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 <asm/unaligned.h>
95 #include <linux/capability.h>
96 #include <linux/errno.h>
97 #include <linux/errqueue.h>
98 #include <linux/types.h>
99 #include <linux/socket.h>
100 #include <linux/in.h>
101 #include <linux/kernel.h>
102 #include <linux/module.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <linux/sched.h>
106 #include <linux/sched/mm.h>
107 #include <linux/timer.h>
108 #include <linux/string.h>
109 #include <linux/sockios.h>
110 #include <linux/net.h>
111 #include <linux/mm.h>
112 #include <linux/slab.h>
113 #include <linux/interrupt.h>
114 #include <linux/poll.h>
115 #include <linux/tcp.h>
116 #include <linux/init.h>
117 #include <linux/highmem.h>
118 #include <linux/user_namespace.h>
119 #include <linux/static_key.h>
120 #include <linux/memcontrol.h>
121 #include <linux/prefetch.h>
123 #include <linux/uaccess.h>
125 #include <linux/netdevice.h>
126 #include <net/protocol.h>
127 #include <linux/skbuff.h>
128 #include <net/net_namespace.h>
129 #include <net/request_sock.h>
130 #include <net/sock.h>
131 #include <linux/net_tstamp.h>
132 #include <net/xfrm.h>
133 #include <linux/ipsec.h>
134 #include <net/cls_cgroup.h>
135 #include <net/netprio_cgroup.h>
136 #include <linux/sock_diag.h>
138 #include <linux/filter.h>
139 #include <net/sock_reuseport.h>
140 #include <net/bpf_sk_storage.h>
142 #include <trace/events/sock.h>
145 #include <net/busy_poll.h>
147 static DEFINE_MUTEX(proto_list_mutex);
148 static LIST_HEAD(proto_list);
150 static void sock_inuse_add(struct net *net, int val);
153 * sk_ns_capable - General socket capability test
154 * @sk: Socket to use a capability on or through
155 * @user_ns: The user namespace of the capability to use
156 * @cap: The capability to use
158 * Test to see if the opener of the socket had when the socket was
159 * created and the current process has the capability @cap in the user
160 * namespace @user_ns.
162 bool sk_ns_capable(const struct sock *sk,
163 struct user_namespace *user_ns, int cap)
165 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
166 ns_capable(user_ns, cap);
168 EXPORT_SYMBOL(sk_ns_capable);
171 * sk_capable - Socket global capability test
172 * @sk: Socket to use a capability on or through
173 * @cap: The global capability to use
175 * Test to see if the opener of the socket had when the socket was
176 * created and the current process has the capability @cap in all user
179 bool sk_capable(const struct sock *sk, int cap)
181 return sk_ns_capable(sk, &init_user_ns, cap);
183 EXPORT_SYMBOL(sk_capable);
186 * sk_net_capable - Network namespace socket capability test
187 * @sk: Socket to use a capability on or through
188 * @cap: The capability to use
190 * Test to see if the opener of the socket had when the socket was created
191 * and the current process has the capability @cap over the network namespace
192 * the socket is a member of.
194 bool sk_net_capable(const struct sock *sk, int cap)
196 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
198 EXPORT_SYMBOL(sk_net_capable);
201 * Each address family might have different locking rules, so we have
202 * one slock key per address family and separate keys for internal and
205 static struct lock_class_key af_family_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_keys[AF_MAX];
207 static struct lock_class_key af_family_slock_keys[AF_MAX];
208 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
211 * Make lock validator output more readable. (we pre-construct these
212 * strings build-time, so that runtime initialization of socket
216 #define _sock_locks(x) \
217 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
218 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
219 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
220 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
221 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
222 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
223 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
224 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
225 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
226 x "27" , x "28" , x "AF_CAN" , \
227 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
228 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
229 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
230 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
231 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
234 static const char *const af_family_key_strings[AF_MAX+1] = {
235 _sock_locks("sk_lock-")
237 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
238 _sock_locks("slock-")
240 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
241 _sock_locks("clock-")
244 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
245 _sock_locks("k-sk_lock-")
247 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
248 _sock_locks("k-slock-")
250 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
251 _sock_locks("k-clock-")
253 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
254 _sock_locks("rlock-")
256 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
257 _sock_locks("wlock-")
259 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
260 _sock_locks("elock-")
264 * sk_callback_lock and sk queues locking rules are per-address-family,
265 * so split the lock classes by using a per-AF key:
267 static struct lock_class_key af_callback_keys[AF_MAX];
268 static struct lock_class_key af_rlock_keys[AF_MAX];
269 static struct lock_class_key af_wlock_keys[AF_MAX];
270 static struct lock_class_key af_elock_keys[AF_MAX];
271 static struct lock_class_key af_kern_callback_keys[AF_MAX];
273 /* Run time adjustable parameters. */
274 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
275 EXPORT_SYMBOL(sysctl_wmem_max);
276 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
277 EXPORT_SYMBOL(sysctl_rmem_max);
278 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
279 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
281 /* Maximal space eaten by iovec or ancillary data plus some space */
282 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
283 EXPORT_SYMBOL(sysctl_optmem_max);
285 int sysctl_tstamp_allow_data __read_mostly = 1;
287 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
288 EXPORT_SYMBOL_GPL(memalloc_socks_key);
291 * sk_set_memalloc - sets %SOCK_MEMALLOC
292 * @sk: socket to set it on
294 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
295 * It's the responsibility of the admin to adjust min_free_kbytes
296 * to meet the requirements
298 void sk_set_memalloc(struct sock *sk)
300 sock_set_flag(sk, SOCK_MEMALLOC);
301 sk->sk_allocation |= __GFP_MEMALLOC;
302 static_branch_inc(&memalloc_socks_key);
304 EXPORT_SYMBOL_GPL(sk_set_memalloc);
306 void sk_clear_memalloc(struct sock *sk)
308 sock_reset_flag(sk, SOCK_MEMALLOC);
309 sk->sk_allocation &= ~__GFP_MEMALLOC;
310 static_branch_dec(&memalloc_socks_key);
313 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
314 * progress of swapping. SOCK_MEMALLOC may be cleared while
315 * it has rmem allocations due to the last swapfile being deactivated
316 * but there is a risk that the socket is unusable due to exceeding
317 * the rmem limits. Reclaim the reserves and obey rmem limits again.
321 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
323 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
326 unsigned int noreclaim_flag;
328 /* these should have been dropped before queueing */
329 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
331 noreclaim_flag = memalloc_noreclaim_save();
332 ret = sk->sk_backlog_rcv(sk, skb);
333 memalloc_noreclaim_restore(noreclaim_flag);
337 EXPORT_SYMBOL(__sk_backlog_rcv);
339 static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
341 struct __kernel_sock_timeval tv;
344 if (timeo == MAX_SCHEDULE_TIMEOUT) {
348 tv.tv_sec = timeo / HZ;
349 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
352 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
353 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
354 *(struct old_timeval32 *)optval = tv32;
359 struct __kernel_old_timeval old_tv;
360 old_tv.tv_sec = tv.tv_sec;
361 old_tv.tv_usec = tv.tv_usec;
362 *(struct __kernel_old_timeval *)optval = old_tv;
363 size = sizeof(old_tv);
365 *(struct __kernel_sock_timeval *)optval = tv;
372 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval)
374 struct __kernel_sock_timeval tv;
376 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
377 struct old_timeval32 tv32;
379 if (optlen < sizeof(tv32))
382 if (copy_from_user(&tv32, optval, sizeof(tv32)))
384 tv.tv_sec = tv32.tv_sec;
385 tv.tv_usec = tv32.tv_usec;
386 } else if (old_timeval) {
387 struct __kernel_old_timeval old_tv;
389 if (optlen < sizeof(old_tv))
391 if (copy_from_user(&old_tv, optval, sizeof(old_tv)))
393 tv.tv_sec = old_tv.tv_sec;
394 tv.tv_usec = old_tv.tv_usec;
396 if (optlen < sizeof(tv))
398 if (copy_from_user(&tv, optval, sizeof(tv)))
401 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
405 static int warned __read_mostly;
408 if (warned < 10 && net_ratelimit()) {
410 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
411 __func__, current->comm, task_pid_nr(current));
415 *timeo_p = MAX_SCHEDULE_TIMEOUT;
416 if (tv.tv_sec == 0 && tv.tv_usec == 0)
418 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
419 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
423 static void sock_warn_obsolete_bsdism(const char *name)
426 static char warncomm[TASK_COMM_LEN];
427 if (strcmp(warncomm, current->comm) && warned < 5) {
428 strcpy(warncomm, current->comm);
429 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
435 static bool sock_needs_netstamp(const struct sock *sk)
437 switch (sk->sk_family) {
446 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
448 if (sk->sk_flags & flags) {
449 sk->sk_flags &= ~flags;
450 if (sock_needs_netstamp(sk) &&
451 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
452 net_disable_timestamp();
457 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
460 struct sk_buff_head *list = &sk->sk_receive_queue;
462 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
463 atomic_inc(&sk->sk_drops);
464 trace_sock_rcvqueue_full(sk, skb);
468 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
469 atomic_inc(&sk->sk_drops);
474 skb_set_owner_r(skb, sk);
476 /* we escape from rcu protected region, make sure we dont leak
481 spin_lock_irqsave(&list->lock, flags);
482 sock_skb_set_dropcount(sk, skb);
483 __skb_queue_tail(list, skb);
484 spin_unlock_irqrestore(&list->lock, flags);
486 if (!sock_flag(sk, SOCK_DEAD))
487 sk->sk_data_ready(sk);
490 EXPORT_SYMBOL(__sock_queue_rcv_skb);
492 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
496 err = sk_filter(sk, skb);
500 return __sock_queue_rcv_skb(sk, skb);
502 EXPORT_SYMBOL(sock_queue_rcv_skb);
504 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
505 const int nested, unsigned int trim_cap, bool refcounted)
507 int rc = NET_RX_SUCCESS;
509 if (sk_filter_trim_cap(sk, skb, trim_cap))
510 goto discard_and_relse;
514 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
515 atomic_inc(&sk->sk_drops);
516 goto discard_and_relse;
519 bh_lock_sock_nested(sk);
522 if (!sock_owned_by_user(sk)) {
524 * trylock + unlock semantics:
526 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
528 rc = sk_backlog_rcv(sk, skb);
530 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
531 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
533 atomic_inc(&sk->sk_drops);
534 goto discard_and_relse;
546 EXPORT_SYMBOL(__sk_receive_skb);
548 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
550 struct dst_entry *dst = __sk_dst_get(sk);
552 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
553 sk_tx_queue_clear(sk);
554 sk->sk_dst_pending_confirm = 0;
555 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
562 EXPORT_SYMBOL(__sk_dst_check);
564 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
566 struct dst_entry *dst = sk_dst_get(sk);
568 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
576 EXPORT_SYMBOL(sk_dst_check);
578 static int sock_setbindtodevice_locked(struct sock *sk, int ifindex)
580 int ret = -ENOPROTOOPT;
581 #ifdef CONFIG_NETDEVICES
582 struct net *net = sock_net(sk);
586 if (!ns_capable(net->user_ns, CAP_NET_RAW))
593 sk->sk_bound_dev_if = ifindex;
594 if (sk->sk_prot->rehash)
595 sk->sk_prot->rehash(sk);
606 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
609 int ret = -ENOPROTOOPT;
610 #ifdef CONFIG_NETDEVICES
611 struct net *net = sock_net(sk);
612 char devname[IFNAMSIZ];
619 /* Bind this socket to a particular device like "eth0",
620 * as specified in the passed interface name. If the
621 * name is "" or the option length is zero the socket
624 if (optlen > IFNAMSIZ - 1)
625 optlen = IFNAMSIZ - 1;
626 memset(devname, 0, sizeof(devname));
629 if (copy_from_user(devname, optval, optlen))
633 if (devname[0] != '\0') {
634 struct net_device *dev;
637 dev = dev_get_by_name_rcu(net, devname);
639 index = dev->ifindex;
647 ret = sock_setbindtodevice_locked(sk, index);
656 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
657 int __user *optlen, int len)
659 int ret = -ENOPROTOOPT;
660 #ifdef CONFIG_NETDEVICES
661 struct net *net = sock_net(sk);
662 char devname[IFNAMSIZ];
664 if (sk->sk_bound_dev_if == 0) {
673 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
677 len = strlen(devname) + 1;
680 if (copy_to_user(optval, devname, len))
685 if (put_user(len, optlen))
696 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
699 sock_set_flag(sk, bit);
701 sock_reset_flag(sk, bit);
704 bool sk_mc_loop(struct sock *sk)
706 if (dev_recursion_level())
710 switch (sk->sk_family) {
712 return inet_sk(sk)->mc_loop;
713 #if IS_ENABLED(CONFIG_IPV6)
715 return inet6_sk(sk)->mc_loop;
721 EXPORT_SYMBOL(sk_mc_loop);
724 * This is meant for all protocols to use and covers goings on
725 * at the socket level. Everything here is generic.
728 int sock_setsockopt(struct socket *sock, int level, int optname,
729 char __user *optval, unsigned int optlen)
731 struct sock_txtime sk_txtime;
732 struct sock *sk = sock->sk;
739 * Options without arguments
742 if (optname == SO_BINDTODEVICE)
743 return sock_setbindtodevice(sk, optval, optlen);
745 if (optlen < sizeof(int))
748 if (get_user(val, (int __user *)optval))
751 valbool = val ? 1 : 0;
757 if (val && !capable(CAP_NET_ADMIN))
760 sock_valbool_flag(sk, SOCK_DBG, valbool);
763 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
766 sk->sk_reuseport = valbool;
775 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
779 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
782 /* Don't error on this BSD doesn't and if you think
783 * about it this is right. Otherwise apps have to
784 * play 'guess the biggest size' games. RCVBUF/SNDBUF
785 * are treated in BSD as hints
787 val = min_t(u32, val, sysctl_wmem_max);
789 /* Ensure val * 2 fits into an int, to prevent max_t()
790 * from treating it as a negative value.
792 val = min_t(int, val, INT_MAX / 2);
793 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
794 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
795 /* Wake up sending tasks if we upped the value. */
796 sk->sk_write_space(sk);
800 if (!capable(CAP_NET_ADMIN)) {
805 /* No negative values (to prevent underflow, as val will be
813 /* Don't error on this BSD doesn't and if you think
814 * about it this is right. Otherwise apps have to
815 * play 'guess the biggest size' games. RCVBUF/SNDBUF
816 * are treated in BSD as hints
818 val = min_t(u32, val, sysctl_rmem_max);
820 /* Ensure val * 2 fits into an int, to prevent max_t()
821 * from treating it as a negative value.
823 val = min_t(int, val, INT_MAX / 2);
824 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
826 * We double it on the way in to account for
827 * "struct sk_buff" etc. overhead. Applications
828 * assume that the SO_RCVBUF setting they make will
829 * allow that much actual data to be received on that
832 * Applications are unaware that "struct sk_buff" and
833 * other overheads allocate from the receive buffer
834 * during socket buffer allocation.
836 * And after considering the possible alternatives,
837 * returning the value we actually used in getsockopt
838 * is the most desirable behavior.
840 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
844 if (!capable(CAP_NET_ADMIN)) {
849 /* No negative values (to prevent underflow, as val will be
857 if (sk->sk_prot->keepalive)
858 sk->sk_prot->keepalive(sk, valbool);
859 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
863 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
867 sk->sk_no_check_tx = valbool;
871 if ((val >= 0 && val <= 6) ||
872 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
873 sk->sk_priority = val;
879 if (optlen < sizeof(ling)) {
880 ret = -EINVAL; /* 1003.1g */
883 if (copy_from_user(&ling, optval, sizeof(ling))) {
888 sock_reset_flag(sk, SOCK_LINGER);
890 #if (BITS_PER_LONG == 32)
891 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
892 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
895 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
896 sock_set_flag(sk, SOCK_LINGER);
901 sock_warn_obsolete_bsdism("setsockopt");
906 set_bit(SOCK_PASSCRED, &sock->flags);
908 clear_bit(SOCK_PASSCRED, &sock->flags);
911 case SO_TIMESTAMP_OLD:
912 case SO_TIMESTAMP_NEW:
913 case SO_TIMESTAMPNS_OLD:
914 case SO_TIMESTAMPNS_NEW:
916 if (optname == SO_TIMESTAMP_NEW || optname == SO_TIMESTAMPNS_NEW)
917 sock_set_flag(sk, SOCK_TSTAMP_NEW);
919 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
921 if (optname == SO_TIMESTAMP_OLD || optname == SO_TIMESTAMP_NEW)
922 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
924 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
925 sock_set_flag(sk, SOCK_RCVTSTAMP);
926 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
928 sock_reset_flag(sk, SOCK_RCVTSTAMP);
929 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
930 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
934 case SO_TIMESTAMPING_NEW:
935 sock_set_flag(sk, SOCK_TSTAMP_NEW);
937 case SO_TIMESTAMPING_OLD:
938 if (val & ~SOF_TIMESTAMPING_MASK) {
943 if (val & SOF_TIMESTAMPING_OPT_ID &&
944 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
945 if (sk->sk_protocol == IPPROTO_TCP &&
946 sk->sk_type == SOCK_STREAM) {
947 if ((1 << sk->sk_state) &
948 (TCPF_CLOSE | TCPF_LISTEN)) {
952 sk->sk_tskey = tcp_sk(sk)->snd_una;
958 if (val & SOF_TIMESTAMPING_OPT_STATS &&
959 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
964 sk->sk_tsflags = val;
965 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
966 sock_enable_timestamp(sk,
967 SOCK_TIMESTAMPING_RX_SOFTWARE);
969 if (optname == SO_TIMESTAMPING_NEW)
970 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
972 sock_disable_timestamp(sk,
973 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
980 if (sock->ops->set_rcvlowat)
981 ret = sock->ops->set_rcvlowat(sk, val);
983 sk->sk_rcvlowat = val ? : 1;
986 case SO_RCVTIMEO_OLD:
987 case SO_RCVTIMEO_NEW:
988 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD);
991 case SO_SNDTIMEO_OLD:
992 case SO_SNDTIMEO_NEW:
993 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD);
996 case SO_ATTACH_FILTER:
998 if (optlen == sizeof(struct sock_fprog)) {
999 struct sock_fprog fprog;
1002 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1005 ret = sk_attach_filter(&fprog, sk);
1011 if (optlen == sizeof(u32)) {
1015 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1018 ret = sk_attach_bpf(ufd, sk);
1022 case SO_ATTACH_REUSEPORT_CBPF:
1024 if (optlen == sizeof(struct sock_fprog)) {
1025 struct sock_fprog fprog;
1028 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1031 ret = sk_reuseport_attach_filter(&fprog, sk);
1035 case SO_ATTACH_REUSEPORT_EBPF:
1037 if (optlen == sizeof(u32)) {
1041 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1044 ret = sk_reuseport_attach_bpf(ufd, sk);
1048 case SO_DETACH_FILTER:
1049 ret = sk_detach_filter(sk);
1052 case SO_LOCK_FILTER:
1053 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1056 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1061 set_bit(SOCK_PASSSEC, &sock->flags);
1063 clear_bit(SOCK_PASSSEC, &sock->flags);
1066 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1068 } else if (val != sk->sk_mark) {
1075 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1078 case SO_WIFI_STATUS:
1079 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1083 if (sock->ops->set_peek_off)
1084 ret = sock->ops->set_peek_off(sk, val);
1090 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1093 case SO_SELECT_ERR_QUEUE:
1094 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1097 #ifdef CONFIG_NET_RX_BUSY_POLL
1099 /* allow unprivileged users to decrease the value */
1100 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1106 sk->sk_ll_usec = val;
1111 case SO_MAX_PACING_RATE:
1113 unsigned long ulval = (val == ~0U) ? ~0UL : val;
1115 if (sizeof(ulval) != sizeof(val) &&
1116 optlen >= sizeof(ulval) &&
1117 get_user(ulval, (unsigned long __user *)optval)) {
1122 cmpxchg(&sk->sk_pacing_status,
1125 sk->sk_max_pacing_rate = ulval;
1126 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1129 case SO_INCOMING_CPU:
1130 sk->sk_incoming_cpu = val;
1135 dst_negative_advice(sk);
1139 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1140 if (!((sk->sk_type == SOCK_STREAM &&
1141 sk->sk_protocol == IPPROTO_TCP) ||
1142 (sk->sk_type == SOCK_DGRAM &&
1143 sk->sk_protocol == IPPROTO_UDP)))
1145 } else if (sk->sk_family != PF_RDS) {
1149 if (val < 0 || val > 1)
1152 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1157 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1159 } else if (optlen != sizeof(struct sock_txtime)) {
1161 } else if (copy_from_user(&sk_txtime, optval,
1162 sizeof(struct sock_txtime))) {
1164 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1167 sock_valbool_flag(sk, SOCK_TXTIME, true);
1168 sk->sk_clockid = sk_txtime.clockid;
1169 sk->sk_txtime_deadline_mode =
1170 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1171 sk->sk_txtime_report_errors =
1172 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1176 case SO_BINDTOIFINDEX:
1177 ret = sock_setbindtodevice_locked(sk, val);
1187 EXPORT_SYMBOL(sock_setsockopt);
1190 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1191 struct ucred *ucred)
1193 ucred->pid = pid_vnr(pid);
1194 ucred->uid = ucred->gid = -1;
1196 struct user_namespace *current_ns = current_user_ns();
1198 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1199 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1203 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1205 struct user_namespace *user_ns = current_user_ns();
1208 for (i = 0; i < src->ngroups; i++)
1209 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1215 int sock_getsockopt(struct socket *sock, int level, int optname,
1216 char __user *optval, int __user *optlen)
1218 struct sock *sk = sock->sk;
1223 unsigned long ulval;
1225 struct old_timeval32 tm32;
1226 struct __kernel_old_timeval tm;
1227 struct __kernel_sock_timeval stm;
1228 struct sock_txtime txtime;
1231 int lv = sizeof(int);
1234 if (get_user(len, optlen))
1239 memset(&v, 0, sizeof(v));
1243 v.val = sock_flag(sk, SOCK_DBG);
1247 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1251 v.val = sock_flag(sk, SOCK_BROADCAST);
1255 v.val = sk->sk_sndbuf;
1259 v.val = sk->sk_rcvbuf;
1263 v.val = sk->sk_reuse;
1267 v.val = sk->sk_reuseport;
1271 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1275 v.val = sk->sk_type;
1279 v.val = sk->sk_protocol;
1283 v.val = sk->sk_family;
1287 v.val = -sock_error(sk);
1289 v.val = xchg(&sk->sk_err_soft, 0);
1293 v.val = sock_flag(sk, SOCK_URGINLINE);
1297 v.val = sk->sk_no_check_tx;
1301 v.val = sk->sk_priority;
1305 lv = sizeof(v.ling);
1306 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1307 v.ling.l_linger = sk->sk_lingertime / HZ;
1311 sock_warn_obsolete_bsdism("getsockopt");
1314 case SO_TIMESTAMP_OLD:
1315 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1316 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1317 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1320 case SO_TIMESTAMPNS_OLD:
1321 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1324 case SO_TIMESTAMP_NEW:
1325 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1328 case SO_TIMESTAMPNS_NEW:
1329 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1332 case SO_TIMESTAMPING_OLD:
1333 v.val = sk->sk_tsflags;
1336 case SO_RCVTIMEO_OLD:
1337 case SO_RCVTIMEO_NEW:
1338 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1341 case SO_SNDTIMEO_OLD:
1342 case SO_SNDTIMEO_NEW:
1343 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1347 v.val = sk->sk_rcvlowat;
1355 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1360 struct ucred peercred;
1361 if (len > sizeof(peercred))
1362 len = sizeof(peercred);
1363 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1364 if (copy_to_user(optval, &peercred, len))
1373 if (!sk->sk_peer_cred)
1376 n = sk->sk_peer_cred->group_info->ngroups;
1377 if (len < n * sizeof(gid_t)) {
1378 len = n * sizeof(gid_t);
1379 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1381 len = n * sizeof(gid_t);
1383 ret = groups_to_user((gid_t __user *)optval,
1384 sk->sk_peer_cred->group_info);
1394 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1399 if (copy_to_user(optval, address, len))
1404 /* Dubious BSD thing... Probably nobody even uses it, but
1405 * the UNIX standard wants it for whatever reason... -DaveM
1408 v.val = sk->sk_state == TCP_LISTEN;
1412 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1416 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1419 v.val = sk->sk_mark;
1423 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1426 case SO_WIFI_STATUS:
1427 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1431 if (!sock->ops->set_peek_off)
1434 v.val = sk->sk_peek_off;
1437 v.val = sock_flag(sk, SOCK_NOFCS);
1440 case SO_BINDTODEVICE:
1441 return sock_getbindtodevice(sk, optval, optlen, len);
1444 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1450 case SO_LOCK_FILTER:
1451 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1454 case SO_BPF_EXTENSIONS:
1455 v.val = bpf_tell_extensions();
1458 case SO_SELECT_ERR_QUEUE:
1459 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1462 #ifdef CONFIG_NET_RX_BUSY_POLL
1464 v.val = sk->sk_ll_usec;
1468 case SO_MAX_PACING_RATE:
1469 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1470 lv = sizeof(v.ulval);
1471 v.ulval = sk->sk_max_pacing_rate;
1474 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1478 case SO_INCOMING_CPU:
1479 v.val = sk->sk_incoming_cpu;
1484 u32 meminfo[SK_MEMINFO_VARS];
1486 if (get_user(len, optlen))
1489 sk_get_meminfo(sk, meminfo);
1491 len = min_t(unsigned int, len, sizeof(meminfo));
1492 if (copy_to_user(optval, &meminfo, len))
1498 #ifdef CONFIG_NET_RX_BUSY_POLL
1499 case SO_INCOMING_NAPI_ID:
1500 v.val = READ_ONCE(sk->sk_napi_id);
1502 /* aggregate non-NAPI IDs down to 0 */
1503 if (v.val < MIN_NAPI_ID)
1513 v.val64 = sock_gen_cookie(sk);
1517 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1521 lv = sizeof(v.txtime);
1522 v.txtime.clockid = sk->sk_clockid;
1523 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1524 SOF_TXTIME_DEADLINE_MODE : 0;
1525 v.txtime.flags |= sk->sk_txtime_report_errors ?
1526 SOF_TXTIME_REPORT_ERRORS : 0;
1529 case SO_BINDTOIFINDEX:
1530 v.val = sk->sk_bound_dev_if;
1534 /* We implement the SO_SNDLOWAT etc to not be settable
1537 return -ENOPROTOOPT;
1542 if (copy_to_user(optval, &v, len))
1545 if (put_user(len, optlen))
1551 * Initialize an sk_lock.
1553 * (We also register the sk_lock with the lock validator.)
1555 static inline void sock_lock_init(struct sock *sk)
1557 if (sk->sk_kern_sock)
1558 sock_lock_init_class_and_name(
1560 af_family_kern_slock_key_strings[sk->sk_family],
1561 af_family_kern_slock_keys + sk->sk_family,
1562 af_family_kern_key_strings[sk->sk_family],
1563 af_family_kern_keys + sk->sk_family);
1565 sock_lock_init_class_and_name(
1567 af_family_slock_key_strings[sk->sk_family],
1568 af_family_slock_keys + sk->sk_family,
1569 af_family_key_strings[sk->sk_family],
1570 af_family_keys + sk->sk_family);
1574 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1575 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1576 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1578 static void sock_copy(struct sock *nsk, const struct sock *osk)
1580 #ifdef CONFIG_SECURITY_NETWORK
1581 void *sptr = nsk->sk_security;
1583 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1585 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1586 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1588 #ifdef CONFIG_SECURITY_NETWORK
1589 nsk->sk_security = sptr;
1590 security_sk_clone(osk, nsk);
1594 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1598 struct kmem_cache *slab;
1602 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1605 if (priority & __GFP_ZERO)
1606 sk_prot_clear_nulls(sk, prot->obj_size);
1608 sk = kmalloc(prot->obj_size, priority);
1611 if (security_sk_alloc(sk, family, priority))
1614 if (!try_module_get(prot->owner))
1616 sk_tx_queue_clear(sk);
1622 security_sk_free(sk);
1625 kmem_cache_free(slab, sk);
1631 static void sk_prot_free(struct proto *prot, struct sock *sk)
1633 struct kmem_cache *slab;
1634 struct module *owner;
1636 owner = prot->owner;
1639 cgroup_sk_free(&sk->sk_cgrp_data);
1640 mem_cgroup_sk_free(sk);
1641 security_sk_free(sk);
1643 kmem_cache_free(slab, sk);
1650 * sk_alloc - All socket objects are allocated here
1651 * @net: the applicable net namespace
1652 * @family: protocol family
1653 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1654 * @prot: struct proto associated with this new sock instance
1655 * @kern: is this to be a kernel socket?
1657 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1658 struct proto *prot, int kern)
1662 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1664 sk->sk_family = family;
1666 * See comment in struct sock definition to understand
1667 * why we need sk_prot_creator -acme
1669 sk->sk_prot = sk->sk_prot_creator = prot;
1670 sk->sk_kern_sock = kern;
1672 sk->sk_net_refcnt = kern ? 0 : 1;
1673 if (likely(sk->sk_net_refcnt)) {
1675 sock_inuse_add(net, 1);
1678 sock_net_set(sk, net);
1679 refcount_set(&sk->sk_wmem_alloc, 1);
1681 mem_cgroup_sk_alloc(sk);
1682 cgroup_sk_alloc(&sk->sk_cgrp_data);
1683 sock_update_classid(&sk->sk_cgrp_data);
1684 sock_update_netprioidx(&sk->sk_cgrp_data);
1689 EXPORT_SYMBOL(sk_alloc);
1691 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1692 * grace period. This is the case for UDP sockets and TCP listeners.
1694 static void __sk_destruct(struct rcu_head *head)
1696 struct sock *sk = container_of(head, struct sock, sk_rcu);
1697 struct sk_filter *filter;
1699 if (sk->sk_destruct)
1700 sk->sk_destruct(sk);
1702 filter = rcu_dereference_check(sk->sk_filter,
1703 refcount_read(&sk->sk_wmem_alloc) == 0);
1705 sk_filter_uncharge(sk, filter);
1706 RCU_INIT_POINTER(sk->sk_filter, NULL);
1708 if (rcu_access_pointer(sk->sk_reuseport_cb))
1709 reuseport_detach_sock(sk);
1711 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1713 #ifdef CONFIG_BPF_SYSCALL
1714 bpf_sk_storage_free(sk);
1717 if (atomic_read(&sk->sk_omem_alloc))
1718 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1719 __func__, atomic_read(&sk->sk_omem_alloc));
1721 if (sk->sk_frag.page) {
1722 put_page(sk->sk_frag.page);
1723 sk->sk_frag.page = NULL;
1726 if (sk->sk_peer_cred)
1727 put_cred(sk->sk_peer_cred);
1728 put_pid(sk->sk_peer_pid);
1729 if (likely(sk->sk_net_refcnt))
1730 put_net(sock_net(sk));
1731 sk_prot_free(sk->sk_prot_creator, sk);
1734 void sk_destruct(struct sock *sk)
1736 if (sock_flag(sk, SOCK_RCU_FREE))
1737 call_rcu(&sk->sk_rcu, __sk_destruct);
1739 __sk_destruct(&sk->sk_rcu);
1742 static void __sk_free(struct sock *sk)
1744 if (likely(sk->sk_net_refcnt))
1745 sock_inuse_add(sock_net(sk), -1);
1747 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1748 sock_diag_broadcast_destroy(sk);
1753 void sk_free(struct sock *sk)
1756 * We subtract one from sk_wmem_alloc and can know if
1757 * some packets are still in some tx queue.
1758 * If not null, sock_wfree() will call __sk_free(sk) later
1760 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1763 EXPORT_SYMBOL(sk_free);
1765 static void sk_init_common(struct sock *sk)
1767 skb_queue_head_init(&sk->sk_receive_queue);
1768 skb_queue_head_init(&sk->sk_write_queue);
1769 skb_queue_head_init(&sk->sk_error_queue);
1771 rwlock_init(&sk->sk_callback_lock);
1772 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1773 af_rlock_keys + sk->sk_family,
1774 af_family_rlock_key_strings[sk->sk_family]);
1775 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1776 af_wlock_keys + sk->sk_family,
1777 af_family_wlock_key_strings[sk->sk_family]);
1778 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1779 af_elock_keys + sk->sk_family,
1780 af_family_elock_key_strings[sk->sk_family]);
1781 lockdep_set_class_and_name(&sk->sk_callback_lock,
1782 af_callback_keys + sk->sk_family,
1783 af_family_clock_key_strings[sk->sk_family]);
1787 * sk_clone_lock - clone a socket, and lock its clone
1788 * @sk: the socket to clone
1789 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1791 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1793 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1796 bool is_charged = true;
1798 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1799 if (newsk != NULL) {
1800 struct sk_filter *filter;
1802 sock_copy(newsk, sk);
1804 newsk->sk_prot_creator = sk->sk_prot;
1807 if (likely(newsk->sk_net_refcnt))
1808 get_net(sock_net(newsk));
1809 sk_node_init(&newsk->sk_node);
1810 sock_lock_init(newsk);
1811 bh_lock_sock(newsk);
1812 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1813 newsk->sk_backlog.len = 0;
1815 atomic_set(&newsk->sk_rmem_alloc, 0);
1817 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1819 refcount_set(&newsk->sk_wmem_alloc, 1);
1820 atomic_set(&newsk->sk_omem_alloc, 0);
1821 sk_init_common(newsk);
1823 newsk->sk_dst_cache = NULL;
1824 newsk->sk_dst_pending_confirm = 0;
1825 newsk->sk_wmem_queued = 0;
1826 newsk->sk_forward_alloc = 0;
1827 atomic_set(&newsk->sk_drops, 0);
1828 newsk->sk_send_head = NULL;
1829 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1830 atomic_set(&newsk->sk_zckey, 0);
1832 sock_reset_flag(newsk, SOCK_DONE);
1833 mem_cgroup_sk_alloc(newsk);
1834 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1837 filter = rcu_dereference(sk->sk_filter);
1839 /* though it's an empty new sock, the charging may fail
1840 * if sysctl_optmem_max was changed between creation of
1841 * original socket and cloning
1843 is_charged = sk_filter_charge(newsk, filter);
1844 RCU_INIT_POINTER(newsk->sk_filter, filter);
1847 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1848 /* We need to make sure that we don't uncharge the new
1849 * socket if we couldn't charge it in the first place
1850 * as otherwise we uncharge the parent's filter.
1853 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1854 sk_free_unlock_clone(newsk);
1858 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1861 newsk->sk_err_soft = 0;
1862 newsk->sk_priority = 0;
1863 newsk->sk_incoming_cpu = raw_smp_processor_id();
1864 if (likely(newsk->sk_net_refcnt))
1865 sock_inuse_add(sock_net(newsk), 1);
1868 * Before updating sk_refcnt, we must commit prior changes to memory
1869 * (Documentation/RCU/rculist_nulls.txt for details)
1872 refcount_set(&newsk->sk_refcnt, 2);
1875 * Increment the counter in the same struct proto as the master
1876 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1877 * is the same as sk->sk_prot->socks, as this field was copied
1880 * This _changes_ the previous behaviour, where
1881 * tcp_create_openreq_child always was incrementing the
1882 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1883 * to be taken into account in all callers. -acme
1885 sk_refcnt_debug_inc(newsk);
1886 sk_set_socket(newsk, NULL);
1887 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1889 if (newsk->sk_prot->sockets_allocated)
1890 sk_sockets_allocated_inc(newsk);
1892 if (sock_needs_netstamp(sk) &&
1893 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1894 net_enable_timestamp();
1899 EXPORT_SYMBOL_GPL(sk_clone_lock);
1901 void sk_free_unlock_clone(struct sock *sk)
1903 /* It is still raw copy of parent, so invalidate
1904 * destructor and make plain sk_free() */
1905 sk->sk_destruct = NULL;
1909 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1911 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1915 sk_dst_set(sk, dst);
1916 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1917 if (sk->sk_route_caps & NETIF_F_GSO)
1918 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1919 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1920 if (sk_can_gso(sk)) {
1921 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1922 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1924 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1925 sk->sk_gso_max_size = dst->dev->gso_max_size;
1926 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1929 sk->sk_gso_max_segs = max_segs;
1931 EXPORT_SYMBOL_GPL(sk_setup_caps);
1934 * Simple resource managers for sockets.
1939 * Write buffer destructor automatically called from kfree_skb.
1941 void sock_wfree(struct sk_buff *skb)
1943 struct sock *sk = skb->sk;
1944 unsigned int len = skb->truesize;
1946 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1948 * Keep a reference on sk_wmem_alloc, this will be released
1949 * after sk_write_space() call
1951 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1952 sk->sk_write_space(sk);
1956 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1957 * could not do because of in-flight packets
1959 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1962 EXPORT_SYMBOL(sock_wfree);
1964 /* This variant of sock_wfree() is used by TCP,
1965 * since it sets SOCK_USE_WRITE_QUEUE.
1967 void __sock_wfree(struct sk_buff *skb)
1969 struct sock *sk = skb->sk;
1971 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1975 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1980 if (unlikely(!sk_fullsock(sk))) {
1981 skb->destructor = sock_edemux;
1986 skb->destructor = sock_wfree;
1987 skb_set_hash_from_sk(skb, sk);
1989 * We used to take a refcount on sk, but following operation
1990 * is enough to guarantee sk_free() wont free this sock until
1991 * all in-flight packets are completed
1993 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1995 EXPORT_SYMBOL(skb_set_owner_w);
1997 /* This helper is used by netem, as it can hold packets in its
1998 * delay queue. We want to allow the owner socket to send more
1999 * packets, as if they were already TX completed by a typical driver.
2000 * But we also want to keep skb->sk set because some packet schedulers
2001 * rely on it (sch_fq for example).
2003 void skb_orphan_partial(struct sk_buff *skb)
2005 if (skb_is_tcp_pure_ack(skb))
2008 if (skb->destructor == sock_wfree
2010 || skb->destructor == tcp_wfree
2013 struct sock *sk = skb->sk;
2015 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2016 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2017 skb->destructor = sock_efree;
2023 EXPORT_SYMBOL(skb_orphan_partial);
2026 * Read buffer destructor automatically called from kfree_skb.
2028 void sock_rfree(struct sk_buff *skb)
2030 struct sock *sk = skb->sk;
2031 unsigned int len = skb->truesize;
2033 atomic_sub(len, &sk->sk_rmem_alloc);
2034 sk_mem_uncharge(sk, len);
2036 EXPORT_SYMBOL(sock_rfree);
2039 * Buffer destructor for skbs that are not used directly in read or write
2040 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2042 void sock_efree(struct sk_buff *skb)
2046 EXPORT_SYMBOL(sock_efree);
2048 kuid_t sock_i_uid(struct sock *sk)
2052 read_lock_bh(&sk->sk_callback_lock);
2053 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2054 read_unlock_bh(&sk->sk_callback_lock);
2057 EXPORT_SYMBOL(sock_i_uid);
2059 unsigned long sock_i_ino(struct sock *sk)
2063 read_lock_bh(&sk->sk_callback_lock);
2064 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2065 read_unlock_bh(&sk->sk_callback_lock);
2068 EXPORT_SYMBOL(sock_i_ino);
2071 * Allocate a skb from the socket's send buffer.
2073 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2076 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2077 struct sk_buff *skb = alloc_skb(size, priority);
2079 skb_set_owner_w(skb, sk);
2085 EXPORT_SYMBOL(sock_wmalloc);
2087 static void sock_ofree(struct sk_buff *skb)
2089 struct sock *sk = skb->sk;
2091 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2094 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2097 struct sk_buff *skb;
2099 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2100 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2104 skb = alloc_skb(size, priority);
2108 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2110 skb->destructor = sock_ofree;
2115 * Allocate a memory block from the socket's option memory buffer.
2117 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2119 if ((unsigned int)size <= sysctl_optmem_max &&
2120 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2122 /* First do the add, to avoid the race if kmalloc
2125 atomic_add(size, &sk->sk_omem_alloc);
2126 mem = kmalloc(size, priority);
2129 atomic_sub(size, &sk->sk_omem_alloc);
2133 EXPORT_SYMBOL(sock_kmalloc);
2135 /* Free an option memory block. Note, we actually want the inline
2136 * here as this allows gcc to detect the nullify and fold away the
2137 * condition entirely.
2139 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2142 if (WARN_ON_ONCE(!mem))
2148 atomic_sub(size, &sk->sk_omem_alloc);
2151 void sock_kfree_s(struct sock *sk, void *mem, int size)
2153 __sock_kfree_s(sk, mem, size, false);
2155 EXPORT_SYMBOL(sock_kfree_s);
2157 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2159 __sock_kfree_s(sk, mem, size, true);
2161 EXPORT_SYMBOL(sock_kzfree_s);
2163 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2164 I think, these locks should be removed for datagram sockets.
2166 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2170 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2174 if (signal_pending(current))
2176 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2177 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2178 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2180 if (sk->sk_shutdown & SEND_SHUTDOWN)
2184 timeo = schedule_timeout(timeo);
2186 finish_wait(sk_sleep(sk), &wait);
2192 * Generic send/receive buffer handlers
2195 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2196 unsigned long data_len, int noblock,
2197 int *errcode, int max_page_order)
2199 struct sk_buff *skb;
2203 timeo = sock_sndtimeo(sk, noblock);
2205 err = sock_error(sk);
2210 if (sk->sk_shutdown & SEND_SHUTDOWN)
2213 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2216 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2217 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2221 if (signal_pending(current))
2223 timeo = sock_wait_for_wmem(sk, timeo);
2225 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2226 errcode, sk->sk_allocation);
2228 skb_set_owner_w(skb, sk);
2232 err = sock_intr_errno(timeo);
2237 EXPORT_SYMBOL(sock_alloc_send_pskb);
2239 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2240 int noblock, int *errcode)
2242 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2244 EXPORT_SYMBOL(sock_alloc_send_skb);
2246 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2247 struct sockcm_cookie *sockc)
2251 switch (cmsg->cmsg_type) {
2253 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2255 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2257 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2259 case SO_TIMESTAMPING_OLD:
2260 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2263 tsflags = *(u32 *)CMSG_DATA(cmsg);
2264 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2267 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2268 sockc->tsflags |= tsflags;
2271 if (!sock_flag(sk, SOCK_TXTIME))
2273 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2275 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2277 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2279 case SCM_CREDENTIALS:
2286 EXPORT_SYMBOL(__sock_cmsg_send);
2288 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2289 struct sockcm_cookie *sockc)
2291 struct cmsghdr *cmsg;
2294 for_each_cmsghdr(cmsg, msg) {
2295 if (!CMSG_OK(msg, cmsg))
2297 if (cmsg->cmsg_level != SOL_SOCKET)
2299 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2305 EXPORT_SYMBOL(sock_cmsg_send);
2307 static void sk_enter_memory_pressure(struct sock *sk)
2309 if (!sk->sk_prot->enter_memory_pressure)
2312 sk->sk_prot->enter_memory_pressure(sk);
2315 static void sk_leave_memory_pressure(struct sock *sk)
2317 if (sk->sk_prot->leave_memory_pressure) {
2318 sk->sk_prot->leave_memory_pressure(sk);
2320 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2322 if (memory_pressure && *memory_pressure)
2323 *memory_pressure = 0;
2327 /* On 32bit arches, an skb frag is limited to 2^15 */
2328 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2331 * skb_page_frag_refill - check that a page_frag contains enough room
2332 * @sz: minimum size of the fragment we want to get
2333 * @pfrag: pointer to page_frag
2334 * @gfp: priority for memory allocation
2336 * Note: While this allocator tries to use high order pages, there is
2337 * no guarantee that allocations succeed. Therefore, @sz MUST be
2338 * less or equal than PAGE_SIZE.
2340 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2343 if (page_ref_count(pfrag->page) == 1) {
2347 if (pfrag->offset + sz <= pfrag->size)
2349 put_page(pfrag->page);
2353 if (SKB_FRAG_PAGE_ORDER) {
2354 /* Avoid direct reclaim but allow kswapd to wake */
2355 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2356 __GFP_COMP | __GFP_NOWARN |
2358 SKB_FRAG_PAGE_ORDER);
2359 if (likely(pfrag->page)) {
2360 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2364 pfrag->page = alloc_page(gfp);
2365 if (likely(pfrag->page)) {
2366 pfrag->size = PAGE_SIZE;
2371 EXPORT_SYMBOL(skb_page_frag_refill);
2373 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2375 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2378 sk_enter_memory_pressure(sk);
2379 sk_stream_moderate_sndbuf(sk);
2382 EXPORT_SYMBOL(sk_page_frag_refill);
2384 static void __lock_sock(struct sock *sk)
2385 __releases(&sk->sk_lock.slock)
2386 __acquires(&sk->sk_lock.slock)
2391 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2392 TASK_UNINTERRUPTIBLE);
2393 spin_unlock_bh(&sk->sk_lock.slock);
2395 spin_lock_bh(&sk->sk_lock.slock);
2396 if (!sock_owned_by_user(sk))
2399 finish_wait(&sk->sk_lock.wq, &wait);
2402 void __release_sock(struct sock *sk)
2403 __releases(&sk->sk_lock.slock)
2404 __acquires(&sk->sk_lock.slock)
2406 struct sk_buff *skb, *next;
2408 while ((skb = sk->sk_backlog.head) != NULL) {
2409 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2411 spin_unlock_bh(&sk->sk_lock.slock);
2416 WARN_ON_ONCE(skb_dst_is_noref(skb));
2417 skb_mark_not_on_list(skb);
2418 sk_backlog_rcv(sk, skb);
2423 } while (skb != NULL);
2425 spin_lock_bh(&sk->sk_lock.slock);
2429 * Doing the zeroing here guarantee we can not loop forever
2430 * while a wild producer attempts to flood us.
2432 sk->sk_backlog.len = 0;
2435 void __sk_flush_backlog(struct sock *sk)
2437 spin_lock_bh(&sk->sk_lock.slock);
2439 spin_unlock_bh(&sk->sk_lock.slock);
2443 * sk_wait_data - wait for data to arrive at sk_receive_queue
2444 * @sk: sock to wait on
2445 * @timeo: for how long
2446 * @skb: last skb seen on sk_receive_queue
2448 * Now socket state including sk->sk_err is changed only under lock,
2449 * hence we may omit checks after joining wait queue.
2450 * We check receive queue before schedule() only as optimization;
2451 * it is very likely that release_sock() added new data.
2453 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2455 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2458 add_wait_queue(sk_sleep(sk), &wait);
2459 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2460 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2461 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2462 remove_wait_queue(sk_sleep(sk), &wait);
2465 EXPORT_SYMBOL(sk_wait_data);
2468 * __sk_mem_raise_allocated - increase memory_allocated
2470 * @size: memory size to allocate
2471 * @amt: pages to allocate
2472 * @kind: allocation type
2474 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2476 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2478 struct proto *prot = sk->sk_prot;
2479 long allocated = sk_memory_allocated_add(sk, amt);
2480 bool charged = true;
2482 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2483 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2484 goto suppress_allocation;
2487 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2488 sk_leave_memory_pressure(sk);
2492 /* Under pressure. */
2493 if (allocated > sk_prot_mem_limits(sk, 1))
2494 sk_enter_memory_pressure(sk);
2496 /* Over hard limit. */
2497 if (allocated > sk_prot_mem_limits(sk, 2))
2498 goto suppress_allocation;
2500 /* guarantee minimum buffer size under pressure */
2501 if (kind == SK_MEM_RECV) {
2502 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2505 } else { /* SK_MEM_SEND */
2506 int wmem0 = sk_get_wmem0(sk, prot);
2508 if (sk->sk_type == SOCK_STREAM) {
2509 if (sk->sk_wmem_queued < wmem0)
2511 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2516 if (sk_has_memory_pressure(sk)) {
2519 if (!sk_under_memory_pressure(sk))
2521 alloc = sk_sockets_allocated_read_positive(sk);
2522 if (sk_prot_mem_limits(sk, 2) > alloc *
2523 sk_mem_pages(sk->sk_wmem_queued +
2524 atomic_read(&sk->sk_rmem_alloc) +
2525 sk->sk_forward_alloc))
2529 suppress_allocation:
2531 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2532 sk_stream_moderate_sndbuf(sk);
2534 /* Fail only if socket is _under_ its sndbuf.
2535 * In this case we cannot block, so that we have to fail.
2537 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2541 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2542 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2544 sk_memory_allocated_sub(sk, amt);
2546 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2547 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2551 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2554 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2556 * @size: memory size to allocate
2557 * @kind: allocation type
2559 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2560 * rmem allocation. This function assumes that protocols which have
2561 * memory_pressure use sk_wmem_queued as write buffer accounting.
2563 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2565 int ret, amt = sk_mem_pages(size);
2567 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2568 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2570 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2573 EXPORT_SYMBOL(__sk_mem_schedule);
2576 * __sk_mem_reduce_allocated - reclaim memory_allocated
2578 * @amount: number of quanta
2580 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2582 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2584 sk_memory_allocated_sub(sk, amount);
2586 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2587 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2589 if (sk_under_memory_pressure(sk) &&
2590 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2591 sk_leave_memory_pressure(sk);
2593 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2596 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2598 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2600 void __sk_mem_reclaim(struct sock *sk, int amount)
2602 amount >>= SK_MEM_QUANTUM_SHIFT;
2603 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2604 __sk_mem_reduce_allocated(sk, amount);
2606 EXPORT_SYMBOL(__sk_mem_reclaim);
2608 int sk_set_peek_off(struct sock *sk, int val)
2610 sk->sk_peek_off = val;
2613 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2616 * Set of default routines for initialising struct proto_ops when
2617 * the protocol does not support a particular function. In certain
2618 * cases where it makes no sense for a protocol to have a "do nothing"
2619 * function, some default processing is provided.
2622 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2626 EXPORT_SYMBOL(sock_no_bind);
2628 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2633 EXPORT_SYMBOL(sock_no_connect);
2635 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2639 EXPORT_SYMBOL(sock_no_socketpair);
2641 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2646 EXPORT_SYMBOL(sock_no_accept);
2648 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2653 EXPORT_SYMBOL(sock_no_getname);
2655 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2659 EXPORT_SYMBOL(sock_no_ioctl);
2661 int sock_no_listen(struct socket *sock, int backlog)
2665 EXPORT_SYMBOL(sock_no_listen);
2667 int sock_no_shutdown(struct socket *sock, int how)
2671 EXPORT_SYMBOL(sock_no_shutdown);
2673 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2674 char __user *optval, unsigned int optlen)
2678 EXPORT_SYMBOL(sock_no_setsockopt);
2680 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2681 char __user *optval, int __user *optlen)
2685 EXPORT_SYMBOL(sock_no_getsockopt);
2687 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2691 EXPORT_SYMBOL(sock_no_sendmsg);
2693 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2697 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2699 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2704 EXPORT_SYMBOL(sock_no_recvmsg);
2706 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2708 /* Mirror missing mmap method error code */
2711 EXPORT_SYMBOL(sock_no_mmap);
2713 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2716 struct msghdr msg = {.msg_flags = flags};
2718 char *kaddr = kmap(page);
2719 iov.iov_base = kaddr + offset;
2721 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2725 EXPORT_SYMBOL(sock_no_sendpage);
2727 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2728 int offset, size_t size, int flags)
2731 struct msghdr msg = {.msg_flags = flags};
2733 char *kaddr = kmap(page);
2735 iov.iov_base = kaddr + offset;
2737 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2741 EXPORT_SYMBOL(sock_no_sendpage_locked);
2744 * Default Socket Callbacks
2747 static void sock_def_wakeup(struct sock *sk)
2749 struct socket_wq *wq;
2752 wq = rcu_dereference(sk->sk_wq);
2753 if (skwq_has_sleeper(wq))
2754 wake_up_interruptible_all(&wq->wait);
2758 static void sock_def_error_report(struct sock *sk)
2760 struct socket_wq *wq;
2763 wq = rcu_dereference(sk->sk_wq);
2764 if (skwq_has_sleeper(wq))
2765 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2766 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2770 static void sock_def_readable(struct sock *sk)
2772 struct socket_wq *wq;
2775 wq = rcu_dereference(sk->sk_wq);
2776 if (skwq_has_sleeper(wq))
2777 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2778 EPOLLRDNORM | EPOLLRDBAND);
2779 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2783 static void sock_def_write_space(struct sock *sk)
2785 struct socket_wq *wq;
2789 /* Do not wake up a writer until he can make "significant"
2792 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2793 wq = rcu_dereference(sk->sk_wq);
2794 if (skwq_has_sleeper(wq))
2795 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2796 EPOLLWRNORM | EPOLLWRBAND);
2798 /* Should agree with poll, otherwise some programs break */
2799 if (sock_writeable(sk))
2800 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2806 static void sock_def_destruct(struct sock *sk)
2810 void sk_send_sigurg(struct sock *sk)
2812 if (sk->sk_socket && sk->sk_socket->file)
2813 if (send_sigurg(&sk->sk_socket->file->f_owner))
2814 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2816 EXPORT_SYMBOL(sk_send_sigurg);
2818 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2819 unsigned long expires)
2821 if (!mod_timer(timer, expires))
2824 EXPORT_SYMBOL(sk_reset_timer);
2826 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2828 if (del_timer(timer))
2831 EXPORT_SYMBOL(sk_stop_timer);
2833 void sock_init_data(struct socket *sock, struct sock *sk)
2836 sk->sk_send_head = NULL;
2838 timer_setup(&sk->sk_timer, NULL, 0);
2840 sk->sk_allocation = GFP_KERNEL;
2841 sk->sk_rcvbuf = sysctl_rmem_default;
2842 sk->sk_sndbuf = sysctl_wmem_default;
2843 sk->sk_state = TCP_CLOSE;
2844 sk_set_socket(sk, sock);
2846 sock_set_flag(sk, SOCK_ZAPPED);
2849 sk->sk_type = sock->type;
2850 RCU_INIT_POINTER(sk->sk_wq, sock->wq);
2852 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2854 RCU_INIT_POINTER(sk->sk_wq, NULL);
2855 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2858 rwlock_init(&sk->sk_callback_lock);
2859 if (sk->sk_kern_sock)
2860 lockdep_set_class_and_name(
2861 &sk->sk_callback_lock,
2862 af_kern_callback_keys + sk->sk_family,
2863 af_family_kern_clock_key_strings[sk->sk_family]);
2865 lockdep_set_class_and_name(
2866 &sk->sk_callback_lock,
2867 af_callback_keys + sk->sk_family,
2868 af_family_clock_key_strings[sk->sk_family]);
2870 sk->sk_state_change = sock_def_wakeup;
2871 sk->sk_data_ready = sock_def_readable;
2872 sk->sk_write_space = sock_def_write_space;
2873 sk->sk_error_report = sock_def_error_report;
2874 sk->sk_destruct = sock_def_destruct;
2876 sk->sk_frag.page = NULL;
2877 sk->sk_frag.offset = 0;
2878 sk->sk_peek_off = -1;
2880 sk->sk_peer_pid = NULL;
2881 sk->sk_peer_cred = NULL;
2882 sk->sk_write_pending = 0;
2883 sk->sk_rcvlowat = 1;
2884 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2885 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2887 sk->sk_stamp = SK_DEFAULT_STAMP;
2888 #if BITS_PER_LONG==32
2889 seqlock_init(&sk->sk_stamp_seq);
2891 atomic_set(&sk->sk_zckey, 0);
2893 #ifdef CONFIG_NET_RX_BUSY_POLL
2895 sk->sk_ll_usec = sysctl_net_busy_read;
2898 sk->sk_max_pacing_rate = ~0UL;
2899 sk->sk_pacing_rate = ~0UL;
2900 sk->sk_pacing_shift = 10;
2901 sk->sk_incoming_cpu = -1;
2903 sk_rx_queue_clear(sk);
2905 * Before updating sk_refcnt, we must commit prior changes to memory
2906 * (Documentation/RCU/rculist_nulls.txt for details)
2909 refcount_set(&sk->sk_refcnt, 1);
2910 atomic_set(&sk->sk_drops, 0);
2912 EXPORT_SYMBOL(sock_init_data);
2914 void lock_sock_nested(struct sock *sk, int subclass)
2917 spin_lock_bh(&sk->sk_lock.slock);
2918 if (sk->sk_lock.owned)
2920 sk->sk_lock.owned = 1;
2921 spin_unlock(&sk->sk_lock.slock);
2923 * The sk_lock has mutex_lock() semantics here:
2925 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2928 EXPORT_SYMBOL(lock_sock_nested);
2930 void release_sock(struct sock *sk)
2932 spin_lock_bh(&sk->sk_lock.slock);
2933 if (sk->sk_backlog.tail)
2936 /* Warning : release_cb() might need to release sk ownership,
2937 * ie call sock_release_ownership(sk) before us.
2939 if (sk->sk_prot->release_cb)
2940 sk->sk_prot->release_cb(sk);
2942 sock_release_ownership(sk);
2943 if (waitqueue_active(&sk->sk_lock.wq))
2944 wake_up(&sk->sk_lock.wq);
2945 spin_unlock_bh(&sk->sk_lock.slock);
2947 EXPORT_SYMBOL(release_sock);
2950 * lock_sock_fast - fast version of lock_sock
2953 * This version should be used for very small section, where process wont block
2954 * return false if fast path is taken:
2956 * sk_lock.slock locked, owned = 0, BH disabled
2958 * return true if slow path is taken:
2960 * sk_lock.slock unlocked, owned = 1, BH enabled
2962 bool lock_sock_fast(struct sock *sk)
2965 spin_lock_bh(&sk->sk_lock.slock);
2967 if (!sk->sk_lock.owned)
2969 * Note : We must disable BH
2974 sk->sk_lock.owned = 1;
2975 spin_unlock(&sk->sk_lock.slock);
2977 * The sk_lock has mutex_lock() semantics here:
2979 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2983 EXPORT_SYMBOL(lock_sock_fast);
2985 int sock_gettstamp(struct socket *sock, void __user *userstamp,
2986 bool timeval, bool time32)
2988 struct sock *sk = sock->sk;
2989 struct timespec64 ts;
2991 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2992 ts = ktime_to_timespec64(sock_read_timestamp(sk));
2993 if (ts.tv_sec == -1)
2995 if (ts.tv_sec == 0) {
2996 ktime_t kt = ktime_get_real();
2997 sock_write_timestamp(sk, kt);;
2998 ts = ktime_to_timespec64(kt);
3004 #ifdef CONFIG_COMPAT_32BIT_TIME
3006 return put_old_timespec32(&ts, userstamp);
3008 #ifdef CONFIG_SPARC64
3009 /* beware of padding in sparc64 timeval */
3010 if (timeval && !in_compat_syscall()) {
3011 struct __kernel_old_timeval __user tv = {
3012 .tv_sec = ts.tv_sec,
3013 .tv_usec = ts.tv_nsec,
3015 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3020 return put_timespec64(&ts, userstamp);
3022 EXPORT_SYMBOL(sock_gettstamp);
3024 void sock_enable_timestamp(struct sock *sk, int flag)
3026 if (!sock_flag(sk, flag)) {
3027 unsigned long previous_flags = sk->sk_flags;
3029 sock_set_flag(sk, flag);
3031 * we just set one of the two flags which require net
3032 * time stamping, but time stamping might have been on
3033 * already because of the other one
3035 if (sock_needs_netstamp(sk) &&
3036 !(previous_flags & SK_FLAGS_TIMESTAMP))
3037 net_enable_timestamp();
3041 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3042 int level, int type)
3044 struct sock_exterr_skb *serr;
3045 struct sk_buff *skb;
3049 skb = sock_dequeue_err_skb(sk);
3055 msg->msg_flags |= MSG_TRUNC;
3058 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3062 sock_recv_timestamp(msg, sk, skb);
3064 serr = SKB_EXT_ERR(skb);
3065 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3067 msg->msg_flags |= MSG_ERRQUEUE;
3075 EXPORT_SYMBOL(sock_recv_errqueue);
3078 * Get a socket option on an socket.
3080 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3081 * asynchronous errors should be reported by getsockopt. We assume
3082 * this means if you specify SO_ERROR (otherwise whats the point of it).
3084 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3085 char __user *optval, int __user *optlen)
3087 struct sock *sk = sock->sk;
3089 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3091 EXPORT_SYMBOL(sock_common_getsockopt);
3093 #ifdef CONFIG_COMPAT
3094 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3095 char __user *optval, int __user *optlen)
3097 struct sock *sk = sock->sk;
3099 if (sk->sk_prot->compat_getsockopt != NULL)
3100 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3102 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3104 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3107 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3110 struct sock *sk = sock->sk;
3114 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3115 flags & ~MSG_DONTWAIT, &addr_len);
3117 msg->msg_namelen = addr_len;
3120 EXPORT_SYMBOL(sock_common_recvmsg);
3123 * Set socket options on an inet socket.
3125 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3126 char __user *optval, unsigned int optlen)
3128 struct sock *sk = sock->sk;
3130 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3132 EXPORT_SYMBOL(sock_common_setsockopt);
3134 #ifdef CONFIG_COMPAT
3135 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3136 char __user *optval, unsigned int optlen)
3138 struct sock *sk = sock->sk;
3140 if (sk->sk_prot->compat_setsockopt != NULL)
3141 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3143 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3145 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3148 void sk_common_release(struct sock *sk)
3150 if (sk->sk_prot->destroy)
3151 sk->sk_prot->destroy(sk);
3154 * Observation: when sock_common_release is called, processes have
3155 * no access to socket. But net still has.
3156 * Step one, detach it from networking:
3158 * A. Remove from hash tables.
3161 sk->sk_prot->unhash(sk);
3164 * In this point socket cannot receive new packets, but it is possible
3165 * that some packets are in flight because some CPU runs receiver and
3166 * did hash table lookup before we unhashed socket. They will achieve
3167 * receive queue and will be purged by socket destructor.
3169 * Also we still have packets pending on receive queue and probably,
3170 * our own packets waiting in device queues. sock_destroy will drain
3171 * receive queue, but transmitted packets will delay socket destruction
3172 * until the last reference will be released.
3177 xfrm_sk_free_policy(sk);
3179 sk_refcnt_debug_release(sk);
3183 EXPORT_SYMBOL(sk_common_release);
3185 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3187 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3189 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3190 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3191 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3192 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3193 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3194 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3195 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3196 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3197 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3200 #ifdef CONFIG_PROC_FS
3201 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3203 int val[PROTO_INUSE_NR];
3206 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3208 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3210 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3212 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3214 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3216 int cpu, idx = prot->inuse_idx;
3219 for_each_possible_cpu(cpu)
3220 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3222 return res >= 0 ? res : 0;
3224 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3226 static void sock_inuse_add(struct net *net, int val)
3228 this_cpu_add(*net->core.sock_inuse, val);
3231 int sock_inuse_get(struct net *net)
3235 for_each_possible_cpu(cpu)
3236 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3241 EXPORT_SYMBOL_GPL(sock_inuse_get);
3243 static int __net_init sock_inuse_init_net(struct net *net)
3245 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3246 if (net->core.prot_inuse == NULL)
3249 net->core.sock_inuse = alloc_percpu(int);
3250 if (net->core.sock_inuse == NULL)
3256 free_percpu(net->core.prot_inuse);
3260 static void __net_exit sock_inuse_exit_net(struct net *net)
3262 free_percpu(net->core.prot_inuse);
3263 free_percpu(net->core.sock_inuse);
3266 static struct pernet_operations net_inuse_ops = {
3267 .init = sock_inuse_init_net,
3268 .exit = sock_inuse_exit_net,
3271 static __init int net_inuse_init(void)
3273 if (register_pernet_subsys(&net_inuse_ops))
3274 panic("Cannot initialize net inuse counters");
3279 core_initcall(net_inuse_init);
3281 static void assign_proto_idx(struct proto *prot)
3283 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3285 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3286 pr_err("PROTO_INUSE_NR exhausted\n");
3290 set_bit(prot->inuse_idx, proto_inuse_idx);
3293 static void release_proto_idx(struct proto *prot)
3295 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3296 clear_bit(prot->inuse_idx, proto_inuse_idx);
3299 static inline void assign_proto_idx(struct proto *prot)
3303 static inline void release_proto_idx(struct proto *prot)
3307 static void sock_inuse_add(struct net *net, int val)
3312 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3316 kfree(rsk_prot->slab_name);
3317 rsk_prot->slab_name = NULL;
3318 kmem_cache_destroy(rsk_prot->slab);
3319 rsk_prot->slab = NULL;
3322 static int req_prot_init(const struct proto *prot)
3324 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3329 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3331 if (!rsk_prot->slab_name)
3334 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3335 rsk_prot->obj_size, 0,
3336 SLAB_ACCOUNT | prot->slab_flags,
3339 if (!rsk_prot->slab) {
3340 pr_crit("%s: Can't create request sock SLAB cache!\n",
3347 int proto_register(struct proto *prot, int alloc_slab)
3350 prot->slab = kmem_cache_create_usercopy(prot->name,
3352 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3354 prot->useroffset, prot->usersize,
3357 if (prot->slab == NULL) {
3358 pr_crit("%s: Can't create sock SLAB cache!\n",
3363 if (req_prot_init(prot))
3364 goto out_free_request_sock_slab;
3366 if (prot->twsk_prot != NULL) {
3367 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3369 if (prot->twsk_prot->twsk_slab_name == NULL)
3370 goto out_free_request_sock_slab;
3372 prot->twsk_prot->twsk_slab =
3373 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3374 prot->twsk_prot->twsk_obj_size,
3379 if (prot->twsk_prot->twsk_slab == NULL)
3380 goto out_free_timewait_sock_slab_name;
3384 mutex_lock(&proto_list_mutex);
3385 list_add(&prot->node, &proto_list);
3386 assign_proto_idx(prot);
3387 mutex_unlock(&proto_list_mutex);
3390 out_free_timewait_sock_slab_name:
3391 kfree(prot->twsk_prot->twsk_slab_name);
3392 out_free_request_sock_slab:
3393 req_prot_cleanup(prot->rsk_prot);
3395 kmem_cache_destroy(prot->slab);
3400 EXPORT_SYMBOL(proto_register);
3402 void proto_unregister(struct proto *prot)
3404 mutex_lock(&proto_list_mutex);
3405 release_proto_idx(prot);
3406 list_del(&prot->node);
3407 mutex_unlock(&proto_list_mutex);
3409 kmem_cache_destroy(prot->slab);
3412 req_prot_cleanup(prot->rsk_prot);
3414 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3415 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3416 kfree(prot->twsk_prot->twsk_slab_name);
3417 prot->twsk_prot->twsk_slab = NULL;
3420 EXPORT_SYMBOL(proto_unregister);
3422 int sock_load_diag_module(int family, int protocol)
3425 if (!sock_is_registered(family))
3428 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3429 NETLINK_SOCK_DIAG, family);
3433 if (family == AF_INET &&
3434 protocol != IPPROTO_RAW &&
3435 !rcu_access_pointer(inet_protos[protocol]))
3439 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3440 NETLINK_SOCK_DIAG, family, protocol);
3442 EXPORT_SYMBOL(sock_load_diag_module);
3444 #ifdef CONFIG_PROC_FS
3445 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3446 __acquires(proto_list_mutex)
3448 mutex_lock(&proto_list_mutex);
3449 return seq_list_start_head(&proto_list, *pos);
3452 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3454 return seq_list_next(v, &proto_list, pos);
3457 static void proto_seq_stop(struct seq_file *seq, void *v)
3458 __releases(proto_list_mutex)
3460 mutex_unlock(&proto_list_mutex);
3463 static char proto_method_implemented(const void *method)
3465 return method == NULL ? 'n' : 'y';
3467 static long sock_prot_memory_allocated(struct proto *proto)
3469 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3472 static char *sock_prot_memory_pressure(struct proto *proto)
3474 return proto->memory_pressure != NULL ?
3475 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3478 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3481 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3482 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3485 sock_prot_inuse_get(seq_file_net(seq), proto),
3486 sock_prot_memory_allocated(proto),
3487 sock_prot_memory_pressure(proto),
3489 proto->slab == NULL ? "no" : "yes",
3490 module_name(proto->owner),
3491 proto_method_implemented(proto->close),
3492 proto_method_implemented(proto->connect),
3493 proto_method_implemented(proto->disconnect),
3494 proto_method_implemented(proto->accept),
3495 proto_method_implemented(proto->ioctl),
3496 proto_method_implemented(proto->init),
3497 proto_method_implemented(proto->destroy),
3498 proto_method_implemented(proto->shutdown),
3499 proto_method_implemented(proto->setsockopt),
3500 proto_method_implemented(proto->getsockopt),
3501 proto_method_implemented(proto->sendmsg),
3502 proto_method_implemented(proto->recvmsg),
3503 proto_method_implemented(proto->sendpage),
3504 proto_method_implemented(proto->bind),
3505 proto_method_implemented(proto->backlog_rcv),
3506 proto_method_implemented(proto->hash),
3507 proto_method_implemented(proto->unhash),
3508 proto_method_implemented(proto->get_port),
3509 proto_method_implemented(proto->enter_memory_pressure));
3512 static int proto_seq_show(struct seq_file *seq, void *v)
3514 if (v == &proto_list)
3515 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3524 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3526 proto_seq_printf(seq, list_entry(v, struct proto, node));
3530 static const struct seq_operations proto_seq_ops = {
3531 .start = proto_seq_start,
3532 .next = proto_seq_next,
3533 .stop = proto_seq_stop,
3534 .show = proto_seq_show,
3537 static __net_init int proto_init_net(struct net *net)
3539 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3540 sizeof(struct seq_net_private)))
3546 static __net_exit void proto_exit_net(struct net *net)
3548 remove_proc_entry("protocols", net->proc_net);
3552 static __net_initdata struct pernet_operations proto_net_ops = {
3553 .init = proto_init_net,
3554 .exit = proto_exit_net,
3557 static int __init proto_init(void)
3559 return register_pernet_subsys(&proto_net_ops);
3562 subsys_initcall(proto_init);
3564 #endif /* PROC_FS */
3566 #ifdef CONFIG_NET_RX_BUSY_POLL
3567 bool sk_busy_loop_end(void *p, unsigned long start_time)
3569 struct sock *sk = p;
3571 return !skb_queue_empty(&sk->sk_receive_queue) ||
3572 sk_busy_loop_timeout(sk, start_time);
3574 EXPORT_SYMBOL(sk_busy_loop_end);
3575 #endif /* CONFIG_NET_RX_BUSY_POLL */