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>
116 #include <asm/uaccess.h>
117 #include <asm/system.h>
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/net_namespace.h>
123 #include <net/request_sock.h>
124 #include <net/sock.h>
125 #include <linux/net_tstamp.h>
126 #include <net/xfrm.h>
127 #include <linux/ipsec.h>
128 #include <net/cls_cgroup.h>
129 #include <net/netprio_cgroup.h>
131 #include <linux/filter.h>
133 #include <trace/events/sock.h>
139 static DEFINE_MUTEX(proto_list_mutex);
140 static LIST_HEAD(proto_list);
142 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
143 int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss)
148 mutex_lock(&proto_list_mutex);
149 list_for_each_entry(proto, &proto_list, node) {
150 if (proto->init_cgroup) {
151 ret = proto->init_cgroup(cgrp, ss);
157 mutex_unlock(&proto_list_mutex);
160 list_for_each_entry_continue_reverse(proto, &proto_list, node)
161 if (proto->destroy_cgroup)
162 proto->destroy_cgroup(cgrp, ss);
163 mutex_unlock(&proto_list_mutex);
167 void mem_cgroup_sockets_destroy(struct cgroup *cgrp, struct cgroup_subsys *ss)
171 mutex_lock(&proto_list_mutex);
172 list_for_each_entry_reverse(proto, &proto_list, node)
173 if (proto->destroy_cgroup)
174 proto->destroy_cgroup(cgrp, ss);
175 mutex_unlock(&proto_list_mutex);
180 * Each address family might have different locking rules, so we have
181 * one slock key per address family:
183 static struct lock_class_key af_family_keys[AF_MAX];
184 static struct lock_class_key af_family_slock_keys[AF_MAX];
186 struct jump_label_key memcg_socket_limit_enabled;
187 EXPORT_SYMBOL(memcg_socket_limit_enabled);
190 * Make lock validator output more readable. (we pre-construct these
191 * strings build-time, so that runtime initialization of socket
194 static const char *const af_family_key_strings[AF_MAX+1] = {
195 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
196 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
197 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
198 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
199 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
200 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
201 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
202 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
203 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
204 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
205 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
206 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
207 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
208 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
210 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
211 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
212 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
213 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
214 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
215 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
216 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
217 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
218 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
219 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
220 "slock-27" , "slock-28" , "slock-AF_CAN" ,
221 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
222 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
223 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
224 "slock-AF_NFC" , "slock-AF_MAX"
226 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
227 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
228 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
229 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
230 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
231 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
232 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
233 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
234 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
235 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
236 "clock-27" , "clock-28" , "clock-AF_CAN" ,
237 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
238 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
239 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
240 "clock-AF_NFC" , "clock-AF_MAX"
244 * sk_callback_lock locking rules are per-address-family,
245 * so split the lock classes by using a per-AF key:
247 static struct lock_class_key af_callback_keys[AF_MAX];
249 /* Take into consideration the size of the struct sk_buff overhead in the
250 * determination of these values, since that is non-constant across
251 * platforms. This makes socket queueing behavior and performance
252 * not depend upon such differences.
254 #define _SK_MEM_PACKETS 256
255 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
256 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
257 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
259 /* Run time adjustable parameters. */
260 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
261 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
262 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
263 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
265 /* Maximal space eaten by iovec or ancillary data plus some space */
266 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
267 EXPORT_SYMBOL(sysctl_optmem_max);
269 #if defined(CONFIG_CGROUPS)
270 #if !defined(CONFIG_NET_CLS_CGROUP)
271 int net_cls_subsys_id = -1;
272 EXPORT_SYMBOL_GPL(net_cls_subsys_id);
274 #if !defined(CONFIG_NETPRIO_CGROUP)
275 int net_prio_subsys_id = -1;
276 EXPORT_SYMBOL_GPL(net_prio_subsys_id);
280 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
284 if (optlen < sizeof(tv))
286 if (copy_from_user(&tv, optval, sizeof(tv)))
288 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
292 static int warned __read_mostly;
295 if (warned < 10 && net_ratelimit()) {
297 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
298 "tries to set negative timeout\n",
299 current->comm, task_pid_nr(current));
303 *timeo_p = MAX_SCHEDULE_TIMEOUT;
304 if (tv.tv_sec == 0 && tv.tv_usec == 0)
306 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
307 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
311 static void sock_warn_obsolete_bsdism(const char *name)
314 static char warncomm[TASK_COMM_LEN];
315 if (strcmp(warncomm, current->comm) && warned < 5) {
316 strcpy(warncomm, current->comm);
317 printk(KERN_WARNING "process `%s' is using obsolete "
318 "%s SO_BSDCOMPAT\n", warncomm, name);
323 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
325 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
327 if (sk->sk_flags & flags) {
328 sk->sk_flags &= ~flags;
329 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
330 net_disable_timestamp();
335 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
340 struct sk_buff_head *list = &sk->sk_receive_queue;
342 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
343 atomic_inc(&sk->sk_drops);
344 trace_sock_rcvqueue_full(sk, skb);
348 err = sk_filter(sk, skb);
352 if (!sk_rmem_schedule(sk, skb->truesize)) {
353 atomic_inc(&sk->sk_drops);
358 skb_set_owner_r(skb, sk);
360 /* Cache the SKB length before we tack it onto the receive
361 * queue. Once it is added it no longer belongs to us and
362 * may be freed by other threads of control pulling packets
367 /* we escape from rcu protected region, make sure we dont leak
372 spin_lock_irqsave(&list->lock, flags);
373 skb->dropcount = atomic_read(&sk->sk_drops);
374 __skb_queue_tail(list, skb);
375 spin_unlock_irqrestore(&list->lock, flags);
377 if (!sock_flag(sk, SOCK_DEAD))
378 sk->sk_data_ready(sk, skb_len);
381 EXPORT_SYMBOL(sock_queue_rcv_skb);
383 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
385 int rc = NET_RX_SUCCESS;
387 if (sk_filter(sk, skb))
388 goto discard_and_relse;
392 if (sk_rcvqueues_full(sk, skb)) {
393 atomic_inc(&sk->sk_drops);
394 goto discard_and_relse;
397 bh_lock_sock_nested(sk);
400 if (!sock_owned_by_user(sk)) {
402 * trylock + unlock semantics:
404 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
406 rc = sk_backlog_rcv(sk, skb);
408 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
409 } else if (sk_add_backlog(sk, skb)) {
411 atomic_inc(&sk->sk_drops);
412 goto discard_and_relse;
423 EXPORT_SYMBOL(sk_receive_skb);
425 void sk_reset_txq(struct sock *sk)
427 sk_tx_queue_clear(sk);
429 EXPORT_SYMBOL(sk_reset_txq);
431 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
433 struct dst_entry *dst = __sk_dst_get(sk);
435 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
436 sk_tx_queue_clear(sk);
437 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
444 EXPORT_SYMBOL(__sk_dst_check);
446 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
448 struct dst_entry *dst = sk_dst_get(sk);
450 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
458 EXPORT_SYMBOL(sk_dst_check);
460 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
462 int ret = -ENOPROTOOPT;
463 #ifdef CONFIG_NETDEVICES
464 struct net *net = sock_net(sk);
465 char devname[IFNAMSIZ];
470 if (!capable(CAP_NET_RAW))
477 /* Bind this socket to a particular device like "eth0",
478 * as specified in the passed interface name. If the
479 * name is "" or the option length is zero the socket
482 if (optlen > IFNAMSIZ - 1)
483 optlen = IFNAMSIZ - 1;
484 memset(devname, 0, sizeof(devname));
487 if (copy_from_user(devname, optval, optlen))
491 if (devname[0] != '\0') {
492 struct net_device *dev;
495 dev = dev_get_by_name_rcu(net, devname);
497 index = dev->ifindex;
505 sk->sk_bound_dev_if = index;
517 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
520 sock_set_flag(sk, bit);
522 sock_reset_flag(sk, bit);
526 * This is meant for all protocols to use and covers goings on
527 * at the socket level. Everything here is generic.
530 int sock_setsockopt(struct socket *sock, int level, int optname,
531 char __user *optval, unsigned int optlen)
533 struct sock *sk = sock->sk;
540 * Options without arguments
543 if (optname == SO_BINDTODEVICE)
544 return sock_bindtodevice(sk, optval, optlen);
546 if (optlen < sizeof(int))
549 if (get_user(val, (int __user *)optval))
552 valbool = val ? 1 : 0;
558 if (val && !capable(CAP_NET_ADMIN))
561 sock_valbool_flag(sk, SOCK_DBG, valbool);
564 sk->sk_reuse = valbool;
573 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
576 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
579 /* Don't error on this BSD doesn't and if you think
580 about it this is right. Otherwise apps have to
581 play 'guess the biggest size' games. RCVBUF/SNDBUF
582 are treated in BSD as hints */
584 if (val > sysctl_wmem_max)
585 val = sysctl_wmem_max;
587 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
588 if ((val * 2) < SOCK_MIN_SNDBUF)
589 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
591 sk->sk_sndbuf = val * 2;
594 * Wake up sending tasks if we
597 sk->sk_write_space(sk);
601 if (!capable(CAP_NET_ADMIN)) {
608 /* Don't error on this BSD doesn't and if you think
609 about it this is right. Otherwise apps have to
610 play 'guess the biggest size' games. RCVBUF/SNDBUF
611 are treated in BSD as hints */
613 if (val > sysctl_rmem_max)
614 val = sysctl_rmem_max;
616 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
618 * We double it on the way in to account for
619 * "struct sk_buff" etc. overhead. Applications
620 * assume that the SO_RCVBUF setting they make will
621 * allow that much actual data to be received on that
624 * Applications are unaware that "struct sk_buff" and
625 * other overheads allocate from the receive buffer
626 * during socket buffer allocation.
628 * And after considering the possible alternatives,
629 * returning the value we actually used in getsockopt
630 * is the most desirable behavior.
632 if ((val * 2) < SOCK_MIN_RCVBUF)
633 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
635 sk->sk_rcvbuf = val * 2;
639 if (!capable(CAP_NET_ADMIN)) {
647 if (sk->sk_protocol == IPPROTO_TCP)
648 tcp_set_keepalive(sk, valbool);
650 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
654 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
658 sk->sk_no_check = valbool;
662 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
663 sk->sk_priority = val;
669 if (optlen < sizeof(ling)) {
670 ret = -EINVAL; /* 1003.1g */
673 if (copy_from_user(&ling, optval, sizeof(ling))) {
678 sock_reset_flag(sk, SOCK_LINGER);
680 #if (BITS_PER_LONG == 32)
681 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
682 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
685 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
686 sock_set_flag(sk, SOCK_LINGER);
691 sock_warn_obsolete_bsdism("setsockopt");
696 set_bit(SOCK_PASSCRED, &sock->flags);
698 clear_bit(SOCK_PASSCRED, &sock->flags);
704 if (optname == SO_TIMESTAMP)
705 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
707 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
708 sock_set_flag(sk, SOCK_RCVTSTAMP);
709 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
711 sock_reset_flag(sk, SOCK_RCVTSTAMP);
712 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
716 case SO_TIMESTAMPING:
717 if (val & ~SOF_TIMESTAMPING_MASK) {
721 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
722 val & SOF_TIMESTAMPING_TX_HARDWARE);
723 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
724 val & SOF_TIMESTAMPING_TX_SOFTWARE);
725 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
726 val & SOF_TIMESTAMPING_RX_HARDWARE);
727 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
728 sock_enable_timestamp(sk,
729 SOCK_TIMESTAMPING_RX_SOFTWARE);
731 sock_disable_timestamp(sk,
732 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
733 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
734 val & SOF_TIMESTAMPING_SOFTWARE);
735 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
736 val & SOF_TIMESTAMPING_SYS_HARDWARE);
737 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
738 val & SOF_TIMESTAMPING_RAW_HARDWARE);
744 sk->sk_rcvlowat = val ? : 1;
748 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
752 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
755 case SO_ATTACH_FILTER:
757 if (optlen == sizeof(struct sock_fprog)) {
758 struct sock_fprog fprog;
761 if (copy_from_user(&fprog, optval, sizeof(fprog)))
764 ret = sk_attach_filter(&fprog, sk);
768 case SO_DETACH_FILTER:
769 ret = sk_detach_filter(sk);
774 set_bit(SOCK_PASSSEC, &sock->flags);
776 clear_bit(SOCK_PASSSEC, &sock->flags);
779 if (!capable(CAP_NET_ADMIN))
785 /* We implement the SO_SNDLOWAT etc to
786 not be settable (1003.1g 5.3) */
788 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
792 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
802 EXPORT_SYMBOL(sock_setsockopt);
805 void cred_to_ucred(struct pid *pid, const struct cred *cred,
808 ucred->pid = pid_vnr(pid);
809 ucred->uid = ucred->gid = -1;
811 struct user_namespace *current_ns = current_user_ns();
813 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
814 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
817 EXPORT_SYMBOL_GPL(cred_to_ucred);
819 int sock_getsockopt(struct socket *sock, int level, int optname,
820 char __user *optval, int __user *optlen)
822 struct sock *sk = sock->sk;
830 int lv = sizeof(int);
833 if (get_user(len, optlen))
838 memset(&v, 0, sizeof(v));
842 v.val = sock_flag(sk, SOCK_DBG);
846 v.val = sock_flag(sk, SOCK_LOCALROUTE);
850 v.val = !!sock_flag(sk, SOCK_BROADCAST);
854 v.val = sk->sk_sndbuf;
858 v.val = sk->sk_rcvbuf;
862 v.val = sk->sk_reuse;
866 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
874 v.val = sk->sk_protocol;
878 v.val = sk->sk_family;
882 v.val = -sock_error(sk);
884 v.val = xchg(&sk->sk_err_soft, 0);
888 v.val = !!sock_flag(sk, SOCK_URGINLINE);
892 v.val = sk->sk_no_check;
896 v.val = sk->sk_priority;
901 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
902 v.ling.l_linger = sk->sk_lingertime / HZ;
906 sock_warn_obsolete_bsdism("getsockopt");
910 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
911 !sock_flag(sk, SOCK_RCVTSTAMPNS);
915 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
918 case SO_TIMESTAMPING:
920 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
921 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
922 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
923 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
924 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
925 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
926 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
927 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
928 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
929 v.val |= SOF_TIMESTAMPING_SOFTWARE;
930 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
931 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
932 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
933 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
937 lv = sizeof(struct timeval);
938 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
942 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
943 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
948 lv = sizeof(struct timeval);
949 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
953 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
954 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
959 v.val = sk->sk_rcvlowat;
967 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
972 struct ucred peercred;
973 if (len > sizeof(peercred))
974 len = sizeof(peercred);
975 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
976 if (copy_to_user(optval, &peercred, len))
985 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
989 if (copy_to_user(optval, address, len))
994 /* Dubious BSD thing... Probably nobody even uses it, but
995 * the UNIX standard wants it for whatever reason... -DaveM
998 v.val = sk->sk_state == TCP_LISTEN;
1002 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
1006 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1009 v.val = sk->sk_mark;
1013 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
1016 case SO_WIFI_STATUS:
1017 v.val = !!sock_flag(sk, SOCK_WIFI_STATUS);
1021 return -ENOPROTOOPT;
1026 if (copy_to_user(optval, &v, len))
1029 if (put_user(len, optlen))
1035 * Initialize an sk_lock.
1037 * (We also register the sk_lock with the lock validator.)
1039 static inline void sock_lock_init(struct sock *sk)
1041 sock_lock_init_class_and_name(sk,
1042 af_family_slock_key_strings[sk->sk_family],
1043 af_family_slock_keys + sk->sk_family,
1044 af_family_key_strings[sk->sk_family],
1045 af_family_keys + sk->sk_family);
1049 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1050 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1051 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1053 static void sock_copy(struct sock *nsk, const struct sock *osk)
1055 #ifdef CONFIG_SECURITY_NETWORK
1056 void *sptr = nsk->sk_security;
1058 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1060 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1061 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1063 #ifdef CONFIG_SECURITY_NETWORK
1064 nsk->sk_security = sptr;
1065 security_sk_clone(osk, nsk);
1070 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1071 * un-modified. Special care is taken when initializing object to zero.
1073 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1075 if (offsetof(struct sock, sk_node.next) != 0)
1076 memset(sk, 0, offsetof(struct sock, sk_node.next));
1077 memset(&sk->sk_node.pprev, 0,
1078 size - offsetof(struct sock, sk_node.pprev));
1081 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1083 unsigned long nulls1, nulls2;
1085 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1086 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1087 if (nulls1 > nulls2)
1088 swap(nulls1, nulls2);
1091 memset((char *)sk, 0, nulls1);
1092 memset((char *)sk + nulls1 + sizeof(void *), 0,
1093 nulls2 - nulls1 - sizeof(void *));
1094 memset((char *)sk + nulls2 + sizeof(void *), 0,
1095 size - nulls2 - sizeof(void *));
1097 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1099 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1103 struct kmem_cache *slab;
1107 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1110 if (priority & __GFP_ZERO) {
1112 prot->clear_sk(sk, prot->obj_size);
1114 sk_prot_clear_nulls(sk, prot->obj_size);
1117 sk = kmalloc(prot->obj_size, priority);
1120 kmemcheck_annotate_bitfield(sk, flags);
1122 if (security_sk_alloc(sk, family, priority))
1125 if (!try_module_get(prot->owner))
1127 sk_tx_queue_clear(sk);
1133 security_sk_free(sk);
1136 kmem_cache_free(slab, sk);
1142 static void sk_prot_free(struct proto *prot, struct sock *sk)
1144 struct kmem_cache *slab;
1145 struct module *owner;
1147 owner = prot->owner;
1150 security_sk_free(sk);
1152 kmem_cache_free(slab, sk);
1158 #ifdef CONFIG_CGROUPS
1159 void sock_update_classid(struct sock *sk)
1163 rcu_read_lock(); /* doing current task, which cannot vanish. */
1164 classid = task_cls_classid(current);
1166 if (classid && classid != sk->sk_classid)
1167 sk->sk_classid = classid;
1169 EXPORT_SYMBOL(sock_update_classid);
1171 void sock_update_netprioidx(struct sock *sk)
1173 struct cgroup_netprio_state *state;
1177 state = task_netprio_state(current);
1178 sk->sk_cgrp_prioidx = state ? state->prioidx : 0;
1181 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1185 * sk_alloc - All socket objects are allocated here
1186 * @net: the applicable net namespace
1187 * @family: protocol family
1188 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1189 * @prot: struct proto associated with this new sock instance
1191 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1196 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1198 sk->sk_family = family;
1200 * See comment in struct sock definition to understand
1201 * why we need sk_prot_creator -acme
1203 sk->sk_prot = sk->sk_prot_creator = prot;
1205 sock_net_set(sk, get_net(net));
1206 atomic_set(&sk->sk_wmem_alloc, 1);
1208 sock_update_classid(sk);
1209 sock_update_netprioidx(sk);
1214 EXPORT_SYMBOL(sk_alloc);
1216 static void __sk_free(struct sock *sk)
1218 struct sk_filter *filter;
1220 if (sk->sk_destruct)
1221 sk->sk_destruct(sk);
1223 filter = rcu_dereference_check(sk->sk_filter,
1224 atomic_read(&sk->sk_wmem_alloc) == 0);
1226 sk_filter_uncharge(sk, filter);
1227 RCU_INIT_POINTER(sk->sk_filter, NULL);
1230 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1232 if (atomic_read(&sk->sk_omem_alloc))
1233 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1234 __func__, atomic_read(&sk->sk_omem_alloc));
1236 if (sk->sk_peer_cred)
1237 put_cred(sk->sk_peer_cred);
1238 put_pid(sk->sk_peer_pid);
1239 put_net(sock_net(sk));
1240 sk_prot_free(sk->sk_prot_creator, sk);
1243 void sk_free(struct sock *sk)
1246 * We subtract one from sk_wmem_alloc and can know if
1247 * some packets are still in some tx queue.
1248 * If not null, sock_wfree() will call __sk_free(sk) later
1250 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1253 EXPORT_SYMBOL(sk_free);
1256 * Last sock_put should drop reference to sk->sk_net. It has already
1257 * been dropped in sk_change_net. Taking reference to stopping namespace
1259 * Take reference to a socket to remove it from hash _alive_ and after that
1260 * destroy it in the context of init_net.
1262 void sk_release_kernel(struct sock *sk)
1264 if (sk == NULL || sk->sk_socket == NULL)
1268 sock_release(sk->sk_socket);
1269 release_net(sock_net(sk));
1270 sock_net_set(sk, get_net(&init_net));
1273 EXPORT_SYMBOL(sk_release_kernel);
1276 * sk_clone_lock - clone a socket, and lock its clone
1277 * @sk: the socket to clone
1278 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1280 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1282 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1286 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1287 if (newsk != NULL) {
1288 struct sk_filter *filter;
1290 sock_copy(newsk, sk);
1293 get_net(sock_net(newsk));
1294 sk_node_init(&newsk->sk_node);
1295 sock_lock_init(newsk);
1296 bh_lock_sock(newsk);
1297 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1298 newsk->sk_backlog.len = 0;
1300 atomic_set(&newsk->sk_rmem_alloc, 0);
1302 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1304 atomic_set(&newsk->sk_wmem_alloc, 1);
1305 atomic_set(&newsk->sk_omem_alloc, 0);
1306 skb_queue_head_init(&newsk->sk_receive_queue);
1307 skb_queue_head_init(&newsk->sk_write_queue);
1308 #ifdef CONFIG_NET_DMA
1309 skb_queue_head_init(&newsk->sk_async_wait_queue);
1312 spin_lock_init(&newsk->sk_dst_lock);
1313 rwlock_init(&newsk->sk_callback_lock);
1314 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1315 af_callback_keys + newsk->sk_family,
1316 af_family_clock_key_strings[newsk->sk_family]);
1318 newsk->sk_dst_cache = NULL;
1319 newsk->sk_wmem_queued = 0;
1320 newsk->sk_forward_alloc = 0;
1321 newsk->sk_send_head = NULL;
1322 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1324 sock_reset_flag(newsk, SOCK_DONE);
1325 skb_queue_head_init(&newsk->sk_error_queue);
1327 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1329 sk_filter_charge(newsk, filter);
1331 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1332 /* It is still raw copy of parent, so invalidate
1333 * destructor and make plain sk_free() */
1334 newsk->sk_destruct = NULL;
1335 bh_unlock_sock(newsk);
1342 newsk->sk_priority = 0;
1344 * Before updating sk_refcnt, we must commit prior changes to memory
1345 * (Documentation/RCU/rculist_nulls.txt for details)
1348 atomic_set(&newsk->sk_refcnt, 2);
1351 * Increment the counter in the same struct proto as the master
1352 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1353 * is the same as sk->sk_prot->socks, as this field was copied
1356 * This _changes_ the previous behaviour, where
1357 * tcp_create_openreq_child always was incrementing the
1358 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1359 * to be taken into account in all callers. -acme
1361 sk_refcnt_debug_inc(newsk);
1362 sk_set_socket(newsk, NULL);
1363 newsk->sk_wq = NULL;
1365 sk_update_clone(sk, newsk);
1367 if (newsk->sk_prot->sockets_allocated)
1368 sk_sockets_allocated_inc(newsk);
1370 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1371 net_enable_timestamp();
1376 EXPORT_SYMBOL_GPL(sk_clone_lock);
1378 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1380 __sk_dst_set(sk, dst);
1381 sk->sk_route_caps = dst->dev->features;
1382 if (sk->sk_route_caps & NETIF_F_GSO)
1383 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1384 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1385 if (sk_can_gso(sk)) {
1386 if (dst->header_len) {
1387 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1389 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1390 sk->sk_gso_max_size = dst->dev->gso_max_size;
1394 EXPORT_SYMBOL_GPL(sk_setup_caps);
1396 void __init sk_init(void)
1398 if (totalram_pages <= 4096) {
1399 sysctl_wmem_max = 32767;
1400 sysctl_rmem_max = 32767;
1401 sysctl_wmem_default = 32767;
1402 sysctl_rmem_default = 32767;
1403 } else if (totalram_pages >= 131072) {
1404 sysctl_wmem_max = 131071;
1405 sysctl_rmem_max = 131071;
1410 * Simple resource managers for sockets.
1415 * Write buffer destructor automatically called from kfree_skb.
1417 void sock_wfree(struct sk_buff *skb)
1419 struct sock *sk = skb->sk;
1420 unsigned int len = skb->truesize;
1422 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1424 * Keep a reference on sk_wmem_alloc, this will be released
1425 * after sk_write_space() call
1427 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1428 sk->sk_write_space(sk);
1432 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1433 * could not do because of in-flight packets
1435 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1438 EXPORT_SYMBOL(sock_wfree);
1441 * Read buffer destructor automatically called from kfree_skb.
1443 void sock_rfree(struct sk_buff *skb)
1445 struct sock *sk = skb->sk;
1446 unsigned int len = skb->truesize;
1448 atomic_sub(len, &sk->sk_rmem_alloc);
1449 sk_mem_uncharge(sk, len);
1451 EXPORT_SYMBOL(sock_rfree);
1454 int sock_i_uid(struct sock *sk)
1458 read_lock_bh(&sk->sk_callback_lock);
1459 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1460 read_unlock_bh(&sk->sk_callback_lock);
1463 EXPORT_SYMBOL(sock_i_uid);
1465 unsigned long sock_i_ino(struct sock *sk)
1469 read_lock_bh(&sk->sk_callback_lock);
1470 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1471 read_unlock_bh(&sk->sk_callback_lock);
1474 EXPORT_SYMBOL(sock_i_ino);
1477 * Allocate a skb from the socket's send buffer.
1479 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1482 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1483 struct sk_buff *skb = alloc_skb(size, priority);
1485 skb_set_owner_w(skb, sk);
1491 EXPORT_SYMBOL(sock_wmalloc);
1494 * Allocate a skb from the socket's receive buffer.
1496 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1499 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1500 struct sk_buff *skb = alloc_skb(size, priority);
1502 skb_set_owner_r(skb, sk);
1510 * Allocate a memory block from the socket's option memory buffer.
1512 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1514 if ((unsigned)size <= sysctl_optmem_max &&
1515 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1517 /* First do the add, to avoid the race if kmalloc
1520 atomic_add(size, &sk->sk_omem_alloc);
1521 mem = kmalloc(size, priority);
1524 atomic_sub(size, &sk->sk_omem_alloc);
1528 EXPORT_SYMBOL(sock_kmalloc);
1531 * Free an option memory block.
1533 void sock_kfree_s(struct sock *sk, void *mem, int size)
1536 atomic_sub(size, &sk->sk_omem_alloc);
1538 EXPORT_SYMBOL(sock_kfree_s);
1540 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1541 I think, these locks should be removed for datagram sockets.
1543 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1547 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1551 if (signal_pending(current))
1553 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1554 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1555 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1557 if (sk->sk_shutdown & SEND_SHUTDOWN)
1561 timeo = schedule_timeout(timeo);
1563 finish_wait(sk_sleep(sk), &wait);
1569 * Generic send/receive buffer handlers
1572 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1573 unsigned long data_len, int noblock,
1576 struct sk_buff *skb;
1581 gfp_mask = sk->sk_allocation;
1582 if (gfp_mask & __GFP_WAIT)
1583 gfp_mask |= __GFP_REPEAT;
1585 timeo = sock_sndtimeo(sk, noblock);
1587 err = sock_error(sk);
1592 if (sk->sk_shutdown & SEND_SHUTDOWN)
1595 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1596 skb = alloc_skb(header_len, gfp_mask);
1601 /* No pages, we're done... */
1605 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1606 skb->truesize += data_len;
1607 skb_shinfo(skb)->nr_frags = npages;
1608 for (i = 0; i < npages; i++) {
1611 page = alloc_pages(sk->sk_allocation, 0);
1614 skb_shinfo(skb)->nr_frags = i;
1619 __skb_fill_page_desc(skb, i,
1621 (data_len >= PAGE_SIZE ?
1624 data_len -= PAGE_SIZE;
1627 /* Full success... */
1633 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1634 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1638 if (signal_pending(current))
1640 timeo = sock_wait_for_wmem(sk, timeo);
1643 skb_set_owner_w(skb, sk);
1647 err = sock_intr_errno(timeo);
1652 EXPORT_SYMBOL(sock_alloc_send_pskb);
1654 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1655 int noblock, int *errcode)
1657 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1659 EXPORT_SYMBOL(sock_alloc_send_skb);
1661 static void __lock_sock(struct sock *sk)
1662 __releases(&sk->sk_lock.slock)
1663 __acquires(&sk->sk_lock.slock)
1668 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1669 TASK_UNINTERRUPTIBLE);
1670 spin_unlock_bh(&sk->sk_lock.slock);
1672 spin_lock_bh(&sk->sk_lock.slock);
1673 if (!sock_owned_by_user(sk))
1676 finish_wait(&sk->sk_lock.wq, &wait);
1679 static void __release_sock(struct sock *sk)
1680 __releases(&sk->sk_lock.slock)
1681 __acquires(&sk->sk_lock.slock)
1683 struct sk_buff *skb = sk->sk_backlog.head;
1686 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1690 struct sk_buff *next = skb->next;
1692 WARN_ON_ONCE(skb_dst_is_noref(skb));
1694 sk_backlog_rcv(sk, skb);
1697 * We are in process context here with softirqs
1698 * disabled, use cond_resched_softirq() to preempt.
1699 * This is safe to do because we've taken the backlog
1702 cond_resched_softirq();
1705 } while (skb != NULL);
1708 } while ((skb = sk->sk_backlog.head) != NULL);
1711 * Doing the zeroing here guarantee we can not loop forever
1712 * while a wild producer attempts to flood us.
1714 sk->sk_backlog.len = 0;
1718 * sk_wait_data - wait for data to arrive at sk_receive_queue
1719 * @sk: sock to wait on
1720 * @timeo: for how long
1722 * Now socket state including sk->sk_err is changed only under lock,
1723 * hence we may omit checks after joining wait queue.
1724 * We check receive queue before schedule() only as optimization;
1725 * it is very likely that release_sock() added new data.
1727 int sk_wait_data(struct sock *sk, long *timeo)
1732 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1733 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1734 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1735 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1736 finish_wait(sk_sleep(sk), &wait);
1739 EXPORT_SYMBOL(sk_wait_data);
1742 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1744 * @size: memory size to allocate
1745 * @kind: allocation type
1747 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1748 * rmem allocation. This function assumes that protocols which have
1749 * memory_pressure use sk_wmem_queued as write buffer accounting.
1751 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1753 struct proto *prot = sk->sk_prot;
1754 int amt = sk_mem_pages(size);
1756 int parent_status = UNDER_LIMIT;
1758 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1760 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1763 if (parent_status == UNDER_LIMIT &&
1764 allocated <= sk_prot_mem_limits(sk, 0)) {
1765 sk_leave_memory_pressure(sk);
1769 /* Under pressure. (we or our parents) */
1770 if ((parent_status > SOFT_LIMIT) ||
1771 allocated > sk_prot_mem_limits(sk, 1))
1772 sk_enter_memory_pressure(sk);
1774 /* Over hard limit (we or our parents) */
1775 if ((parent_status == OVER_LIMIT) ||
1776 (allocated > sk_prot_mem_limits(sk, 2)))
1777 goto suppress_allocation;
1779 /* guarantee minimum buffer size under pressure */
1780 if (kind == SK_MEM_RECV) {
1781 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1784 } else { /* SK_MEM_SEND */
1785 if (sk->sk_type == SOCK_STREAM) {
1786 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1788 } else if (atomic_read(&sk->sk_wmem_alloc) <
1789 prot->sysctl_wmem[0])
1793 if (sk_has_memory_pressure(sk)) {
1796 if (!sk_under_memory_pressure(sk))
1798 alloc = sk_sockets_allocated_read_positive(sk);
1799 if (sk_prot_mem_limits(sk, 2) > alloc *
1800 sk_mem_pages(sk->sk_wmem_queued +
1801 atomic_read(&sk->sk_rmem_alloc) +
1802 sk->sk_forward_alloc))
1806 suppress_allocation:
1808 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1809 sk_stream_moderate_sndbuf(sk);
1811 /* Fail only if socket is _under_ its sndbuf.
1812 * In this case we cannot block, so that we have to fail.
1814 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1818 trace_sock_exceed_buf_limit(sk, prot, allocated);
1820 /* Alas. Undo changes. */
1821 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1823 sk_memory_allocated_sub(sk, amt, parent_status);
1827 EXPORT_SYMBOL(__sk_mem_schedule);
1830 * __sk_reclaim - reclaim memory_allocated
1833 void __sk_mem_reclaim(struct sock *sk)
1835 sk_memory_allocated_sub(sk,
1836 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT, 0);
1837 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1839 if (sk_under_memory_pressure(sk) &&
1840 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1841 sk_leave_memory_pressure(sk);
1843 EXPORT_SYMBOL(__sk_mem_reclaim);
1847 * Set of default routines for initialising struct proto_ops when
1848 * the protocol does not support a particular function. In certain
1849 * cases where it makes no sense for a protocol to have a "do nothing"
1850 * function, some default processing is provided.
1853 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1857 EXPORT_SYMBOL(sock_no_bind);
1859 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1864 EXPORT_SYMBOL(sock_no_connect);
1866 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1870 EXPORT_SYMBOL(sock_no_socketpair);
1872 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1876 EXPORT_SYMBOL(sock_no_accept);
1878 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1883 EXPORT_SYMBOL(sock_no_getname);
1885 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1889 EXPORT_SYMBOL(sock_no_poll);
1891 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1895 EXPORT_SYMBOL(sock_no_ioctl);
1897 int sock_no_listen(struct socket *sock, int backlog)
1901 EXPORT_SYMBOL(sock_no_listen);
1903 int sock_no_shutdown(struct socket *sock, int how)
1907 EXPORT_SYMBOL(sock_no_shutdown);
1909 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1910 char __user *optval, unsigned int optlen)
1914 EXPORT_SYMBOL(sock_no_setsockopt);
1916 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1917 char __user *optval, int __user *optlen)
1921 EXPORT_SYMBOL(sock_no_getsockopt);
1923 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1928 EXPORT_SYMBOL(sock_no_sendmsg);
1930 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1931 size_t len, int flags)
1935 EXPORT_SYMBOL(sock_no_recvmsg);
1937 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1939 /* Mirror missing mmap method error code */
1942 EXPORT_SYMBOL(sock_no_mmap);
1944 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1947 struct msghdr msg = {.msg_flags = flags};
1949 char *kaddr = kmap(page);
1950 iov.iov_base = kaddr + offset;
1952 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1956 EXPORT_SYMBOL(sock_no_sendpage);
1959 * Default Socket Callbacks
1962 static void sock_def_wakeup(struct sock *sk)
1964 struct socket_wq *wq;
1967 wq = rcu_dereference(sk->sk_wq);
1968 if (wq_has_sleeper(wq))
1969 wake_up_interruptible_all(&wq->wait);
1973 static void sock_def_error_report(struct sock *sk)
1975 struct socket_wq *wq;
1978 wq = rcu_dereference(sk->sk_wq);
1979 if (wq_has_sleeper(wq))
1980 wake_up_interruptible_poll(&wq->wait, POLLERR);
1981 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1985 static void sock_def_readable(struct sock *sk, int len)
1987 struct socket_wq *wq;
1990 wq = rcu_dereference(sk->sk_wq);
1991 if (wq_has_sleeper(wq))
1992 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1993 POLLRDNORM | POLLRDBAND);
1994 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1998 static void sock_def_write_space(struct sock *sk)
2000 struct socket_wq *wq;
2004 /* Do not wake up a writer until he can make "significant"
2007 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2008 wq = rcu_dereference(sk->sk_wq);
2009 if (wq_has_sleeper(wq))
2010 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2011 POLLWRNORM | POLLWRBAND);
2013 /* Should agree with poll, otherwise some programs break */
2014 if (sock_writeable(sk))
2015 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2021 static void sock_def_destruct(struct sock *sk)
2023 kfree(sk->sk_protinfo);
2026 void sk_send_sigurg(struct sock *sk)
2028 if (sk->sk_socket && sk->sk_socket->file)
2029 if (send_sigurg(&sk->sk_socket->file->f_owner))
2030 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2032 EXPORT_SYMBOL(sk_send_sigurg);
2034 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2035 unsigned long expires)
2037 if (!mod_timer(timer, expires))
2040 EXPORT_SYMBOL(sk_reset_timer);
2042 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2044 if (timer_pending(timer) && del_timer(timer))
2047 EXPORT_SYMBOL(sk_stop_timer);
2049 void sock_init_data(struct socket *sock, struct sock *sk)
2051 skb_queue_head_init(&sk->sk_receive_queue);
2052 skb_queue_head_init(&sk->sk_write_queue);
2053 skb_queue_head_init(&sk->sk_error_queue);
2054 #ifdef CONFIG_NET_DMA
2055 skb_queue_head_init(&sk->sk_async_wait_queue);
2058 sk->sk_send_head = NULL;
2060 init_timer(&sk->sk_timer);
2062 sk->sk_allocation = GFP_KERNEL;
2063 sk->sk_rcvbuf = sysctl_rmem_default;
2064 sk->sk_sndbuf = sysctl_wmem_default;
2065 sk->sk_state = TCP_CLOSE;
2066 sk_set_socket(sk, sock);
2068 sock_set_flag(sk, SOCK_ZAPPED);
2071 sk->sk_type = sock->type;
2072 sk->sk_wq = sock->wq;
2077 spin_lock_init(&sk->sk_dst_lock);
2078 rwlock_init(&sk->sk_callback_lock);
2079 lockdep_set_class_and_name(&sk->sk_callback_lock,
2080 af_callback_keys + sk->sk_family,
2081 af_family_clock_key_strings[sk->sk_family]);
2083 sk->sk_state_change = sock_def_wakeup;
2084 sk->sk_data_ready = sock_def_readable;
2085 sk->sk_write_space = sock_def_write_space;
2086 sk->sk_error_report = sock_def_error_report;
2087 sk->sk_destruct = sock_def_destruct;
2089 sk->sk_sndmsg_page = NULL;
2090 sk->sk_sndmsg_off = 0;
2092 sk->sk_peer_pid = NULL;
2093 sk->sk_peer_cred = NULL;
2094 sk->sk_write_pending = 0;
2095 sk->sk_rcvlowat = 1;
2096 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2097 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2099 sk->sk_stamp = ktime_set(-1L, 0);
2102 * Before updating sk_refcnt, we must commit prior changes to memory
2103 * (Documentation/RCU/rculist_nulls.txt for details)
2106 atomic_set(&sk->sk_refcnt, 1);
2107 atomic_set(&sk->sk_drops, 0);
2109 EXPORT_SYMBOL(sock_init_data);
2111 void lock_sock_nested(struct sock *sk, int subclass)
2114 spin_lock_bh(&sk->sk_lock.slock);
2115 if (sk->sk_lock.owned)
2117 sk->sk_lock.owned = 1;
2118 spin_unlock(&sk->sk_lock.slock);
2120 * The sk_lock has mutex_lock() semantics here:
2122 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2125 EXPORT_SYMBOL(lock_sock_nested);
2127 void release_sock(struct sock *sk)
2130 * The sk_lock has mutex_unlock() semantics:
2132 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2134 spin_lock_bh(&sk->sk_lock.slock);
2135 if (sk->sk_backlog.tail)
2137 sk->sk_lock.owned = 0;
2138 if (waitqueue_active(&sk->sk_lock.wq))
2139 wake_up(&sk->sk_lock.wq);
2140 spin_unlock_bh(&sk->sk_lock.slock);
2142 EXPORT_SYMBOL(release_sock);
2145 * lock_sock_fast - fast version of lock_sock
2148 * This version should be used for very small section, where process wont block
2149 * return false if fast path is taken
2150 * sk_lock.slock locked, owned = 0, BH disabled
2151 * return true if slow path is taken
2152 * sk_lock.slock unlocked, owned = 1, BH enabled
2154 bool lock_sock_fast(struct sock *sk)
2157 spin_lock_bh(&sk->sk_lock.slock);
2159 if (!sk->sk_lock.owned)
2161 * Note : We must disable BH
2166 sk->sk_lock.owned = 1;
2167 spin_unlock(&sk->sk_lock.slock);
2169 * The sk_lock has mutex_lock() semantics here:
2171 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2175 EXPORT_SYMBOL(lock_sock_fast);
2177 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2180 if (!sock_flag(sk, SOCK_TIMESTAMP))
2181 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2182 tv = ktime_to_timeval(sk->sk_stamp);
2183 if (tv.tv_sec == -1)
2185 if (tv.tv_sec == 0) {
2186 sk->sk_stamp = ktime_get_real();
2187 tv = ktime_to_timeval(sk->sk_stamp);
2189 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2191 EXPORT_SYMBOL(sock_get_timestamp);
2193 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2196 if (!sock_flag(sk, SOCK_TIMESTAMP))
2197 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2198 ts = ktime_to_timespec(sk->sk_stamp);
2199 if (ts.tv_sec == -1)
2201 if (ts.tv_sec == 0) {
2202 sk->sk_stamp = ktime_get_real();
2203 ts = ktime_to_timespec(sk->sk_stamp);
2205 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2207 EXPORT_SYMBOL(sock_get_timestampns);
2209 void sock_enable_timestamp(struct sock *sk, int flag)
2211 if (!sock_flag(sk, flag)) {
2212 unsigned long previous_flags = sk->sk_flags;
2214 sock_set_flag(sk, flag);
2216 * we just set one of the two flags which require net
2217 * time stamping, but time stamping might have been on
2218 * already because of the other one
2220 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2221 net_enable_timestamp();
2226 * Get a socket option on an socket.
2228 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2229 * asynchronous errors should be reported by getsockopt. We assume
2230 * this means if you specify SO_ERROR (otherwise whats the point of it).
2232 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2233 char __user *optval, int __user *optlen)
2235 struct sock *sk = sock->sk;
2237 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2239 EXPORT_SYMBOL(sock_common_getsockopt);
2241 #ifdef CONFIG_COMPAT
2242 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2243 char __user *optval, int __user *optlen)
2245 struct sock *sk = sock->sk;
2247 if (sk->sk_prot->compat_getsockopt != NULL)
2248 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2250 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2252 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2255 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2256 struct msghdr *msg, size_t size, int flags)
2258 struct sock *sk = sock->sk;
2262 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2263 flags & ~MSG_DONTWAIT, &addr_len);
2265 msg->msg_namelen = addr_len;
2268 EXPORT_SYMBOL(sock_common_recvmsg);
2271 * Set socket options on an inet socket.
2273 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2274 char __user *optval, unsigned int optlen)
2276 struct sock *sk = sock->sk;
2278 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2280 EXPORT_SYMBOL(sock_common_setsockopt);
2282 #ifdef CONFIG_COMPAT
2283 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2284 char __user *optval, unsigned int optlen)
2286 struct sock *sk = sock->sk;
2288 if (sk->sk_prot->compat_setsockopt != NULL)
2289 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2291 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2293 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2296 void sk_common_release(struct sock *sk)
2298 if (sk->sk_prot->destroy)
2299 sk->sk_prot->destroy(sk);
2302 * Observation: when sock_common_release is called, processes have
2303 * no access to socket. But net still has.
2304 * Step one, detach it from networking:
2306 * A. Remove from hash tables.
2309 sk->sk_prot->unhash(sk);
2312 * In this point socket cannot receive new packets, but it is possible
2313 * that some packets are in flight because some CPU runs receiver and
2314 * did hash table lookup before we unhashed socket. They will achieve
2315 * receive queue and will be purged by socket destructor.
2317 * Also we still have packets pending on receive queue and probably,
2318 * our own packets waiting in device queues. sock_destroy will drain
2319 * receive queue, but transmitted packets will delay socket destruction
2320 * until the last reference will be released.
2325 xfrm_sk_free_policy(sk);
2327 sk_refcnt_debug_release(sk);
2330 EXPORT_SYMBOL(sk_common_release);
2332 #ifdef CONFIG_PROC_FS
2333 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2335 int val[PROTO_INUSE_NR];
2338 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2340 #ifdef CONFIG_NET_NS
2341 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2343 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2345 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2347 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2349 int cpu, idx = prot->inuse_idx;
2352 for_each_possible_cpu(cpu)
2353 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2355 return res >= 0 ? res : 0;
2357 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2359 static int __net_init sock_inuse_init_net(struct net *net)
2361 net->core.inuse = alloc_percpu(struct prot_inuse);
2362 return net->core.inuse ? 0 : -ENOMEM;
2365 static void __net_exit sock_inuse_exit_net(struct net *net)
2367 free_percpu(net->core.inuse);
2370 static struct pernet_operations net_inuse_ops = {
2371 .init = sock_inuse_init_net,
2372 .exit = sock_inuse_exit_net,
2375 static __init int net_inuse_init(void)
2377 if (register_pernet_subsys(&net_inuse_ops))
2378 panic("Cannot initialize net inuse counters");
2383 core_initcall(net_inuse_init);
2385 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2387 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2389 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2391 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2393 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2395 int cpu, idx = prot->inuse_idx;
2398 for_each_possible_cpu(cpu)
2399 res += per_cpu(prot_inuse, cpu).val[idx];
2401 return res >= 0 ? res : 0;
2403 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2406 static void assign_proto_idx(struct proto *prot)
2408 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2410 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2411 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2415 set_bit(prot->inuse_idx, proto_inuse_idx);
2418 static void release_proto_idx(struct proto *prot)
2420 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2421 clear_bit(prot->inuse_idx, proto_inuse_idx);
2424 static inline void assign_proto_idx(struct proto *prot)
2428 static inline void release_proto_idx(struct proto *prot)
2433 int proto_register(struct proto *prot, int alloc_slab)
2436 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2437 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2440 if (prot->slab == NULL) {
2441 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2446 if (prot->rsk_prot != NULL) {
2447 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2448 if (prot->rsk_prot->slab_name == NULL)
2449 goto out_free_sock_slab;
2451 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2452 prot->rsk_prot->obj_size, 0,
2453 SLAB_HWCACHE_ALIGN, NULL);
2455 if (prot->rsk_prot->slab == NULL) {
2456 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2458 goto out_free_request_sock_slab_name;
2462 if (prot->twsk_prot != NULL) {
2463 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2465 if (prot->twsk_prot->twsk_slab_name == NULL)
2466 goto out_free_request_sock_slab;
2468 prot->twsk_prot->twsk_slab =
2469 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2470 prot->twsk_prot->twsk_obj_size,
2472 SLAB_HWCACHE_ALIGN |
2475 if (prot->twsk_prot->twsk_slab == NULL)
2476 goto out_free_timewait_sock_slab_name;
2480 mutex_lock(&proto_list_mutex);
2481 list_add(&prot->node, &proto_list);
2482 assign_proto_idx(prot);
2483 mutex_unlock(&proto_list_mutex);
2486 out_free_timewait_sock_slab_name:
2487 kfree(prot->twsk_prot->twsk_slab_name);
2488 out_free_request_sock_slab:
2489 if (prot->rsk_prot && prot->rsk_prot->slab) {
2490 kmem_cache_destroy(prot->rsk_prot->slab);
2491 prot->rsk_prot->slab = NULL;
2493 out_free_request_sock_slab_name:
2495 kfree(prot->rsk_prot->slab_name);
2497 kmem_cache_destroy(prot->slab);
2502 EXPORT_SYMBOL(proto_register);
2504 void proto_unregister(struct proto *prot)
2506 mutex_lock(&proto_list_mutex);
2507 release_proto_idx(prot);
2508 list_del(&prot->node);
2509 mutex_unlock(&proto_list_mutex);
2511 if (prot->slab != NULL) {
2512 kmem_cache_destroy(prot->slab);
2516 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2517 kmem_cache_destroy(prot->rsk_prot->slab);
2518 kfree(prot->rsk_prot->slab_name);
2519 prot->rsk_prot->slab = NULL;
2522 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2523 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2524 kfree(prot->twsk_prot->twsk_slab_name);
2525 prot->twsk_prot->twsk_slab = NULL;
2528 EXPORT_SYMBOL(proto_unregister);
2530 #ifdef CONFIG_PROC_FS
2531 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2532 __acquires(proto_list_mutex)
2534 mutex_lock(&proto_list_mutex);
2535 return seq_list_start_head(&proto_list, *pos);
2538 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2540 return seq_list_next(v, &proto_list, pos);
2543 static void proto_seq_stop(struct seq_file *seq, void *v)
2544 __releases(proto_list_mutex)
2546 mutex_unlock(&proto_list_mutex);
2549 static char proto_method_implemented(const void *method)
2551 return method == NULL ? 'n' : 'y';
2553 static long sock_prot_memory_allocated(struct proto *proto)
2555 return proto->memory_allocated != NULL ? proto_memory_allocated(proto): -1L;
2558 static char *sock_prot_memory_pressure(struct proto *proto)
2560 return proto->memory_pressure != NULL ?
2561 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2564 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2567 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2568 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2571 sock_prot_inuse_get(seq_file_net(seq), proto),
2572 sock_prot_memory_allocated(proto),
2573 sock_prot_memory_pressure(proto),
2575 proto->slab == NULL ? "no" : "yes",
2576 module_name(proto->owner),
2577 proto_method_implemented(proto->close),
2578 proto_method_implemented(proto->connect),
2579 proto_method_implemented(proto->disconnect),
2580 proto_method_implemented(proto->accept),
2581 proto_method_implemented(proto->ioctl),
2582 proto_method_implemented(proto->init),
2583 proto_method_implemented(proto->destroy),
2584 proto_method_implemented(proto->shutdown),
2585 proto_method_implemented(proto->setsockopt),
2586 proto_method_implemented(proto->getsockopt),
2587 proto_method_implemented(proto->sendmsg),
2588 proto_method_implemented(proto->recvmsg),
2589 proto_method_implemented(proto->sendpage),
2590 proto_method_implemented(proto->bind),
2591 proto_method_implemented(proto->backlog_rcv),
2592 proto_method_implemented(proto->hash),
2593 proto_method_implemented(proto->unhash),
2594 proto_method_implemented(proto->get_port),
2595 proto_method_implemented(proto->enter_memory_pressure));
2598 static int proto_seq_show(struct seq_file *seq, void *v)
2600 if (v == &proto_list)
2601 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2610 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2612 proto_seq_printf(seq, list_entry(v, struct proto, node));
2616 static const struct seq_operations proto_seq_ops = {
2617 .start = proto_seq_start,
2618 .next = proto_seq_next,
2619 .stop = proto_seq_stop,
2620 .show = proto_seq_show,
2623 static int proto_seq_open(struct inode *inode, struct file *file)
2625 return seq_open_net(inode, file, &proto_seq_ops,
2626 sizeof(struct seq_net_private));
2629 static const struct file_operations proto_seq_fops = {
2630 .owner = THIS_MODULE,
2631 .open = proto_seq_open,
2633 .llseek = seq_lseek,
2634 .release = seq_release_net,
2637 static __net_init int proto_init_net(struct net *net)
2639 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2645 static __net_exit void proto_exit_net(struct net *net)
2647 proc_net_remove(net, "protocols");
2651 static __net_initdata struct pernet_operations proto_net_ops = {
2652 .init = proto_init_net,
2653 .exit = proto_exit_net,
2656 static int __init proto_init(void)
2658 return register_pernet_subsys(&proto_net_ops);
2661 subsys_initcall(proto_init);
2663 #endif /* PROC_FS */