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>
141 #include <trace/events/sock.h>
144 #include <net/busy_poll.h>
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
149 static void sock_inuse_add(struct net *net, int val);
152 * sk_ns_capable - General socket capability test
153 * @sk: Socket to use a capability on or through
154 * @user_ns: The user namespace of the capability to use
155 * @cap: The capability to use
157 * Test to see if the opener of the socket had when the socket was
158 * created and the current process has the capability @cap in the user
159 * namespace @user_ns.
161 bool sk_ns_capable(const struct sock *sk,
162 struct user_namespace *user_ns, int cap)
164 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
165 ns_capable(user_ns, cap);
167 EXPORT_SYMBOL(sk_ns_capable);
170 * sk_capable - Socket global capability test
171 * @sk: Socket to use a capability on or through
172 * @cap: The global capability to use
174 * Test to see if the opener of the socket had when the socket was
175 * created and the current process has the capability @cap in all user
178 bool sk_capable(const struct sock *sk, int cap)
180 return sk_ns_capable(sk, &init_user_ns, cap);
182 EXPORT_SYMBOL(sk_capable);
185 * sk_net_capable - Network namespace socket capability test
186 * @sk: Socket to use a capability on or through
187 * @cap: The capability to use
189 * Test to see if the opener of the socket had when the socket was created
190 * and the current process has the capability @cap over the network namespace
191 * the socket is a member of.
193 bool sk_net_capable(const struct sock *sk, int cap)
195 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
197 EXPORT_SYMBOL(sk_net_capable);
200 * Each address family might have different locking rules, so we have
201 * one slock key per address family and separate keys for internal and
204 static struct lock_class_key af_family_keys[AF_MAX];
205 static struct lock_class_key af_family_kern_keys[AF_MAX];
206 static struct lock_class_key af_family_slock_keys[AF_MAX];
207 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
210 * Make lock validator output more readable. (we pre-construct these
211 * strings build-time, so that runtime initialization of socket
215 #define _sock_locks(x) \
216 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
217 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
218 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
219 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
220 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
221 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
222 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
223 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
224 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
225 x "27" , x "28" , x "AF_CAN" , \
226 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
227 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
228 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
229 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
230 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
233 static const char *const af_family_key_strings[AF_MAX+1] = {
234 _sock_locks("sk_lock-")
236 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
237 _sock_locks("slock-")
239 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
240 _sock_locks("clock-")
243 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
244 _sock_locks("k-sk_lock-")
246 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
247 _sock_locks("k-slock-")
249 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
250 _sock_locks("k-clock-")
252 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
253 _sock_locks("rlock-")
255 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
256 _sock_locks("wlock-")
258 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
259 _sock_locks("elock-")
263 * sk_callback_lock and sk queues 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];
267 static struct lock_class_key af_rlock_keys[AF_MAX];
268 static struct lock_class_key af_wlock_keys[AF_MAX];
269 static struct lock_class_key af_elock_keys[AF_MAX];
270 static struct lock_class_key af_kern_callback_keys[AF_MAX];
272 /* Run time adjustable parameters. */
273 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
274 EXPORT_SYMBOL(sysctl_wmem_max);
275 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
276 EXPORT_SYMBOL(sysctl_rmem_max);
277 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
278 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
280 /* Maximal space eaten by iovec or ancillary data plus some space */
281 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
282 EXPORT_SYMBOL(sysctl_optmem_max);
284 int sysctl_tstamp_allow_data __read_mostly = 1;
286 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
287 EXPORT_SYMBOL_GPL(memalloc_socks_key);
290 * sk_set_memalloc - sets %SOCK_MEMALLOC
291 * @sk: socket to set it on
293 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
294 * It's the responsibility of the admin to adjust min_free_kbytes
295 * to meet the requirements
297 void sk_set_memalloc(struct sock *sk)
299 sock_set_flag(sk, SOCK_MEMALLOC);
300 sk->sk_allocation |= __GFP_MEMALLOC;
301 static_branch_inc(&memalloc_socks_key);
303 EXPORT_SYMBOL_GPL(sk_set_memalloc);
305 void sk_clear_memalloc(struct sock *sk)
307 sock_reset_flag(sk, SOCK_MEMALLOC);
308 sk->sk_allocation &= ~__GFP_MEMALLOC;
309 static_branch_dec(&memalloc_socks_key);
312 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
313 * progress of swapping. SOCK_MEMALLOC may be cleared while
314 * it has rmem allocations due to the last swapfile being deactivated
315 * but there is a risk that the socket is unusable due to exceeding
316 * the rmem limits. Reclaim the reserves and obey rmem limits again.
320 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
322 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
325 unsigned int noreclaim_flag;
327 /* these should have been dropped before queueing */
328 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
330 noreclaim_flag = memalloc_noreclaim_save();
331 ret = sk->sk_backlog_rcv(sk, skb);
332 memalloc_noreclaim_restore(noreclaim_flag);
336 EXPORT_SYMBOL(__sk_backlog_rcv);
338 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
342 if (optlen < sizeof(tv))
344 if (copy_from_user(&tv, optval, sizeof(tv)))
346 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
350 static int warned __read_mostly;
353 if (warned < 10 && net_ratelimit()) {
355 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
356 __func__, current->comm, task_pid_nr(current));
360 *timeo_p = MAX_SCHEDULE_TIMEOUT;
361 if (tv.tv_sec == 0 && tv.tv_usec == 0)
363 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
364 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
368 static void sock_warn_obsolete_bsdism(const char *name)
371 static char warncomm[TASK_COMM_LEN];
372 if (strcmp(warncomm, current->comm) && warned < 5) {
373 strcpy(warncomm, current->comm);
374 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
380 static bool sock_needs_netstamp(const struct sock *sk)
382 switch (sk->sk_family) {
391 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
393 if (sk->sk_flags & flags) {
394 sk->sk_flags &= ~flags;
395 if (sock_needs_netstamp(sk) &&
396 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
397 net_disable_timestamp();
402 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
405 struct sk_buff_head *list = &sk->sk_receive_queue;
407 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
408 atomic_inc(&sk->sk_drops);
409 trace_sock_rcvqueue_full(sk, skb);
413 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
414 atomic_inc(&sk->sk_drops);
419 skb_set_owner_r(skb, sk);
421 /* we escape from rcu protected region, make sure we dont leak
426 spin_lock_irqsave(&list->lock, flags);
427 sock_skb_set_dropcount(sk, skb);
428 __skb_queue_tail(list, skb);
429 spin_unlock_irqrestore(&list->lock, flags);
431 if (!sock_flag(sk, SOCK_DEAD))
432 sk->sk_data_ready(sk);
435 EXPORT_SYMBOL(__sock_queue_rcv_skb);
437 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
441 err = sk_filter(sk, skb);
445 return __sock_queue_rcv_skb(sk, skb);
447 EXPORT_SYMBOL(sock_queue_rcv_skb);
449 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
450 const int nested, unsigned int trim_cap, bool refcounted)
452 int rc = NET_RX_SUCCESS;
454 if (sk_filter_trim_cap(sk, skb, trim_cap))
455 goto discard_and_relse;
459 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
460 atomic_inc(&sk->sk_drops);
461 goto discard_and_relse;
464 bh_lock_sock_nested(sk);
467 if (!sock_owned_by_user(sk)) {
469 * trylock + unlock semantics:
471 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
473 rc = sk_backlog_rcv(sk, skb);
475 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
476 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
478 atomic_inc(&sk->sk_drops);
479 goto discard_and_relse;
491 EXPORT_SYMBOL(__sk_receive_skb);
493 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
495 struct dst_entry *dst = __sk_dst_get(sk);
497 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
498 sk_tx_queue_clear(sk);
499 sk->sk_dst_pending_confirm = 0;
500 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
507 EXPORT_SYMBOL(__sk_dst_check);
509 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
511 struct dst_entry *dst = sk_dst_get(sk);
513 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
521 EXPORT_SYMBOL(sk_dst_check);
523 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
526 int ret = -ENOPROTOOPT;
527 #ifdef CONFIG_NETDEVICES
528 struct net *net = sock_net(sk);
529 char devname[IFNAMSIZ];
534 if (!ns_capable(net->user_ns, CAP_NET_RAW))
541 /* Bind this socket to a particular device like "eth0",
542 * as specified in the passed interface name. If the
543 * name is "" or the option length is zero the socket
546 if (optlen > IFNAMSIZ - 1)
547 optlen = IFNAMSIZ - 1;
548 memset(devname, 0, sizeof(devname));
551 if (copy_from_user(devname, optval, optlen))
555 if (devname[0] != '\0') {
556 struct net_device *dev;
559 dev = dev_get_by_name_rcu(net, devname);
561 index = dev->ifindex;
569 sk->sk_bound_dev_if = index;
581 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
582 int __user *optlen, int len)
584 int ret = -ENOPROTOOPT;
585 #ifdef CONFIG_NETDEVICES
586 struct net *net = sock_net(sk);
587 char devname[IFNAMSIZ];
589 if (sk->sk_bound_dev_if == 0) {
598 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
602 len = strlen(devname) + 1;
605 if (copy_to_user(optval, devname, len))
610 if (put_user(len, optlen))
621 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
624 sock_set_flag(sk, bit);
626 sock_reset_flag(sk, bit);
629 bool sk_mc_loop(struct sock *sk)
631 if (dev_recursion_level())
635 switch (sk->sk_family) {
637 return inet_sk(sk)->mc_loop;
638 #if IS_ENABLED(CONFIG_IPV6)
640 return inet6_sk(sk)->mc_loop;
646 EXPORT_SYMBOL(sk_mc_loop);
649 * This is meant for all protocols to use and covers goings on
650 * at the socket level. Everything here is generic.
653 int sock_setsockopt(struct socket *sock, int level, int optname,
654 char __user *optval, unsigned int optlen)
656 struct sock_txtime sk_txtime;
657 struct sock *sk = sock->sk;
664 * Options without arguments
667 if (optname == SO_BINDTODEVICE)
668 return sock_setbindtodevice(sk, optval, optlen);
670 if (optlen < sizeof(int))
673 if (get_user(val, (int __user *)optval))
676 valbool = val ? 1 : 0;
682 if (val && !capable(CAP_NET_ADMIN))
685 sock_valbool_flag(sk, SOCK_DBG, valbool);
688 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
691 sk->sk_reuseport = valbool;
700 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
703 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
706 /* Don't error on this BSD doesn't and if you think
707 * about it this is right. Otherwise apps have to
708 * play 'guess the biggest size' games. RCVBUF/SNDBUF
709 * are treated in BSD as hints
711 val = min_t(u32, val, sysctl_wmem_max);
713 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
714 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
715 /* Wake up sending tasks if we upped the value. */
716 sk->sk_write_space(sk);
720 if (!capable(CAP_NET_ADMIN)) {
727 /* Don't error on this BSD doesn't and if you think
728 * about it this is right. Otherwise apps have to
729 * play 'guess the biggest size' games. RCVBUF/SNDBUF
730 * are treated in BSD as hints
732 val = min_t(u32, val, sysctl_rmem_max);
734 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
736 * We double it on the way in to account for
737 * "struct sk_buff" etc. overhead. Applications
738 * assume that the SO_RCVBUF setting they make will
739 * allow that much actual data to be received on that
742 * Applications are unaware that "struct sk_buff" and
743 * other overheads allocate from the receive buffer
744 * during socket buffer allocation.
746 * And after considering the possible alternatives,
747 * returning the value we actually used in getsockopt
748 * is the most desirable behavior.
750 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
754 if (!capable(CAP_NET_ADMIN)) {
761 if (sk->sk_prot->keepalive)
762 sk->sk_prot->keepalive(sk, valbool);
763 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
767 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
771 sk->sk_no_check_tx = valbool;
775 if ((val >= 0 && val <= 6) ||
776 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
777 sk->sk_priority = val;
783 if (optlen < sizeof(ling)) {
784 ret = -EINVAL; /* 1003.1g */
787 if (copy_from_user(&ling, optval, sizeof(ling))) {
792 sock_reset_flag(sk, SOCK_LINGER);
794 #if (BITS_PER_LONG == 32)
795 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
796 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
799 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
800 sock_set_flag(sk, SOCK_LINGER);
805 sock_warn_obsolete_bsdism("setsockopt");
810 set_bit(SOCK_PASSCRED, &sock->flags);
812 clear_bit(SOCK_PASSCRED, &sock->flags);
818 if (optname == SO_TIMESTAMP)
819 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
821 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
822 sock_set_flag(sk, SOCK_RCVTSTAMP);
823 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
825 sock_reset_flag(sk, SOCK_RCVTSTAMP);
826 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
830 case SO_TIMESTAMPING:
831 if (val & ~SOF_TIMESTAMPING_MASK) {
836 if (val & SOF_TIMESTAMPING_OPT_ID &&
837 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
838 if (sk->sk_protocol == IPPROTO_TCP &&
839 sk->sk_type == SOCK_STREAM) {
840 if ((1 << sk->sk_state) &
841 (TCPF_CLOSE | TCPF_LISTEN)) {
845 sk->sk_tskey = tcp_sk(sk)->snd_una;
851 if (val & SOF_TIMESTAMPING_OPT_STATS &&
852 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
857 sk->sk_tsflags = val;
858 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
859 sock_enable_timestamp(sk,
860 SOCK_TIMESTAMPING_RX_SOFTWARE);
862 sock_disable_timestamp(sk,
863 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
869 if (sock->ops->set_rcvlowat)
870 ret = sock->ops->set_rcvlowat(sk, val);
872 sk->sk_rcvlowat = val ? : 1;
876 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
880 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
883 case SO_ATTACH_FILTER:
885 if (optlen == sizeof(struct sock_fprog)) {
886 struct sock_fprog fprog;
889 if (copy_from_user(&fprog, optval, sizeof(fprog)))
892 ret = sk_attach_filter(&fprog, sk);
898 if (optlen == sizeof(u32)) {
902 if (copy_from_user(&ufd, optval, sizeof(ufd)))
905 ret = sk_attach_bpf(ufd, sk);
909 case SO_ATTACH_REUSEPORT_CBPF:
911 if (optlen == sizeof(struct sock_fprog)) {
912 struct sock_fprog fprog;
915 if (copy_from_user(&fprog, optval, sizeof(fprog)))
918 ret = sk_reuseport_attach_filter(&fprog, sk);
922 case SO_ATTACH_REUSEPORT_EBPF:
924 if (optlen == sizeof(u32)) {
928 if (copy_from_user(&ufd, optval, sizeof(ufd)))
931 ret = sk_reuseport_attach_bpf(ufd, sk);
935 case SO_DETACH_FILTER:
936 ret = sk_detach_filter(sk);
940 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
943 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
948 set_bit(SOCK_PASSSEC, &sock->flags);
950 clear_bit(SOCK_PASSSEC, &sock->flags);
953 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
960 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
964 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
968 if (sock->ops->set_peek_off)
969 ret = sock->ops->set_peek_off(sk, val);
975 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
978 case SO_SELECT_ERR_QUEUE:
979 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
982 #ifdef CONFIG_NET_RX_BUSY_POLL
984 /* allow unprivileged users to decrease the value */
985 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
991 sk->sk_ll_usec = val;
996 case SO_MAX_PACING_RATE:
998 cmpxchg(&sk->sk_pacing_status,
1001 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
1002 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1003 sk->sk_max_pacing_rate);
1006 case SO_INCOMING_CPU:
1007 sk->sk_incoming_cpu = val;
1012 dst_negative_advice(sk);
1016 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1017 if (sk->sk_protocol != IPPROTO_TCP)
1019 } else if (sk->sk_family != PF_RDS) {
1023 if (val < 0 || val > 1)
1026 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1031 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1033 } else if (optlen != sizeof(struct sock_txtime)) {
1035 } else if (copy_from_user(&sk_txtime, optval,
1036 sizeof(struct sock_txtime))) {
1038 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1041 sock_valbool_flag(sk, SOCK_TXTIME, true);
1042 sk->sk_clockid = sk_txtime.clockid;
1043 sk->sk_txtime_deadline_mode =
1044 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1045 sk->sk_txtime_report_errors =
1046 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1057 EXPORT_SYMBOL(sock_setsockopt);
1060 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1061 struct ucred *ucred)
1063 ucred->pid = pid_vnr(pid);
1064 ucred->uid = ucred->gid = -1;
1066 struct user_namespace *current_ns = current_user_ns();
1068 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1069 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1073 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1075 struct user_namespace *user_ns = current_user_ns();
1078 for (i = 0; i < src->ngroups; i++)
1079 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1085 int sock_getsockopt(struct socket *sock, int level, int optname,
1086 char __user *optval, int __user *optlen)
1088 struct sock *sk = sock->sk;
1095 struct sock_txtime txtime;
1098 int lv = sizeof(int);
1101 if (get_user(len, optlen))
1106 memset(&v, 0, sizeof(v));
1110 v.val = sock_flag(sk, SOCK_DBG);
1114 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1118 v.val = sock_flag(sk, SOCK_BROADCAST);
1122 v.val = sk->sk_sndbuf;
1126 v.val = sk->sk_rcvbuf;
1130 v.val = sk->sk_reuse;
1134 v.val = sk->sk_reuseport;
1138 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1142 v.val = sk->sk_type;
1146 v.val = sk->sk_protocol;
1150 v.val = sk->sk_family;
1154 v.val = -sock_error(sk);
1156 v.val = xchg(&sk->sk_err_soft, 0);
1160 v.val = sock_flag(sk, SOCK_URGINLINE);
1164 v.val = sk->sk_no_check_tx;
1168 v.val = sk->sk_priority;
1172 lv = sizeof(v.ling);
1173 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1174 v.ling.l_linger = sk->sk_lingertime / HZ;
1178 sock_warn_obsolete_bsdism("getsockopt");
1182 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1183 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1186 case SO_TIMESTAMPNS:
1187 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1190 case SO_TIMESTAMPING:
1191 v.val = sk->sk_tsflags;
1195 lv = sizeof(struct timeval);
1196 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1200 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1201 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1206 lv = sizeof(struct timeval);
1207 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1211 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1212 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1217 v.val = sk->sk_rcvlowat;
1225 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1230 struct ucred peercred;
1231 if (len > sizeof(peercred))
1232 len = sizeof(peercred);
1233 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1234 if (copy_to_user(optval, &peercred, len))
1243 if (!sk->sk_peer_cred)
1246 n = sk->sk_peer_cred->group_info->ngroups;
1247 if (len < n * sizeof(gid_t)) {
1248 len = n * sizeof(gid_t);
1249 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1251 len = n * sizeof(gid_t);
1253 ret = groups_to_user((gid_t __user *)optval,
1254 sk->sk_peer_cred->group_info);
1264 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1269 if (copy_to_user(optval, address, len))
1274 /* Dubious BSD thing... Probably nobody even uses it, but
1275 * the UNIX standard wants it for whatever reason... -DaveM
1278 v.val = sk->sk_state == TCP_LISTEN;
1282 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1286 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1289 v.val = sk->sk_mark;
1293 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1296 case SO_WIFI_STATUS:
1297 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1301 if (!sock->ops->set_peek_off)
1304 v.val = sk->sk_peek_off;
1307 v.val = sock_flag(sk, SOCK_NOFCS);
1310 case SO_BINDTODEVICE:
1311 return sock_getbindtodevice(sk, optval, optlen, len);
1314 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1320 case SO_LOCK_FILTER:
1321 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1324 case SO_BPF_EXTENSIONS:
1325 v.val = bpf_tell_extensions();
1328 case SO_SELECT_ERR_QUEUE:
1329 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1332 #ifdef CONFIG_NET_RX_BUSY_POLL
1334 v.val = sk->sk_ll_usec;
1338 case SO_MAX_PACING_RATE:
1340 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1343 case SO_INCOMING_CPU:
1344 v.val = sk->sk_incoming_cpu;
1349 u32 meminfo[SK_MEMINFO_VARS];
1351 if (get_user(len, optlen))
1354 sk_get_meminfo(sk, meminfo);
1356 len = min_t(unsigned int, len, sizeof(meminfo));
1357 if (copy_to_user(optval, &meminfo, len))
1363 #ifdef CONFIG_NET_RX_BUSY_POLL
1364 case SO_INCOMING_NAPI_ID:
1365 v.val = READ_ONCE(sk->sk_napi_id);
1367 /* aggregate non-NAPI IDs down to 0 */
1368 if (v.val < MIN_NAPI_ID)
1378 v.val64 = sock_gen_cookie(sk);
1382 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1386 lv = sizeof(v.txtime);
1387 v.txtime.clockid = sk->sk_clockid;
1388 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1389 SOF_TXTIME_DEADLINE_MODE : 0;
1390 v.txtime.flags |= sk->sk_txtime_report_errors ?
1391 SOF_TXTIME_REPORT_ERRORS : 0;
1395 /* We implement the SO_SNDLOWAT etc to not be settable
1398 return -ENOPROTOOPT;
1403 if (copy_to_user(optval, &v, len))
1406 if (put_user(len, optlen))
1412 * Initialize an sk_lock.
1414 * (We also register the sk_lock with the lock validator.)
1416 static inline void sock_lock_init(struct sock *sk)
1418 if (sk->sk_kern_sock)
1419 sock_lock_init_class_and_name(
1421 af_family_kern_slock_key_strings[sk->sk_family],
1422 af_family_kern_slock_keys + sk->sk_family,
1423 af_family_kern_key_strings[sk->sk_family],
1424 af_family_kern_keys + sk->sk_family);
1426 sock_lock_init_class_and_name(
1428 af_family_slock_key_strings[sk->sk_family],
1429 af_family_slock_keys + sk->sk_family,
1430 af_family_key_strings[sk->sk_family],
1431 af_family_keys + sk->sk_family);
1435 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1436 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1437 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1439 static void sock_copy(struct sock *nsk, const struct sock *osk)
1441 #ifdef CONFIG_SECURITY_NETWORK
1442 void *sptr = nsk->sk_security;
1444 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1446 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1447 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1449 #ifdef CONFIG_SECURITY_NETWORK
1450 nsk->sk_security = sptr;
1451 security_sk_clone(osk, nsk);
1455 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1459 struct kmem_cache *slab;
1463 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1466 if (priority & __GFP_ZERO)
1467 sk_prot_clear_nulls(sk, prot->obj_size);
1469 sk = kmalloc(prot->obj_size, priority);
1472 if (security_sk_alloc(sk, family, priority))
1475 if (!try_module_get(prot->owner))
1477 sk_tx_queue_clear(sk);
1483 security_sk_free(sk);
1486 kmem_cache_free(slab, sk);
1492 static void sk_prot_free(struct proto *prot, struct sock *sk)
1494 struct kmem_cache *slab;
1495 struct module *owner;
1497 owner = prot->owner;
1500 cgroup_sk_free(&sk->sk_cgrp_data);
1501 mem_cgroup_sk_free(sk);
1502 security_sk_free(sk);
1504 kmem_cache_free(slab, sk);
1511 * sk_alloc - All socket objects are allocated here
1512 * @net: the applicable net namespace
1513 * @family: protocol family
1514 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1515 * @prot: struct proto associated with this new sock instance
1516 * @kern: is this to be a kernel socket?
1518 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1519 struct proto *prot, int kern)
1523 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1525 sk->sk_family = family;
1527 * See comment in struct sock definition to understand
1528 * why we need sk_prot_creator -acme
1530 sk->sk_prot = sk->sk_prot_creator = prot;
1531 sk->sk_kern_sock = kern;
1533 sk->sk_net_refcnt = kern ? 0 : 1;
1534 if (likely(sk->sk_net_refcnt)) {
1536 sock_inuse_add(net, 1);
1539 sock_net_set(sk, net);
1540 refcount_set(&sk->sk_wmem_alloc, 1);
1542 mem_cgroup_sk_alloc(sk);
1543 cgroup_sk_alloc(&sk->sk_cgrp_data);
1544 sock_update_classid(&sk->sk_cgrp_data);
1545 sock_update_netprioidx(&sk->sk_cgrp_data);
1550 EXPORT_SYMBOL(sk_alloc);
1552 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1553 * grace period. This is the case for UDP sockets and TCP listeners.
1555 static void __sk_destruct(struct rcu_head *head)
1557 struct sock *sk = container_of(head, struct sock, sk_rcu);
1558 struct sk_filter *filter;
1560 if (sk->sk_destruct)
1561 sk->sk_destruct(sk);
1563 filter = rcu_dereference_check(sk->sk_filter,
1564 refcount_read(&sk->sk_wmem_alloc) == 0);
1566 sk_filter_uncharge(sk, filter);
1567 RCU_INIT_POINTER(sk->sk_filter, NULL);
1569 if (rcu_access_pointer(sk->sk_reuseport_cb))
1570 reuseport_detach_sock(sk);
1572 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1574 if (atomic_read(&sk->sk_omem_alloc))
1575 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1576 __func__, atomic_read(&sk->sk_omem_alloc));
1578 if (sk->sk_frag.page) {
1579 put_page(sk->sk_frag.page);
1580 sk->sk_frag.page = NULL;
1583 if (sk->sk_peer_cred)
1584 put_cred(sk->sk_peer_cred);
1585 put_pid(sk->sk_peer_pid);
1586 if (likely(sk->sk_net_refcnt))
1587 put_net(sock_net(sk));
1588 sk_prot_free(sk->sk_prot_creator, sk);
1591 void sk_destruct(struct sock *sk)
1593 if (sock_flag(sk, SOCK_RCU_FREE))
1594 call_rcu(&sk->sk_rcu, __sk_destruct);
1596 __sk_destruct(&sk->sk_rcu);
1599 static void __sk_free(struct sock *sk)
1601 if (likely(sk->sk_net_refcnt))
1602 sock_inuse_add(sock_net(sk), -1);
1604 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1605 sock_diag_broadcast_destroy(sk);
1610 void sk_free(struct sock *sk)
1613 * We subtract one from sk_wmem_alloc and can know if
1614 * some packets are still in some tx queue.
1615 * If not null, sock_wfree() will call __sk_free(sk) later
1617 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1620 EXPORT_SYMBOL(sk_free);
1622 static void sk_init_common(struct sock *sk)
1624 skb_queue_head_init(&sk->sk_receive_queue);
1625 skb_queue_head_init(&sk->sk_write_queue);
1626 skb_queue_head_init(&sk->sk_error_queue);
1628 rwlock_init(&sk->sk_callback_lock);
1629 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1630 af_rlock_keys + sk->sk_family,
1631 af_family_rlock_key_strings[sk->sk_family]);
1632 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1633 af_wlock_keys + sk->sk_family,
1634 af_family_wlock_key_strings[sk->sk_family]);
1635 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1636 af_elock_keys + sk->sk_family,
1637 af_family_elock_key_strings[sk->sk_family]);
1638 lockdep_set_class_and_name(&sk->sk_callback_lock,
1639 af_callback_keys + sk->sk_family,
1640 af_family_clock_key_strings[sk->sk_family]);
1644 * sk_clone_lock - clone a socket, and lock its clone
1645 * @sk: the socket to clone
1646 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1648 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1650 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1653 bool is_charged = true;
1655 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1656 if (newsk != NULL) {
1657 struct sk_filter *filter;
1659 sock_copy(newsk, sk);
1661 newsk->sk_prot_creator = sk->sk_prot;
1664 if (likely(newsk->sk_net_refcnt))
1665 get_net(sock_net(newsk));
1666 sk_node_init(&newsk->sk_node);
1667 sock_lock_init(newsk);
1668 bh_lock_sock(newsk);
1669 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1670 newsk->sk_backlog.len = 0;
1672 atomic_set(&newsk->sk_rmem_alloc, 0);
1674 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1676 refcount_set(&newsk->sk_wmem_alloc, 1);
1677 atomic_set(&newsk->sk_omem_alloc, 0);
1678 sk_init_common(newsk);
1680 newsk->sk_dst_cache = NULL;
1681 newsk->sk_dst_pending_confirm = 0;
1682 newsk->sk_wmem_queued = 0;
1683 newsk->sk_forward_alloc = 0;
1684 atomic_set(&newsk->sk_drops, 0);
1685 newsk->sk_send_head = NULL;
1686 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1687 atomic_set(&newsk->sk_zckey, 0);
1689 sock_reset_flag(newsk, SOCK_DONE);
1690 mem_cgroup_sk_alloc(newsk);
1691 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1694 filter = rcu_dereference(sk->sk_filter);
1696 /* though it's an empty new sock, the charging may fail
1697 * if sysctl_optmem_max was changed between creation of
1698 * original socket and cloning
1700 is_charged = sk_filter_charge(newsk, filter);
1701 RCU_INIT_POINTER(newsk->sk_filter, filter);
1704 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1705 /* We need to make sure that we don't uncharge the new
1706 * socket if we couldn't charge it in the first place
1707 * as otherwise we uncharge the parent's filter.
1710 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1711 sk_free_unlock_clone(newsk);
1715 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1718 newsk->sk_err_soft = 0;
1719 newsk->sk_priority = 0;
1720 newsk->sk_incoming_cpu = raw_smp_processor_id();
1721 atomic64_set(&newsk->sk_cookie, 0);
1722 if (likely(newsk->sk_net_refcnt))
1723 sock_inuse_add(sock_net(newsk), 1);
1726 * Before updating sk_refcnt, we must commit prior changes to memory
1727 * (Documentation/RCU/rculist_nulls.txt for details)
1730 refcount_set(&newsk->sk_refcnt, 2);
1733 * Increment the counter in the same struct proto as the master
1734 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1735 * is the same as sk->sk_prot->socks, as this field was copied
1738 * This _changes_ the previous behaviour, where
1739 * tcp_create_openreq_child always was incrementing the
1740 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1741 * to be taken into account in all callers. -acme
1743 sk_refcnt_debug_inc(newsk);
1744 sk_set_socket(newsk, NULL);
1745 newsk->sk_wq = NULL;
1747 if (newsk->sk_prot->sockets_allocated)
1748 sk_sockets_allocated_inc(newsk);
1750 if (sock_needs_netstamp(sk) &&
1751 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1752 net_enable_timestamp();
1757 EXPORT_SYMBOL_GPL(sk_clone_lock);
1759 void sk_free_unlock_clone(struct sock *sk)
1761 /* It is still raw copy of parent, so invalidate
1762 * destructor and make plain sk_free() */
1763 sk->sk_destruct = NULL;
1767 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1769 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1773 sk_dst_set(sk, dst);
1774 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1775 if (sk->sk_route_caps & NETIF_F_GSO)
1776 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1777 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1778 if (sk_can_gso(sk)) {
1779 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1780 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1782 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1783 sk->sk_gso_max_size = dst->dev->gso_max_size;
1784 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1787 sk->sk_gso_max_segs = max_segs;
1789 EXPORT_SYMBOL_GPL(sk_setup_caps);
1792 * Simple resource managers for sockets.
1797 * Write buffer destructor automatically called from kfree_skb.
1799 void sock_wfree(struct sk_buff *skb)
1801 struct sock *sk = skb->sk;
1802 unsigned int len = skb->truesize;
1804 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1806 * Keep a reference on sk_wmem_alloc, this will be released
1807 * after sk_write_space() call
1809 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1810 sk->sk_write_space(sk);
1814 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1815 * could not do because of in-flight packets
1817 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1820 EXPORT_SYMBOL(sock_wfree);
1822 /* This variant of sock_wfree() is used by TCP,
1823 * since it sets SOCK_USE_WRITE_QUEUE.
1825 void __sock_wfree(struct sk_buff *skb)
1827 struct sock *sk = skb->sk;
1829 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1833 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1838 if (unlikely(!sk_fullsock(sk))) {
1839 skb->destructor = sock_edemux;
1844 skb->destructor = sock_wfree;
1845 skb_set_hash_from_sk(skb, sk);
1847 * We used to take a refcount on sk, but following operation
1848 * is enough to guarantee sk_free() wont free this sock until
1849 * all in-flight packets are completed
1851 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1853 EXPORT_SYMBOL(skb_set_owner_w);
1855 /* This helper is used by netem, as it can hold packets in its
1856 * delay queue. We want to allow the owner socket to send more
1857 * packets, as if they were already TX completed by a typical driver.
1858 * But we also want to keep skb->sk set because some packet schedulers
1859 * rely on it (sch_fq for example).
1861 void skb_orphan_partial(struct sk_buff *skb)
1863 if (skb_is_tcp_pure_ack(skb))
1866 if (skb->destructor == sock_wfree
1868 || skb->destructor == tcp_wfree
1871 struct sock *sk = skb->sk;
1873 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1874 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
1875 skb->destructor = sock_efree;
1881 EXPORT_SYMBOL(skb_orphan_partial);
1884 * Read buffer destructor automatically called from kfree_skb.
1886 void sock_rfree(struct sk_buff *skb)
1888 struct sock *sk = skb->sk;
1889 unsigned int len = skb->truesize;
1891 atomic_sub(len, &sk->sk_rmem_alloc);
1892 sk_mem_uncharge(sk, len);
1894 EXPORT_SYMBOL(sock_rfree);
1897 * Buffer destructor for skbs that are not used directly in read or write
1898 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1900 void sock_efree(struct sk_buff *skb)
1904 EXPORT_SYMBOL(sock_efree);
1906 kuid_t sock_i_uid(struct sock *sk)
1910 read_lock_bh(&sk->sk_callback_lock);
1911 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1912 read_unlock_bh(&sk->sk_callback_lock);
1915 EXPORT_SYMBOL(sock_i_uid);
1917 unsigned long sock_i_ino(struct sock *sk)
1921 read_lock_bh(&sk->sk_callback_lock);
1922 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1923 read_unlock_bh(&sk->sk_callback_lock);
1926 EXPORT_SYMBOL(sock_i_ino);
1929 * Allocate a skb from the socket's send buffer.
1931 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1934 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1935 struct sk_buff *skb = alloc_skb(size, priority);
1937 skb_set_owner_w(skb, sk);
1943 EXPORT_SYMBOL(sock_wmalloc);
1945 static void sock_ofree(struct sk_buff *skb)
1947 struct sock *sk = skb->sk;
1949 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
1952 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1955 struct sk_buff *skb;
1957 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
1958 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
1962 skb = alloc_skb(size, priority);
1966 atomic_add(skb->truesize, &sk->sk_omem_alloc);
1968 skb->destructor = sock_ofree;
1973 * Allocate a memory block from the socket's option memory buffer.
1975 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1977 if ((unsigned int)size <= sysctl_optmem_max &&
1978 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1980 /* First do the add, to avoid the race if kmalloc
1983 atomic_add(size, &sk->sk_omem_alloc);
1984 mem = kmalloc(size, priority);
1987 atomic_sub(size, &sk->sk_omem_alloc);
1991 EXPORT_SYMBOL(sock_kmalloc);
1993 /* Free an option memory block. Note, we actually want the inline
1994 * here as this allows gcc to detect the nullify and fold away the
1995 * condition entirely.
1997 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2000 if (WARN_ON_ONCE(!mem))
2006 atomic_sub(size, &sk->sk_omem_alloc);
2009 void sock_kfree_s(struct sock *sk, void *mem, int size)
2011 __sock_kfree_s(sk, mem, size, false);
2013 EXPORT_SYMBOL(sock_kfree_s);
2015 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2017 __sock_kfree_s(sk, mem, size, true);
2019 EXPORT_SYMBOL(sock_kzfree_s);
2021 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2022 I think, these locks should be removed for datagram sockets.
2024 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2028 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2032 if (signal_pending(current))
2034 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2035 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2036 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2038 if (sk->sk_shutdown & SEND_SHUTDOWN)
2042 timeo = schedule_timeout(timeo);
2044 finish_wait(sk_sleep(sk), &wait);
2050 * Generic send/receive buffer handlers
2053 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2054 unsigned long data_len, int noblock,
2055 int *errcode, int max_page_order)
2057 struct sk_buff *skb;
2061 timeo = sock_sndtimeo(sk, noblock);
2063 err = sock_error(sk);
2068 if (sk->sk_shutdown & SEND_SHUTDOWN)
2071 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2074 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2075 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2079 if (signal_pending(current))
2081 timeo = sock_wait_for_wmem(sk, timeo);
2083 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2084 errcode, sk->sk_allocation);
2086 skb_set_owner_w(skb, sk);
2090 err = sock_intr_errno(timeo);
2095 EXPORT_SYMBOL(sock_alloc_send_pskb);
2097 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2098 int noblock, int *errcode)
2100 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2102 EXPORT_SYMBOL(sock_alloc_send_skb);
2104 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2105 struct sockcm_cookie *sockc)
2109 switch (cmsg->cmsg_type) {
2111 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2113 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2115 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2117 case SO_TIMESTAMPING:
2118 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2121 tsflags = *(u32 *)CMSG_DATA(cmsg);
2122 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2125 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2126 sockc->tsflags |= tsflags;
2129 if (!sock_flag(sk, SOCK_TXTIME))
2131 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2133 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2135 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2137 case SCM_CREDENTIALS:
2144 EXPORT_SYMBOL(__sock_cmsg_send);
2146 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2147 struct sockcm_cookie *sockc)
2149 struct cmsghdr *cmsg;
2152 for_each_cmsghdr(cmsg, msg) {
2153 if (!CMSG_OK(msg, cmsg))
2155 if (cmsg->cmsg_level != SOL_SOCKET)
2157 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2163 EXPORT_SYMBOL(sock_cmsg_send);
2165 static void sk_enter_memory_pressure(struct sock *sk)
2167 if (!sk->sk_prot->enter_memory_pressure)
2170 sk->sk_prot->enter_memory_pressure(sk);
2173 static void sk_leave_memory_pressure(struct sock *sk)
2175 if (sk->sk_prot->leave_memory_pressure) {
2176 sk->sk_prot->leave_memory_pressure(sk);
2178 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2180 if (memory_pressure && *memory_pressure)
2181 *memory_pressure = 0;
2185 /* On 32bit arches, an skb frag is limited to 2^15 */
2186 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2189 * skb_page_frag_refill - check that a page_frag contains enough room
2190 * @sz: minimum size of the fragment we want to get
2191 * @pfrag: pointer to page_frag
2192 * @gfp: priority for memory allocation
2194 * Note: While this allocator tries to use high order pages, there is
2195 * no guarantee that allocations succeed. Therefore, @sz MUST be
2196 * less or equal than PAGE_SIZE.
2198 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2201 if (page_ref_count(pfrag->page) == 1) {
2205 if (pfrag->offset + sz <= pfrag->size)
2207 put_page(pfrag->page);
2211 if (SKB_FRAG_PAGE_ORDER) {
2212 /* Avoid direct reclaim but allow kswapd to wake */
2213 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2214 __GFP_COMP | __GFP_NOWARN |
2216 SKB_FRAG_PAGE_ORDER);
2217 if (likely(pfrag->page)) {
2218 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2222 pfrag->page = alloc_page(gfp);
2223 if (likely(pfrag->page)) {
2224 pfrag->size = PAGE_SIZE;
2229 EXPORT_SYMBOL(skb_page_frag_refill);
2231 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2233 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2236 sk_enter_memory_pressure(sk);
2237 sk_stream_moderate_sndbuf(sk);
2240 EXPORT_SYMBOL(sk_page_frag_refill);
2242 static void __lock_sock(struct sock *sk)
2243 __releases(&sk->sk_lock.slock)
2244 __acquires(&sk->sk_lock.slock)
2249 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2250 TASK_UNINTERRUPTIBLE);
2251 spin_unlock_bh(&sk->sk_lock.slock);
2253 spin_lock_bh(&sk->sk_lock.slock);
2254 if (!sock_owned_by_user(sk))
2257 finish_wait(&sk->sk_lock.wq, &wait);
2260 void __release_sock(struct sock *sk)
2261 __releases(&sk->sk_lock.slock)
2262 __acquires(&sk->sk_lock.slock)
2264 struct sk_buff *skb, *next;
2266 while ((skb = sk->sk_backlog.head) != NULL) {
2267 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2269 spin_unlock_bh(&sk->sk_lock.slock);
2274 WARN_ON_ONCE(skb_dst_is_noref(skb));
2275 skb_mark_not_on_list(skb);
2276 sk_backlog_rcv(sk, skb);
2281 } while (skb != NULL);
2283 spin_lock_bh(&sk->sk_lock.slock);
2287 * Doing the zeroing here guarantee we can not loop forever
2288 * while a wild producer attempts to flood us.
2290 sk->sk_backlog.len = 0;
2293 void __sk_flush_backlog(struct sock *sk)
2295 spin_lock_bh(&sk->sk_lock.slock);
2297 spin_unlock_bh(&sk->sk_lock.slock);
2301 * sk_wait_data - wait for data to arrive at sk_receive_queue
2302 * @sk: sock to wait on
2303 * @timeo: for how long
2304 * @skb: last skb seen on sk_receive_queue
2306 * Now socket state including sk->sk_err is changed only under lock,
2307 * hence we may omit checks after joining wait queue.
2308 * We check receive queue before schedule() only as optimization;
2309 * it is very likely that release_sock() added new data.
2311 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2313 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2316 add_wait_queue(sk_sleep(sk), &wait);
2317 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2318 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2319 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2320 remove_wait_queue(sk_sleep(sk), &wait);
2323 EXPORT_SYMBOL(sk_wait_data);
2326 * __sk_mem_raise_allocated - increase memory_allocated
2328 * @size: memory size to allocate
2329 * @amt: pages to allocate
2330 * @kind: allocation type
2332 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2334 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2336 struct proto *prot = sk->sk_prot;
2337 long allocated = sk_memory_allocated_add(sk, amt);
2338 bool charged = true;
2340 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2341 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2342 goto suppress_allocation;
2345 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2346 sk_leave_memory_pressure(sk);
2350 /* Under pressure. */
2351 if (allocated > sk_prot_mem_limits(sk, 1))
2352 sk_enter_memory_pressure(sk);
2354 /* Over hard limit. */
2355 if (allocated > sk_prot_mem_limits(sk, 2))
2356 goto suppress_allocation;
2358 /* guarantee minimum buffer size under pressure */
2359 if (kind == SK_MEM_RECV) {
2360 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2363 } else { /* SK_MEM_SEND */
2364 int wmem0 = sk_get_wmem0(sk, prot);
2366 if (sk->sk_type == SOCK_STREAM) {
2367 if (sk->sk_wmem_queued < wmem0)
2369 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2374 if (sk_has_memory_pressure(sk)) {
2377 if (!sk_under_memory_pressure(sk))
2379 alloc = sk_sockets_allocated_read_positive(sk);
2380 if (sk_prot_mem_limits(sk, 2) > alloc *
2381 sk_mem_pages(sk->sk_wmem_queued +
2382 atomic_read(&sk->sk_rmem_alloc) +
2383 sk->sk_forward_alloc))
2387 suppress_allocation:
2389 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2390 sk_stream_moderate_sndbuf(sk);
2392 /* Fail only if socket is _under_ its sndbuf.
2393 * In this case we cannot block, so that we have to fail.
2395 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2399 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2400 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2402 sk_memory_allocated_sub(sk, amt);
2404 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2405 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2409 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2412 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2414 * @size: memory size to allocate
2415 * @kind: allocation type
2417 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2418 * rmem allocation. This function assumes that protocols which have
2419 * memory_pressure use sk_wmem_queued as write buffer accounting.
2421 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2423 int ret, amt = sk_mem_pages(size);
2425 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2426 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2428 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2431 EXPORT_SYMBOL(__sk_mem_schedule);
2434 * __sk_mem_reduce_allocated - reclaim memory_allocated
2436 * @amount: number of quanta
2438 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2440 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2442 sk_memory_allocated_sub(sk, amount);
2444 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2445 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2447 if (sk_under_memory_pressure(sk) &&
2448 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2449 sk_leave_memory_pressure(sk);
2451 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2454 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2456 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2458 void __sk_mem_reclaim(struct sock *sk, int amount)
2460 amount >>= SK_MEM_QUANTUM_SHIFT;
2461 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2462 __sk_mem_reduce_allocated(sk, amount);
2464 EXPORT_SYMBOL(__sk_mem_reclaim);
2466 int sk_set_peek_off(struct sock *sk, int val)
2468 sk->sk_peek_off = val;
2471 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2474 * Set of default routines for initialising struct proto_ops when
2475 * the protocol does not support a particular function. In certain
2476 * cases where it makes no sense for a protocol to have a "do nothing"
2477 * function, some default processing is provided.
2480 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2484 EXPORT_SYMBOL(sock_no_bind);
2486 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2491 EXPORT_SYMBOL(sock_no_connect);
2493 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2497 EXPORT_SYMBOL(sock_no_socketpair);
2499 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2504 EXPORT_SYMBOL(sock_no_accept);
2506 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2511 EXPORT_SYMBOL(sock_no_getname);
2513 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2517 EXPORT_SYMBOL(sock_no_ioctl);
2519 int sock_no_listen(struct socket *sock, int backlog)
2523 EXPORT_SYMBOL(sock_no_listen);
2525 int sock_no_shutdown(struct socket *sock, int how)
2529 EXPORT_SYMBOL(sock_no_shutdown);
2531 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2532 char __user *optval, unsigned int optlen)
2536 EXPORT_SYMBOL(sock_no_setsockopt);
2538 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2539 char __user *optval, int __user *optlen)
2543 EXPORT_SYMBOL(sock_no_getsockopt);
2545 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2549 EXPORT_SYMBOL(sock_no_sendmsg);
2551 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2555 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2557 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2562 EXPORT_SYMBOL(sock_no_recvmsg);
2564 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2566 /* Mirror missing mmap method error code */
2569 EXPORT_SYMBOL(sock_no_mmap);
2571 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2574 struct msghdr msg = {.msg_flags = flags};
2576 char *kaddr = kmap(page);
2577 iov.iov_base = kaddr + offset;
2579 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2583 EXPORT_SYMBOL(sock_no_sendpage);
2585 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2586 int offset, size_t size, int flags)
2589 struct msghdr msg = {.msg_flags = flags};
2591 char *kaddr = kmap(page);
2593 iov.iov_base = kaddr + offset;
2595 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2599 EXPORT_SYMBOL(sock_no_sendpage_locked);
2602 * Default Socket Callbacks
2605 static void sock_def_wakeup(struct sock *sk)
2607 struct socket_wq *wq;
2610 wq = rcu_dereference(sk->sk_wq);
2611 if (skwq_has_sleeper(wq))
2612 wake_up_interruptible_all(&wq->wait);
2616 static void sock_def_error_report(struct sock *sk)
2618 struct socket_wq *wq;
2621 wq = rcu_dereference(sk->sk_wq);
2622 if (skwq_has_sleeper(wq))
2623 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2624 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2628 static void sock_def_readable(struct sock *sk)
2630 struct socket_wq *wq;
2633 wq = rcu_dereference(sk->sk_wq);
2634 if (skwq_has_sleeper(wq))
2635 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2636 EPOLLRDNORM | EPOLLRDBAND);
2637 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2641 static void sock_def_write_space(struct sock *sk)
2643 struct socket_wq *wq;
2647 /* Do not wake up a writer until he can make "significant"
2650 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2651 wq = rcu_dereference(sk->sk_wq);
2652 if (skwq_has_sleeper(wq))
2653 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2654 EPOLLWRNORM | EPOLLWRBAND);
2656 /* Should agree with poll, otherwise some programs break */
2657 if (sock_writeable(sk))
2658 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2664 static void sock_def_destruct(struct sock *sk)
2668 void sk_send_sigurg(struct sock *sk)
2670 if (sk->sk_socket && sk->sk_socket->file)
2671 if (send_sigurg(&sk->sk_socket->file->f_owner))
2672 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2674 EXPORT_SYMBOL(sk_send_sigurg);
2676 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2677 unsigned long expires)
2679 if (!mod_timer(timer, expires))
2682 EXPORT_SYMBOL(sk_reset_timer);
2684 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2686 if (del_timer(timer))
2689 EXPORT_SYMBOL(sk_stop_timer);
2691 void sock_init_data(struct socket *sock, struct sock *sk)
2694 sk->sk_send_head = NULL;
2696 timer_setup(&sk->sk_timer, NULL, 0);
2698 sk->sk_allocation = GFP_KERNEL;
2699 sk->sk_rcvbuf = sysctl_rmem_default;
2700 sk->sk_sndbuf = sysctl_wmem_default;
2701 sk->sk_state = TCP_CLOSE;
2702 sk_set_socket(sk, sock);
2704 sock_set_flag(sk, SOCK_ZAPPED);
2707 sk->sk_type = sock->type;
2708 sk->sk_wq = sock->wq;
2710 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2713 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2716 rwlock_init(&sk->sk_callback_lock);
2717 if (sk->sk_kern_sock)
2718 lockdep_set_class_and_name(
2719 &sk->sk_callback_lock,
2720 af_kern_callback_keys + sk->sk_family,
2721 af_family_kern_clock_key_strings[sk->sk_family]);
2723 lockdep_set_class_and_name(
2724 &sk->sk_callback_lock,
2725 af_callback_keys + sk->sk_family,
2726 af_family_clock_key_strings[sk->sk_family]);
2728 sk->sk_state_change = sock_def_wakeup;
2729 sk->sk_data_ready = sock_def_readable;
2730 sk->sk_write_space = sock_def_write_space;
2731 sk->sk_error_report = sock_def_error_report;
2732 sk->sk_destruct = sock_def_destruct;
2734 sk->sk_frag.page = NULL;
2735 sk->sk_frag.offset = 0;
2736 sk->sk_peek_off = -1;
2738 sk->sk_peer_pid = NULL;
2739 sk->sk_peer_cred = NULL;
2740 sk->sk_write_pending = 0;
2741 sk->sk_rcvlowat = 1;
2742 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2743 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2745 sk->sk_stamp = SK_DEFAULT_STAMP;
2746 atomic_set(&sk->sk_zckey, 0);
2748 #ifdef CONFIG_NET_RX_BUSY_POLL
2750 sk->sk_ll_usec = sysctl_net_busy_read;
2753 sk->sk_max_pacing_rate = ~0UL;
2754 sk->sk_pacing_rate = ~0UL;
2755 sk->sk_pacing_shift = 10;
2756 sk->sk_incoming_cpu = -1;
2758 sk_rx_queue_clear(sk);
2760 * Before updating sk_refcnt, we must commit prior changes to memory
2761 * (Documentation/RCU/rculist_nulls.txt for details)
2764 refcount_set(&sk->sk_refcnt, 1);
2765 atomic_set(&sk->sk_drops, 0);
2767 EXPORT_SYMBOL(sock_init_data);
2769 void lock_sock_nested(struct sock *sk, int subclass)
2772 spin_lock_bh(&sk->sk_lock.slock);
2773 if (sk->sk_lock.owned)
2775 sk->sk_lock.owned = 1;
2776 spin_unlock(&sk->sk_lock.slock);
2778 * The sk_lock has mutex_lock() semantics here:
2780 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2783 EXPORT_SYMBOL(lock_sock_nested);
2785 void release_sock(struct sock *sk)
2787 spin_lock_bh(&sk->sk_lock.slock);
2788 if (sk->sk_backlog.tail)
2791 /* Warning : release_cb() might need to release sk ownership,
2792 * ie call sock_release_ownership(sk) before us.
2794 if (sk->sk_prot->release_cb)
2795 sk->sk_prot->release_cb(sk);
2797 sock_release_ownership(sk);
2798 if (waitqueue_active(&sk->sk_lock.wq))
2799 wake_up(&sk->sk_lock.wq);
2800 spin_unlock_bh(&sk->sk_lock.slock);
2802 EXPORT_SYMBOL(release_sock);
2805 * lock_sock_fast - fast version of lock_sock
2808 * This version should be used for very small section, where process wont block
2809 * return false if fast path is taken:
2811 * sk_lock.slock locked, owned = 0, BH disabled
2813 * return true if slow path is taken:
2815 * sk_lock.slock unlocked, owned = 1, BH enabled
2817 bool lock_sock_fast(struct sock *sk)
2820 spin_lock_bh(&sk->sk_lock.slock);
2822 if (!sk->sk_lock.owned)
2824 * Note : We must disable BH
2829 sk->sk_lock.owned = 1;
2830 spin_unlock(&sk->sk_lock.slock);
2832 * The sk_lock has mutex_lock() semantics here:
2834 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2838 EXPORT_SYMBOL(lock_sock_fast);
2840 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2844 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2845 tv = ktime_to_timeval(sk->sk_stamp);
2846 if (tv.tv_sec == -1)
2848 if (tv.tv_sec == 0) {
2849 sk->sk_stamp = ktime_get_real();
2850 tv = ktime_to_timeval(sk->sk_stamp);
2852 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2854 EXPORT_SYMBOL(sock_get_timestamp);
2856 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2860 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2861 ts = ktime_to_timespec(sk->sk_stamp);
2862 if (ts.tv_sec == -1)
2864 if (ts.tv_sec == 0) {
2865 sk->sk_stamp = ktime_get_real();
2866 ts = ktime_to_timespec(sk->sk_stamp);
2868 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2870 EXPORT_SYMBOL(sock_get_timestampns);
2872 void sock_enable_timestamp(struct sock *sk, int flag)
2874 if (!sock_flag(sk, flag)) {
2875 unsigned long previous_flags = sk->sk_flags;
2877 sock_set_flag(sk, flag);
2879 * we just set one of the two flags which require net
2880 * time stamping, but time stamping might have been on
2881 * already because of the other one
2883 if (sock_needs_netstamp(sk) &&
2884 !(previous_flags & SK_FLAGS_TIMESTAMP))
2885 net_enable_timestamp();
2889 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2890 int level, int type)
2892 struct sock_exterr_skb *serr;
2893 struct sk_buff *skb;
2897 skb = sock_dequeue_err_skb(sk);
2903 msg->msg_flags |= MSG_TRUNC;
2906 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2910 sock_recv_timestamp(msg, sk, skb);
2912 serr = SKB_EXT_ERR(skb);
2913 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2915 msg->msg_flags |= MSG_ERRQUEUE;
2923 EXPORT_SYMBOL(sock_recv_errqueue);
2926 * Get a socket option on an socket.
2928 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2929 * asynchronous errors should be reported by getsockopt. We assume
2930 * this means if you specify SO_ERROR (otherwise whats the point of it).
2932 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2933 char __user *optval, int __user *optlen)
2935 struct sock *sk = sock->sk;
2937 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2939 EXPORT_SYMBOL(sock_common_getsockopt);
2941 #ifdef CONFIG_COMPAT
2942 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2943 char __user *optval, int __user *optlen)
2945 struct sock *sk = sock->sk;
2947 if (sk->sk_prot->compat_getsockopt != NULL)
2948 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2950 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2952 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2955 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2958 struct sock *sk = sock->sk;
2962 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2963 flags & ~MSG_DONTWAIT, &addr_len);
2965 msg->msg_namelen = addr_len;
2968 EXPORT_SYMBOL(sock_common_recvmsg);
2971 * Set socket options on an inet socket.
2973 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2974 char __user *optval, unsigned int optlen)
2976 struct sock *sk = sock->sk;
2978 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2980 EXPORT_SYMBOL(sock_common_setsockopt);
2982 #ifdef CONFIG_COMPAT
2983 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2984 char __user *optval, unsigned int optlen)
2986 struct sock *sk = sock->sk;
2988 if (sk->sk_prot->compat_setsockopt != NULL)
2989 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2991 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2993 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2996 void sk_common_release(struct sock *sk)
2998 if (sk->sk_prot->destroy)
2999 sk->sk_prot->destroy(sk);
3002 * Observation: when sock_common_release is called, processes have
3003 * no access to socket. But net still has.
3004 * Step one, detach it from networking:
3006 * A. Remove from hash tables.
3009 sk->sk_prot->unhash(sk);
3012 * In this point socket cannot receive new packets, but it is possible
3013 * that some packets are in flight because some CPU runs receiver and
3014 * did hash table lookup before we unhashed socket. They will achieve
3015 * receive queue and will be purged by socket destructor.
3017 * Also we still have packets pending on receive queue and probably,
3018 * our own packets waiting in device queues. sock_destroy will drain
3019 * receive queue, but transmitted packets will delay socket destruction
3020 * until the last reference will be released.
3025 xfrm_sk_free_policy(sk);
3027 sk_refcnt_debug_release(sk);
3031 EXPORT_SYMBOL(sk_common_release);
3033 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3035 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3037 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3038 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3039 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3040 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3041 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3042 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3043 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3044 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3045 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3048 #ifdef CONFIG_PROC_FS
3049 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3051 int val[PROTO_INUSE_NR];
3054 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3056 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3058 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3060 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3062 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3064 int cpu, idx = prot->inuse_idx;
3067 for_each_possible_cpu(cpu)
3068 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3070 return res >= 0 ? res : 0;
3072 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3074 static void sock_inuse_add(struct net *net, int val)
3076 this_cpu_add(*net->core.sock_inuse, val);
3079 int sock_inuse_get(struct net *net)
3083 for_each_possible_cpu(cpu)
3084 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3089 EXPORT_SYMBOL_GPL(sock_inuse_get);
3091 static int __net_init sock_inuse_init_net(struct net *net)
3093 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3094 if (net->core.prot_inuse == NULL)
3097 net->core.sock_inuse = alloc_percpu(int);
3098 if (net->core.sock_inuse == NULL)
3104 free_percpu(net->core.prot_inuse);
3108 static void __net_exit sock_inuse_exit_net(struct net *net)
3110 free_percpu(net->core.prot_inuse);
3111 free_percpu(net->core.sock_inuse);
3114 static struct pernet_operations net_inuse_ops = {
3115 .init = sock_inuse_init_net,
3116 .exit = sock_inuse_exit_net,
3119 static __init int net_inuse_init(void)
3121 if (register_pernet_subsys(&net_inuse_ops))
3122 panic("Cannot initialize net inuse counters");
3127 core_initcall(net_inuse_init);
3129 static void assign_proto_idx(struct proto *prot)
3131 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3133 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3134 pr_err("PROTO_INUSE_NR exhausted\n");
3138 set_bit(prot->inuse_idx, proto_inuse_idx);
3141 static void release_proto_idx(struct proto *prot)
3143 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3144 clear_bit(prot->inuse_idx, proto_inuse_idx);
3147 static inline void assign_proto_idx(struct proto *prot)
3151 static inline void release_proto_idx(struct proto *prot)
3155 static void sock_inuse_add(struct net *net, int val)
3160 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3164 kfree(rsk_prot->slab_name);
3165 rsk_prot->slab_name = NULL;
3166 kmem_cache_destroy(rsk_prot->slab);
3167 rsk_prot->slab = NULL;
3170 static int req_prot_init(const struct proto *prot)
3172 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3177 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3179 if (!rsk_prot->slab_name)
3182 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3183 rsk_prot->obj_size, 0,
3184 SLAB_ACCOUNT | prot->slab_flags,
3187 if (!rsk_prot->slab) {
3188 pr_crit("%s: Can't create request sock SLAB cache!\n",
3195 int proto_register(struct proto *prot, int alloc_slab)
3198 prot->slab = kmem_cache_create_usercopy(prot->name,
3200 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3202 prot->useroffset, prot->usersize,
3205 if (prot->slab == NULL) {
3206 pr_crit("%s: Can't create sock SLAB cache!\n",
3211 if (req_prot_init(prot))
3212 goto out_free_request_sock_slab;
3214 if (prot->twsk_prot != NULL) {
3215 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3217 if (prot->twsk_prot->twsk_slab_name == NULL)
3218 goto out_free_request_sock_slab;
3220 prot->twsk_prot->twsk_slab =
3221 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3222 prot->twsk_prot->twsk_obj_size,
3227 if (prot->twsk_prot->twsk_slab == NULL)
3228 goto out_free_timewait_sock_slab_name;
3232 mutex_lock(&proto_list_mutex);
3233 list_add(&prot->node, &proto_list);
3234 assign_proto_idx(prot);
3235 mutex_unlock(&proto_list_mutex);
3238 out_free_timewait_sock_slab_name:
3239 kfree(prot->twsk_prot->twsk_slab_name);
3240 out_free_request_sock_slab:
3241 req_prot_cleanup(prot->rsk_prot);
3243 kmem_cache_destroy(prot->slab);
3248 EXPORT_SYMBOL(proto_register);
3250 void proto_unregister(struct proto *prot)
3252 mutex_lock(&proto_list_mutex);
3253 release_proto_idx(prot);
3254 list_del(&prot->node);
3255 mutex_unlock(&proto_list_mutex);
3257 kmem_cache_destroy(prot->slab);
3260 req_prot_cleanup(prot->rsk_prot);
3262 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3263 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3264 kfree(prot->twsk_prot->twsk_slab_name);
3265 prot->twsk_prot->twsk_slab = NULL;
3268 EXPORT_SYMBOL(proto_unregister);
3270 int sock_load_diag_module(int family, int protocol)
3273 if (!sock_is_registered(family))
3276 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3277 NETLINK_SOCK_DIAG, family);
3281 if (family == AF_INET &&
3282 protocol != IPPROTO_RAW &&
3283 !rcu_access_pointer(inet_protos[protocol]))
3287 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3288 NETLINK_SOCK_DIAG, family, protocol);
3290 EXPORT_SYMBOL(sock_load_diag_module);
3292 #ifdef CONFIG_PROC_FS
3293 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3294 __acquires(proto_list_mutex)
3296 mutex_lock(&proto_list_mutex);
3297 return seq_list_start_head(&proto_list, *pos);
3300 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3302 return seq_list_next(v, &proto_list, pos);
3305 static void proto_seq_stop(struct seq_file *seq, void *v)
3306 __releases(proto_list_mutex)
3308 mutex_unlock(&proto_list_mutex);
3311 static char proto_method_implemented(const void *method)
3313 return method == NULL ? 'n' : 'y';
3315 static long sock_prot_memory_allocated(struct proto *proto)
3317 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3320 static char *sock_prot_memory_pressure(struct proto *proto)
3322 return proto->memory_pressure != NULL ?
3323 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3326 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3329 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3330 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3333 sock_prot_inuse_get(seq_file_net(seq), proto),
3334 sock_prot_memory_allocated(proto),
3335 sock_prot_memory_pressure(proto),
3337 proto->slab == NULL ? "no" : "yes",
3338 module_name(proto->owner),
3339 proto_method_implemented(proto->close),
3340 proto_method_implemented(proto->connect),
3341 proto_method_implemented(proto->disconnect),
3342 proto_method_implemented(proto->accept),
3343 proto_method_implemented(proto->ioctl),
3344 proto_method_implemented(proto->init),
3345 proto_method_implemented(proto->destroy),
3346 proto_method_implemented(proto->shutdown),
3347 proto_method_implemented(proto->setsockopt),
3348 proto_method_implemented(proto->getsockopt),
3349 proto_method_implemented(proto->sendmsg),
3350 proto_method_implemented(proto->recvmsg),
3351 proto_method_implemented(proto->sendpage),
3352 proto_method_implemented(proto->bind),
3353 proto_method_implemented(proto->backlog_rcv),
3354 proto_method_implemented(proto->hash),
3355 proto_method_implemented(proto->unhash),
3356 proto_method_implemented(proto->get_port),
3357 proto_method_implemented(proto->enter_memory_pressure));
3360 static int proto_seq_show(struct seq_file *seq, void *v)
3362 if (v == &proto_list)
3363 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3372 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3374 proto_seq_printf(seq, list_entry(v, struct proto, node));
3378 static const struct seq_operations proto_seq_ops = {
3379 .start = proto_seq_start,
3380 .next = proto_seq_next,
3381 .stop = proto_seq_stop,
3382 .show = proto_seq_show,
3385 static __net_init int proto_init_net(struct net *net)
3387 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3388 sizeof(struct seq_net_private)))
3394 static __net_exit void proto_exit_net(struct net *net)
3396 remove_proc_entry("protocols", net->proc_net);
3400 static __net_initdata struct pernet_operations proto_net_ops = {
3401 .init = proto_init_net,
3402 .exit = proto_exit_net,
3405 static int __init proto_init(void)
3407 return register_pernet_subsys(&proto_net_ops);
3410 subsys_initcall(proto_init);
3412 #endif /* PROC_FS */
3414 #ifdef CONFIG_NET_RX_BUSY_POLL
3415 bool sk_busy_loop_end(void *p, unsigned long start_time)
3417 struct sock *sk = p;
3419 return !skb_queue_empty(&sk->sk_receive_queue) ||
3420 sk_busy_loop_timeout(sk, start_time);
3422 EXPORT_SYMBOL(sk_busy_loop_end);
3423 #endif /* CONFIG_NET_RX_BUSY_POLL */