2
* INET An implementation of the TCP/IP protocol suite for the LINUX
3
* operating system. INET is implemented using the BSD Socket
4
* interface as the means of communication with the user level.
6
* Generic socket support routines. Memory allocators, socket lock/release
7
* handler for protocols to use and generic option handler.
11
* Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12
* Florian La Roche, <flla@stud.uni-sb.de>
13
* Alan Cox, <A.Cox@swansea.ac.uk>
16
* Alan Cox : Numerous verify_area() problems
17
* Alan Cox : Connecting on a connecting socket
18
* now returns an error for tcp.
19
* Alan Cox : sock->protocol is set correctly.
20
* and is not sometimes left as 0.
21
* Alan Cox : connect handles icmp errors on a
22
* connect properly. Unfortunately there
23
* is a restart syscall nasty there. I
24
* can't match BSD without hacking the C
25
* library. Ideas urgently sought!
26
* Alan Cox : Disallow bind() to addresses that are
27
* not ours - especially broadcast ones!!
28
* Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29
* Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30
* instead they leave that for the DESTROY timer.
31
* Alan Cox : Clean up error flag in accept
32
* Alan Cox : TCP ack handling is buggy, the DESTROY timer
33
* was buggy. Put a remove_sock() in the handler
34
* for memory when we hit 0. Also altered the timer
35
* code. The ACK stuff can wait and needs major
37
* Alan Cox : Fixed TCP ack bug, removed remove sock
38
* and fixed timer/inet_bh race.
39
* Alan Cox : Added zapped flag for TCP
40
* Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41
* Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42
* Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43
* Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44
* Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45
* Rick Sladkey : Relaxed UDP rules for matching packets.
46
* C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47
* Pauline Middelink : identd support
48
* Alan Cox : Fixed connect() taking signals I think.
49
* Alan Cox : SO_LINGER supported
50
* Alan Cox : Error reporting fixes
51
* Anonymous : inet_create tidied up (sk->reuse setting)
52
* Alan Cox : inet sockets don't set sk->type!
53
* Alan Cox : Split socket option code
54
* Alan Cox : Callbacks
55
* Alan Cox : Nagle flag for Charles & Johannes stuff
56
* Alex : Removed restriction on inet fioctl
57
* Alan Cox : Splitting INET from NET core
58
* Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59
* Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60
* Alan Cox : Split IP from generic code
61
* Alan Cox : New kfree_skbmem()
62
* Alan Cox : Make SO_DEBUG superuser only.
63
* Alan Cox : Allow anyone to clear SO_DEBUG
65
* Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66
* Alan Cox : Allocator for a socket is settable.
67
* Alan Cox : SO_ERROR includes soft errors.
68
* Alan Cox : Allow NULL arguments on some SO_ opts
69
* Alan Cox : Generic socket allocation to make hooks
70
* easier (suggested by Craig Metz).
71
* Michael Pall : SO_ERROR returns positive errno again
72
* Steve Whitehouse: Added default destructor to free
73
* protocol private data.
74
* Steve Whitehouse: Added various other default routines
75
* common to several socket families.
76
* Chris Evans : Call suser() check last on F_SETOWN
77
* Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78
* Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79
* Andi Kleen : Fix write_space callback
80
* Chris Evans : Security fixes - signedness again
81
* Arnaldo C. Melo : cleanups, use skb_queue_purge
86
* This program is free software; you can redistribute it and/or
87
* modify it under the terms of the GNU General Public License
88
* as published by the Free Software Foundation; either version
89
* 2 of the License, or (at your option) any later version.
92
#include <linux/capability.h>
93
#include <linux/errno.h>
94
#include <linux/types.h>
95
#include <linux/socket.h>
97
#include <linux/kernel.h>
98
#include <linux/module.h>
99
#include <linux/proc_fs.h>
100
#include <linux/seq_file.h>
101
#include <linux/sched.h>
102
#include <linux/timer.h>
103
#include <linux/string.h>
104
#include <linux/sockios.h>
105
#include <linux/net.h>
106
#include <linux/mm.h>
107
#include <linux/slab.h>
108
#include <linux/interrupt.h>
109
#include <linux/poll.h>
110
#include <linux/tcp.h>
111
#include <linux/init.h>
112
#include <linux/highmem.h>
113
#include <linux/user_namespace.h>
115
#include <asm/uaccess.h>
116
#include <asm/system.h>
118
#include <linux/netdevice.h>
119
#include <net/protocol.h>
120
#include <linux/skbuff.h>
121
#include <net/net_namespace.h>
122
#include <net/request_sock.h>
123
#include <net/sock.h>
124
#include <linux/net_tstamp.h>
125
#include <net/xfrm.h>
126
#include <linux/ipsec.h>
127
#include <net/cls_cgroup.h>
129
#include <linux/filter.h>
131
#include <trace/events/sock.h>
138
* Each address family might have different locking rules, so we have
139
* one slock key per address family:
141
static struct lock_class_key af_family_keys[AF_MAX];
142
static struct lock_class_key af_family_slock_keys[AF_MAX];
145
* Make lock validator output more readable. (we pre-construct these
146
* strings build-time, so that runtime initialization of socket
149
static const char *const af_family_key_strings[AF_MAX+1] = {
150
"sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
151
"sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
152
"sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
153
"sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
154
"sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
155
"sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
156
"sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
157
"sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
158
"sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
159
"sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
160
"sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
161
"sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
162
"sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
163
"sk_lock-AF_NFC" , "sk_lock-AF_MAX"
165
static const char *const af_family_slock_key_strings[AF_MAX+1] = {
166
"slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
167
"slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
168
"slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
169
"slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
170
"slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
171
"slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
172
"slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
173
"slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
174
"slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
175
"slock-27" , "slock-28" , "slock-AF_CAN" ,
176
"slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
177
"slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
178
"slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
179
"slock-AF_NFC" , "slock-AF_MAX"
181
static const char *const af_family_clock_key_strings[AF_MAX+1] = {
182
"clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
183
"clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
184
"clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
185
"clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
186
"clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
187
"clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
188
"clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
189
"clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
190
"clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
191
"clock-27" , "clock-28" , "clock-AF_CAN" ,
192
"clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
193
"clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
194
"clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
195
"clock-AF_NFC" , "clock-AF_MAX"
199
* sk_callback_lock locking rules are per-address-family,
200
* so split the lock classes by using a per-AF key:
202
static struct lock_class_key af_callback_keys[AF_MAX];
204
/* Take into consideration the size of the struct sk_buff overhead in the
205
* determination of these values, since that is non-constant across
206
* platforms. This makes socket queueing behavior and performance
207
* not depend upon such differences.
209
#define _SK_MEM_PACKETS 256
210
#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
211
#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
212
#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
214
/* Run time adjustable parameters. */
215
__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
216
__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
217
__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
218
__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
220
/* Maximal space eaten by iovec or ancillary data plus some space */
221
int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
222
EXPORT_SYMBOL(sysctl_optmem_max);
224
#if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
225
int net_cls_subsys_id = -1;
226
EXPORT_SYMBOL_GPL(net_cls_subsys_id);
229
static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
233
if (optlen < sizeof(tv))
235
if (copy_from_user(&tv, optval, sizeof(tv)))
237
if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
241
static int warned __read_mostly;
244
if (warned < 10 && net_ratelimit()) {
246
printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
247
"tries to set negative timeout\n",
248
current->comm, task_pid_nr(current));
252
*timeo_p = MAX_SCHEDULE_TIMEOUT;
253
if (tv.tv_sec == 0 && tv.tv_usec == 0)
255
if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
256
*timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
260
static void sock_warn_obsolete_bsdism(const char *name)
263
static char warncomm[TASK_COMM_LEN];
264
if (strcmp(warncomm, current->comm) && warned < 5) {
265
strcpy(warncomm, current->comm);
266
printk(KERN_WARNING "process `%s' is using obsolete "
267
"%s SO_BSDCOMPAT\n", warncomm, name);
272
static void sock_disable_timestamp(struct sock *sk, int flag)
274
if (sock_flag(sk, flag)) {
275
sock_reset_flag(sk, flag);
276
if (!sock_flag(sk, SOCK_TIMESTAMP) &&
277
!sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
278
net_disable_timestamp();
284
int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
289
struct sk_buff_head *list = &sk->sk_receive_queue;
291
if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
292
atomic_inc(&sk->sk_drops);
293
trace_sock_rcvqueue_full(sk, skb);
297
err = sk_filter(sk, skb);
301
if (!sk_rmem_schedule(sk, skb->truesize)) {
302
atomic_inc(&sk->sk_drops);
307
skb_set_owner_r(skb, sk);
309
/* Cache the SKB length before we tack it onto the receive
310
* queue. Once it is added it no longer belongs to us and
311
* may be freed by other threads of control pulling packets
316
/* we escape from rcu protected region, make sure we dont leak
321
spin_lock_irqsave(&list->lock, flags);
322
skb->dropcount = atomic_read(&sk->sk_drops);
323
__skb_queue_tail(list, skb);
324
spin_unlock_irqrestore(&list->lock, flags);
326
if (!sock_flag(sk, SOCK_DEAD))
327
sk->sk_data_ready(sk, skb_len);
330
EXPORT_SYMBOL(sock_queue_rcv_skb);
332
int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
334
int rc = NET_RX_SUCCESS;
336
if (sk_filter(sk, skb))
337
goto discard_and_relse;
341
if (sk_rcvqueues_full(sk, skb)) {
342
atomic_inc(&sk->sk_drops);
343
goto discard_and_relse;
346
bh_lock_sock_nested(sk);
349
if (!sock_owned_by_user(sk)) {
351
* trylock + unlock semantics:
353
mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
355
rc = sk_backlog_rcv(sk, skb);
357
mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
358
} else if (sk_add_backlog(sk, skb)) {
360
atomic_inc(&sk->sk_drops);
361
goto discard_and_relse;
372
EXPORT_SYMBOL(sk_receive_skb);
374
void sk_reset_txq(struct sock *sk)
376
sk_tx_queue_clear(sk);
378
EXPORT_SYMBOL(sk_reset_txq);
380
struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
382
struct dst_entry *dst = __sk_dst_get(sk);
384
if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
385
sk_tx_queue_clear(sk);
386
RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
393
EXPORT_SYMBOL(__sk_dst_check);
395
struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
397
struct dst_entry *dst = sk_dst_get(sk);
399
if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
407
EXPORT_SYMBOL(sk_dst_check);
409
static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
411
int ret = -ENOPROTOOPT;
412
#ifdef CONFIG_NETDEVICES
413
struct net *net = sock_net(sk);
414
char devname[IFNAMSIZ];
419
if (!capable(CAP_NET_RAW))
426
/* Bind this socket to a particular device like "eth0",
427
* as specified in the passed interface name. If the
428
* name is "" or the option length is zero the socket
431
if (optlen > IFNAMSIZ - 1)
432
optlen = IFNAMSIZ - 1;
433
memset(devname, 0, sizeof(devname));
436
if (copy_from_user(devname, optval, optlen))
440
if (devname[0] != '\0') {
441
struct net_device *dev;
444
dev = dev_get_by_name_rcu(net, devname);
446
index = dev->ifindex;
454
sk->sk_bound_dev_if = index;
466
static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
469
sock_set_flag(sk, bit);
471
sock_reset_flag(sk, bit);
475
* This is meant for all protocols to use and covers goings on
476
* at the socket level. Everything here is generic.
479
int sock_setsockopt(struct socket *sock, int level, int optname,
480
char __user *optval, unsigned int optlen)
482
struct sock *sk = sock->sk;
489
* Options without arguments
492
if (optname == SO_BINDTODEVICE)
493
return sock_bindtodevice(sk, optval, optlen);
495
if (optlen < sizeof(int))
498
if (get_user(val, (int __user *)optval))
501
valbool = val ? 1 : 0;
507
if (val && !capable(CAP_NET_ADMIN))
510
sock_valbool_flag(sk, SOCK_DBG, valbool);
513
sk->sk_reuse = valbool;
522
sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
525
sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
528
/* Don't error on this BSD doesn't and if you think
529
about it this is right. Otherwise apps have to
530
play 'guess the biggest size' games. RCVBUF/SNDBUF
531
are treated in BSD as hints */
533
if (val > sysctl_wmem_max)
534
val = sysctl_wmem_max;
536
sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
537
if ((val * 2) < SOCK_MIN_SNDBUF)
538
sk->sk_sndbuf = SOCK_MIN_SNDBUF;
540
sk->sk_sndbuf = val * 2;
543
* Wake up sending tasks if we
546
sk->sk_write_space(sk);
550
if (!capable(CAP_NET_ADMIN)) {
557
/* Don't error on this BSD doesn't and if you think
558
about it this is right. Otherwise apps have to
559
play 'guess the biggest size' games. RCVBUF/SNDBUF
560
are treated in BSD as hints */
562
if (val > sysctl_rmem_max)
563
val = sysctl_rmem_max;
565
sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
567
* We double it on the way in to account for
568
* "struct sk_buff" etc. overhead. Applications
569
* assume that the SO_RCVBUF setting they make will
570
* allow that much actual data to be received on that
573
* Applications are unaware that "struct sk_buff" and
574
* other overheads allocate from the receive buffer
575
* during socket buffer allocation.
577
* And after considering the possible alternatives,
578
* returning the value we actually used in getsockopt
579
* is the most desirable behavior.
581
if ((val * 2) < SOCK_MIN_RCVBUF)
582
sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
584
sk->sk_rcvbuf = val * 2;
588
if (!capable(CAP_NET_ADMIN)) {
596
if (sk->sk_protocol == IPPROTO_TCP)
597
tcp_set_keepalive(sk, valbool);
599
sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
603
sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
607
sk->sk_no_check = valbool;
611
if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
612
sk->sk_priority = val;
618
if (optlen < sizeof(ling)) {
619
ret = -EINVAL; /* 1003.1g */
622
if (copy_from_user(&ling, optval, sizeof(ling))) {
627
sock_reset_flag(sk, SOCK_LINGER);
629
#if (BITS_PER_LONG == 32)
630
if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
631
sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
634
sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
635
sock_set_flag(sk, SOCK_LINGER);
640
sock_warn_obsolete_bsdism("setsockopt");
645
set_bit(SOCK_PASSCRED, &sock->flags);
647
clear_bit(SOCK_PASSCRED, &sock->flags);
653
if (optname == SO_TIMESTAMP)
654
sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
656
sock_set_flag(sk, SOCK_RCVTSTAMPNS);
657
sock_set_flag(sk, SOCK_RCVTSTAMP);
658
sock_enable_timestamp(sk, SOCK_TIMESTAMP);
660
sock_reset_flag(sk, SOCK_RCVTSTAMP);
661
sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
665
case SO_TIMESTAMPING:
666
if (val & ~SOF_TIMESTAMPING_MASK) {
670
sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
671
val & SOF_TIMESTAMPING_TX_HARDWARE);
672
sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
673
val & SOF_TIMESTAMPING_TX_SOFTWARE);
674
sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
675
val & SOF_TIMESTAMPING_RX_HARDWARE);
676
if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
677
sock_enable_timestamp(sk,
678
SOCK_TIMESTAMPING_RX_SOFTWARE);
680
sock_disable_timestamp(sk,
681
SOCK_TIMESTAMPING_RX_SOFTWARE);
682
sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
683
val & SOF_TIMESTAMPING_SOFTWARE);
684
sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
685
val & SOF_TIMESTAMPING_SYS_HARDWARE);
686
sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
687
val & SOF_TIMESTAMPING_RAW_HARDWARE);
693
sk->sk_rcvlowat = val ? : 1;
697
ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
701
ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
704
case SO_ATTACH_FILTER:
706
if (optlen == sizeof(struct sock_fprog)) {
707
struct sock_fprog fprog;
710
if (copy_from_user(&fprog, optval, sizeof(fprog)))
713
ret = sk_attach_filter(&fprog, sk);
717
case SO_DETACH_FILTER:
718
ret = sk_detach_filter(sk);
723
set_bit(SOCK_PASSSEC, &sock->flags);
725
clear_bit(SOCK_PASSSEC, &sock->flags);
728
if (!capable(CAP_NET_ADMIN))
734
/* We implement the SO_SNDLOWAT etc to
735
not be settable (1003.1g 5.3) */
737
sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
746
EXPORT_SYMBOL(sock_setsockopt);
749
void cred_to_ucred(struct pid *pid, const struct cred *cred,
752
ucred->pid = pid_vnr(pid);
753
ucred->uid = ucred->gid = -1;
755
struct user_namespace *current_ns = current_user_ns();
757
ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
758
ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
761
EXPORT_SYMBOL_GPL(cred_to_ucred);
763
int sock_getsockopt(struct socket *sock, int level, int optname,
764
char __user *optval, int __user *optlen)
766
struct sock *sk = sock->sk;
774
int lv = sizeof(int);
777
if (get_user(len, optlen))
782
memset(&v, 0, sizeof(v));
786
v.val = sock_flag(sk, SOCK_DBG);
790
v.val = sock_flag(sk, SOCK_LOCALROUTE);
794
v.val = !!sock_flag(sk, SOCK_BROADCAST);
798
v.val = sk->sk_sndbuf;
802
v.val = sk->sk_rcvbuf;
806
v.val = sk->sk_reuse;
810
v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
818
v.val = sk->sk_protocol;
822
v.val = sk->sk_family;
826
v.val = -sock_error(sk);
828
v.val = xchg(&sk->sk_err_soft, 0);
832
v.val = !!sock_flag(sk, SOCK_URGINLINE);
836
v.val = sk->sk_no_check;
840
v.val = sk->sk_priority;
845
v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
846
v.ling.l_linger = sk->sk_lingertime / HZ;
850
sock_warn_obsolete_bsdism("getsockopt");
854
v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
855
!sock_flag(sk, SOCK_RCVTSTAMPNS);
859
v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
862
case SO_TIMESTAMPING:
864
if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
865
v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
866
if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
867
v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
868
if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
869
v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
870
if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
871
v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
872
if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
873
v.val |= SOF_TIMESTAMPING_SOFTWARE;
874
if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
875
v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
876
if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
877
v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
881
lv = sizeof(struct timeval);
882
if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
886
v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
887
v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
892
lv = sizeof(struct timeval);
893
if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
897
v.tm.tv_sec = sk->sk_sndtimeo / HZ;
898
v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
903
v.val = sk->sk_rcvlowat;
911
v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
916
struct ucred peercred;
917
if (len > sizeof(peercred))
918
len = sizeof(peercred);
919
cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
920
if (copy_to_user(optval, &peercred, len))
929
if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
933
if (copy_to_user(optval, address, len))
938
/* Dubious BSD thing... Probably nobody even uses it, but
939
* the UNIX standard wants it for whatever reason... -DaveM
942
v.val = sk->sk_state == TCP_LISTEN;
946
v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
950
return security_socket_getpeersec_stream(sock, optval, optlen, len);
957
v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
966
if (copy_to_user(optval, &v, len))
969
if (put_user(len, optlen))
975
* Initialize an sk_lock.
977
* (We also register the sk_lock with the lock validator.)
979
static inline void sock_lock_init(struct sock *sk)
981
sock_lock_init_class_and_name(sk,
982
af_family_slock_key_strings[sk->sk_family],
983
af_family_slock_keys + sk->sk_family,
984
af_family_key_strings[sk->sk_family],
985
af_family_keys + sk->sk_family);
989
* Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
990
* even temporarly, because of RCU lookups. sk_node should also be left as is.
991
* We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
993
static void sock_copy(struct sock *nsk, const struct sock *osk)
995
#ifdef CONFIG_SECURITY_NETWORK
996
void *sptr = nsk->sk_security;
998
memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1000
memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1001
osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1003
#ifdef CONFIG_SECURITY_NETWORK
1004
nsk->sk_security = sptr;
1005
security_sk_clone(osk, nsk);
1010
* caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1011
* un-modified. Special care is taken when initializing object to zero.
1013
static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1015
if (offsetof(struct sock, sk_node.next) != 0)
1016
memset(sk, 0, offsetof(struct sock, sk_node.next));
1017
memset(&sk->sk_node.pprev, 0,
1018
size - offsetof(struct sock, sk_node.pprev));
1021
void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1023
unsigned long nulls1, nulls2;
1025
nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1026
nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1027
if (nulls1 > nulls2)
1028
swap(nulls1, nulls2);
1031
memset((char *)sk, 0, nulls1);
1032
memset((char *)sk + nulls1 + sizeof(void *), 0,
1033
nulls2 - nulls1 - sizeof(void *));
1034
memset((char *)sk + nulls2 + sizeof(void *), 0,
1035
size - nulls2 - sizeof(void *));
1037
EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1039
static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1043
struct kmem_cache *slab;
1047
sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1050
if (priority & __GFP_ZERO) {
1052
prot->clear_sk(sk, prot->obj_size);
1054
sk_prot_clear_nulls(sk, prot->obj_size);
1057
sk = kmalloc(prot->obj_size, priority);
1060
kmemcheck_annotate_bitfield(sk, flags);
1062
if (security_sk_alloc(sk, family, priority))
1065
if (!try_module_get(prot->owner))
1067
sk_tx_queue_clear(sk);
1073
security_sk_free(sk);
1076
kmem_cache_free(slab, sk);
1082
static void sk_prot_free(struct proto *prot, struct sock *sk)
1084
struct kmem_cache *slab;
1085
struct module *owner;
1087
owner = prot->owner;
1090
security_sk_free(sk);
1092
kmem_cache_free(slab, sk);
1098
#ifdef CONFIG_CGROUPS
1099
void sock_update_classid(struct sock *sk)
1103
rcu_read_lock(); /* doing current task, which cannot vanish. */
1104
classid = task_cls_classid(current);
1106
if (classid && classid != sk->sk_classid)
1107
sk->sk_classid = classid;
1109
EXPORT_SYMBOL(sock_update_classid);
1113
* sk_alloc - All socket objects are allocated here
1114
* @net: the applicable net namespace
1115
* @family: protocol family
1116
* @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1117
* @prot: struct proto associated with this new sock instance
1119
struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1124
sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1126
sk->sk_family = family;
1128
* See comment in struct sock definition to understand
1129
* why we need sk_prot_creator -acme
1131
sk->sk_prot = sk->sk_prot_creator = prot;
1133
sock_net_set(sk, get_net(net));
1134
atomic_set(&sk->sk_wmem_alloc, 1);
1136
sock_update_classid(sk);
1141
EXPORT_SYMBOL(sk_alloc);
1143
static void __sk_free(struct sock *sk)
1145
struct sk_filter *filter;
1147
if (sk->sk_destruct)
1148
sk->sk_destruct(sk);
1150
filter = rcu_dereference_check(sk->sk_filter,
1151
atomic_read(&sk->sk_wmem_alloc) == 0);
1153
sk_filter_uncharge(sk, filter);
1154
RCU_INIT_POINTER(sk->sk_filter, NULL);
1157
sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1158
sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1160
if (atomic_read(&sk->sk_omem_alloc))
1161
printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1162
__func__, atomic_read(&sk->sk_omem_alloc));
1164
if (sk->sk_peer_cred)
1165
put_cred(sk->sk_peer_cred);
1166
put_pid(sk->sk_peer_pid);
1167
put_net(sock_net(sk));
1168
sk_prot_free(sk->sk_prot_creator, sk);
1171
void sk_free(struct sock *sk)
1174
* We subtract one from sk_wmem_alloc and can know if
1175
* some packets are still in some tx queue.
1176
* If not null, sock_wfree() will call __sk_free(sk) later
1178
if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1181
EXPORT_SYMBOL(sk_free);
1184
* Last sock_put should drop reference to sk->sk_net. It has already
1185
* been dropped in sk_change_net. Taking reference to stopping namespace
1187
* Take reference to a socket to remove it from hash _alive_ and after that
1188
* destroy it in the context of init_net.
1190
void sk_release_kernel(struct sock *sk)
1192
if (sk == NULL || sk->sk_socket == NULL)
1196
sock_release(sk->sk_socket);
1197
release_net(sock_net(sk));
1198
sock_net_set(sk, get_net(&init_net));
1201
EXPORT_SYMBOL(sk_release_kernel);
1203
struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1207
newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1208
if (newsk != NULL) {
1209
struct sk_filter *filter;
1211
sock_copy(newsk, sk);
1214
get_net(sock_net(newsk));
1215
sk_node_init(&newsk->sk_node);
1216
sock_lock_init(newsk);
1217
bh_lock_sock(newsk);
1218
newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1219
newsk->sk_backlog.len = 0;
1221
atomic_set(&newsk->sk_rmem_alloc, 0);
1223
* sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1225
atomic_set(&newsk->sk_wmem_alloc, 1);
1226
atomic_set(&newsk->sk_omem_alloc, 0);
1227
skb_queue_head_init(&newsk->sk_receive_queue);
1228
skb_queue_head_init(&newsk->sk_write_queue);
1229
#ifdef CONFIG_NET_DMA
1230
skb_queue_head_init(&newsk->sk_async_wait_queue);
1233
spin_lock_init(&newsk->sk_dst_lock);
1234
rwlock_init(&newsk->sk_callback_lock);
1235
lockdep_set_class_and_name(&newsk->sk_callback_lock,
1236
af_callback_keys + newsk->sk_family,
1237
af_family_clock_key_strings[newsk->sk_family]);
1239
newsk->sk_dst_cache = NULL;
1240
newsk->sk_wmem_queued = 0;
1241
newsk->sk_forward_alloc = 0;
1242
newsk->sk_send_head = NULL;
1243
newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1245
sock_reset_flag(newsk, SOCK_DONE);
1246
skb_queue_head_init(&newsk->sk_error_queue);
1248
filter = rcu_dereference_protected(newsk->sk_filter, 1);
1250
sk_filter_charge(newsk, filter);
1252
if (unlikely(xfrm_sk_clone_policy(newsk))) {
1253
/* It is still raw copy of parent, so invalidate
1254
* destructor and make plain sk_free() */
1255
newsk->sk_destruct = NULL;
1256
bh_unlock_sock(newsk);
1263
newsk->sk_priority = 0;
1265
* Before updating sk_refcnt, we must commit prior changes to memory
1266
* (Documentation/RCU/rculist_nulls.txt for details)
1269
atomic_set(&newsk->sk_refcnt, 2);
1272
* Increment the counter in the same struct proto as the master
1273
* sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1274
* is the same as sk->sk_prot->socks, as this field was copied
1277
* This _changes_ the previous behaviour, where
1278
* tcp_create_openreq_child always was incrementing the
1279
* equivalent to tcp_prot->socks (inet_sock_nr), so this have
1280
* to be taken into account in all callers. -acme
1282
sk_refcnt_debug_inc(newsk);
1283
sk_set_socket(newsk, NULL);
1284
newsk->sk_wq = NULL;
1286
if (newsk->sk_prot->sockets_allocated)
1287
percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1289
if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1290
sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1291
net_enable_timestamp();
1296
EXPORT_SYMBOL_GPL(sk_clone);
1298
void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1300
__sk_dst_set(sk, dst);
1301
sk->sk_route_caps = dst->dev->features;
1302
if (sk->sk_route_caps & NETIF_F_GSO)
1303
sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1304
sk->sk_route_caps &= ~sk->sk_route_nocaps;
1305
if (sk_can_gso(sk)) {
1306
if (dst->header_len) {
1307
sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1309
sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1310
sk->sk_gso_max_size = dst->dev->gso_max_size;
1314
EXPORT_SYMBOL_GPL(sk_setup_caps);
1316
void __init sk_init(void)
1318
if (totalram_pages <= 4096) {
1319
sysctl_wmem_max = 32767;
1320
sysctl_rmem_max = 32767;
1321
sysctl_wmem_default = 32767;
1322
sysctl_rmem_default = 32767;
1323
} else if (totalram_pages >= 131072) {
1324
sysctl_wmem_max = 131071;
1325
sysctl_rmem_max = 131071;
1330
* Simple resource managers for sockets.
1335
* Write buffer destructor automatically called from kfree_skb.
1337
void sock_wfree(struct sk_buff *skb)
1339
struct sock *sk = skb->sk;
1340
unsigned int len = skb->truesize;
1342
if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1344
* Keep a reference on sk_wmem_alloc, this will be released
1345
* after sk_write_space() call
1347
atomic_sub(len - 1, &sk->sk_wmem_alloc);
1348
sk->sk_write_space(sk);
1352
* if sk_wmem_alloc reaches 0, we must finish what sk_free()
1353
* could not do because of in-flight packets
1355
if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1358
EXPORT_SYMBOL(sock_wfree);
1361
* Read buffer destructor automatically called from kfree_skb.
1363
void sock_rfree(struct sk_buff *skb)
1365
struct sock *sk = skb->sk;
1366
unsigned int len = skb->truesize;
1368
atomic_sub(len, &sk->sk_rmem_alloc);
1369
sk_mem_uncharge(sk, len);
1371
EXPORT_SYMBOL(sock_rfree);
1374
int sock_i_uid(struct sock *sk)
1378
read_lock_bh(&sk->sk_callback_lock);
1379
uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1380
read_unlock_bh(&sk->sk_callback_lock);
1383
EXPORT_SYMBOL(sock_i_uid);
1385
unsigned long sock_i_ino(struct sock *sk)
1389
read_lock_bh(&sk->sk_callback_lock);
1390
ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1391
read_unlock_bh(&sk->sk_callback_lock);
1394
EXPORT_SYMBOL(sock_i_ino);
1397
* Allocate a skb from the socket's send buffer.
1399
struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1402
if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1403
struct sk_buff *skb = alloc_skb(size, priority);
1405
skb_set_owner_w(skb, sk);
1411
EXPORT_SYMBOL(sock_wmalloc);
1414
* Allocate a skb from the socket's receive buffer.
1416
struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1419
if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1420
struct sk_buff *skb = alloc_skb(size, priority);
1422
skb_set_owner_r(skb, sk);
1430
* Allocate a memory block from the socket's option memory buffer.
1432
void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1434
if ((unsigned)size <= sysctl_optmem_max &&
1435
atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1437
/* First do the add, to avoid the race if kmalloc
1440
atomic_add(size, &sk->sk_omem_alloc);
1441
mem = kmalloc(size, priority);
1444
atomic_sub(size, &sk->sk_omem_alloc);
1448
EXPORT_SYMBOL(sock_kmalloc);
1451
* Free an option memory block.
1453
void sock_kfree_s(struct sock *sk, void *mem, int size)
1456
atomic_sub(size, &sk->sk_omem_alloc);
1458
EXPORT_SYMBOL(sock_kfree_s);
1460
/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1461
I think, these locks should be removed for datagram sockets.
1463
static long sock_wait_for_wmem(struct sock *sk, long timeo)
1467
clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1471
if (signal_pending(current))
1473
set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1474
prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1475
if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1477
if (sk->sk_shutdown & SEND_SHUTDOWN)
1481
timeo = schedule_timeout(timeo);
1483
finish_wait(sk_sleep(sk), &wait);
1489
* Generic send/receive buffer handlers
1492
struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1493
unsigned long data_len, int noblock,
1496
struct sk_buff *skb;
1501
gfp_mask = sk->sk_allocation;
1502
if (gfp_mask & __GFP_WAIT)
1503
gfp_mask |= __GFP_REPEAT;
1505
timeo = sock_sndtimeo(sk, noblock);
1507
err = sock_error(sk);
1512
if (sk->sk_shutdown & SEND_SHUTDOWN)
1515
if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1516
skb = alloc_skb(header_len, gfp_mask);
1521
/* No pages, we're done... */
1525
npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1526
skb->truesize += data_len;
1527
skb_shinfo(skb)->nr_frags = npages;
1528
for (i = 0; i < npages; i++) {
1531
page = alloc_pages(sk->sk_allocation, 0);
1534
skb_shinfo(skb)->nr_frags = i;
1539
__skb_fill_page_desc(skb, i,
1541
(data_len >= PAGE_SIZE ?
1544
data_len -= PAGE_SIZE;
1547
/* Full success... */
1553
set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1554
set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1558
if (signal_pending(current))
1560
timeo = sock_wait_for_wmem(sk, timeo);
1563
skb_set_owner_w(skb, sk);
1567
err = sock_intr_errno(timeo);
1572
EXPORT_SYMBOL(sock_alloc_send_pskb);
1574
struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1575
int noblock, int *errcode)
1577
return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1579
EXPORT_SYMBOL(sock_alloc_send_skb);
1581
static void __lock_sock(struct sock *sk)
1582
__releases(&sk->sk_lock.slock)
1583
__acquires(&sk->sk_lock.slock)
1588
prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1589
TASK_UNINTERRUPTIBLE);
1590
spin_unlock_bh(&sk->sk_lock.slock);
1592
spin_lock_bh(&sk->sk_lock.slock);
1593
if (!sock_owned_by_user(sk))
1596
finish_wait(&sk->sk_lock.wq, &wait);
1599
static void __release_sock(struct sock *sk)
1600
__releases(&sk->sk_lock.slock)
1601
__acquires(&sk->sk_lock.slock)
1603
struct sk_buff *skb = sk->sk_backlog.head;
1606
sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1610
struct sk_buff *next = skb->next;
1612
WARN_ON_ONCE(skb_dst_is_noref(skb));
1614
sk_backlog_rcv(sk, skb);
1617
* We are in process context here with softirqs
1618
* disabled, use cond_resched_softirq() to preempt.
1619
* This is safe to do because we've taken the backlog
1622
cond_resched_softirq();
1625
} while (skb != NULL);
1628
} while ((skb = sk->sk_backlog.head) != NULL);
1631
* Doing the zeroing here guarantee we can not loop forever
1632
* while a wild producer attempts to flood us.
1634
sk->sk_backlog.len = 0;
1638
* sk_wait_data - wait for data to arrive at sk_receive_queue
1639
* @sk: sock to wait on
1640
* @timeo: for how long
1642
* Now socket state including sk->sk_err is changed only under lock,
1643
* hence we may omit checks after joining wait queue.
1644
* We check receive queue before schedule() only as optimization;
1645
* it is very likely that release_sock() added new data.
1647
int sk_wait_data(struct sock *sk, long *timeo)
1652
prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1653
set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1654
rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1655
clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1656
finish_wait(sk_sleep(sk), &wait);
1659
EXPORT_SYMBOL(sk_wait_data);
1662
* __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1664
* @size: memory size to allocate
1665
* @kind: allocation type
1667
* If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1668
* rmem allocation. This function assumes that protocols which have
1669
* memory_pressure use sk_wmem_queued as write buffer accounting.
1671
int __sk_mem_schedule(struct sock *sk, int size, int kind)
1673
struct proto *prot = sk->sk_prot;
1674
int amt = sk_mem_pages(size);
1677
sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1678
allocated = atomic_long_add_return(amt, prot->memory_allocated);
1681
if (allocated <= prot->sysctl_mem[0]) {
1682
if (prot->memory_pressure && *prot->memory_pressure)
1683
*prot->memory_pressure = 0;
1687
/* Under pressure. */
1688
if (allocated > prot->sysctl_mem[1])
1689
if (prot->enter_memory_pressure)
1690
prot->enter_memory_pressure(sk);
1692
/* Over hard limit. */
1693
if (allocated > prot->sysctl_mem[2])
1694
goto suppress_allocation;
1696
/* guarantee minimum buffer size under pressure */
1697
if (kind == SK_MEM_RECV) {
1698
if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1700
} else { /* SK_MEM_SEND */
1701
if (sk->sk_type == SOCK_STREAM) {
1702
if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1704
} else if (atomic_read(&sk->sk_wmem_alloc) <
1705
prot->sysctl_wmem[0])
1709
if (prot->memory_pressure) {
1712
if (!*prot->memory_pressure)
1714
alloc = percpu_counter_read_positive(prot->sockets_allocated);
1715
if (prot->sysctl_mem[2] > alloc *
1716
sk_mem_pages(sk->sk_wmem_queued +
1717
atomic_read(&sk->sk_rmem_alloc) +
1718
sk->sk_forward_alloc))
1722
suppress_allocation:
1724
if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1725
sk_stream_moderate_sndbuf(sk);
1727
/* Fail only if socket is _under_ its sndbuf.
1728
* In this case we cannot block, so that we have to fail.
1730
if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1734
trace_sock_exceed_buf_limit(sk, prot, allocated);
1736
/* Alas. Undo changes. */
1737
sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1738
atomic_long_sub(amt, prot->memory_allocated);
1741
EXPORT_SYMBOL(__sk_mem_schedule);
1744
* __sk_reclaim - reclaim memory_allocated
1747
void __sk_mem_reclaim(struct sock *sk)
1749
struct proto *prot = sk->sk_prot;
1751
atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1752
prot->memory_allocated);
1753
sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1755
if (prot->memory_pressure && *prot->memory_pressure &&
1756
(atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1757
*prot->memory_pressure = 0;
1759
EXPORT_SYMBOL(__sk_mem_reclaim);
1763
* Set of default routines for initialising struct proto_ops when
1764
* the protocol does not support a particular function. In certain
1765
* cases where it makes no sense for a protocol to have a "do nothing"
1766
* function, some default processing is provided.
1769
int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1773
EXPORT_SYMBOL(sock_no_bind);
1775
int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1780
EXPORT_SYMBOL(sock_no_connect);
1782
int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1786
EXPORT_SYMBOL(sock_no_socketpair);
1788
int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1792
EXPORT_SYMBOL(sock_no_accept);
1794
int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1799
EXPORT_SYMBOL(sock_no_getname);
1801
unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1805
EXPORT_SYMBOL(sock_no_poll);
1807
int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1811
EXPORT_SYMBOL(sock_no_ioctl);
1813
int sock_no_listen(struct socket *sock, int backlog)
1817
EXPORT_SYMBOL(sock_no_listen);
1819
int sock_no_shutdown(struct socket *sock, int how)
1823
EXPORT_SYMBOL(sock_no_shutdown);
1825
int sock_no_setsockopt(struct socket *sock, int level, int optname,
1826
char __user *optval, unsigned int optlen)
1830
EXPORT_SYMBOL(sock_no_setsockopt);
1832
int sock_no_getsockopt(struct socket *sock, int level, int optname,
1833
char __user *optval, int __user *optlen)
1837
EXPORT_SYMBOL(sock_no_getsockopt);
1839
int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1844
EXPORT_SYMBOL(sock_no_sendmsg);
1846
int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1847
size_t len, int flags)
1851
EXPORT_SYMBOL(sock_no_recvmsg);
1853
int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1855
/* Mirror missing mmap method error code */
1858
EXPORT_SYMBOL(sock_no_mmap);
1860
ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1863
struct msghdr msg = {.msg_flags = flags};
1865
char *kaddr = kmap(page);
1866
iov.iov_base = kaddr + offset;
1868
res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1872
EXPORT_SYMBOL(sock_no_sendpage);
1875
* Default Socket Callbacks
1878
static void sock_def_wakeup(struct sock *sk)
1880
struct socket_wq *wq;
1883
wq = rcu_dereference(sk->sk_wq);
1884
if (wq_has_sleeper(wq))
1885
wake_up_interruptible_all(&wq->wait);
1889
static void sock_def_error_report(struct sock *sk)
1891
struct socket_wq *wq;
1894
wq = rcu_dereference(sk->sk_wq);
1895
if (wq_has_sleeper(wq))
1896
wake_up_interruptible_poll(&wq->wait, POLLERR);
1897
sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1901
static void sock_def_readable(struct sock *sk, int len)
1903
struct socket_wq *wq;
1906
wq = rcu_dereference(sk->sk_wq);
1907
if (wq_has_sleeper(wq))
1908
wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1909
POLLRDNORM | POLLRDBAND);
1910
sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1914
static void sock_def_write_space(struct sock *sk)
1916
struct socket_wq *wq;
1920
/* Do not wake up a writer until he can make "significant"
1923
if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1924
wq = rcu_dereference(sk->sk_wq);
1925
if (wq_has_sleeper(wq))
1926
wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1927
POLLWRNORM | POLLWRBAND);
1929
/* Should agree with poll, otherwise some programs break */
1930
if (sock_writeable(sk))
1931
sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1937
static void sock_def_destruct(struct sock *sk)
1939
kfree(sk->sk_protinfo);
1942
void sk_send_sigurg(struct sock *sk)
1944
if (sk->sk_socket && sk->sk_socket->file)
1945
if (send_sigurg(&sk->sk_socket->file->f_owner))
1946
sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1948
EXPORT_SYMBOL(sk_send_sigurg);
1950
void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1951
unsigned long expires)
1953
if (!mod_timer(timer, expires))
1956
EXPORT_SYMBOL(sk_reset_timer);
1958
void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1960
if (timer_pending(timer) && del_timer(timer))
1963
EXPORT_SYMBOL(sk_stop_timer);
1965
void sock_init_data(struct socket *sock, struct sock *sk)
1967
skb_queue_head_init(&sk->sk_receive_queue);
1968
skb_queue_head_init(&sk->sk_write_queue);
1969
skb_queue_head_init(&sk->sk_error_queue);
1970
#ifdef CONFIG_NET_DMA
1971
skb_queue_head_init(&sk->sk_async_wait_queue);
1974
sk->sk_send_head = NULL;
1976
init_timer(&sk->sk_timer);
1978
sk->sk_allocation = GFP_KERNEL;
1979
sk->sk_rcvbuf = sysctl_rmem_default;
1980
sk->sk_sndbuf = sysctl_wmem_default;
1981
sk->sk_state = TCP_CLOSE;
1982
sk_set_socket(sk, sock);
1984
sock_set_flag(sk, SOCK_ZAPPED);
1987
sk->sk_type = sock->type;
1988
sk->sk_wq = sock->wq;
1993
spin_lock_init(&sk->sk_dst_lock);
1994
rwlock_init(&sk->sk_callback_lock);
1995
lockdep_set_class_and_name(&sk->sk_callback_lock,
1996
af_callback_keys + sk->sk_family,
1997
af_family_clock_key_strings[sk->sk_family]);
1999
sk->sk_state_change = sock_def_wakeup;
2000
sk->sk_data_ready = sock_def_readable;
2001
sk->sk_write_space = sock_def_write_space;
2002
sk->sk_error_report = sock_def_error_report;
2003
sk->sk_destruct = sock_def_destruct;
2005
sk->sk_sndmsg_page = NULL;
2006
sk->sk_sndmsg_off = 0;
2008
sk->sk_peer_pid = NULL;
2009
sk->sk_peer_cred = NULL;
2010
sk->sk_write_pending = 0;
2011
sk->sk_rcvlowat = 1;
2012
sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2013
sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2015
sk->sk_stamp = ktime_set(-1L, 0);
2018
* Before updating sk_refcnt, we must commit prior changes to memory
2019
* (Documentation/RCU/rculist_nulls.txt for details)
2022
atomic_set(&sk->sk_refcnt, 1);
2023
atomic_set(&sk->sk_drops, 0);
2025
EXPORT_SYMBOL(sock_init_data);
2027
void lock_sock_nested(struct sock *sk, int subclass)
2030
spin_lock_bh(&sk->sk_lock.slock);
2031
if (sk->sk_lock.owned)
2033
sk->sk_lock.owned = 1;
2034
spin_unlock(&sk->sk_lock.slock);
2036
* The sk_lock has mutex_lock() semantics here:
2038
mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2041
EXPORT_SYMBOL(lock_sock_nested);
2043
void release_sock(struct sock *sk)
2046
* The sk_lock has mutex_unlock() semantics:
2048
mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2050
spin_lock_bh(&sk->sk_lock.slock);
2051
if (sk->sk_backlog.tail)
2053
sk->sk_lock.owned = 0;
2054
if (waitqueue_active(&sk->sk_lock.wq))
2055
wake_up(&sk->sk_lock.wq);
2056
spin_unlock_bh(&sk->sk_lock.slock);
2058
EXPORT_SYMBOL(release_sock);
2061
* lock_sock_fast - fast version of lock_sock
2064
* This version should be used for very small section, where process wont block
2065
* return false if fast path is taken
2066
* sk_lock.slock locked, owned = 0, BH disabled
2067
* return true if slow path is taken
2068
* sk_lock.slock unlocked, owned = 1, BH enabled
2070
bool lock_sock_fast(struct sock *sk)
2073
spin_lock_bh(&sk->sk_lock.slock);
2075
if (!sk->sk_lock.owned)
2077
* Note : We must disable BH
2082
sk->sk_lock.owned = 1;
2083
spin_unlock(&sk->sk_lock.slock);
2085
* The sk_lock has mutex_lock() semantics here:
2087
mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2091
EXPORT_SYMBOL(lock_sock_fast);
2093
int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2096
if (!sock_flag(sk, SOCK_TIMESTAMP))
2097
sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2098
tv = ktime_to_timeval(sk->sk_stamp);
2099
if (tv.tv_sec == -1)
2101
if (tv.tv_sec == 0) {
2102
sk->sk_stamp = ktime_get_real();
2103
tv = ktime_to_timeval(sk->sk_stamp);
2105
return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2107
EXPORT_SYMBOL(sock_get_timestamp);
2109
int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2112
if (!sock_flag(sk, SOCK_TIMESTAMP))
2113
sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2114
ts = ktime_to_timespec(sk->sk_stamp);
2115
if (ts.tv_sec == -1)
2117
if (ts.tv_sec == 0) {
2118
sk->sk_stamp = ktime_get_real();
2119
ts = ktime_to_timespec(sk->sk_stamp);
2121
return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2123
EXPORT_SYMBOL(sock_get_timestampns);
2125
void sock_enable_timestamp(struct sock *sk, int flag)
2127
if (!sock_flag(sk, flag)) {
2128
sock_set_flag(sk, flag);
2130
* we just set one of the two flags which require net
2131
* time stamping, but time stamping might have been on
2132
* already because of the other one
2135
flag == SOCK_TIMESTAMP ?
2136
SOCK_TIMESTAMPING_RX_SOFTWARE :
2138
net_enable_timestamp();
2143
* Get a socket option on an socket.
2145
* FIX: POSIX 1003.1g is very ambiguous here. It states that
2146
* asynchronous errors should be reported by getsockopt. We assume
2147
* this means if you specify SO_ERROR (otherwise whats the point of it).
2149
int sock_common_getsockopt(struct socket *sock, int level, int optname,
2150
char __user *optval, int __user *optlen)
2152
struct sock *sk = sock->sk;
2154
return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2156
EXPORT_SYMBOL(sock_common_getsockopt);
2158
#ifdef CONFIG_COMPAT
2159
int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2160
char __user *optval, int __user *optlen)
2162
struct sock *sk = sock->sk;
2164
if (sk->sk_prot->compat_getsockopt != NULL)
2165
return sk->sk_prot->compat_getsockopt(sk, level, optname,
2167
return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2169
EXPORT_SYMBOL(compat_sock_common_getsockopt);
2172
int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2173
struct msghdr *msg, size_t size, int flags)
2175
struct sock *sk = sock->sk;
2179
err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2180
flags & ~MSG_DONTWAIT, &addr_len);
2182
msg->msg_namelen = addr_len;
2185
EXPORT_SYMBOL(sock_common_recvmsg);
2188
* Set socket options on an inet socket.
2190
int sock_common_setsockopt(struct socket *sock, int level, int optname,
2191
char __user *optval, unsigned int optlen)
2193
struct sock *sk = sock->sk;
2195
return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2197
EXPORT_SYMBOL(sock_common_setsockopt);
2199
#ifdef CONFIG_COMPAT
2200
int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2201
char __user *optval, unsigned int optlen)
2203
struct sock *sk = sock->sk;
2205
if (sk->sk_prot->compat_setsockopt != NULL)
2206
return sk->sk_prot->compat_setsockopt(sk, level, optname,
2208
return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2210
EXPORT_SYMBOL(compat_sock_common_setsockopt);
2213
void sk_common_release(struct sock *sk)
2215
if (sk->sk_prot->destroy)
2216
sk->sk_prot->destroy(sk);
2219
* Observation: when sock_common_release is called, processes have
2220
* no access to socket. But net still has.
2221
* Step one, detach it from networking:
2223
* A. Remove from hash tables.
2226
sk->sk_prot->unhash(sk);
2229
* In this point socket cannot receive new packets, but it is possible
2230
* that some packets are in flight because some CPU runs receiver and
2231
* did hash table lookup before we unhashed socket. They will achieve
2232
* receive queue and will be purged by socket destructor.
2234
* Also we still have packets pending on receive queue and probably,
2235
* our own packets waiting in device queues. sock_destroy will drain
2236
* receive queue, but transmitted packets will delay socket destruction
2237
* until the last reference will be released.
2242
xfrm_sk_free_policy(sk);
2244
sk_refcnt_debug_release(sk);
2247
EXPORT_SYMBOL(sk_common_release);
2249
static DEFINE_RWLOCK(proto_list_lock);
2250
static LIST_HEAD(proto_list);
2252
#ifdef CONFIG_PROC_FS
2253
#define PROTO_INUSE_NR 64 /* should be enough for the first time */
2255
int val[PROTO_INUSE_NR];
2258
static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2260
#ifdef CONFIG_NET_NS
2261
void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2263
__this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2265
EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2267
int sock_prot_inuse_get(struct net *net, struct proto *prot)
2269
int cpu, idx = prot->inuse_idx;
2272
for_each_possible_cpu(cpu)
2273
res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2275
return res >= 0 ? res : 0;
2277
EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2279
static int __net_init sock_inuse_init_net(struct net *net)
2281
net->core.inuse = alloc_percpu(struct prot_inuse);
2282
return net->core.inuse ? 0 : -ENOMEM;
2285
static void __net_exit sock_inuse_exit_net(struct net *net)
2287
free_percpu(net->core.inuse);
2290
static struct pernet_operations net_inuse_ops = {
2291
.init = sock_inuse_init_net,
2292
.exit = sock_inuse_exit_net,
2295
static __init int net_inuse_init(void)
2297
if (register_pernet_subsys(&net_inuse_ops))
2298
panic("Cannot initialize net inuse counters");
2303
core_initcall(net_inuse_init);
2305
static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2307
void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2309
__this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2311
EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2313
int sock_prot_inuse_get(struct net *net, struct proto *prot)
2315
int cpu, idx = prot->inuse_idx;
2318
for_each_possible_cpu(cpu)
2319
res += per_cpu(prot_inuse, cpu).val[idx];
2321
return res >= 0 ? res : 0;
2323
EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2326
static void assign_proto_idx(struct proto *prot)
2328
prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2330
if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2331
printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2335
set_bit(prot->inuse_idx, proto_inuse_idx);
2338
static void release_proto_idx(struct proto *prot)
2340
if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2341
clear_bit(prot->inuse_idx, proto_inuse_idx);
2344
static inline void assign_proto_idx(struct proto *prot)
2348
static inline void release_proto_idx(struct proto *prot)
2353
int proto_register(struct proto *prot, int alloc_slab)
2356
prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2357
SLAB_HWCACHE_ALIGN | prot->slab_flags,
2360
if (prot->slab == NULL) {
2361
printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2366
if (prot->rsk_prot != NULL) {
2367
prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2368
if (prot->rsk_prot->slab_name == NULL)
2369
goto out_free_sock_slab;
2371
prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2372
prot->rsk_prot->obj_size, 0,
2373
SLAB_HWCACHE_ALIGN, NULL);
2375
if (prot->rsk_prot->slab == NULL) {
2376
printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2378
goto out_free_request_sock_slab_name;
2382
if (prot->twsk_prot != NULL) {
2383
prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2385
if (prot->twsk_prot->twsk_slab_name == NULL)
2386
goto out_free_request_sock_slab;
2388
prot->twsk_prot->twsk_slab =
2389
kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2390
prot->twsk_prot->twsk_obj_size,
2392
SLAB_HWCACHE_ALIGN |
2395
if (prot->twsk_prot->twsk_slab == NULL)
2396
goto out_free_timewait_sock_slab_name;
2400
write_lock(&proto_list_lock);
2401
list_add(&prot->node, &proto_list);
2402
assign_proto_idx(prot);
2403
write_unlock(&proto_list_lock);
2406
out_free_timewait_sock_slab_name:
2407
kfree(prot->twsk_prot->twsk_slab_name);
2408
out_free_request_sock_slab:
2409
if (prot->rsk_prot && prot->rsk_prot->slab) {
2410
kmem_cache_destroy(prot->rsk_prot->slab);
2411
prot->rsk_prot->slab = NULL;
2413
out_free_request_sock_slab_name:
2415
kfree(prot->rsk_prot->slab_name);
2417
kmem_cache_destroy(prot->slab);
2422
EXPORT_SYMBOL(proto_register);
2424
void proto_unregister(struct proto *prot)
2426
write_lock(&proto_list_lock);
2427
release_proto_idx(prot);
2428
list_del(&prot->node);
2429
write_unlock(&proto_list_lock);
2431
if (prot->slab != NULL) {
2432
kmem_cache_destroy(prot->slab);
2436
if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2437
kmem_cache_destroy(prot->rsk_prot->slab);
2438
kfree(prot->rsk_prot->slab_name);
2439
prot->rsk_prot->slab = NULL;
2442
if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2443
kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2444
kfree(prot->twsk_prot->twsk_slab_name);
2445
prot->twsk_prot->twsk_slab = NULL;
2448
EXPORT_SYMBOL(proto_unregister);
2450
#ifdef CONFIG_PROC_FS
2451
static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2452
__acquires(proto_list_lock)
2454
read_lock(&proto_list_lock);
2455
return seq_list_start_head(&proto_list, *pos);
2458
static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2460
return seq_list_next(v, &proto_list, pos);
2463
static void proto_seq_stop(struct seq_file *seq, void *v)
2464
__releases(proto_list_lock)
2466
read_unlock(&proto_list_lock);
2469
static char proto_method_implemented(const void *method)
2471
return method == NULL ? 'n' : 'y';
2474
static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2476
seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2477
"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2480
sock_prot_inuse_get(seq_file_net(seq), proto),
2481
proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2482
proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2484
proto->slab == NULL ? "no" : "yes",
2485
module_name(proto->owner),
2486
proto_method_implemented(proto->close),
2487
proto_method_implemented(proto->connect),
2488
proto_method_implemented(proto->disconnect),
2489
proto_method_implemented(proto->accept),
2490
proto_method_implemented(proto->ioctl),
2491
proto_method_implemented(proto->init),
2492
proto_method_implemented(proto->destroy),
2493
proto_method_implemented(proto->shutdown),
2494
proto_method_implemented(proto->setsockopt),
2495
proto_method_implemented(proto->getsockopt),
2496
proto_method_implemented(proto->sendmsg),
2497
proto_method_implemented(proto->recvmsg),
2498
proto_method_implemented(proto->sendpage),
2499
proto_method_implemented(proto->bind),
2500
proto_method_implemented(proto->backlog_rcv),
2501
proto_method_implemented(proto->hash),
2502
proto_method_implemented(proto->unhash),
2503
proto_method_implemented(proto->get_port),
2504
proto_method_implemented(proto->enter_memory_pressure));
2507
static int proto_seq_show(struct seq_file *seq, void *v)
2509
if (v == &proto_list)
2510
seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2519
"cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2521
proto_seq_printf(seq, list_entry(v, struct proto, node));
2525
static const struct seq_operations proto_seq_ops = {
2526
.start = proto_seq_start,
2527
.next = proto_seq_next,
2528
.stop = proto_seq_stop,
2529
.show = proto_seq_show,
2532
static int proto_seq_open(struct inode *inode, struct file *file)
2534
return seq_open_net(inode, file, &proto_seq_ops,
2535
sizeof(struct seq_net_private));
2538
static const struct file_operations proto_seq_fops = {
2539
.owner = THIS_MODULE,
2540
.open = proto_seq_open,
2542
.llseek = seq_lseek,
2543
.release = seq_release_net,
2546
static __net_init int proto_init_net(struct net *net)
2548
if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2554
static __net_exit void proto_exit_net(struct net *net)
2556
proc_net_remove(net, "protocols");
2560
static __net_initdata struct pernet_operations proto_net_ops = {
2561
.init = proto_init_net,
2562
.exit = proto_exit_net,
2565
static int __init proto_init(void)
2567
return register_pernet_subsys(&proto_net_ops);
2570
subsys_initcall(proto_init);
2572
#endif /* PROC_FS */