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.
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
121 #include <linux/uaccess.h>
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
139 #include <trace/events/sock.h>
145 #include <net/busy_poll.h>
147 static DEFINE_MUTEX(proto_list_mutex);
148 static LIST_HEAD(proto_list);
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
160 bool sk_ns_capable(const struct sock *sk,
161 struct user_namespace *user_ns, int cap)
163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
164 ns_capable(user_ns, cap);
166 EXPORT_SYMBOL(sk_ns_capable);
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
177 bool sk_capable(const struct sock *sk, int cap)
179 return sk_ns_capable(sk, &init_user_ns, cap);
181 EXPORT_SYMBOL(sk_capable);
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
192 bool sk_net_capable(const struct sock *sk, int cap)
194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
196 EXPORT_SYMBOL(sk_net_capable);
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family:
202 static struct lock_class_key af_family_keys[AF_MAX];
203 static struct lock_class_key af_family_slock_keys[AF_MAX];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
210 static const char *const af_family_key_strings[AF_MAX+1] = {
211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_KCM" ,
225 "sk_lock-AF_QIPCRTR", "sk_lock-AF_SMC" , "sk_lock-AF_MAX"
227 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
228 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
229 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
230 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
231 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
232 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
233 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
234 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
235 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
236 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
237 "slock-27" , "slock-28" , "slock-AF_CAN" ,
238 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
239 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
240 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
241 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" ,
242 "slock-AF_QIPCRTR", "slock-AF_SMC" , "slock-AF_MAX"
244 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
245 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
246 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
247 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
248 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
249 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
250 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
251 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
252 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
253 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
254 "clock-27" , "clock-28" , "clock-AF_CAN" ,
255 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
256 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
257 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
258 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" ,
259 "clock-AF_QIPCRTR", "clock-AF_SMC" , "clock-AF_MAX"
263 * sk_callback_lock locking rules are per-address-family,
264 * so split the lock classes by using a per-AF key:
266 static struct lock_class_key af_callback_keys[AF_MAX];
268 /* Take into consideration the size of the struct sk_buff overhead in the
269 * determination of these values, since that is non-constant across
270 * platforms. This makes socket queueing behavior and performance
271 * not depend upon such differences.
273 #define _SK_MEM_PACKETS 256
274 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
275 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
276 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
278 /* Run time adjustable parameters. */
279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
280 EXPORT_SYMBOL(sysctl_wmem_max);
281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
282 EXPORT_SYMBOL(sysctl_rmem_max);
283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
286 /* Maximal space eaten by iovec or ancillary data plus some space */
287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
288 EXPORT_SYMBOL(sysctl_optmem_max);
290 int sysctl_tstamp_allow_data __read_mostly = 1;
292 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
293 EXPORT_SYMBOL_GPL(memalloc_socks);
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
303 void sk_set_memalloc(struct sock *sk)
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
307 static_key_slow_inc(&memalloc_socks);
309 EXPORT_SYMBOL_GPL(sk_set_memalloc);
311 void sk_clear_memalloc(struct sock *sk)
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
315 static_key_slow_dec(&memalloc_socks);
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
326 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
331 unsigned long pflags = current->flags;
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
336 current->flags |= PF_MEMALLOC;
337 ret = sk->sk_backlog_rcv(sk, skb);
338 tsk_restore_flags(current, pflags, PF_MEMALLOC);
342 EXPORT_SYMBOL(__sk_backlog_rcv);
344 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
348 if (optlen < sizeof(tv))
350 if (copy_from_user(&tv, optval, sizeof(tv)))
352 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
356 static int warned __read_mostly;
359 if (warned < 10 && net_ratelimit()) {
361 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
362 __func__, current->comm, task_pid_nr(current));
366 *timeo_p = MAX_SCHEDULE_TIMEOUT;
367 if (tv.tv_sec == 0 && tv.tv_usec == 0)
369 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
370 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
374 static void sock_warn_obsolete_bsdism(const char *name)
377 static char warncomm[TASK_COMM_LEN];
378 if (strcmp(warncomm, current->comm) && warned < 5) {
379 strcpy(warncomm, current->comm);
380 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
386 static bool sock_needs_netstamp(const struct sock *sk)
388 switch (sk->sk_family) {
397 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
399 if (sk->sk_flags & flags) {
400 sk->sk_flags &= ~flags;
401 if (sock_needs_netstamp(sk) &&
402 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
403 net_disable_timestamp();
408 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
411 struct sk_buff_head *list = &sk->sk_receive_queue;
413 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
414 atomic_inc(&sk->sk_drops);
415 trace_sock_rcvqueue_full(sk, skb);
419 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
420 atomic_inc(&sk->sk_drops);
425 skb_set_owner_r(skb, sk);
427 /* we escape from rcu protected region, make sure we dont leak
432 spin_lock_irqsave(&list->lock, flags);
433 sock_skb_set_dropcount(sk, skb);
434 __skb_queue_tail(list, skb);
435 spin_unlock_irqrestore(&list->lock, flags);
437 if (!sock_flag(sk, SOCK_DEAD))
438 sk->sk_data_ready(sk);
441 EXPORT_SYMBOL(__sock_queue_rcv_skb);
443 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
447 err = sk_filter(sk, skb);
451 return __sock_queue_rcv_skb(sk, skb);
453 EXPORT_SYMBOL(sock_queue_rcv_skb);
455 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
456 const int nested, unsigned int trim_cap, bool refcounted)
458 int rc = NET_RX_SUCCESS;
460 if (sk_filter_trim_cap(sk, skb, trim_cap))
461 goto discard_and_relse;
465 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
466 atomic_inc(&sk->sk_drops);
467 goto discard_and_relse;
470 bh_lock_sock_nested(sk);
473 if (!sock_owned_by_user(sk)) {
475 * trylock + unlock semantics:
477 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
479 rc = sk_backlog_rcv(sk, skb);
481 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
482 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
484 atomic_inc(&sk->sk_drops);
485 goto discard_and_relse;
497 EXPORT_SYMBOL(__sk_receive_skb);
499 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
501 struct dst_entry *dst = __sk_dst_get(sk);
503 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
504 sk_tx_queue_clear(sk);
505 sk->sk_dst_pending_confirm = 0;
506 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
513 EXPORT_SYMBOL(__sk_dst_check);
515 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
517 struct dst_entry *dst = sk_dst_get(sk);
519 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
527 EXPORT_SYMBOL(sk_dst_check);
529 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
532 int ret = -ENOPROTOOPT;
533 #ifdef CONFIG_NETDEVICES
534 struct net *net = sock_net(sk);
535 char devname[IFNAMSIZ];
540 if (!ns_capable(net->user_ns, CAP_NET_RAW))
547 /* Bind this socket to a particular device like "eth0",
548 * as specified in the passed interface name. If the
549 * name is "" or the option length is zero the socket
552 if (optlen > IFNAMSIZ - 1)
553 optlen = IFNAMSIZ - 1;
554 memset(devname, 0, sizeof(devname));
557 if (copy_from_user(devname, optval, optlen))
561 if (devname[0] != '\0') {
562 struct net_device *dev;
565 dev = dev_get_by_name_rcu(net, devname);
567 index = dev->ifindex;
575 sk->sk_bound_dev_if = index;
587 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
588 int __user *optlen, int len)
590 int ret = -ENOPROTOOPT;
591 #ifdef CONFIG_NETDEVICES
592 struct net *net = sock_net(sk);
593 char devname[IFNAMSIZ];
595 if (sk->sk_bound_dev_if == 0) {
604 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
608 len = strlen(devname) + 1;
611 if (copy_to_user(optval, devname, len))
616 if (put_user(len, optlen))
627 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
630 sock_set_flag(sk, bit);
632 sock_reset_flag(sk, bit);
635 bool sk_mc_loop(struct sock *sk)
637 if (dev_recursion_level())
641 switch (sk->sk_family) {
643 return inet_sk(sk)->mc_loop;
644 #if IS_ENABLED(CONFIG_IPV6)
646 return inet6_sk(sk)->mc_loop;
652 EXPORT_SYMBOL(sk_mc_loop);
655 * This is meant for all protocols to use and covers goings on
656 * at the socket level. Everything here is generic.
659 int sock_setsockopt(struct socket *sock, int level, int optname,
660 char __user *optval, unsigned int optlen)
662 struct sock *sk = sock->sk;
669 * Options without arguments
672 if (optname == SO_BINDTODEVICE)
673 return sock_setbindtodevice(sk, optval, optlen);
675 if (optlen < sizeof(int))
678 if (get_user(val, (int __user *)optval))
681 valbool = val ? 1 : 0;
687 if (val && !capable(CAP_NET_ADMIN))
690 sock_valbool_flag(sk, SOCK_DBG, valbool);
693 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
696 sk->sk_reuseport = valbool;
705 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
708 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
711 /* Don't error on this BSD doesn't and if you think
712 * about it this is right. Otherwise apps have to
713 * play 'guess the biggest size' games. RCVBUF/SNDBUF
714 * are treated in BSD as hints
716 val = min_t(u32, val, sysctl_wmem_max);
718 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
719 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
720 /* Wake up sending tasks if we upped the value. */
721 sk->sk_write_space(sk);
725 if (!capable(CAP_NET_ADMIN)) {
732 /* Don't error on this BSD doesn't and if you think
733 * about it this is right. Otherwise apps have to
734 * play 'guess the biggest size' games. RCVBUF/SNDBUF
735 * are treated in BSD as hints
737 val = min_t(u32, val, sysctl_rmem_max);
739 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
741 * We double it on the way in to account for
742 * "struct sk_buff" etc. overhead. Applications
743 * assume that the SO_RCVBUF setting they make will
744 * allow that much actual data to be received on that
747 * Applications are unaware that "struct sk_buff" and
748 * other overheads allocate from the receive buffer
749 * during socket buffer allocation.
751 * And after considering the possible alternatives,
752 * returning the value we actually used in getsockopt
753 * is the most desirable behavior.
755 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
759 if (!capable(CAP_NET_ADMIN)) {
766 if (sk->sk_prot->keepalive)
767 sk->sk_prot->keepalive(sk, valbool);
768 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
772 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
776 sk->sk_no_check_tx = valbool;
780 if ((val >= 0 && val <= 6) ||
781 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
782 sk->sk_priority = val;
788 if (optlen < sizeof(ling)) {
789 ret = -EINVAL; /* 1003.1g */
792 if (copy_from_user(&ling, optval, sizeof(ling))) {
797 sock_reset_flag(sk, SOCK_LINGER);
799 #if (BITS_PER_LONG == 32)
800 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
801 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
804 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
805 sock_set_flag(sk, SOCK_LINGER);
810 sock_warn_obsolete_bsdism("setsockopt");
815 set_bit(SOCK_PASSCRED, &sock->flags);
817 clear_bit(SOCK_PASSCRED, &sock->flags);
823 if (optname == SO_TIMESTAMP)
824 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
826 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
827 sock_set_flag(sk, SOCK_RCVTSTAMP);
828 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
830 sock_reset_flag(sk, SOCK_RCVTSTAMP);
831 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
835 case SO_TIMESTAMPING:
836 if (val & ~SOF_TIMESTAMPING_MASK) {
841 if (val & SOF_TIMESTAMPING_OPT_ID &&
842 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
843 if (sk->sk_protocol == IPPROTO_TCP &&
844 sk->sk_type == SOCK_STREAM) {
845 if ((1 << sk->sk_state) &
846 (TCPF_CLOSE | TCPF_LISTEN)) {
850 sk->sk_tskey = tcp_sk(sk)->snd_una;
856 if (val & SOF_TIMESTAMPING_OPT_STATS &&
857 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
862 sk->sk_tsflags = val;
863 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
864 sock_enable_timestamp(sk,
865 SOCK_TIMESTAMPING_RX_SOFTWARE);
867 sock_disable_timestamp(sk,
868 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
874 sk->sk_rcvlowat = val ? : 1;
878 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
882 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
885 case SO_ATTACH_FILTER:
887 if (optlen == sizeof(struct sock_fprog)) {
888 struct sock_fprog fprog;
891 if (copy_from_user(&fprog, optval, sizeof(fprog)))
894 ret = sk_attach_filter(&fprog, sk);
900 if (optlen == sizeof(u32)) {
904 if (copy_from_user(&ufd, optval, sizeof(ufd)))
907 ret = sk_attach_bpf(ufd, sk);
911 case SO_ATTACH_REUSEPORT_CBPF:
913 if (optlen == sizeof(struct sock_fprog)) {
914 struct sock_fprog fprog;
917 if (copy_from_user(&fprog, optval, sizeof(fprog)))
920 ret = sk_reuseport_attach_filter(&fprog, sk);
924 case SO_ATTACH_REUSEPORT_EBPF:
926 if (optlen == sizeof(u32)) {
930 if (copy_from_user(&ufd, optval, sizeof(ufd)))
933 ret = sk_reuseport_attach_bpf(ufd, sk);
937 case SO_DETACH_FILTER:
938 ret = sk_detach_filter(sk);
942 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
945 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
950 set_bit(SOCK_PASSSEC, &sock->flags);
952 clear_bit(SOCK_PASSSEC, &sock->flags);
955 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
962 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
966 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
970 if (sock->ops->set_peek_off)
971 ret = sock->ops->set_peek_off(sk, val);
977 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
980 case SO_SELECT_ERR_QUEUE:
981 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
984 #ifdef CONFIG_NET_RX_BUSY_POLL
986 /* allow unprivileged users to decrease the value */
987 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
993 sk->sk_ll_usec = val;
998 case SO_MAX_PACING_RATE:
999 sk->sk_max_pacing_rate = val;
1000 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1001 sk->sk_max_pacing_rate);
1004 case SO_INCOMING_CPU:
1005 sk->sk_incoming_cpu = val;
1010 dst_negative_advice(sk);
1019 EXPORT_SYMBOL(sock_setsockopt);
1022 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1023 struct ucred *ucred)
1025 ucred->pid = pid_vnr(pid);
1026 ucred->uid = ucred->gid = -1;
1028 struct user_namespace *current_ns = current_user_ns();
1030 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1031 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1035 int sock_getsockopt(struct socket *sock, int level, int optname,
1036 char __user *optval, int __user *optlen)
1038 struct sock *sk = sock->sk;
1046 int lv = sizeof(int);
1049 if (get_user(len, optlen))
1054 memset(&v, 0, sizeof(v));
1058 v.val = sock_flag(sk, SOCK_DBG);
1062 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1066 v.val = sock_flag(sk, SOCK_BROADCAST);
1070 v.val = sk->sk_sndbuf;
1074 v.val = sk->sk_rcvbuf;
1078 v.val = sk->sk_reuse;
1082 v.val = sk->sk_reuseport;
1086 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1090 v.val = sk->sk_type;
1094 v.val = sk->sk_protocol;
1098 v.val = sk->sk_family;
1102 v.val = -sock_error(sk);
1104 v.val = xchg(&sk->sk_err_soft, 0);
1108 v.val = sock_flag(sk, SOCK_URGINLINE);
1112 v.val = sk->sk_no_check_tx;
1116 v.val = sk->sk_priority;
1120 lv = sizeof(v.ling);
1121 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1122 v.ling.l_linger = sk->sk_lingertime / HZ;
1126 sock_warn_obsolete_bsdism("getsockopt");
1130 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1131 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1134 case SO_TIMESTAMPNS:
1135 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1138 case SO_TIMESTAMPING:
1139 v.val = sk->sk_tsflags;
1143 lv = sizeof(struct timeval);
1144 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1148 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1149 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1154 lv = sizeof(struct timeval);
1155 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1159 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1160 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1165 v.val = sk->sk_rcvlowat;
1173 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1178 struct ucred peercred;
1179 if (len > sizeof(peercred))
1180 len = sizeof(peercred);
1181 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1182 if (copy_to_user(optval, &peercred, len))
1191 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1195 if (copy_to_user(optval, address, len))
1200 /* Dubious BSD thing... Probably nobody even uses it, but
1201 * the UNIX standard wants it for whatever reason... -DaveM
1204 v.val = sk->sk_state == TCP_LISTEN;
1208 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1212 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1215 v.val = sk->sk_mark;
1219 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1222 case SO_WIFI_STATUS:
1223 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1227 if (!sock->ops->set_peek_off)
1230 v.val = sk->sk_peek_off;
1233 v.val = sock_flag(sk, SOCK_NOFCS);
1236 case SO_BINDTODEVICE:
1237 return sock_getbindtodevice(sk, optval, optlen, len);
1240 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1246 case SO_LOCK_FILTER:
1247 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1250 case SO_BPF_EXTENSIONS:
1251 v.val = bpf_tell_extensions();
1254 case SO_SELECT_ERR_QUEUE:
1255 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1258 #ifdef CONFIG_NET_RX_BUSY_POLL
1260 v.val = sk->sk_ll_usec;
1264 case SO_MAX_PACING_RATE:
1265 v.val = sk->sk_max_pacing_rate;
1268 case SO_INCOMING_CPU:
1269 v.val = sk->sk_incoming_cpu;
1273 /* We implement the SO_SNDLOWAT etc to not be settable
1276 return -ENOPROTOOPT;
1281 if (copy_to_user(optval, &v, len))
1284 if (put_user(len, optlen))
1290 * Initialize an sk_lock.
1292 * (We also register the sk_lock with the lock validator.)
1294 static inline void sock_lock_init(struct sock *sk)
1296 sock_lock_init_class_and_name(sk,
1297 af_family_slock_key_strings[sk->sk_family],
1298 af_family_slock_keys + sk->sk_family,
1299 af_family_key_strings[sk->sk_family],
1300 af_family_keys + sk->sk_family);
1304 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1305 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1306 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1308 static void sock_copy(struct sock *nsk, const struct sock *osk)
1310 #ifdef CONFIG_SECURITY_NETWORK
1311 void *sptr = nsk->sk_security;
1313 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1315 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1316 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1318 #ifdef CONFIG_SECURITY_NETWORK
1319 nsk->sk_security = sptr;
1320 security_sk_clone(osk, nsk);
1324 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1328 struct kmem_cache *slab;
1332 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1335 if (priority & __GFP_ZERO)
1336 sk_prot_clear_nulls(sk, prot->obj_size);
1338 sk = kmalloc(prot->obj_size, priority);
1341 kmemcheck_annotate_bitfield(sk, flags);
1343 if (security_sk_alloc(sk, family, priority))
1346 if (!try_module_get(prot->owner))
1348 sk_tx_queue_clear(sk);
1354 security_sk_free(sk);
1357 kmem_cache_free(slab, sk);
1363 static void sk_prot_free(struct proto *prot, struct sock *sk)
1365 struct kmem_cache *slab;
1366 struct module *owner;
1368 owner = prot->owner;
1371 cgroup_sk_free(&sk->sk_cgrp_data);
1372 mem_cgroup_sk_free(sk);
1373 security_sk_free(sk);
1375 kmem_cache_free(slab, sk);
1382 * sk_alloc - All socket objects are allocated here
1383 * @net: the applicable net namespace
1384 * @family: protocol family
1385 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1386 * @prot: struct proto associated with this new sock instance
1387 * @kern: is this to be a kernel socket?
1389 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1390 struct proto *prot, int kern)
1394 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1396 sk->sk_family = family;
1398 * See comment in struct sock definition to understand
1399 * why we need sk_prot_creator -acme
1401 sk->sk_prot = sk->sk_prot_creator = prot;
1403 sk->sk_net_refcnt = kern ? 0 : 1;
1404 if (likely(sk->sk_net_refcnt))
1406 sock_net_set(sk, net);
1407 atomic_set(&sk->sk_wmem_alloc, 1);
1409 mem_cgroup_sk_alloc(sk);
1410 cgroup_sk_alloc(&sk->sk_cgrp_data);
1411 sock_update_classid(&sk->sk_cgrp_data);
1412 sock_update_netprioidx(&sk->sk_cgrp_data);
1417 EXPORT_SYMBOL(sk_alloc);
1419 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1420 * grace period. This is the case for UDP sockets and TCP listeners.
1422 static void __sk_destruct(struct rcu_head *head)
1424 struct sock *sk = container_of(head, struct sock, sk_rcu);
1425 struct sk_filter *filter;
1427 if (sk->sk_destruct)
1428 sk->sk_destruct(sk);
1430 filter = rcu_dereference_check(sk->sk_filter,
1431 atomic_read(&sk->sk_wmem_alloc) == 0);
1433 sk_filter_uncharge(sk, filter);
1434 RCU_INIT_POINTER(sk->sk_filter, NULL);
1436 if (rcu_access_pointer(sk->sk_reuseport_cb))
1437 reuseport_detach_sock(sk);
1439 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1441 if (atomic_read(&sk->sk_omem_alloc))
1442 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1443 __func__, atomic_read(&sk->sk_omem_alloc));
1445 if (sk->sk_peer_cred)
1446 put_cred(sk->sk_peer_cred);
1447 put_pid(sk->sk_peer_pid);
1448 if (likely(sk->sk_net_refcnt))
1449 put_net(sock_net(sk));
1450 sk_prot_free(sk->sk_prot_creator, sk);
1453 void sk_destruct(struct sock *sk)
1455 if (sock_flag(sk, SOCK_RCU_FREE))
1456 call_rcu(&sk->sk_rcu, __sk_destruct);
1458 __sk_destruct(&sk->sk_rcu);
1461 static void __sk_free(struct sock *sk)
1463 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1464 sock_diag_broadcast_destroy(sk);
1469 void sk_free(struct sock *sk)
1472 * We subtract one from sk_wmem_alloc and can know if
1473 * some packets are still in some tx queue.
1474 * If not null, sock_wfree() will call __sk_free(sk) later
1476 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1479 EXPORT_SYMBOL(sk_free);
1482 * sk_clone_lock - clone a socket, and lock its clone
1483 * @sk: the socket to clone
1484 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1486 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1488 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1491 bool is_charged = true;
1493 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1494 if (newsk != NULL) {
1495 struct sk_filter *filter;
1497 sock_copy(newsk, sk);
1500 if (likely(newsk->sk_net_refcnt))
1501 get_net(sock_net(newsk));
1502 sk_node_init(&newsk->sk_node);
1503 sock_lock_init(newsk);
1504 bh_lock_sock(newsk);
1505 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1506 newsk->sk_backlog.len = 0;
1508 atomic_set(&newsk->sk_rmem_alloc, 0);
1510 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1512 atomic_set(&newsk->sk_wmem_alloc, 1);
1513 atomic_set(&newsk->sk_omem_alloc, 0);
1514 skb_queue_head_init(&newsk->sk_receive_queue);
1515 skb_queue_head_init(&newsk->sk_write_queue);
1517 rwlock_init(&newsk->sk_callback_lock);
1518 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1519 af_callback_keys + newsk->sk_family,
1520 af_family_clock_key_strings[newsk->sk_family]);
1522 newsk->sk_dst_cache = NULL;
1523 newsk->sk_dst_pending_confirm = 0;
1524 newsk->sk_wmem_queued = 0;
1525 newsk->sk_forward_alloc = 0;
1526 atomic_set(&newsk->sk_drops, 0);
1527 newsk->sk_send_head = NULL;
1528 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1530 sock_reset_flag(newsk, SOCK_DONE);
1531 skb_queue_head_init(&newsk->sk_error_queue);
1533 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1535 /* though it's an empty new sock, the charging may fail
1536 * if sysctl_optmem_max was changed between creation of
1537 * original socket and cloning
1539 is_charged = sk_filter_charge(newsk, filter);
1541 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1542 sk_free_unlock_clone(newsk);
1546 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1549 newsk->sk_err_soft = 0;
1550 newsk->sk_priority = 0;
1551 newsk->sk_incoming_cpu = raw_smp_processor_id();
1552 atomic64_set(&newsk->sk_cookie, 0);
1554 mem_cgroup_sk_alloc(newsk);
1555 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1558 * Before updating sk_refcnt, we must commit prior changes to memory
1559 * (Documentation/RCU/rculist_nulls.txt for details)
1562 atomic_set(&newsk->sk_refcnt, 2);
1565 * Increment the counter in the same struct proto as the master
1566 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1567 * is the same as sk->sk_prot->socks, as this field was copied
1570 * This _changes_ the previous behaviour, where
1571 * tcp_create_openreq_child always was incrementing the
1572 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1573 * to be taken into account in all callers. -acme
1575 sk_refcnt_debug_inc(newsk);
1576 sk_set_socket(newsk, NULL);
1577 newsk->sk_wq = NULL;
1579 if (newsk->sk_prot->sockets_allocated)
1580 sk_sockets_allocated_inc(newsk);
1582 if (sock_needs_netstamp(sk) &&
1583 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1584 net_enable_timestamp();
1589 EXPORT_SYMBOL_GPL(sk_clone_lock);
1591 void sk_free_unlock_clone(struct sock *sk)
1593 /* It is still raw copy of parent, so invalidate
1594 * destructor and make plain sk_free() */
1595 sk->sk_destruct = NULL;
1599 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1601 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1605 sk_dst_set(sk, dst);
1606 sk->sk_route_caps = dst->dev->features;
1607 if (sk->sk_route_caps & NETIF_F_GSO)
1608 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1609 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1610 if (sk_can_gso(sk)) {
1611 if (dst->header_len) {
1612 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1614 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1615 sk->sk_gso_max_size = dst->dev->gso_max_size;
1616 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1619 sk->sk_gso_max_segs = max_segs;
1621 EXPORT_SYMBOL_GPL(sk_setup_caps);
1624 * Simple resource managers for sockets.
1629 * Write buffer destructor automatically called from kfree_skb.
1631 void sock_wfree(struct sk_buff *skb)
1633 struct sock *sk = skb->sk;
1634 unsigned int len = skb->truesize;
1636 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1638 * Keep a reference on sk_wmem_alloc, this will be released
1639 * after sk_write_space() call
1641 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1642 sk->sk_write_space(sk);
1646 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1647 * could not do because of in-flight packets
1649 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1652 EXPORT_SYMBOL(sock_wfree);
1654 /* This variant of sock_wfree() is used by TCP,
1655 * since it sets SOCK_USE_WRITE_QUEUE.
1657 void __sock_wfree(struct sk_buff *skb)
1659 struct sock *sk = skb->sk;
1661 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1665 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1670 if (unlikely(!sk_fullsock(sk))) {
1671 skb->destructor = sock_edemux;
1676 skb->destructor = sock_wfree;
1677 skb_set_hash_from_sk(skb, sk);
1679 * We used to take a refcount on sk, but following operation
1680 * is enough to guarantee sk_free() wont free this sock until
1681 * all in-flight packets are completed
1683 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1685 EXPORT_SYMBOL(skb_set_owner_w);
1687 /* This helper is used by netem, as it can hold packets in its
1688 * delay queue. We want to allow the owner socket to send more
1689 * packets, as if they were already TX completed by a typical driver.
1690 * But we also want to keep skb->sk set because some packet schedulers
1691 * rely on it (sch_fq for example). So we set skb->truesize to a small
1692 * amount (1) and decrease sk_wmem_alloc accordingly.
1694 void skb_orphan_partial(struct sk_buff *skb)
1696 /* If this skb is a TCP pure ACK or already went here,
1697 * we have nothing to do. 2 is already a very small truesize.
1699 if (skb->truesize <= 2)
1702 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1703 * so we do not completely orphan skb, but transfert all
1704 * accounted bytes but one, to avoid unexpected reorders.
1706 if (skb->destructor == sock_wfree
1708 || skb->destructor == tcp_wfree
1711 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1717 EXPORT_SYMBOL(skb_orphan_partial);
1720 * Read buffer destructor automatically called from kfree_skb.
1722 void sock_rfree(struct sk_buff *skb)
1724 struct sock *sk = skb->sk;
1725 unsigned int len = skb->truesize;
1727 atomic_sub(len, &sk->sk_rmem_alloc);
1728 sk_mem_uncharge(sk, len);
1730 EXPORT_SYMBOL(sock_rfree);
1733 * Buffer destructor for skbs that are not used directly in read or write
1734 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1736 void sock_efree(struct sk_buff *skb)
1740 EXPORT_SYMBOL(sock_efree);
1742 kuid_t sock_i_uid(struct sock *sk)
1746 read_lock_bh(&sk->sk_callback_lock);
1747 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1748 read_unlock_bh(&sk->sk_callback_lock);
1751 EXPORT_SYMBOL(sock_i_uid);
1753 unsigned long sock_i_ino(struct sock *sk)
1757 read_lock_bh(&sk->sk_callback_lock);
1758 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1759 read_unlock_bh(&sk->sk_callback_lock);
1762 EXPORT_SYMBOL(sock_i_ino);
1765 * Allocate a skb from the socket's send buffer.
1767 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1770 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1771 struct sk_buff *skb = alloc_skb(size, priority);
1773 skb_set_owner_w(skb, sk);
1779 EXPORT_SYMBOL(sock_wmalloc);
1782 * Allocate a memory block from the socket's option memory buffer.
1784 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1786 if ((unsigned int)size <= sysctl_optmem_max &&
1787 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1789 /* First do the add, to avoid the race if kmalloc
1792 atomic_add(size, &sk->sk_omem_alloc);
1793 mem = kmalloc(size, priority);
1796 atomic_sub(size, &sk->sk_omem_alloc);
1800 EXPORT_SYMBOL(sock_kmalloc);
1802 /* Free an option memory block. Note, we actually want the inline
1803 * here as this allows gcc to detect the nullify and fold away the
1804 * condition entirely.
1806 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1809 if (WARN_ON_ONCE(!mem))
1815 atomic_sub(size, &sk->sk_omem_alloc);
1818 void sock_kfree_s(struct sock *sk, void *mem, int size)
1820 __sock_kfree_s(sk, mem, size, false);
1822 EXPORT_SYMBOL(sock_kfree_s);
1824 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1826 __sock_kfree_s(sk, mem, size, true);
1828 EXPORT_SYMBOL(sock_kzfree_s);
1830 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1831 I think, these locks should be removed for datagram sockets.
1833 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1837 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1841 if (signal_pending(current))
1843 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1844 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1845 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1847 if (sk->sk_shutdown & SEND_SHUTDOWN)
1851 timeo = schedule_timeout(timeo);
1853 finish_wait(sk_sleep(sk), &wait);
1859 * Generic send/receive buffer handlers
1862 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1863 unsigned long data_len, int noblock,
1864 int *errcode, int max_page_order)
1866 struct sk_buff *skb;
1870 timeo = sock_sndtimeo(sk, noblock);
1872 err = sock_error(sk);
1877 if (sk->sk_shutdown & SEND_SHUTDOWN)
1880 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1883 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1884 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1888 if (signal_pending(current))
1890 timeo = sock_wait_for_wmem(sk, timeo);
1892 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1893 errcode, sk->sk_allocation);
1895 skb_set_owner_w(skb, sk);
1899 err = sock_intr_errno(timeo);
1904 EXPORT_SYMBOL(sock_alloc_send_pskb);
1906 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1907 int noblock, int *errcode)
1909 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1911 EXPORT_SYMBOL(sock_alloc_send_skb);
1913 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1914 struct sockcm_cookie *sockc)
1918 switch (cmsg->cmsg_type) {
1920 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1922 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1924 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1926 case SO_TIMESTAMPING:
1927 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1930 tsflags = *(u32 *)CMSG_DATA(cmsg);
1931 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1934 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1935 sockc->tsflags |= tsflags;
1937 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1939 case SCM_CREDENTIALS:
1946 EXPORT_SYMBOL(__sock_cmsg_send);
1948 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1949 struct sockcm_cookie *sockc)
1951 struct cmsghdr *cmsg;
1954 for_each_cmsghdr(cmsg, msg) {
1955 if (!CMSG_OK(msg, cmsg))
1957 if (cmsg->cmsg_level != SOL_SOCKET)
1959 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1965 EXPORT_SYMBOL(sock_cmsg_send);
1967 /* On 32bit arches, an skb frag is limited to 2^15 */
1968 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1971 * skb_page_frag_refill - check that a page_frag contains enough room
1972 * @sz: minimum size of the fragment we want to get
1973 * @pfrag: pointer to page_frag
1974 * @gfp: priority for memory allocation
1976 * Note: While this allocator tries to use high order pages, there is
1977 * no guarantee that allocations succeed. Therefore, @sz MUST be
1978 * less or equal than PAGE_SIZE.
1980 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1983 if (page_ref_count(pfrag->page) == 1) {
1987 if (pfrag->offset + sz <= pfrag->size)
1989 put_page(pfrag->page);
1993 if (SKB_FRAG_PAGE_ORDER) {
1994 /* Avoid direct reclaim but allow kswapd to wake */
1995 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1996 __GFP_COMP | __GFP_NOWARN |
1998 SKB_FRAG_PAGE_ORDER);
1999 if (likely(pfrag->page)) {
2000 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2004 pfrag->page = alloc_page(gfp);
2005 if (likely(pfrag->page)) {
2006 pfrag->size = PAGE_SIZE;
2011 EXPORT_SYMBOL(skb_page_frag_refill);
2013 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2015 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2018 sk_enter_memory_pressure(sk);
2019 sk_stream_moderate_sndbuf(sk);
2022 EXPORT_SYMBOL(sk_page_frag_refill);
2024 static void __lock_sock(struct sock *sk)
2025 __releases(&sk->sk_lock.slock)
2026 __acquires(&sk->sk_lock.slock)
2031 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2032 TASK_UNINTERRUPTIBLE);
2033 spin_unlock_bh(&sk->sk_lock.slock);
2035 spin_lock_bh(&sk->sk_lock.slock);
2036 if (!sock_owned_by_user(sk))
2039 finish_wait(&sk->sk_lock.wq, &wait);
2042 static void __release_sock(struct sock *sk)
2043 __releases(&sk->sk_lock.slock)
2044 __acquires(&sk->sk_lock.slock)
2046 struct sk_buff *skb, *next;
2048 while ((skb = sk->sk_backlog.head) != NULL) {
2049 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2051 spin_unlock_bh(&sk->sk_lock.slock);
2056 WARN_ON_ONCE(skb_dst_is_noref(skb));
2058 sk_backlog_rcv(sk, skb);
2063 } while (skb != NULL);
2065 spin_lock_bh(&sk->sk_lock.slock);
2069 * Doing the zeroing here guarantee we can not loop forever
2070 * while a wild producer attempts to flood us.
2072 sk->sk_backlog.len = 0;
2075 void __sk_flush_backlog(struct sock *sk)
2077 spin_lock_bh(&sk->sk_lock.slock);
2079 spin_unlock_bh(&sk->sk_lock.slock);
2083 * sk_wait_data - wait for data to arrive at sk_receive_queue
2084 * @sk: sock to wait on
2085 * @timeo: for how long
2086 * @skb: last skb seen on sk_receive_queue
2088 * Now socket state including sk->sk_err is changed only under lock,
2089 * hence we may omit checks after joining wait queue.
2090 * We check receive queue before schedule() only as optimization;
2091 * it is very likely that release_sock() added new data.
2093 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2095 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2098 add_wait_queue(sk_sleep(sk), &wait);
2099 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2100 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2101 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2102 remove_wait_queue(sk_sleep(sk), &wait);
2105 EXPORT_SYMBOL(sk_wait_data);
2108 * __sk_mem_raise_allocated - increase memory_allocated
2110 * @size: memory size to allocate
2111 * @amt: pages to allocate
2112 * @kind: allocation type
2114 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2116 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2118 struct proto *prot = sk->sk_prot;
2119 long allocated = sk_memory_allocated_add(sk, amt);
2121 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2122 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2123 goto suppress_allocation;
2126 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2127 sk_leave_memory_pressure(sk);
2131 /* Under pressure. */
2132 if (allocated > sk_prot_mem_limits(sk, 1))
2133 sk_enter_memory_pressure(sk);
2135 /* Over hard limit. */
2136 if (allocated > sk_prot_mem_limits(sk, 2))
2137 goto suppress_allocation;
2139 /* guarantee minimum buffer size under pressure */
2140 if (kind == SK_MEM_RECV) {
2141 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2144 } else { /* SK_MEM_SEND */
2145 if (sk->sk_type == SOCK_STREAM) {
2146 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2148 } else if (atomic_read(&sk->sk_wmem_alloc) <
2149 prot->sysctl_wmem[0])
2153 if (sk_has_memory_pressure(sk)) {
2156 if (!sk_under_memory_pressure(sk))
2158 alloc = sk_sockets_allocated_read_positive(sk);
2159 if (sk_prot_mem_limits(sk, 2) > alloc *
2160 sk_mem_pages(sk->sk_wmem_queued +
2161 atomic_read(&sk->sk_rmem_alloc) +
2162 sk->sk_forward_alloc))
2166 suppress_allocation:
2168 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2169 sk_stream_moderate_sndbuf(sk);
2171 /* Fail only if socket is _under_ its sndbuf.
2172 * In this case we cannot block, so that we have to fail.
2174 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2178 trace_sock_exceed_buf_limit(sk, prot, allocated);
2180 sk_memory_allocated_sub(sk, amt);
2182 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2183 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2187 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2190 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2192 * @size: memory size to allocate
2193 * @kind: allocation type
2195 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2196 * rmem allocation. This function assumes that protocols which have
2197 * memory_pressure use sk_wmem_queued as write buffer accounting.
2199 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2201 int ret, amt = sk_mem_pages(size);
2203 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2204 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2206 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2209 EXPORT_SYMBOL(__sk_mem_schedule);
2212 * __sk_mem_reduce_allocated - reclaim memory_allocated
2214 * @amount: number of quanta
2216 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2218 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2220 sk_memory_allocated_sub(sk, amount);
2222 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2223 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2225 if (sk_under_memory_pressure(sk) &&
2226 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2227 sk_leave_memory_pressure(sk);
2229 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2232 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2234 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2236 void __sk_mem_reclaim(struct sock *sk, int amount)
2238 amount >>= SK_MEM_QUANTUM_SHIFT;
2239 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2240 __sk_mem_reduce_allocated(sk, amount);
2242 EXPORT_SYMBOL(__sk_mem_reclaim);
2244 int sk_set_peek_off(struct sock *sk, int val)
2249 sk->sk_peek_off = val;
2252 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2255 * Set of default routines for initialising struct proto_ops when
2256 * the protocol does not support a particular function. In certain
2257 * cases where it makes no sense for a protocol to have a "do nothing"
2258 * function, some default processing is provided.
2261 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2265 EXPORT_SYMBOL(sock_no_bind);
2267 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2272 EXPORT_SYMBOL(sock_no_connect);
2274 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2278 EXPORT_SYMBOL(sock_no_socketpair);
2280 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2284 EXPORT_SYMBOL(sock_no_accept);
2286 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2291 EXPORT_SYMBOL(sock_no_getname);
2293 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2297 EXPORT_SYMBOL(sock_no_poll);
2299 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2303 EXPORT_SYMBOL(sock_no_ioctl);
2305 int sock_no_listen(struct socket *sock, int backlog)
2309 EXPORT_SYMBOL(sock_no_listen);
2311 int sock_no_shutdown(struct socket *sock, int how)
2315 EXPORT_SYMBOL(sock_no_shutdown);
2317 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2318 char __user *optval, unsigned int optlen)
2322 EXPORT_SYMBOL(sock_no_setsockopt);
2324 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2325 char __user *optval, int __user *optlen)
2329 EXPORT_SYMBOL(sock_no_getsockopt);
2331 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2335 EXPORT_SYMBOL(sock_no_sendmsg);
2337 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2342 EXPORT_SYMBOL(sock_no_recvmsg);
2344 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2346 /* Mirror missing mmap method error code */
2349 EXPORT_SYMBOL(sock_no_mmap);
2351 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2354 struct msghdr msg = {.msg_flags = flags};
2356 char *kaddr = kmap(page);
2357 iov.iov_base = kaddr + offset;
2359 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2363 EXPORT_SYMBOL(sock_no_sendpage);
2366 * Default Socket Callbacks
2369 static void sock_def_wakeup(struct sock *sk)
2371 struct socket_wq *wq;
2374 wq = rcu_dereference(sk->sk_wq);
2375 if (skwq_has_sleeper(wq))
2376 wake_up_interruptible_all(&wq->wait);
2380 static void sock_def_error_report(struct sock *sk)
2382 struct socket_wq *wq;
2385 wq = rcu_dereference(sk->sk_wq);
2386 if (skwq_has_sleeper(wq))
2387 wake_up_interruptible_poll(&wq->wait, POLLERR);
2388 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2392 static void sock_def_readable(struct sock *sk)
2394 struct socket_wq *wq;
2397 wq = rcu_dereference(sk->sk_wq);
2398 if (skwq_has_sleeper(wq))
2399 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2400 POLLRDNORM | POLLRDBAND);
2401 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2405 static void sock_def_write_space(struct sock *sk)
2407 struct socket_wq *wq;
2411 /* Do not wake up a writer until he can make "significant"
2414 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2415 wq = rcu_dereference(sk->sk_wq);
2416 if (skwq_has_sleeper(wq))
2417 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2418 POLLWRNORM | POLLWRBAND);
2420 /* Should agree with poll, otherwise some programs break */
2421 if (sock_writeable(sk))
2422 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2428 static void sock_def_destruct(struct sock *sk)
2432 void sk_send_sigurg(struct sock *sk)
2434 if (sk->sk_socket && sk->sk_socket->file)
2435 if (send_sigurg(&sk->sk_socket->file->f_owner))
2436 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2438 EXPORT_SYMBOL(sk_send_sigurg);
2440 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2441 unsigned long expires)
2443 if (!mod_timer(timer, expires))
2446 EXPORT_SYMBOL(sk_reset_timer);
2448 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2450 if (del_timer(timer))
2453 EXPORT_SYMBOL(sk_stop_timer);
2455 void sock_init_data(struct socket *sock, struct sock *sk)
2457 skb_queue_head_init(&sk->sk_receive_queue);
2458 skb_queue_head_init(&sk->sk_write_queue);
2459 skb_queue_head_init(&sk->sk_error_queue);
2461 sk->sk_send_head = NULL;
2463 init_timer(&sk->sk_timer);
2465 sk->sk_allocation = GFP_KERNEL;
2466 sk->sk_rcvbuf = sysctl_rmem_default;
2467 sk->sk_sndbuf = sysctl_wmem_default;
2468 sk->sk_state = TCP_CLOSE;
2469 sk_set_socket(sk, sock);
2471 sock_set_flag(sk, SOCK_ZAPPED);
2474 sk->sk_type = sock->type;
2475 sk->sk_wq = sock->wq;
2477 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2480 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2483 rwlock_init(&sk->sk_callback_lock);
2484 lockdep_set_class_and_name(&sk->sk_callback_lock,
2485 af_callback_keys + sk->sk_family,
2486 af_family_clock_key_strings[sk->sk_family]);
2488 sk->sk_state_change = sock_def_wakeup;
2489 sk->sk_data_ready = sock_def_readable;
2490 sk->sk_write_space = sock_def_write_space;
2491 sk->sk_error_report = sock_def_error_report;
2492 sk->sk_destruct = sock_def_destruct;
2494 sk->sk_frag.page = NULL;
2495 sk->sk_frag.offset = 0;
2496 sk->sk_peek_off = -1;
2498 sk->sk_peer_pid = NULL;
2499 sk->sk_peer_cred = NULL;
2500 sk->sk_write_pending = 0;
2501 sk->sk_rcvlowat = 1;
2502 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2503 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2505 sk->sk_stamp = ktime_set(-1L, 0);
2507 #ifdef CONFIG_NET_RX_BUSY_POLL
2509 sk->sk_ll_usec = sysctl_net_busy_read;
2512 sk->sk_max_pacing_rate = ~0U;
2513 sk->sk_pacing_rate = ~0U;
2514 sk->sk_incoming_cpu = -1;
2516 * Before updating sk_refcnt, we must commit prior changes to memory
2517 * (Documentation/RCU/rculist_nulls.txt for details)
2520 atomic_set(&sk->sk_refcnt, 1);
2521 atomic_set(&sk->sk_drops, 0);
2523 EXPORT_SYMBOL(sock_init_data);
2525 void lock_sock_nested(struct sock *sk, int subclass)
2528 spin_lock_bh(&sk->sk_lock.slock);
2529 if (sk->sk_lock.owned)
2531 sk->sk_lock.owned = 1;
2532 spin_unlock(&sk->sk_lock.slock);
2534 * The sk_lock has mutex_lock() semantics here:
2536 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2539 EXPORT_SYMBOL(lock_sock_nested);
2541 void release_sock(struct sock *sk)
2543 spin_lock_bh(&sk->sk_lock.slock);
2544 if (sk->sk_backlog.tail)
2547 /* Warning : release_cb() might need to release sk ownership,
2548 * ie call sock_release_ownership(sk) before us.
2550 if (sk->sk_prot->release_cb)
2551 sk->sk_prot->release_cb(sk);
2553 sock_release_ownership(sk);
2554 if (waitqueue_active(&sk->sk_lock.wq))
2555 wake_up(&sk->sk_lock.wq);
2556 spin_unlock_bh(&sk->sk_lock.slock);
2558 EXPORT_SYMBOL(release_sock);
2561 * lock_sock_fast - fast version of lock_sock
2564 * This version should be used for very small section, where process wont block
2565 * return false if fast path is taken
2566 * sk_lock.slock locked, owned = 0, BH disabled
2567 * return true if slow path is taken
2568 * sk_lock.slock unlocked, owned = 1, BH enabled
2570 bool lock_sock_fast(struct sock *sk)
2573 spin_lock_bh(&sk->sk_lock.slock);
2575 if (!sk->sk_lock.owned)
2577 * Note : We must disable BH
2582 sk->sk_lock.owned = 1;
2583 spin_unlock(&sk->sk_lock.slock);
2585 * The sk_lock has mutex_lock() semantics here:
2587 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2591 EXPORT_SYMBOL(lock_sock_fast);
2593 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2596 if (!sock_flag(sk, SOCK_TIMESTAMP))
2597 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2598 tv = ktime_to_timeval(sk->sk_stamp);
2599 if (tv.tv_sec == -1)
2601 if (tv.tv_sec == 0) {
2602 sk->sk_stamp = ktime_get_real();
2603 tv = ktime_to_timeval(sk->sk_stamp);
2605 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2607 EXPORT_SYMBOL(sock_get_timestamp);
2609 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2612 if (!sock_flag(sk, SOCK_TIMESTAMP))
2613 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2614 ts = ktime_to_timespec(sk->sk_stamp);
2615 if (ts.tv_sec == -1)
2617 if (ts.tv_sec == 0) {
2618 sk->sk_stamp = ktime_get_real();
2619 ts = ktime_to_timespec(sk->sk_stamp);
2621 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2623 EXPORT_SYMBOL(sock_get_timestampns);
2625 void sock_enable_timestamp(struct sock *sk, int flag)
2627 if (!sock_flag(sk, flag)) {
2628 unsigned long previous_flags = sk->sk_flags;
2630 sock_set_flag(sk, flag);
2632 * we just set one of the two flags which require net
2633 * time stamping, but time stamping might have been on
2634 * already because of the other one
2636 if (sock_needs_netstamp(sk) &&
2637 !(previous_flags & SK_FLAGS_TIMESTAMP))
2638 net_enable_timestamp();
2642 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2643 int level, int type)
2645 struct sock_exterr_skb *serr;
2646 struct sk_buff *skb;
2650 skb = sock_dequeue_err_skb(sk);
2656 msg->msg_flags |= MSG_TRUNC;
2659 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2663 sock_recv_timestamp(msg, sk, skb);
2665 serr = SKB_EXT_ERR(skb);
2666 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2668 msg->msg_flags |= MSG_ERRQUEUE;
2676 EXPORT_SYMBOL(sock_recv_errqueue);
2679 * Get a socket option on an socket.
2681 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2682 * asynchronous errors should be reported by getsockopt. We assume
2683 * this means if you specify SO_ERROR (otherwise whats the point of it).
2685 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2686 char __user *optval, int __user *optlen)
2688 struct sock *sk = sock->sk;
2690 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2692 EXPORT_SYMBOL(sock_common_getsockopt);
2694 #ifdef CONFIG_COMPAT
2695 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2696 char __user *optval, int __user *optlen)
2698 struct sock *sk = sock->sk;
2700 if (sk->sk_prot->compat_getsockopt != NULL)
2701 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2703 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2705 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2708 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2711 struct sock *sk = sock->sk;
2715 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2716 flags & ~MSG_DONTWAIT, &addr_len);
2718 msg->msg_namelen = addr_len;
2721 EXPORT_SYMBOL(sock_common_recvmsg);
2724 * Set socket options on an inet socket.
2726 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2727 char __user *optval, unsigned int optlen)
2729 struct sock *sk = sock->sk;
2731 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2733 EXPORT_SYMBOL(sock_common_setsockopt);
2735 #ifdef CONFIG_COMPAT
2736 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2737 char __user *optval, unsigned int optlen)
2739 struct sock *sk = sock->sk;
2741 if (sk->sk_prot->compat_setsockopt != NULL)
2742 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2744 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2746 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2749 void sk_common_release(struct sock *sk)
2751 if (sk->sk_prot->destroy)
2752 sk->sk_prot->destroy(sk);
2755 * Observation: when sock_common_release is called, processes have
2756 * no access to socket. But net still has.
2757 * Step one, detach it from networking:
2759 * A. Remove from hash tables.
2762 sk->sk_prot->unhash(sk);
2765 * In this point socket cannot receive new packets, but it is possible
2766 * that some packets are in flight because some CPU runs receiver and
2767 * did hash table lookup before we unhashed socket. They will achieve
2768 * receive queue and will be purged by socket destructor.
2770 * Also we still have packets pending on receive queue and probably,
2771 * our own packets waiting in device queues. sock_destroy will drain
2772 * receive queue, but transmitted packets will delay socket destruction
2773 * until the last reference will be released.
2778 xfrm_sk_free_policy(sk);
2780 sk_refcnt_debug_release(sk);
2782 if (sk->sk_frag.page) {
2783 put_page(sk->sk_frag.page);
2784 sk->sk_frag.page = NULL;
2789 EXPORT_SYMBOL(sk_common_release);
2791 #ifdef CONFIG_PROC_FS
2792 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2794 int val[PROTO_INUSE_NR];
2797 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2799 #ifdef CONFIG_NET_NS
2800 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2802 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2804 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2806 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2808 int cpu, idx = prot->inuse_idx;
2811 for_each_possible_cpu(cpu)
2812 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2814 return res >= 0 ? res : 0;
2816 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2818 static int __net_init sock_inuse_init_net(struct net *net)
2820 net->core.inuse = alloc_percpu(struct prot_inuse);
2821 return net->core.inuse ? 0 : -ENOMEM;
2824 static void __net_exit sock_inuse_exit_net(struct net *net)
2826 free_percpu(net->core.inuse);
2829 static struct pernet_operations net_inuse_ops = {
2830 .init = sock_inuse_init_net,
2831 .exit = sock_inuse_exit_net,
2834 static __init int net_inuse_init(void)
2836 if (register_pernet_subsys(&net_inuse_ops))
2837 panic("Cannot initialize net inuse counters");
2842 core_initcall(net_inuse_init);
2844 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2846 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2848 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2850 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2852 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2854 int cpu, idx = prot->inuse_idx;
2857 for_each_possible_cpu(cpu)
2858 res += per_cpu(prot_inuse, cpu).val[idx];
2860 return res >= 0 ? res : 0;
2862 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2865 static void assign_proto_idx(struct proto *prot)
2867 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2869 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2870 pr_err("PROTO_INUSE_NR exhausted\n");
2874 set_bit(prot->inuse_idx, proto_inuse_idx);
2877 static void release_proto_idx(struct proto *prot)
2879 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2880 clear_bit(prot->inuse_idx, proto_inuse_idx);
2883 static inline void assign_proto_idx(struct proto *prot)
2887 static inline void release_proto_idx(struct proto *prot)
2892 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2896 kfree(rsk_prot->slab_name);
2897 rsk_prot->slab_name = NULL;
2898 kmem_cache_destroy(rsk_prot->slab);
2899 rsk_prot->slab = NULL;
2902 static int req_prot_init(const struct proto *prot)
2904 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2909 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2911 if (!rsk_prot->slab_name)
2914 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2915 rsk_prot->obj_size, 0,
2916 prot->slab_flags, NULL);
2918 if (!rsk_prot->slab) {
2919 pr_crit("%s: Can't create request sock SLAB cache!\n",
2926 int proto_register(struct proto *prot, int alloc_slab)
2929 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2930 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2933 if (prot->slab == NULL) {
2934 pr_crit("%s: Can't create sock SLAB cache!\n",
2939 if (req_prot_init(prot))
2940 goto out_free_request_sock_slab;
2942 if (prot->twsk_prot != NULL) {
2943 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2945 if (prot->twsk_prot->twsk_slab_name == NULL)
2946 goto out_free_request_sock_slab;
2948 prot->twsk_prot->twsk_slab =
2949 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2950 prot->twsk_prot->twsk_obj_size,
2954 if (prot->twsk_prot->twsk_slab == NULL)
2955 goto out_free_timewait_sock_slab_name;
2959 mutex_lock(&proto_list_mutex);
2960 list_add(&prot->node, &proto_list);
2961 assign_proto_idx(prot);
2962 mutex_unlock(&proto_list_mutex);
2965 out_free_timewait_sock_slab_name:
2966 kfree(prot->twsk_prot->twsk_slab_name);
2967 out_free_request_sock_slab:
2968 req_prot_cleanup(prot->rsk_prot);
2970 kmem_cache_destroy(prot->slab);
2975 EXPORT_SYMBOL(proto_register);
2977 void proto_unregister(struct proto *prot)
2979 mutex_lock(&proto_list_mutex);
2980 release_proto_idx(prot);
2981 list_del(&prot->node);
2982 mutex_unlock(&proto_list_mutex);
2984 kmem_cache_destroy(prot->slab);
2987 req_prot_cleanup(prot->rsk_prot);
2989 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2990 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2991 kfree(prot->twsk_prot->twsk_slab_name);
2992 prot->twsk_prot->twsk_slab = NULL;
2995 EXPORT_SYMBOL(proto_unregister);
2997 #ifdef CONFIG_PROC_FS
2998 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2999 __acquires(proto_list_mutex)
3001 mutex_lock(&proto_list_mutex);
3002 return seq_list_start_head(&proto_list, *pos);
3005 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3007 return seq_list_next(v, &proto_list, pos);
3010 static void proto_seq_stop(struct seq_file *seq, void *v)
3011 __releases(proto_list_mutex)
3013 mutex_unlock(&proto_list_mutex);
3016 static char proto_method_implemented(const void *method)
3018 return method == NULL ? 'n' : 'y';
3020 static long sock_prot_memory_allocated(struct proto *proto)
3022 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3025 static char *sock_prot_memory_pressure(struct proto *proto)
3027 return proto->memory_pressure != NULL ?
3028 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3031 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3034 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3035 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3038 sock_prot_inuse_get(seq_file_net(seq), proto),
3039 sock_prot_memory_allocated(proto),
3040 sock_prot_memory_pressure(proto),
3042 proto->slab == NULL ? "no" : "yes",
3043 module_name(proto->owner),
3044 proto_method_implemented(proto->close),
3045 proto_method_implemented(proto->connect),
3046 proto_method_implemented(proto->disconnect),
3047 proto_method_implemented(proto->accept),
3048 proto_method_implemented(proto->ioctl),
3049 proto_method_implemented(proto->init),
3050 proto_method_implemented(proto->destroy),
3051 proto_method_implemented(proto->shutdown),
3052 proto_method_implemented(proto->setsockopt),
3053 proto_method_implemented(proto->getsockopt),
3054 proto_method_implemented(proto->sendmsg),
3055 proto_method_implemented(proto->recvmsg),
3056 proto_method_implemented(proto->sendpage),
3057 proto_method_implemented(proto->bind),
3058 proto_method_implemented(proto->backlog_rcv),
3059 proto_method_implemented(proto->hash),
3060 proto_method_implemented(proto->unhash),
3061 proto_method_implemented(proto->get_port),
3062 proto_method_implemented(proto->enter_memory_pressure));
3065 static int proto_seq_show(struct seq_file *seq, void *v)
3067 if (v == &proto_list)
3068 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3077 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3079 proto_seq_printf(seq, list_entry(v, struct proto, node));
3083 static const struct seq_operations proto_seq_ops = {
3084 .start = proto_seq_start,
3085 .next = proto_seq_next,
3086 .stop = proto_seq_stop,
3087 .show = proto_seq_show,
3090 static int proto_seq_open(struct inode *inode, struct file *file)
3092 return seq_open_net(inode, file, &proto_seq_ops,
3093 sizeof(struct seq_net_private));
3096 static const struct file_operations proto_seq_fops = {
3097 .owner = THIS_MODULE,
3098 .open = proto_seq_open,
3100 .llseek = seq_lseek,
3101 .release = seq_release_net,
3104 static __net_init int proto_init_net(struct net *net)
3106 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3112 static __net_exit void proto_exit_net(struct net *net)
3114 remove_proc_entry("protocols", net->proc_net);
3118 static __net_initdata struct pernet_operations proto_net_ops = {
3119 .init = proto_init_net,
3120 .exit = proto_exit_net,
3123 static int __init proto_init(void)
3125 return register_pernet_subsys(&proto_net_ops);
3128 subsys_initcall(proto_init);
3130 #endif /* PROC_FS */