init(&newsk->sk_async_wait_queue);
1272 spin_lock_init(&newsk->sk_dst_lock);
1273 rwlock_init(&newsk->sk_callback_lock);
1274 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1275 af_callback_keys + newsk->sk_family,
1276 af_family_clock_key_strings[newsk->sk_family]);
1278 newsk->sk_dst_cache = NULL;
1279 newsk->sk_wmem_queued = 0;
1280 newsk->sk_forward_alloc = 0;
1281 newsk->sk_send_head = NULL;
1282 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1284 sock_reset_flag(newsk, SOCK_DONE);
1285 skb_queue_head_init(&newsk->sk_error_queue);
1287 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1289 sk_filter_charge(newsk, filter);
1291 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1292 /* It is still raw copy of parent, so invalidate
1293 * destructor and make plain sk_free() */
1294 newsk->sk_destruct = NULL;
1295 bh_unlock_sock(newsk);
1302 newsk->sk_priority = 0;
1304 * Before updating sk_refcnt, we must commit prior changes to memory
1305 * (Documentation/RCU/rculist_nulls.txt for details)
1308 atomic_set(&newsk->sk_refcnt, 2);
1311 * Increment the counter in the same struct proto as the master
1312 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1313 * is the same as sk->sk_prot->socks, as this field was copied
1316 * This _changes_ the previous behaviour, where
1317 * tcp_create_openreq_child always was incrementing the
1318 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1319 * to be taken into account in all callers. -acme
1321 sk_refcnt_debug_inc(newsk);
1322 sk_set_socket(newsk, NULL);
1323 newsk->sk_wq = NULL;
1325 if (newsk->sk_prot->sockets_allocated)
1326 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1328 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1329 net_enable_timestamp();
1334 EXPORT_SYMBOL_GPL(sk_clone_lock);
1336 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1338 __sk_dst_set(sk, dst);
1339 sk->sk_route_caps = dst->dev->features;
1340 if (sk->sk_route_caps & NETIF_F_GSO)
1341 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1342 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1343 if (sk_can_gso(sk)) {
1344 if (dst->header_len) {
1345 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1347 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1348 sk->sk_gso_max_size = dst->dev->gso_max_size;
1352 EXPORT_SYMBOL_GPL(sk_setup_caps);
1354 void __init sk_init(void)
1356 if (totalram_pages <= 4096) {
1357 sysctl_wmem_max = 32767;
1358 sysctl_rmem_max = 32767;
1359 sysctl_wmem_default = 32767;
1360 sysctl_rmem_default = 32767;
1361 } else if (totalram_pages >= 131072) {
1362 sysctl_wmem_max = 131071;
1363 sysctl_rmem_max = 131071;
1368 * Simple resource managers for sockets.
1373 * Write buffer destructor automatically called from kfree_skb.
1375 void sock_wfree(struct sk_buff *skb)
1377 struct sock *sk = skb->sk;
1378 unsigned int len = skb->truesize;
1380 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1382 * Keep a reference on sk_wmem_alloc, this will be released
1383 * after sk_write_space() call
1385 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1386 sk->sk_write_space(sk);
1390 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1391 * could not do because of in-flight packets
1393 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1396 EXPORT_SYMBOL(sock_wfree);
1399 * Read buffer destructor automatically called from kfree_skb.
1401 void sock_rfree(struct sk_buff *skb)
1403 struct sock *sk = skb->sk;
1404 unsigned int len = skb->truesize;
1406 atomic_sub(len, &sk->sk_rmem_alloc);
1407 sk_mem_uncharge(sk, len);
1409 EXPORT_SYMBOL(sock_rfree);
1412 int sock_i_uid(struct sock *sk)
1416 read_lock_bh(&sk->sk_callback_lock);
1417 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1418 read_unlock_bh(&sk->sk_callback_lock);
1421 EXPORT_SYMBOL(sock_i_uid);
1423 unsigned long sock_i_ino(struct sock *sk)
1427 read_lock_bh(&sk->sk_callback_lock);
1428 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1429 read_unlock_bh(&sk->sk_callback_lock);
1432 EXPORT_SYMBOL(sock_i_ino);
1435 * Allocate a skb from the socket's send buffer.
1437 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1440 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1441 struct sk_buff *skb = alloc_skb(size, priority);
1443 skb_set_owner_w(skb, sk);
1449 EXPORT_SYMBOL(sock_wmalloc);
1452 * Allocate a skb from the socket's receive buffer.
1454 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1457 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1458 struct sk_buff *skb = alloc_skb(size, priority);
1460 skb_set_owner_r(skb, sk);
1468 * Allocate a memory block from the socket's option memory buffer.
1470 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1472 if ((unsigned)size <= sysctl_optmem_max &&
1473 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1475 /* First do the add, to avoid the race if kmalloc
1478 atomic_add(size, &sk->sk_omem_alloc);
1479 mem = kmalloc(size, priority);
1482 atomic_sub(size, &sk->sk_omem_alloc);
1486 EXPORT_SYMBOL(sock_kmalloc);
1489 * Free an option memory block.
1491 void sock_kfree_s(struct sock *sk, void *mem, int size)
1494 atomic_sub(size, &sk->sk_omem_alloc);
1496 EXPORT_SYMBOL(sock_kfree_s);
1498 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1499 I think, these locks should be removed for datagram sockets.
1501 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1505 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1509 if (signal_pending(current))
1511 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1512 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1513 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1515 if (sk->sk_shutdown & SEND_SHUTDOWN)
1519 timeo = schedule_timeout(timeo);
1521 finish_wait(sk_sleep(sk), &wait);
1527 * Generic send/receive buffer handlers
1530 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1531 unsigned long data_len, int noblock,
1534 struct sk_buff *skb;
1539 gfp_mask = sk->sk_allocation;
1540 if (gfp_mask & __GFP_WAIT)
1541 gfp_mask |= __GFP_REPEAT;
1543 timeo = sock_sndtimeo(sk, noblock);
1545 err = sock_error(sk);
1550 if (sk->sk_shutdown & SEND_SHUTDOWN)
1553 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1554 skb = alloc_skb(header_len, gfp_mask);
1559 /* No pages, we're done... */
1563 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1564 skb->truesize += data_len;
1565 skb_shinfo(skb)->nr_frags = npages;
1566 for (i = 0; i < npages; i++) {
1569 page = alloc_pages(sk->sk_allocation, 0);
1572 skb_shinfo(skb)->nr_frags = i;
1577 __skb_fill_page_desc(skb, i,
1579 (data_len >= PAGE_SIZE ?
1582 data_len -= PAGE_SIZE;
1585 /* Full success... */
1591 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1592 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1596 if (signal_pending(current))
1598 timeo = sock_wait_for_wmem(sk, timeo);
1601 skb_set_owner_w(skb, sk);
1605 err = sock_intr_errno(timeo);
1610 EXPORT_SYMBOL(sock_alloc_send_pskb);
1612 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1613 int noblock, int *errcode)
1615 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1617 EXPORT_SYMBOL(sock_alloc_send_skb);
1619 static void __lock_sock(struct sock *sk)
1620 __releases(&sk->sk_lock.slock)
1621 __acquires(&sk->sk_lock.slock)
1626 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1627 TASK_UNINTERRUPTIBLE);
1628 spin_unlock_bh(&sk->sk_lock.slock);
1630 spin_lock_bh(&sk->sk_lock.slock);
1631 if (!sock_owned_by_user(sk))
1634 finish_wait(&sk->sk_lock.wq, &wait);
1637 static void __release_sock(struct sock *sk)
1638 __releases(&sk->sk_lock.slock)
1639 __acquires(&sk->sk_lock.slock)
1641 struct sk_buff *skb = sk->sk_backlog.head;
1644 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1648 struct sk_buff *next = skb->next;
1650 WARN_ON_ONCE(skb_dst_is_noref(skb));
1652 sk_backlog_rcv(sk, skb);
1655 * We are in process context here with softirqs
1656 * disabled, use cond_resched_softirq() to preempt.
1657 * This is safe to do because we've taken the backlog
1660 cond_resched_softirq();
1663 } while (skb != NULL);
1666 } while ((skb = sk->sk_backlog.head) != NULL);
1669 * Doing the zeroing here guarantee we can not loop forever
1670 * while a wild producer attempts to flood us.
1672 sk->sk_backlog.len = 0;
1676 * sk_wait_data - wait for data to arrive at sk_receive_queue
1677 * @sk: sock to wait on
1678 * @timeo: for how long
1680 * Now socket state including sk->sk_err is changed only under lock,
1681 * hence we may omit checks after joining wait queue.
1682 * We check receive queue before schedule() only as optimization;
1683 * it is very likely that release_sock() added new data.
1685 int sk_wait_data(struct sock *sk, long *timeo)
1690 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1691 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1692 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1693 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1694 finish_wait(sk_sleep(sk), &wait);
1697 EXPORT_SYMBOL(sk_wait_data);
1700 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1702 * @size: memory size to allocate
1703 * @kind: allocation type
1705 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1706 * rmem allocation. This function assumes that protocols which have
1707 * memory_pressure use sk_wmem_queued as write buffer accounting.
1709 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1711 struct proto *prot = sk->sk_prot;
1712 int amt = sk_mem_pages(size);
1715 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1716 allocated = atomic_long_add_return(amt, prot->memory_allocated);
1719 if (allocated <= prot->sysctl_mem[0]) {
1720 if (prot->memory_pressure && *prot->memory_pressure)
1721 *prot->memory_pressure = 0;
1725 /* Under pressure. */
1726 if (allocated > prot->sysctl_mem[1])
1727 if (prot->enter_memory_pressure)
1728 prot->enter_memory_pressure(sk);
1730 /* Over hard limit. */
1731 if (allocated > prot->sysctl_mem[2])
1732 goto suppress_allocation;
1734 /* guarantee minimum buffer size under pressure */
1735 if (kind == SK_MEM_RECV) {
1736 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1738 } else { /* SK_MEM_SEND */
1739 if (sk->sk_type == SOCK_STREAM) {
1740 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1742 } else if (atomic_read(&sk->sk_wmem_alloc) <
1743 prot->sysctl_wmem[0])
1747 if (prot->memory_pressure) {
1750 if (!*prot->memory_pressure)
1752 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1753 if (prot->sysctl_mem[2] > alloc *
1754 sk_mem_pages(sk->sk_wmem_queued +
1755 atomic_read(&sk->sk_rmem_alloc) +
1756 sk->sk_forward_alloc))
1760 suppress_allocation:
1762 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1763 sk_stream_moderate_sndbuf(sk);
1765 /* Fail only if socket is _under_ its sndbuf.
1766 * In this case we cannot block, so that we have to fail.
1768 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1772 trace_sock_exceed_buf_limit(sk, prot, allocated);
1774 /* Alas. Undo changes. */
1775 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1776 atomic_long_sub(amt, prot->memory_allocated);
1779 EXPORT_SYMBOL(__sk_mem_schedule);
1782 * __sk_reclaim - reclaim memory_allocated
1785 void __sk_mem_reclaim(struct sock *sk)
1787 struct proto *prot = sk->sk_prot;
1789 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1790 prot->memory_allocated);
1791 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1793 if (prot->memory_pressure && *prot->memory_pressure &&
1794 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1795 *prot->memory_pressure = 0;
1797 EXPORT_SYMBOL(__sk_mem_reclaim);
1801 * Set of default routines for initialising struct proto_ops when
1802 * the protocol does not support a particular function. In certain
1803 * cases where it makes no sense for a protocol to have a "do nothing"
1804 * function, some default processing is provided.
1807 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1811 EXPORT_SYMBOL(sock_no_bind);
1813 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1818 EXPORT_SYMBOL(sock_no_connect);
1820 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1824 EXPORT_SYMBOL(sock_no_socketpair);
1826 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1830 EXPORT_SYMBOL(sock_no_accept);
1832 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1837 EXPORT_SYMBOL(sock_no_getname);
1839 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1843 EXPORT_SYMBOL(sock_no_poll);
1845 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1849 EXPORT_SYMBOL(sock_no_ioctl);
1851 int sock_no_listen(struct socket *sock, int backlog)
1855 EXPORT_SYMBOL(sock_no_listen);
1857 int sock_no_shutdown(struct socket *sock, int how)
1861 EXPORT_SYMBOL(sock_no_shutdown);
1863 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1864 char __user *optval, unsigned int optlen)
1868 EXPORT_SYMBOL(sock_no_setsockopt);
1870 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1871 char __user *optval, int __user *optlen)
1875 EXPORT_SYMBOL(sock_no_getsockopt);
1877 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1882 EXPORT_SYMBOL(sock_no_sendmsg);
1884 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1885 size_t len, int flags)
1889 EXPORT_SYMBOL(sock_no_recvmsg);
1891 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1893 /* Mirror missing mmap method error code */
1896 EXPORT_SYMBOL(sock_no_mmap);
1898 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1901 struct msghdr msg = {.msg_flags = flags};
1903 char *kaddr = kmap(page);
1904 iov.iov_base = kaddr + offset;
1906 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1910 EXPORT_SYMBOL(sock_no_sendpage);
1913 * Default Socket Callbacks
1916 static void sock_def_wakeup(struct sock *sk)
1918 struct socket_wq *wq;
1921 wq = rcu_dereference(sk->sk_wq);
1922 if (wq_has_sleeper(wq))
1923 wake_up_interruptible_all(&wq->wait);
1927 static void sock_def_error_report(struct sock *sk)
1929 struct socket_wq *wq;
1932 wq = rcu_dereference(sk->sk_wq);
1933 if (wq_has_sleeper(wq))
1934 wake_up_interruptible_poll(&wq->wait, POLLERR);
1935 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1939 static void sock_def_readable(struct sock *sk, int len)
1941 struct socket_wq *wq;
1944 wq = rcu_dereference(sk->sk_wq);
1945 if (wq_has_sleeper(wq))
1946 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1947 POLLRDNORM | POLLRDBAND);
1948 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1952 static void sock_def_write_space(struct sock *sk)
1954 struct socket_wq *wq;
1958 /* Do not wake up a writer until he can make "significant"
1961 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1962 wq = rcu_dereference(sk->sk_wq);
1963 if (wq_has_sleeper(wq))
1964 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1965 POLLWRNORM | POLLWRBAND);
1967 /* Should agree with poll, otherwise some programs break */
1968 if (sock_writeable(sk))
1969 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1975 static void sock_def_destruct(struct sock *sk)
1977 kfree(sk->sk_protinfo);
1980 void sk_send_sigurg(struct sock *sk)
1982 if (sk->sk_socket && sk->sk_socket->file)
1983 if (send_sigurg(&sk->sk_socket->file->f_owner))
1984 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1986 EXPORT_SYMBOL(sk_send_sigurg);
1988 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1989 unsigned long expires)
1991 if (!mod_timer(timer, expires))
1994 EXPORT_SYMBOL(sk_reset_timer);
1996 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1998 if (timer_pending(timer) && del_timer(timer))
2001 EXPORT_SYMBOL(sk_stop_timer);
2003 void sock_init_data(struct socket *sock, struct sock *sk)
2005 skb_queue_head_init(&sk->sk_receive_queue);
2006 skb_queue_head_init(&sk->sk_write_queue);
2007 skb_queue_head_init(&sk->sk_error_queue);
2008 #ifdef CONFIG_NET_DMA
2009 skb_queue_head_init(&sk->sk_async_wait_queue);
2012 sk->sk_send_head = NULL;
2014 init_timer(&sk->sk_timer);
2016 sk->sk_allocation = GFP_KERNEL;
2017 sk->sk_rcvbuf = sysctl_rmem_default;
2018 sk->sk_sndbuf = sysctl_wmem_default;
2019 sk->sk_state = TCP_CLOSE;
2020 sk_set_socket(sk, sock);
2022 sock_set_flag(sk, SOCK_ZAPPED);
2025 sk->sk_type = sock->type;
2026 sk->sk_wq = sock->wq;
2031 spin_lock_init(&sk->sk_dst_lock);
2032 rwlock_init(&sk->sk_callback_lock);
2033 lockdep_set_class_and_name(&sk->sk_callback_lock,
2034 af_callback_keys + sk->sk_family,
2035 af_family_clock_key_strings[sk->sk_family]);
2037 sk->sk_state_change = sock_def_wakeup;
2038 sk->sk_data_ready = sock_def_readable;
2039 sk->sk_write_space = sock_def_write_space;
2040 sk->sk_error_report = sock_def_error_report;
2041 sk->sk_destruct = sock_def_destruct;
2043 sk->sk_sndmsg_page = NULL;
2044 sk->sk_sndmsg_off = 0;
2046 sk->sk_peer_pid = NULL;
2047 sk->sk_peer_cred = NULL;
2048 sk->sk_write_pending = 0;
2049 sk->sk_rcvlowat = 1;
2050 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2051 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2053 sk->sk_stamp = ktime_set(-1L, 0);
2056 * Before updating sk_refcnt, we must commit prior changes to memory
2057 * (Documentation/RCU/rculist_nulls.txt for details)
2060 atomic_set(&sk->sk_refcnt, 1);
2061 atomic_set(&sk->sk_drops, 0);
2063 EXPORT_SYMBOL(sock_init_data);
2065 void lock_sock_nested(struct sock *sk, int subclass)
2068 spin_lock_bh(&sk->sk_lock.slock);
2069 if (sk->sk_lock.owned)
2071 sk->sk_lock.owned = 1;
2072 spin_unlock(&sk->sk_lock.slock);
2074 * The sk_lock has mutex_lock() semantics here:
2076 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2079 EXPORT_SYMBOL(lock_sock_nested);
2081 void release_sock(struct sock *sk)
2084 * The sk_lock has mutex_unlock() semantics:
2086 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2088 spin_lock_bh(&sk->sk_lock.slock);
2089 if (sk->sk_backlog.tail)
2091 sk->sk_lock.owned = 0;
2092 if (waitqueue_active(&sk->sk_lock.wq))
2093 wake_up(&sk->sk_lock.wq);
2094 spin_unlock_bh(&sk->sk_lock.slock);
2096 EXPORT_SYMBOL(release_sock);
2099 * lock_sock_fast - fast version of lock_sock
2102 * This version should be used for very small section, where process wont block
2103 * return false if fast path is taken
2104 * sk_lock.slock locked, owned = 0, BH disabled
2105 * return true if slow path is taken
2106 * sk_lock.slock unlocked, owned = 1, BH enabled
2108 bool lock_sock_fast(struct sock *sk)
2111 spin_lock_bh(&sk->sk_lock.slock);
2113 if (!sk->sk_lock.owned)
2115 * Note : We must disable BH
2120 sk->sk_lock.owned = 1;
2121 spin_unlock(&sk->sk_lock.slock);
2123 * The sk_lock has mutex_lock() semantics here:
2125 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2129 EXPORT_SYMBOL(lock_sock_fast);
2131 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2134 if (!sock_flag(sk, SOCK_TIMESTAMP))
2135 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2136 tv = ktime_to_timeval(sk->sk_stamp);
2137 if (tv.tv_sec == -1)
2139 if (tv.tv_sec == 0) {
2140 sk->sk_stamp = ktime_get_real();
2141 tv = ktime_to_timeval(sk->sk_stamp);
2143 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2145 EXPORT_SYMBOL(sock_get_timestamp);
2147 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2150 if (!sock_flag(sk, SOCK_TIMESTAMP))
2151 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2152 ts = ktime_to_timespec(sk->sk_stamp);
2153 if (ts.tv_sec == -1)
2155 if (ts.tv_sec == 0) {
2156 sk->sk_stamp = ktime_get_real();
2157 ts = ktime_to_timespec(sk->sk_stamp);
2159 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2161 EXPORT_SYMBOL(sock_get_timestampns);
2163 void sock_enable_timestamp(struct sock *sk, int flag)
2165 if (!sock_flag(sk, flag)) {
2166 unsigned long previous_flags = sk->sk_flags;
2168 sock_set_flag(sk, flag);
2170 * we just set one of the two flags which require net
2171 * time stamping, but time stamping might have been on
2172 * already because of the other one
2174 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2175 net_enable_timestamp();
2180 * Get a socket option on an socket.
2182 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2183 * asynchronous errors should be reported by getsockopt. We assume
2184 * this means if you specify SO_ERROR (otherwise whats the point of it).
2186 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2187 char __user *optval, int __user *optlen)
2189 struct sock *sk = sock->sk;
2191 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2193 EXPORT_SYMBOL(sock_common_getsockopt);
2195 #ifdef CONFIG_COMPAT
2196 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2197 char __user *optval, int __user *optlen)
2199 struct sock *sk = sock->sk;
2201 if (sk->sk_prot->compat_getsockopt != NULL)
2202 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2204 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2206 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2209 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2210 struct msghdr *msg, size_t size, int flags)
2212 struct sock *sk = sock->sk;
2216 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2217 flags & ~MSG_DONTWAIT, &addr_len);
2219 msg->msg_namelen = addr_len;
2222 EXPORT_SYMBOL(sock_common_recvmsg);
2225 * Set socket options on an inet socket.
2227 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2228 char __user *optval, unsigned int optlen)
2230 struct sock *sk = sock->sk;
2232 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2234 EXPORT_SYMBOL(sock_common_setsockopt);
2236 #ifdef CONFIG_COMPAT
2237 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2238 char __user *optval, unsigned int optlen)
2240 struct sock *sk = sock->sk;
2242 if (sk->sk_prot->compat_setsockopt != NULL)
2243 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2245 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2247 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2250 void sk_common_release(struct sock *sk)
2252 if (sk->sk_prot->destroy)
2253 sk->sk_prot->destroy(sk);
2256 * Observation: when sock_common_release is called, processes have
2257 * no access to socket. But net still has.
2258 * Step one, detach it from networking:
2260 * A. Remove from hash tables.
2263 sk->sk_prot->unhash(sk);
2266 * In this point socket cannot receive new packets, but it is possible
2267 * that some packets are in flight because some CPU runs receiver and
2268 * did hash table lookup before we unhashed socket. They will achieve
2269 * receive queue and will be purged by socket destructor.
2271 * Also we still have packets pending on receive queue and probably,
2272 * our own packets waiting in device queues. sock_destroy will drain
2273 * receive queue, but transmitted packets will delay socket destruction
2274 * until the last reference will be released.
2279 xfrm_sk_free_policy(sk);
2281 sk_refcnt_debug_release(sk);
2284 EXPORT_SYMBOL(sk_common_release);
2286 static DEFINE_RWLOCK(proto_list_lock);
2287 static LIST_HEAD(proto_list);
2289 #ifdef CONFIG_PROC_FS
2290 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2292 int val[PROTO_INUSE_NR];
2295 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2297 #ifdef CONFIG_NET_NS
2298 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2300 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2302 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2304 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2306 int cpu, idx = prot->inuse_idx;
2309 for_each_possible_cpu(cpu)
2310 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2312 return res >= 0 ? res : 0;
2314 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2316 static int __net_init sock_inuse_init_net(struct net *net)
2318 net->core.inuse = alloc_percpu(struct prot_inuse);
2319 return net->core.inuse ? 0 : -ENOMEM;
2322 static void __net_exit sock_inuse_exit_net(struct net *net)
2324 free_percpu(net->core.inuse);
2327 static struct pernet_operations net_inuse_ops = {
2328 .init = sock_inuse_init_net,
2329 .exit = sock_inuse_exit_net,
2332 static __init int net_inuse_init(void)
2334 if (register_pernet_subsys(&net_inuse_ops))
2335 panic("Cannot initialize net inuse counters");
2340 core_initcall(net_inuse_init);
2342 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2344 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2346 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2348 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2350 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2352 int cpu, idx = prot->inuse_idx;
2355 for_each_possible_cpu(cpu)
2356 res += per_cpu(prot_inuse, cpu).val[idx];
2358 return res >= 0 ? res : 0;
2360 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2363 static void assign_proto_idx(struct proto *prot)
2365 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2367 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2368 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2372 set_bit(prot->inuse_idx, proto_inuse_idx);
2375 static void release_proto_idx(struct proto *prot)
2377 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2378 clear_bit(prot->inuse_idx, proto_inuse_idx);
2381 static inline void assign_proto_idx(struct proto *prot)
2385 static inline void release_proto_idx(struct proto *prot)
2390 int proto_register(struct proto *prot, int alloc_slab)
2393 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2394 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2397 if (prot->slab == NULL) {
2398 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2403 if (prot->rsk_prot != NULL) {
2404 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2405 if (prot->rsk_prot->slab_name == NULL)
2406 goto out_free_sock_slab;
2408 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2409 prot->rsk_prot->obj_size, 0,
2410 SLAB_HWCACHE_ALIGN, NULL);
2412 if (prot->rsk_prot->slab == NULL) {
2413 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2415 goto out_free_request_sock_slab_name;
2419 if (prot->twsk_prot != NULL) {
2420 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2422 if (prot->twsk_prot->twsk_slab_name == NULL)
2423 goto out_free_request_sock_slab;
2425 prot->twsk_prot->twsk_slab =
2426 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2427 prot->twsk_prot->twsk_obj_size,
2429 SLAB_HWCACHE_ALIGN |
2432 if (prot->twsk_prot->twsk_slab == NULL)
2433 goto out_free_timewait_sock_slab_name;
2437 write_lock(&proto_list_lock);
2438 list_add(&prot->node, &proto_list);
2439 assign_proto_idx(prot);
2440 write_unlock(&proto_list_lock);
2443 out_free_timewait_sock_slab_name:
2444 kfree(prot->twsk_prot->twsk_slab_name);
2445 out_free_request_sock_slab:
2446 if (prot->rsk_prot && prot->rsk_prot->slab) {
2447 kmem_cache_destroy(prot->rsk_prot->slab);
2448 prot->rsk_prot->slab = NULL;
2450 out_free_request_sock_slab_name:
2452 kfree(prot->rsk_prot->slab_name);
2454 kmem_cache_destroy(prot->slab);
2459 EXPORT_SYMBOL(proto_register);
2461 void proto_unregister(struct proto *prot)
2463 write_lock(&proto_list_lock);
2464 release_proto_idx(prot);
2465 list_del(&prot->node);
2466 write_unlock(&proto_list_lock);
2468 if (prot->slab != NULL) {
2469 kmem_cache_destroy(prot->slab);
2473 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2474 kmem_cache_destroy(prot->rsk_prot->slab);
2475 kfree(prot->rsk_prot->slab_name);
2476 prot->rsk_prot->slab = NULL;
2479 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2480 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2481 kfree(prot->twsk_prot->twsk_slab_name);
2482 prot->twsk_prot->twsk_slab = NULL;
2485 EXPORT_SYMBOL(proto_unregister);
2487 #ifdef CONFIG_PROC_FS
2488 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2489 __acquires(proto_list_lock)
2491 read_lock(&proto_list_lock);
2492 return seq_list_start_head(&proto_list, *pos);
2495 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2497 return seq_list_next(v, &proto_list, pos);
2500 static void proto_seq_stop(struct seq_file *seq, void *v)
2501 __releases(proto_list_lock)
2503 read_unlock(&proto_list_lock);
2506 static char proto_method_implemented(const void *method)
2508 return method == NULL ? 'n' : 'y';
2511 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2513 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2514 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2517 sock_prot_inuse_get(seq_file_net(seq), proto),
2518 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2519 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2521 proto->slab == NULL ? "no" : "yes",
2522 module_name(proto->owner),
2523 proto_method_implemented(proto->close),
2524 proto_method_implemented(proto->connect),
2525 proto_method_implemented(proto->disconnect),
2526 proto_method_implemented(proto->accept),
2527 proto_method_implemented(proto->ioctl),
2528 proto_method_implemented(proto->init),
2529 proto_method_implemented(proto->destroy),
2530 proto_method_implemented(proto->shutdown),
2531 proto_method_implemented(proto->setsockopt),
2532 proto_method_implemented(proto->getsockopt),
2533 proto_method_implemented(proto->sendmsg),
2534 proto_method_implemented(proto->recvmsg),
2535 proto_method_implemented(proto->sendpage),
2536 proto_method_implemented(proto->bind),
2537 proto_method_implemented(proto->backlog_rcv),
2538 proto_method_implemented(proto->hash),
2539 proto_method_implemented(proto->unhash),
2540 proto_method_implemented(proto->get_port),
2541 proto_method_implemented(proto->enter_memory_pressure));
2544 static int proto_seq_show(struct seq_file *seq, void *v)
2546 if (v == &proto_list)
2547 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2556 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2558 proto_seq_printf(seq, list_entry(v, struct proto, node));
2562 static const struct seq_operations proto_seq_ops = {
2563 .start = proto_seq_start,
2564 .next = proto_seq_next,
2565 .stop = proto_seq_stop,
2566 .show = proto_seq_show,
2569 static int proto_seq_open(struct inode *inode, struct file *file)
2571 return seq_open_net(inode, file, &proto_seq_ops,
2572 sizeof(struct seq_net_private));
2575 static const struct file_operations proto_seq_fops = {
2576 .owner = THIS_MODULE,
2577 .open = proto_seq_open,
2579 .llseek = seq_lseek,
2580 .release = seq_release_net,
2583 static __net_init int proto_init_net(struct net *net)
2585 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2591 static __net_exit void proto_exit_net(struct net *net)
2593 proc_net_remove(net, "protocols");
2597 static __net_initdata struct pernet_operations proto_net_ops = {
2598 .init = proto_init_net,
2599 .exit = proto_exit_net,
2602 static int __init proto_init(void)
2604 return register_pernet_subsys(&proto_net_ops);
2607 subsys_initcall(proto_init);
2609 #endif /* PROC_FS */