1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
118 #include <linux/uaccess.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
137 #include <trace/events/sock.h>
140 #include <net/busy_poll.h>
142 static DEFINE_MUTEX(proto_list_mutex);
143 static LIST_HEAD(proto_list);
145 static void sock_inuse_add(struct net *net, int val);
148 * sk_ns_capable - General socket capability test
149 * @sk: Socket to use a capability on or through
150 * @user_ns: The user namespace of the capability to use
151 * @cap: The capability to use
153 * Test to see if the opener of the socket had when the socket was
154 * created and the current process has the capability @cap in the user
155 * namespace @user_ns.
157 bool sk_ns_capable(const struct sock *sk,
158 struct user_namespace *user_ns, int cap)
160 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
161 ns_capable(user_ns, cap);
163 EXPORT_SYMBOL(sk_ns_capable);
166 * sk_capable - Socket global capability test
167 * @sk: Socket to use a capability on or through
168 * @cap: The global capability to use
170 * Test to see if the opener of the socket had when the socket was
171 * created and the current process has the capability @cap in all user
174 bool sk_capable(const struct sock *sk, int cap)
176 return sk_ns_capable(sk, &init_user_ns, cap);
178 EXPORT_SYMBOL(sk_capable);
181 * sk_net_capable - Network namespace socket capability test
182 * @sk: Socket to use a capability on or through
183 * @cap: The capability to use
185 * Test to see if the opener of the socket had when the socket was created
186 * and the current process has the capability @cap over the network namespace
187 * the socket is a member of.
189 bool sk_net_capable(const struct sock *sk, int cap)
191 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193 EXPORT_SYMBOL(sk_net_capable);
196 * Each address family might have different locking rules, so we have
197 * one slock key per address family and separate keys for internal and
200 static struct lock_class_key af_family_keys[AF_MAX];
201 static struct lock_class_key af_family_kern_keys[AF_MAX];
202 static struct lock_class_key af_family_slock_keys[AF_MAX];
203 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
211 #define _sock_locks(x) \
212 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
213 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
214 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
215 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
216 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
217 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
218 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
219 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
220 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
221 x "27" , x "28" , x "AF_CAN" , \
222 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
223 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
224 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
225 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
226 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
229 static const char *const af_family_key_strings[AF_MAX+1] = {
230 _sock_locks("sk_lock-")
232 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
233 _sock_locks("slock-")
235 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
236 _sock_locks("clock-")
239 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
240 _sock_locks("k-sk_lock-")
242 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("k-slock-")
245 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-clock-")
248 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
249 _sock_locks("rlock-")
251 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
252 _sock_locks("wlock-")
254 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
255 _sock_locks("elock-")
259 * sk_callback_lock and sk queues locking rules are per-address-family,
260 * so split the lock classes by using a per-AF key:
262 static struct lock_class_key af_callback_keys[AF_MAX];
263 static struct lock_class_key af_rlock_keys[AF_MAX];
264 static struct lock_class_key af_wlock_keys[AF_MAX];
265 static struct lock_class_key af_elock_keys[AF_MAX];
266 static struct lock_class_key af_kern_callback_keys[AF_MAX];
268 /* Run time adjustable parameters. */
269 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
270 EXPORT_SYMBOL(sysctl_wmem_max);
271 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
272 EXPORT_SYMBOL(sysctl_rmem_max);
273 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
274 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
276 /* Maximal space eaten by iovec or ancillary data plus some space */
277 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
278 EXPORT_SYMBOL(sysctl_optmem_max);
280 int sysctl_tstamp_allow_data __read_mostly = 1;
282 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
283 EXPORT_SYMBOL_GPL(memalloc_socks_key);
286 * sk_set_memalloc - sets %SOCK_MEMALLOC
287 * @sk: socket to set it on
289 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
290 * It's the responsibility of the admin to adjust min_free_kbytes
291 * to meet the requirements
293 void sk_set_memalloc(struct sock *sk)
295 sock_set_flag(sk, SOCK_MEMALLOC);
296 sk->sk_allocation |= __GFP_MEMALLOC;
297 static_branch_inc(&memalloc_socks_key);
299 EXPORT_SYMBOL_GPL(sk_set_memalloc);
301 void sk_clear_memalloc(struct sock *sk)
303 sock_reset_flag(sk, SOCK_MEMALLOC);
304 sk->sk_allocation &= ~__GFP_MEMALLOC;
305 static_branch_dec(&memalloc_socks_key);
308 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
309 * progress of swapping. SOCK_MEMALLOC may be cleared while
310 * it has rmem allocations due to the last swapfile being deactivated
311 * but there is a risk that the socket is unusable due to exceeding
312 * the rmem limits. Reclaim the reserves and obey rmem limits again.
316 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
318 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
321 unsigned int noreclaim_flag;
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
326 noreclaim_flag = memalloc_noreclaim_save();
327 ret = sk->sk_backlog_rcv(sk, skb);
328 memalloc_noreclaim_restore(noreclaim_flag);
332 EXPORT_SYMBOL(__sk_backlog_rcv);
334 static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
336 struct __kernel_sock_timeval tv;
338 if (timeo == MAX_SCHEDULE_TIMEOUT) {
342 tv.tv_sec = timeo / HZ;
343 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
346 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
347 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
348 *(struct old_timeval32 *)optval = tv32;
353 struct __kernel_old_timeval old_tv;
354 old_tv.tv_sec = tv.tv_sec;
355 old_tv.tv_usec = tv.tv_usec;
356 *(struct __kernel_old_timeval *)optval = old_tv;
357 return sizeof(old_tv);
360 *(struct __kernel_sock_timeval *)optval = tv;
364 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
367 struct __kernel_sock_timeval tv;
369 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
370 struct old_timeval32 tv32;
372 if (optlen < sizeof(tv32))
375 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
377 tv.tv_sec = tv32.tv_sec;
378 tv.tv_usec = tv32.tv_usec;
379 } else if (old_timeval) {
380 struct __kernel_old_timeval old_tv;
382 if (optlen < sizeof(old_tv))
384 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
386 tv.tv_sec = old_tv.tv_sec;
387 tv.tv_usec = old_tv.tv_usec;
389 if (optlen < sizeof(tv))
391 if (copy_from_sockptr(&tv, optval, sizeof(tv)))
394 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
398 static int warned __read_mostly;
401 if (warned < 10 && net_ratelimit()) {
403 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
404 __func__, current->comm, task_pid_nr(current));
408 *timeo_p = MAX_SCHEDULE_TIMEOUT;
409 if (tv.tv_sec == 0 && tv.tv_usec == 0)
411 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
412 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
416 static bool sock_needs_netstamp(const struct sock *sk)
418 switch (sk->sk_family) {
427 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
429 if (sk->sk_flags & flags) {
430 sk->sk_flags &= ~flags;
431 if (sock_needs_netstamp(sk) &&
432 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
433 net_disable_timestamp();
438 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
441 struct sk_buff_head *list = &sk->sk_receive_queue;
443 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
444 atomic_inc(&sk->sk_drops);
445 trace_sock_rcvqueue_full(sk, skb);
449 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
450 atomic_inc(&sk->sk_drops);
455 skb_set_owner_r(skb, sk);
457 /* we escape from rcu protected region, make sure we dont leak
462 spin_lock_irqsave(&list->lock, flags);
463 sock_skb_set_dropcount(sk, skb);
464 __skb_queue_tail(list, skb);
465 spin_unlock_irqrestore(&list->lock, flags);
467 if (!sock_flag(sk, SOCK_DEAD))
468 sk->sk_data_ready(sk);
471 EXPORT_SYMBOL(__sock_queue_rcv_skb);
473 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
477 err = sk_filter(sk, skb);
481 return __sock_queue_rcv_skb(sk, skb);
483 EXPORT_SYMBOL(sock_queue_rcv_skb);
485 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
486 const int nested, unsigned int trim_cap, bool refcounted)
488 int rc = NET_RX_SUCCESS;
490 if (sk_filter_trim_cap(sk, skb, trim_cap))
491 goto discard_and_relse;
495 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
496 atomic_inc(&sk->sk_drops);
497 goto discard_and_relse;
500 bh_lock_sock_nested(sk);
503 if (!sock_owned_by_user(sk)) {
505 * trylock + unlock semantics:
507 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
509 rc = sk_backlog_rcv(sk, skb);
511 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
512 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
514 atomic_inc(&sk->sk_drops);
515 goto discard_and_relse;
527 EXPORT_SYMBOL(__sk_receive_skb);
529 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
531 struct dst_entry *dst = __sk_dst_get(sk);
533 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
534 sk_tx_queue_clear(sk);
535 sk->sk_dst_pending_confirm = 0;
536 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
543 EXPORT_SYMBOL(__sk_dst_check);
545 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
547 struct dst_entry *dst = sk_dst_get(sk);
549 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
557 EXPORT_SYMBOL(sk_dst_check);
559 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
561 int ret = -ENOPROTOOPT;
562 #ifdef CONFIG_NETDEVICES
563 struct net *net = sock_net(sk);
567 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
574 sk->sk_bound_dev_if = ifindex;
575 if (sk->sk_prot->rehash)
576 sk->sk_prot->rehash(sk);
587 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
593 ret = sock_bindtoindex_locked(sk, ifindex);
599 EXPORT_SYMBOL(sock_bindtoindex);
601 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
603 int ret = -ENOPROTOOPT;
604 #ifdef CONFIG_NETDEVICES
605 struct net *net = sock_net(sk);
606 char devname[IFNAMSIZ];
613 /* Bind this socket to a particular device like "eth0",
614 * as specified in the passed interface name. If the
615 * name is "" or the option length is zero the socket
618 if (optlen > IFNAMSIZ - 1)
619 optlen = IFNAMSIZ - 1;
620 memset(devname, 0, sizeof(devname));
623 if (copy_from_sockptr(devname, optval, optlen))
627 if (devname[0] != '\0') {
628 struct net_device *dev;
631 dev = dev_get_by_name_rcu(net, devname);
633 index = dev->ifindex;
640 return sock_bindtoindex(sk, index, true);
647 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
648 int __user *optlen, int len)
650 int ret = -ENOPROTOOPT;
651 #ifdef CONFIG_NETDEVICES
652 struct net *net = sock_net(sk);
653 char devname[IFNAMSIZ];
655 if (sk->sk_bound_dev_if == 0) {
664 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
668 len = strlen(devname) + 1;
671 if (copy_to_user(optval, devname, len))
676 if (put_user(len, optlen))
687 bool sk_mc_loop(struct sock *sk)
689 if (dev_recursion_level())
693 switch (sk->sk_family) {
695 return inet_sk(sk)->mc_loop;
696 #if IS_ENABLED(CONFIG_IPV6)
698 return inet6_sk(sk)->mc_loop;
704 EXPORT_SYMBOL(sk_mc_loop);
706 void sock_set_reuseaddr(struct sock *sk)
709 sk->sk_reuse = SK_CAN_REUSE;
712 EXPORT_SYMBOL(sock_set_reuseaddr);
714 void sock_set_reuseport(struct sock *sk)
717 sk->sk_reuseport = true;
720 EXPORT_SYMBOL(sock_set_reuseport);
722 void sock_no_linger(struct sock *sk)
725 sk->sk_lingertime = 0;
726 sock_set_flag(sk, SOCK_LINGER);
729 EXPORT_SYMBOL(sock_no_linger);
731 void sock_set_priority(struct sock *sk, u32 priority)
734 sk->sk_priority = priority;
737 EXPORT_SYMBOL(sock_set_priority);
739 void sock_set_sndtimeo(struct sock *sk, s64 secs)
742 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
743 sk->sk_sndtimeo = secs * HZ;
745 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
748 EXPORT_SYMBOL(sock_set_sndtimeo);
750 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
753 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
754 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
755 sock_set_flag(sk, SOCK_RCVTSTAMP);
756 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
758 sock_reset_flag(sk, SOCK_RCVTSTAMP);
759 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
763 void sock_enable_timestamps(struct sock *sk)
766 __sock_set_timestamps(sk, true, false, true);
769 EXPORT_SYMBOL(sock_enable_timestamps);
771 void sock_set_keepalive(struct sock *sk)
774 if (sk->sk_prot->keepalive)
775 sk->sk_prot->keepalive(sk, true);
776 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
779 EXPORT_SYMBOL(sock_set_keepalive);
781 static void __sock_set_rcvbuf(struct sock *sk, int val)
783 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
784 * as a negative value.
786 val = min_t(int, val, INT_MAX / 2);
787 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
789 /* We double it on the way in to account for "struct sk_buff" etc.
790 * overhead. Applications assume that the SO_RCVBUF setting they make
791 * will allow that much actual data to be received on that socket.
793 * Applications are unaware that "struct sk_buff" and other overheads
794 * allocate from the receive buffer during socket buffer allocation.
796 * And after considering the possible alternatives, returning the value
797 * we actually used in getsockopt is the most desirable behavior.
799 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
802 void sock_set_rcvbuf(struct sock *sk, int val)
805 __sock_set_rcvbuf(sk, val);
808 EXPORT_SYMBOL(sock_set_rcvbuf);
810 void sock_set_mark(struct sock *sk, u32 val)
816 EXPORT_SYMBOL(sock_set_mark);
819 * This is meant for all protocols to use and covers goings on
820 * at the socket level. Everything here is generic.
823 int sock_setsockopt(struct socket *sock, int level, int optname,
824 sockptr_t optval, unsigned int optlen)
826 struct sock_txtime sk_txtime;
827 struct sock *sk = sock->sk;
834 * Options without arguments
837 if (optname == SO_BINDTODEVICE)
838 return sock_setbindtodevice(sk, optval, optlen);
840 if (optlen < sizeof(int))
843 if (copy_from_sockptr(&val, optval, sizeof(val)))
846 valbool = val ? 1 : 0;
852 if (val && !capable(CAP_NET_ADMIN))
855 sock_valbool_flag(sk, SOCK_DBG, valbool);
858 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
861 sk->sk_reuseport = valbool;
870 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
874 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
877 /* Don't error on this BSD doesn't and if you think
878 * about it this is right. Otherwise apps have to
879 * play 'guess the biggest size' games. RCVBUF/SNDBUF
880 * are treated in BSD as hints
882 val = min_t(u32, val, sysctl_wmem_max);
884 /* Ensure val * 2 fits into an int, to prevent max_t()
885 * from treating it as a negative value.
887 val = min_t(int, val, INT_MAX / 2);
888 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
889 WRITE_ONCE(sk->sk_sndbuf,
890 max_t(int, val * 2, SOCK_MIN_SNDBUF));
891 /* Wake up sending tasks if we upped the value. */
892 sk->sk_write_space(sk);
896 if (!capable(CAP_NET_ADMIN)) {
901 /* No negative values (to prevent underflow, as val will be
909 /* Don't error on this BSD doesn't and if you think
910 * about it this is right. Otherwise apps have to
911 * play 'guess the biggest size' games. RCVBUF/SNDBUF
912 * are treated in BSD as hints
914 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
918 if (!capable(CAP_NET_ADMIN)) {
923 /* No negative values (to prevent underflow, as val will be
926 __sock_set_rcvbuf(sk, max(val, 0));
930 if (sk->sk_prot->keepalive)
931 sk->sk_prot->keepalive(sk, valbool);
932 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
936 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
940 sk->sk_no_check_tx = valbool;
944 if ((val >= 0 && val <= 6) ||
945 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
946 sk->sk_priority = val;
952 if (optlen < sizeof(ling)) {
953 ret = -EINVAL; /* 1003.1g */
956 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
961 sock_reset_flag(sk, SOCK_LINGER);
963 #if (BITS_PER_LONG == 32)
964 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
965 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
968 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
969 sock_set_flag(sk, SOCK_LINGER);
978 set_bit(SOCK_PASSCRED, &sock->flags);
980 clear_bit(SOCK_PASSCRED, &sock->flags);
983 case SO_TIMESTAMP_OLD:
984 __sock_set_timestamps(sk, valbool, false, false);
986 case SO_TIMESTAMP_NEW:
987 __sock_set_timestamps(sk, valbool, true, false);
989 case SO_TIMESTAMPNS_OLD:
990 __sock_set_timestamps(sk, valbool, false, true);
992 case SO_TIMESTAMPNS_NEW:
993 __sock_set_timestamps(sk, valbool, true, true);
995 case SO_TIMESTAMPING_NEW:
996 case SO_TIMESTAMPING_OLD:
997 if (val & ~SOF_TIMESTAMPING_MASK) {
1002 if (val & SOF_TIMESTAMPING_OPT_ID &&
1003 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
1004 if (sk->sk_protocol == IPPROTO_TCP &&
1005 sk->sk_type == SOCK_STREAM) {
1006 if ((1 << sk->sk_state) &
1007 (TCPF_CLOSE | TCPF_LISTEN)) {
1011 sk->sk_tskey = tcp_sk(sk)->snd_una;
1017 if (val & SOF_TIMESTAMPING_OPT_STATS &&
1018 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
1023 sk->sk_tsflags = val;
1024 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
1026 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
1027 sock_enable_timestamp(sk,
1028 SOCK_TIMESTAMPING_RX_SOFTWARE);
1030 sock_disable_timestamp(sk,
1031 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
1037 if (sock->ops->set_rcvlowat)
1038 ret = sock->ops->set_rcvlowat(sk, val);
1040 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1043 case SO_RCVTIMEO_OLD:
1044 case SO_RCVTIMEO_NEW:
1045 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1046 optlen, optname == SO_RCVTIMEO_OLD);
1049 case SO_SNDTIMEO_OLD:
1050 case SO_SNDTIMEO_NEW:
1051 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1052 optlen, optname == SO_SNDTIMEO_OLD);
1055 case SO_ATTACH_FILTER: {
1056 struct sock_fprog fprog;
1058 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1060 ret = sk_attach_filter(&fprog, sk);
1065 if (optlen == sizeof(u32)) {
1069 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1072 ret = sk_attach_bpf(ufd, sk);
1076 case SO_ATTACH_REUSEPORT_CBPF: {
1077 struct sock_fprog fprog;
1079 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1081 ret = sk_reuseport_attach_filter(&fprog, sk);
1084 case SO_ATTACH_REUSEPORT_EBPF:
1086 if (optlen == sizeof(u32)) {
1090 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1093 ret = sk_reuseport_attach_bpf(ufd, sk);
1097 case SO_DETACH_REUSEPORT_BPF:
1098 ret = reuseport_detach_prog(sk);
1101 case SO_DETACH_FILTER:
1102 ret = sk_detach_filter(sk);
1105 case SO_LOCK_FILTER:
1106 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1109 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1114 set_bit(SOCK_PASSSEC, &sock->flags);
1116 clear_bit(SOCK_PASSSEC, &sock->flags);
1119 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1121 } else if (val != sk->sk_mark) {
1128 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1131 case SO_WIFI_STATUS:
1132 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1136 if (sock->ops->set_peek_off)
1137 ret = sock->ops->set_peek_off(sk, val);
1143 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1146 case SO_SELECT_ERR_QUEUE:
1147 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1150 #ifdef CONFIG_NET_RX_BUSY_POLL
1152 /* allow unprivileged users to decrease the value */
1153 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1159 sk->sk_ll_usec = val;
1162 case SO_PREFER_BUSY_POLL:
1163 if (valbool && !capable(CAP_NET_ADMIN))
1166 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1168 case SO_BUSY_POLL_BUDGET:
1169 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1172 if (val < 0 || val > U16_MAX)
1175 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1180 case SO_MAX_PACING_RATE:
1182 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1184 if (sizeof(ulval) != sizeof(val) &&
1185 optlen >= sizeof(ulval) &&
1186 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1191 cmpxchg(&sk->sk_pacing_status,
1194 sk->sk_max_pacing_rate = ulval;
1195 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1198 case SO_INCOMING_CPU:
1199 WRITE_ONCE(sk->sk_incoming_cpu, val);
1204 dst_negative_advice(sk);
1208 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1209 if (!((sk->sk_type == SOCK_STREAM &&
1210 sk->sk_protocol == IPPROTO_TCP) ||
1211 (sk->sk_type == SOCK_DGRAM &&
1212 sk->sk_protocol == IPPROTO_UDP)))
1214 } else if (sk->sk_family != PF_RDS) {
1218 if (val < 0 || val > 1)
1221 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1226 if (optlen != sizeof(struct sock_txtime)) {
1229 } else if (copy_from_sockptr(&sk_txtime, optval,
1230 sizeof(struct sock_txtime))) {
1233 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1237 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1238 * scheduler has enough safe guards.
1240 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1241 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1245 sock_valbool_flag(sk, SOCK_TXTIME, true);
1246 sk->sk_clockid = sk_txtime.clockid;
1247 sk->sk_txtime_deadline_mode =
1248 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1249 sk->sk_txtime_report_errors =
1250 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1253 case SO_BINDTOIFINDEX:
1254 ret = sock_bindtoindex_locked(sk, val);
1264 EXPORT_SYMBOL(sock_setsockopt);
1267 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1268 struct ucred *ucred)
1270 ucred->pid = pid_vnr(pid);
1271 ucred->uid = ucred->gid = -1;
1273 struct user_namespace *current_ns = current_user_ns();
1275 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1276 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1280 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1282 struct user_namespace *user_ns = current_user_ns();
1285 for (i = 0; i < src->ngroups; i++)
1286 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1292 int sock_getsockopt(struct socket *sock, int level, int optname,
1293 char __user *optval, int __user *optlen)
1295 struct sock *sk = sock->sk;
1300 unsigned long ulval;
1302 struct old_timeval32 tm32;
1303 struct __kernel_old_timeval tm;
1304 struct __kernel_sock_timeval stm;
1305 struct sock_txtime txtime;
1308 int lv = sizeof(int);
1311 if (get_user(len, optlen))
1316 memset(&v, 0, sizeof(v));
1320 v.val = sock_flag(sk, SOCK_DBG);
1324 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1328 v.val = sock_flag(sk, SOCK_BROADCAST);
1332 v.val = sk->sk_sndbuf;
1336 v.val = sk->sk_rcvbuf;
1340 v.val = sk->sk_reuse;
1344 v.val = sk->sk_reuseport;
1348 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1352 v.val = sk->sk_type;
1356 v.val = sk->sk_protocol;
1360 v.val = sk->sk_family;
1364 v.val = -sock_error(sk);
1366 v.val = xchg(&sk->sk_err_soft, 0);
1370 v.val = sock_flag(sk, SOCK_URGINLINE);
1374 v.val = sk->sk_no_check_tx;
1378 v.val = sk->sk_priority;
1382 lv = sizeof(v.ling);
1383 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1384 v.ling.l_linger = sk->sk_lingertime / HZ;
1390 case SO_TIMESTAMP_OLD:
1391 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1392 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1393 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1396 case SO_TIMESTAMPNS_OLD:
1397 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1400 case SO_TIMESTAMP_NEW:
1401 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1404 case SO_TIMESTAMPNS_NEW:
1405 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1408 case SO_TIMESTAMPING_OLD:
1409 v.val = sk->sk_tsflags;
1412 case SO_RCVTIMEO_OLD:
1413 case SO_RCVTIMEO_NEW:
1414 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1417 case SO_SNDTIMEO_OLD:
1418 case SO_SNDTIMEO_NEW:
1419 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1423 v.val = sk->sk_rcvlowat;
1431 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1436 struct ucred peercred;
1437 if (len > sizeof(peercred))
1438 len = sizeof(peercred);
1439 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1440 if (copy_to_user(optval, &peercred, len))
1449 if (!sk->sk_peer_cred)
1452 n = sk->sk_peer_cred->group_info->ngroups;
1453 if (len < n * sizeof(gid_t)) {
1454 len = n * sizeof(gid_t);
1455 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1457 len = n * sizeof(gid_t);
1459 ret = groups_to_user((gid_t __user *)optval,
1460 sk->sk_peer_cred->group_info);
1470 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1475 if (copy_to_user(optval, address, len))
1480 /* Dubious BSD thing... Probably nobody even uses it, but
1481 * the UNIX standard wants it for whatever reason... -DaveM
1484 v.val = sk->sk_state == TCP_LISTEN;
1488 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1492 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1495 v.val = sk->sk_mark;
1499 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1502 case SO_WIFI_STATUS:
1503 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1507 if (!sock->ops->set_peek_off)
1510 v.val = sk->sk_peek_off;
1513 v.val = sock_flag(sk, SOCK_NOFCS);
1516 case SO_BINDTODEVICE:
1517 return sock_getbindtodevice(sk, optval, optlen, len);
1520 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1526 case SO_LOCK_FILTER:
1527 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1530 case SO_BPF_EXTENSIONS:
1531 v.val = bpf_tell_extensions();
1534 case SO_SELECT_ERR_QUEUE:
1535 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1538 #ifdef CONFIG_NET_RX_BUSY_POLL
1540 v.val = sk->sk_ll_usec;
1542 case SO_PREFER_BUSY_POLL:
1543 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1547 case SO_MAX_PACING_RATE:
1548 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1549 lv = sizeof(v.ulval);
1550 v.ulval = sk->sk_max_pacing_rate;
1553 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1557 case SO_INCOMING_CPU:
1558 v.val = READ_ONCE(sk->sk_incoming_cpu);
1563 u32 meminfo[SK_MEMINFO_VARS];
1565 sk_get_meminfo(sk, meminfo);
1567 len = min_t(unsigned int, len, sizeof(meminfo));
1568 if (copy_to_user(optval, &meminfo, len))
1574 #ifdef CONFIG_NET_RX_BUSY_POLL
1575 case SO_INCOMING_NAPI_ID:
1576 v.val = READ_ONCE(sk->sk_napi_id);
1578 /* aggregate non-NAPI IDs down to 0 */
1579 if (v.val < MIN_NAPI_ID)
1589 v.val64 = sock_gen_cookie(sk);
1593 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1597 lv = sizeof(v.txtime);
1598 v.txtime.clockid = sk->sk_clockid;
1599 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1600 SOF_TXTIME_DEADLINE_MODE : 0;
1601 v.txtime.flags |= sk->sk_txtime_report_errors ?
1602 SOF_TXTIME_REPORT_ERRORS : 0;
1605 case SO_BINDTOIFINDEX:
1606 v.val = sk->sk_bound_dev_if;
1610 /* We implement the SO_SNDLOWAT etc to not be settable
1613 return -ENOPROTOOPT;
1618 if (copy_to_user(optval, &v, len))
1621 if (put_user(len, optlen))
1627 * Initialize an sk_lock.
1629 * (We also register the sk_lock with the lock validator.)
1631 static inline void sock_lock_init(struct sock *sk)
1633 if (sk->sk_kern_sock)
1634 sock_lock_init_class_and_name(
1636 af_family_kern_slock_key_strings[sk->sk_family],
1637 af_family_kern_slock_keys + sk->sk_family,
1638 af_family_kern_key_strings[sk->sk_family],
1639 af_family_kern_keys + sk->sk_family);
1641 sock_lock_init_class_and_name(
1643 af_family_slock_key_strings[sk->sk_family],
1644 af_family_slock_keys + sk->sk_family,
1645 af_family_key_strings[sk->sk_family],
1646 af_family_keys + sk->sk_family);
1650 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1651 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1652 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1654 static void sock_copy(struct sock *nsk, const struct sock *osk)
1656 const struct proto *prot = READ_ONCE(osk->sk_prot);
1657 #ifdef CONFIG_SECURITY_NETWORK
1658 void *sptr = nsk->sk_security;
1661 /* If we move sk_tx_queue_mapping out of the private section,
1662 * we must check if sk_tx_queue_clear() is called after
1663 * sock_copy() in sk_clone_lock().
1665 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1666 offsetof(struct sock, sk_dontcopy_begin) ||
1667 offsetof(struct sock, sk_tx_queue_mapping) >=
1668 offsetof(struct sock, sk_dontcopy_end));
1670 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1672 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1673 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1675 #ifdef CONFIG_SECURITY_NETWORK
1676 nsk->sk_security = sptr;
1677 security_sk_clone(osk, nsk);
1681 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1685 struct kmem_cache *slab;
1689 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1692 if (want_init_on_alloc(priority))
1693 sk_prot_clear_nulls(sk, prot->obj_size);
1695 sk = kmalloc(prot->obj_size, priority);
1698 if (security_sk_alloc(sk, family, priority))
1701 if (!try_module_get(prot->owner))
1708 security_sk_free(sk);
1711 kmem_cache_free(slab, sk);
1717 static void sk_prot_free(struct proto *prot, struct sock *sk)
1719 struct kmem_cache *slab;
1720 struct module *owner;
1722 owner = prot->owner;
1725 cgroup_sk_free(&sk->sk_cgrp_data);
1726 mem_cgroup_sk_free(sk);
1727 security_sk_free(sk);
1729 kmem_cache_free(slab, sk);
1736 * sk_alloc - All socket objects are allocated here
1737 * @net: the applicable net namespace
1738 * @family: protocol family
1739 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1740 * @prot: struct proto associated with this new sock instance
1741 * @kern: is this to be a kernel socket?
1743 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1744 struct proto *prot, int kern)
1748 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1750 sk->sk_family = family;
1752 * See comment in struct sock definition to understand
1753 * why we need sk_prot_creator -acme
1755 sk->sk_prot = sk->sk_prot_creator = prot;
1756 sk->sk_kern_sock = kern;
1758 sk->sk_net_refcnt = kern ? 0 : 1;
1759 if (likely(sk->sk_net_refcnt)) {
1761 sock_inuse_add(net, 1);
1764 sock_net_set(sk, net);
1765 refcount_set(&sk->sk_wmem_alloc, 1);
1767 mem_cgroup_sk_alloc(sk);
1768 cgroup_sk_alloc(&sk->sk_cgrp_data);
1769 sock_update_classid(&sk->sk_cgrp_data);
1770 sock_update_netprioidx(&sk->sk_cgrp_data);
1771 sk_tx_queue_clear(sk);
1776 EXPORT_SYMBOL(sk_alloc);
1778 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1779 * grace period. This is the case for UDP sockets and TCP listeners.
1781 static void __sk_destruct(struct rcu_head *head)
1783 struct sock *sk = container_of(head, struct sock, sk_rcu);
1784 struct sk_filter *filter;
1786 if (sk->sk_destruct)
1787 sk->sk_destruct(sk);
1789 filter = rcu_dereference_check(sk->sk_filter,
1790 refcount_read(&sk->sk_wmem_alloc) == 0);
1792 sk_filter_uncharge(sk, filter);
1793 RCU_INIT_POINTER(sk->sk_filter, NULL);
1796 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1798 #ifdef CONFIG_BPF_SYSCALL
1799 bpf_sk_storage_free(sk);
1802 if (atomic_read(&sk->sk_omem_alloc))
1803 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1804 __func__, atomic_read(&sk->sk_omem_alloc));
1806 if (sk->sk_frag.page) {
1807 put_page(sk->sk_frag.page);
1808 sk->sk_frag.page = NULL;
1811 if (sk->sk_peer_cred)
1812 put_cred(sk->sk_peer_cred);
1813 put_pid(sk->sk_peer_pid);
1814 if (likely(sk->sk_net_refcnt))
1815 put_net(sock_net(sk));
1816 sk_prot_free(sk->sk_prot_creator, sk);
1819 void sk_destruct(struct sock *sk)
1821 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1823 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1824 reuseport_detach_sock(sk);
1825 use_call_rcu = true;
1829 call_rcu(&sk->sk_rcu, __sk_destruct);
1831 __sk_destruct(&sk->sk_rcu);
1834 static void __sk_free(struct sock *sk)
1836 if (likely(sk->sk_net_refcnt))
1837 sock_inuse_add(sock_net(sk), -1);
1839 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1840 sock_diag_broadcast_destroy(sk);
1845 void sk_free(struct sock *sk)
1848 * We subtract one from sk_wmem_alloc and can know if
1849 * some packets are still in some tx queue.
1850 * If not null, sock_wfree() will call __sk_free(sk) later
1852 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1855 EXPORT_SYMBOL(sk_free);
1857 static void sk_init_common(struct sock *sk)
1859 skb_queue_head_init(&sk->sk_receive_queue);
1860 skb_queue_head_init(&sk->sk_write_queue);
1861 skb_queue_head_init(&sk->sk_error_queue);
1863 rwlock_init(&sk->sk_callback_lock);
1864 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1865 af_rlock_keys + sk->sk_family,
1866 af_family_rlock_key_strings[sk->sk_family]);
1867 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1868 af_wlock_keys + sk->sk_family,
1869 af_family_wlock_key_strings[sk->sk_family]);
1870 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1871 af_elock_keys + sk->sk_family,
1872 af_family_elock_key_strings[sk->sk_family]);
1873 lockdep_set_class_and_name(&sk->sk_callback_lock,
1874 af_callback_keys + sk->sk_family,
1875 af_family_clock_key_strings[sk->sk_family]);
1879 * sk_clone_lock - clone a socket, and lock its clone
1880 * @sk: the socket to clone
1881 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1883 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1885 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1887 struct proto *prot = READ_ONCE(sk->sk_prot);
1888 struct sk_filter *filter;
1889 bool is_charged = true;
1892 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
1896 sock_copy(newsk, sk);
1898 newsk->sk_prot_creator = prot;
1901 if (likely(newsk->sk_net_refcnt))
1902 get_net(sock_net(newsk));
1903 sk_node_init(&newsk->sk_node);
1904 sock_lock_init(newsk);
1905 bh_lock_sock(newsk);
1906 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1907 newsk->sk_backlog.len = 0;
1909 atomic_set(&newsk->sk_rmem_alloc, 0);
1911 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
1912 refcount_set(&newsk->sk_wmem_alloc, 1);
1914 atomic_set(&newsk->sk_omem_alloc, 0);
1915 sk_init_common(newsk);
1917 newsk->sk_dst_cache = NULL;
1918 newsk->sk_dst_pending_confirm = 0;
1919 newsk->sk_wmem_queued = 0;
1920 newsk->sk_forward_alloc = 0;
1921 atomic_set(&newsk->sk_drops, 0);
1922 newsk->sk_send_head = NULL;
1923 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1924 atomic_set(&newsk->sk_zckey, 0);
1926 sock_reset_flag(newsk, SOCK_DONE);
1928 /* sk->sk_memcg will be populated at accept() time */
1929 newsk->sk_memcg = NULL;
1931 cgroup_sk_clone(&newsk->sk_cgrp_data);
1934 filter = rcu_dereference(sk->sk_filter);
1936 /* though it's an empty new sock, the charging may fail
1937 * if sysctl_optmem_max was changed between creation of
1938 * original socket and cloning
1940 is_charged = sk_filter_charge(newsk, filter);
1941 RCU_INIT_POINTER(newsk->sk_filter, filter);
1944 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1945 /* We need to make sure that we don't uncharge the new
1946 * socket if we couldn't charge it in the first place
1947 * as otherwise we uncharge the parent's filter.
1950 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1951 sk_free_unlock_clone(newsk);
1955 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1957 if (bpf_sk_storage_clone(sk, newsk)) {
1958 sk_free_unlock_clone(newsk);
1963 /* Clear sk_user_data if parent had the pointer tagged
1964 * as not suitable for copying when cloning.
1966 if (sk_user_data_is_nocopy(newsk))
1967 newsk->sk_user_data = NULL;
1970 newsk->sk_err_soft = 0;
1971 newsk->sk_priority = 0;
1972 newsk->sk_incoming_cpu = raw_smp_processor_id();
1973 if (likely(newsk->sk_net_refcnt))
1974 sock_inuse_add(sock_net(newsk), 1);
1976 /* Before updating sk_refcnt, we must commit prior changes to memory
1977 * (Documentation/RCU/rculist_nulls.rst for details)
1980 refcount_set(&newsk->sk_refcnt, 2);
1982 /* Increment the counter in the same struct proto as the master
1983 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1984 * is the same as sk->sk_prot->socks, as this field was copied
1987 * This _changes_ the previous behaviour, where
1988 * tcp_create_openreq_child always was incrementing the
1989 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1990 * to be taken into account in all callers. -acme
1992 sk_refcnt_debug_inc(newsk);
1993 sk_set_socket(newsk, NULL);
1994 sk_tx_queue_clear(newsk);
1995 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1997 if (newsk->sk_prot->sockets_allocated)
1998 sk_sockets_allocated_inc(newsk);
2000 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2001 net_enable_timestamp();
2005 EXPORT_SYMBOL_GPL(sk_clone_lock);
2007 void sk_free_unlock_clone(struct sock *sk)
2009 /* It is still raw copy of parent, so invalidate
2010 * destructor and make plain sk_free() */
2011 sk->sk_destruct = NULL;
2015 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2017 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2021 sk_dst_set(sk, dst);
2022 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2023 if (sk->sk_route_caps & NETIF_F_GSO)
2024 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2025 sk->sk_route_caps &= ~sk->sk_route_nocaps;
2026 if (sk_can_gso(sk)) {
2027 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2028 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2030 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2031 sk->sk_gso_max_size = dst->dev->gso_max_size;
2032 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2035 sk->sk_gso_max_segs = max_segs;
2037 EXPORT_SYMBOL_GPL(sk_setup_caps);
2040 * Simple resource managers for sockets.
2045 * Write buffer destructor automatically called from kfree_skb.
2047 void sock_wfree(struct sk_buff *skb)
2049 struct sock *sk = skb->sk;
2050 unsigned int len = skb->truesize;
2052 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2054 * Keep a reference on sk_wmem_alloc, this will be released
2055 * after sk_write_space() call
2057 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2058 sk->sk_write_space(sk);
2062 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2063 * could not do because of in-flight packets
2065 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2068 EXPORT_SYMBOL(sock_wfree);
2070 /* This variant of sock_wfree() is used by TCP,
2071 * since it sets SOCK_USE_WRITE_QUEUE.
2073 void __sock_wfree(struct sk_buff *skb)
2075 struct sock *sk = skb->sk;
2077 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2081 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2086 if (unlikely(!sk_fullsock(sk))) {
2087 skb->destructor = sock_edemux;
2092 skb->destructor = sock_wfree;
2093 skb_set_hash_from_sk(skb, sk);
2095 * We used to take a refcount on sk, but following operation
2096 * is enough to guarantee sk_free() wont free this sock until
2097 * all in-flight packets are completed
2099 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2101 EXPORT_SYMBOL(skb_set_owner_w);
2103 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2105 #ifdef CONFIG_TLS_DEVICE
2106 /* Drivers depend on in-order delivery for crypto offload,
2107 * partial orphan breaks out-of-order-OK logic.
2112 return (skb->destructor == sock_wfree ||
2113 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2116 /* This helper is used by netem, as it can hold packets in its
2117 * delay queue. We want to allow the owner socket to send more
2118 * packets, as if they were already TX completed by a typical driver.
2119 * But we also want to keep skb->sk set because some packet schedulers
2120 * rely on it (sch_fq for example).
2122 void skb_orphan_partial(struct sk_buff *skb)
2124 if (skb_is_tcp_pure_ack(skb))
2127 if (can_skb_orphan_partial(skb)) {
2128 struct sock *sk = skb->sk;
2130 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2131 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2132 skb->destructor = sock_efree;
2138 EXPORT_SYMBOL(skb_orphan_partial);
2141 * Read buffer destructor automatically called from kfree_skb.
2143 void sock_rfree(struct sk_buff *skb)
2145 struct sock *sk = skb->sk;
2146 unsigned int len = skb->truesize;
2148 atomic_sub(len, &sk->sk_rmem_alloc);
2149 sk_mem_uncharge(sk, len);
2151 EXPORT_SYMBOL(sock_rfree);
2154 * Buffer destructor for skbs that are not used directly in read or write
2155 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2157 void sock_efree(struct sk_buff *skb)
2161 EXPORT_SYMBOL(sock_efree);
2163 /* Buffer destructor for prefetch/receive path where reference count may
2164 * not be held, e.g. for listen sockets.
2167 void sock_pfree(struct sk_buff *skb)
2169 if (sk_is_refcounted(skb->sk))
2170 sock_gen_put(skb->sk);
2172 EXPORT_SYMBOL(sock_pfree);
2173 #endif /* CONFIG_INET */
2175 kuid_t sock_i_uid(struct sock *sk)
2179 read_lock_bh(&sk->sk_callback_lock);
2180 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2181 read_unlock_bh(&sk->sk_callback_lock);
2184 EXPORT_SYMBOL(sock_i_uid);
2186 unsigned long sock_i_ino(struct sock *sk)
2190 read_lock_bh(&sk->sk_callback_lock);
2191 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2192 read_unlock_bh(&sk->sk_callback_lock);
2195 EXPORT_SYMBOL(sock_i_ino);
2198 * Allocate a skb from the socket's send buffer.
2200 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2204 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2205 struct sk_buff *skb = alloc_skb(size, priority);
2208 skb_set_owner_w(skb, sk);
2214 EXPORT_SYMBOL(sock_wmalloc);
2216 static void sock_ofree(struct sk_buff *skb)
2218 struct sock *sk = skb->sk;
2220 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2223 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2226 struct sk_buff *skb;
2228 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2229 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2233 skb = alloc_skb(size, priority);
2237 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2239 skb->destructor = sock_ofree;
2244 * Allocate a memory block from the socket's option memory buffer.
2246 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2248 if ((unsigned int)size <= sysctl_optmem_max &&
2249 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2251 /* First do the add, to avoid the race if kmalloc
2254 atomic_add(size, &sk->sk_omem_alloc);
2255 mem = kmalloc(size, priority);
2258 atomic_sub(size, &sk->sk_omem_alloc);
2262 EXPORT_SYMBOL(sock_kmalloc);
2264 /* Free an option memory block. Note, we actually want the inline
2265 * here as this allows gcc to detect the nullify and fold away the
2266 * condition entirely.
2268 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2271 if (WARN_ON_ONCE(!mem))
2274 kfree_sensitive(mem);
2277 atomic_sub(size, &sk->sk_omem_alloc);
2280 void sock_kfree_s(struct sock *sk, void *mem, int size)
2282 __sock_kfree_s(sk, mem, size, false);
2284 EXPORT_SYMBOL(sock_kfree_s);
2286 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2288 __sock_kfree_s(sk, mem, size, true);
2290 EXPORT_SYMBOL(sock_kzfree_s);
2292 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2293 I think, these locks should be removed for datagram sockets.
2295 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2299 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2303 if (signal_pending(current))
2305 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2306 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2307 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2309 if (sk->sk_shutdown & SEND_SHUTDOWN)
2313 timeo = schedule_timeout(timeo);
2315 finish_wait(sk_sleep(sk), &wait);
2321 * Generic send/receive buffer handlers
2324 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2325 unsigned long data_len, int noblock,
2326 int *errcode, int max_page_order)
2328 struct sk_buff *skb;
2332 timeo = sock_sndtimeo(sk, noblock);
2334 err = sock_error(sk);
2339 if (sk->sk_shutdown & SEND_SHUTDOWN)
2342 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2345 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2346 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2350 if (signal_pending(current))
2352 timeo = sock_wait_for_wmem(sk, timeo);
2354 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2355 errcode, sk->sk_allocation);
2357 skb_set_owner_w(skb, sk);
2361 err = sock_intr_errno(timeo);
2366 EXPORT_SYMBOL(sock_alloc_send_pskb);
2368 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2369 int noblock, int *errcode)
2371 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2373 EXPORT_SYMBOL(sock_alloc_send_skb);
2375 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2376 struct sockcm_cookie *sockc)
2380 switch (cmsg->cmsg_type) {
2382 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2384 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2386 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2388 case SO_TIMESTAMPING_OLD:
2389 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2392 tsflags = *(u32 *)CMSG_DATA(cmsg);
2393 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2396 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2397 sockc->tsflags |= tsflags;
2400 if (!sock_flag(sk, SOCK_TXTIME))
2402 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2404 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2406 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2408 case SCM_CREDENTIALS:
2415 EXPORT_SYMBOL(__sock_cmsg_send);
2417 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2418 struct sockcm_cookie *sockc)
2420 struct cmsghdr *cmsg;
2423 for_each_cmsghdr(cmsg, msg) {
2424 if (!CMSG_OK(msg, cmsg))
2426 if (cmsg->cmsg_level != SOL_SOCKET)
2428 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2434 EXPORT_SYMBOL(sock_cmsg_send);
2436 static void sk_enter_memory_pressure(struct sock *sk)
2438 if (!sk->sk_prot->enter_memory_pressure)
2441 sk->sk_prot->enter_memory_pressure(sk);
2444 static void sk_leave_memory_pressure(struct sock *sk)
2446 if (sk->sk_prot->leave_memory_pressure) {
2447 sk->sk_prot->leave_memory_pressure(sk);
2449 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2451 if (memory_pressure && READ_ONCE(*memory_pressure))
2452 WRITE_ONCE(*memory_pressure, 0);
2456 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2457 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2460 * skb_page_frag_refill - check that a page_frag contains enough room
2461 * @sz: minimum size of the fragment we want to get
2462 * @pfrag: pointer to page_frag
2463 * @gfp: priority for memory allocation
2465 * Note: While this allocator tries to use high order pages, there is
2466 * no guarantee that allocations succeed. Therefore, @sz MUST be
2467 * less or equal than PAGE_SIZE.
2469 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2472 if (page_ref_count(pfrag->page) == 1) {
2476 if (pfrag->offset + sz <= pfrag->size)
2478 put_page(pfrag->page);
2482 if (SKB_FRAG_PAGE_ORDER &&
2483 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2484 /* Avoid direct reclaim but allow kswapd to wake */
2485 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2486 __GFP_COMP | __GFP_NOWARN |
2488 SKB_FRAG_PAGE_ORDER);
2489 if (likely(pfrag->page)) {
2490 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2494 pfrag->page = alloc_page(gfp);
2495 if (likely(pfrag->page)) {
2496 pfrag->size = PAGE_SIZE;
2501 EXPORT_SYMBOL(skb_page_frag_refill);
2503 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2505 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2508 sk_enter_memory_pressure(sk);
2509 sk_stream_moderate_sndbuf(sk);
2512 EXPORT_SYMBOL(sk_page_frag_refill);
2514 void __lock_sock(struct sock *sk)
2515 __releases(&sk->sk_lock.slock)
2516 __acquires(&sk->sk_lock.slock)
2521 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2522 TASK_UNINTERRUPTIBLE);
2523 spin_unlock_bh(&sk->sk_lock.slock);
2525 spin_lock_bh(&sk->sk_lock.slock);
2526 if (!sock_owned_by_user(sk))
2529 finish_wait(&sk->sk_lock.wq, &wait);
2532 void __release_sock(struct sock *sk)
2533 __releases(&sk->sk_lock.slock)
2534 __acquires(&sk->sk_lock.slock)
2536 struct sk_buff *skb, *next;
2538 while ((skb = sk->sk_backlog.head) != NULL) {
2539 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2541 spin_unlock_bh(&sk->sk_lock.slock);
2546 WARN_ON_ONCE(skb_dst_is_noref(skb));
2547 skb_mark_not_on_list(skb);
2548 sk_backlog_rcv(sk, skb);
2553 } while (skb != NULL);
2555 spin_lock_bh(&sk->sk_lock.slock);
2559 * Doing the zeroing here guarantee we can not loop forever
2560 * while a wild producer attempts to flood us.
2562 sk->sk_backlog.len = 0;
2565 void __sk_flush_backlog(struct sock *sk)
2567 spin_lock_bh(&sk->sk_lock.slock);
2569 spin_unlock_bh(&sk->sk_lock.slock);
2573 * sk_wait_data - wait for data to arrive at sk_receive_queue
2574 * @sk: sock to wait on
2575 * @timeo: for how long
2576 * @skb: last skb seen on sk_receive_queue
2578 * Now socket state including sk->sk_err is changed only under lock,
2579 * hence we may omit checks after joining wait queue.
2580 * We check receive queue before schedule() only as optimization;
2581 * it is very likely that release_sock() added new data.
2583 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2585 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2588 add_wait_queue(sk_sleep(sk), &wait);
2589 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2590 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2591 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2592 remove_wait_queue(sk_sleep(sk), &wait);
2595 EXPORT_SYMBOL(sk_wait_data);
2598 * __sk_mem_raise_allocated - increase memory_allocated
2600 * @size: memory size to allocate
2601 * @amt: pages to allocate
2602 * @kind: allocation type
2604 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2606 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2608 struct proto *prot = sk->sk_prot;
2609 long allocated = sk_memory_allocated_add(sk, amt);
2610 bool charged = true;
2612 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2613 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2614 goto suppress_allocation;
2617 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2618 sk_leave_memory_pressure(sk);
2622 /* Under pressure. */
2623 if (allocated > sk_prot_mem_limits(sk, 1))
2624 sk_enter_memory_pressure(sk);
2626 /* Over hard limit. */
2627 if (allocated > sk_prot_mem_limits(sk, 2))
2628 goto suppress_allocation;
2630 /* guarantee minimum buffer size under pressure */
2631 if (kind == SK_MEM_RECV) {
2632 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2635 } else { /* SK_MEM_SEND */
2636 int wmem0 = sk_get_wmem0(sk, prot);
2638 if (sk->sk_type == SOCK_STREAM) {
2639 if (sk->sk_wmem_queued < wmem0)
2641 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2646 if (sk_has_memory_pressure(sk)) {
2649 if (!sk_under_memory_pressure(sk))
2651 alloc = sk_sockets_allocated_read_positive(sk);
2652 if (sk_prot_mem_limits(sk, 2) > alloc *
2653 sk_mem_pages(sk->sk_wmem_queued +
2654 atomic_read(&sk->sk_rmem_alloc) +
2655 sk->sk_forward_alloc))
2659 suppress_allocation:
2661 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2662 sk_stream_moderate_sndbuf(sk);
2664 /* Fail only if socket is _under_ its sndbuf.
2665 * In this case we cannot block, so that we have to fail.
2667 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2671 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2672 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2674 sk_memory_allocated_sub(sk, amt);
2676 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2677 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2681 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2684 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2686 * @size: memory size to allocate
2687 * @kind: allocation type
2689 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2690 * rmem allocation. This function assumes that protocols which have
2691 * memory_pressure use sk_wmem_queued as write buffer accounting.
2693 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2695 int ret, amt = sk_mem_pages(size);
2697 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2698 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2700 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2703 EXPORT_SYMBOL(__sk_mem_schedule);
2706 * __sk_mem_reduce_allocated - reclaim memory_allocated
2708 * @amount: number of quanta
2710 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2712 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2714 sk_memory_allocated_sub(sk, amount);
2716 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2717 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2719 if (sk_under_memory_pressure(sk) &&
2720 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2721 sk_leave_memory_pressure(sk);
2723 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2726 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2728 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2730 void __sk_mem_reclaim(struct sock *sk, int amount)
2732 amount >>= SK_MEM_QUANTUM_SHIFT;
2733 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2734 __sk_mem_reduce_allocated(sk, amount);
2736 EXPORT_SYMBOL(__sk_mem_reclaim);
2738 int sk_set_peek_off(struct sock *sk, int val)
2740 sk->sk_peek_off = val;
2743 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2746 * Set of default routines for initialising struct proto_ops when
2747 * the protocol does not support a particular function. In certain
2748 * cases where it makes no sense for a protocol to have a "do nothing"
2749 * function, some default processing is provided.
2752 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2756 EXPORT_SYMBOL(sock_no_bind);
2758 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2763 EXPORT_SYMBOL(sock_no_connect);
2765 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2769 EXPORT_SYMBOL(sock_no_socketpair);
2771 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2776 EXPORT_SYMBOL(sock_no_accept);
2778 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2783 EXPORT_SYMBOL(sock_no_getname);
2785 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2789 EXPORT_SYMBOL(sock_no_ioctl);
2791 int sock_no_listen(struct socket *sock, int backlog)
2795 EXPORT_SYMBOL(sock_no_listen);
2797 int sock_no_shutdown(struct socket *sock, int how)
2801 EXPORT_SYMBOL(sock_no_shutdown);
2803 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2807 EXPORT_SYMBOL(sock_no_sendmsg);
2809 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2813 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2815 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2820 EXPORT_SYMBOL(sock_no_recvmsg);
2822 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2824 /* Mirror missing mmap method error code */
2827 EXPORT_SYMBOL(sock_no_mmap);
2830 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2831 * various sock-based usage counts.
2833 void __receive_sock(struct file *file)
2835 struct socket *sock;
2837 sock = sock_from_file(file);
2839 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2840 sock_update_classid(&sock->sk->sk_cgrp_data);
2844 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2847 struct msghdr msg = {.msg_flags = flags};
2849 char *kaddr = kmap(page);
2850 iov.iov_base = kaddr + offset;
2852 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2856 EXPORT_SYMBOL(sock_no_sendpage);
2858 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2859 int offset, size_t size, int flags)
2862 struct msghdr msg = {.msg_flags = flags};
2864 char *kaddr = kmap(page);
2866 iov.iov_base = kaddr + offset;
2868 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2872 EXPORT_SYMBOL(sock_no_sendpage_locked);
2875 * Default Socket Callbacks
2878 static void sock_def_wakeup(struct sock *sk)
2880 struct socket_wq *wq;
2883 wq = rcu_dereference(sk->sk_wq);
2884 if (skwq_has_sleeper(wq))
2885 wake_up_interruptible_all(&wq->wait);
2889 static void sock_def_error_report(struct sock *sk)
2891 struct socket_wq *wq;
2894 wq = rcu_dereference(sk->sk_wq);
2895 if (skwq_has_sleeper(wq))
2896 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2897 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2901 void sock_def_readable(struct sock *sk)
2903 struct socket_wq *wq;
2906 wq = rcu_dereference(sk->sk_wq);
2907 if (skwq_has_sleeper(wq))
2908 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2909 EPOLLRDNORM | EPOLLRDBAND);
2910 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2914 static void sock_def_write_space(struct sock *sk)
2916 struct socket_wq *wq;
2920 /* Do not wake up a writer until he can make "significant"
2923 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2924 wq = rcu_dereference(sk->sk_wq);
2925 if (skwq_has_sleeper(wq))
2926 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2927 EPOLLWRNORM | EPOLLWRBAND);
2929 /* Should agree with poll, otherwise some programs break */
2930 if (sock_writeable(sk))
2931 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2937 static void sock_def_destruct(struct sock *sk)
2941 void sk_send_sigurg(struct sock *sk)
2943 if (sk->sk_socket && sk->sk_socket->file)
2944 if (send_sigurg(&sk->sk_socket->file->f_owner))
2945 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2947 EXPORT_SYMBOL(sk_send_sigurg);
2949 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2950 unsigned long expires)
2952 if (!mod_timer(timer, expires))
2955 EXPORT_SYMBOL(sk_reset_timer);
2957 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2959 if (del_timer(timer))
2962 EXPORT_SYMBOL(sk_stop_timer);
2964 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
2966 if (del_timer_sync(timer))
2969 EXPORT_SYMBOL(sk_stop_timer_sync);
2971 void sock_init_data(struct socket *sock, struct sock *sk)
2974 sk->sk_send_head = NULL;
2976 timer_setup(&sk->sk_timer, NULL, 0);
2978 sk->sk_allocation = GFP_KERNEL;
2979 sk->sk_rcvbuf = sysctl_rmem_default;
2980 sk->sk_sndbuf = sysctl_wmem_default;
2981 sk->sk_state = TCP_CLOSE;
2982 sk_set_socket(sk, sock);
2984 sock_set_flag(sk, SOCK_ZAPPED);
2987 sk->sk_type = sock->type;
2988 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2990 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2992 RCU_INIT_POINTER(sk->sk_wq, NULL);
2993 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2996 rwlock_init(&sk->sk_callback_lock);
2997 if (sk->sk_kern_sock)
2998 lockdep_set_class_and_name(
2999 &sk->sk_callback_lock,
3000 af_kern_callback_keys + sk->sk_family,
3001 af_family_kern_clock_key_strings[sk->sk_family]);
3003 lockdep_set_class_and_name(
3004 &sk->sk_callback_lock,
3005 af_callback_keys + sk->sk_family,
3006 af_family_clock_key_strings[sk->sk_family]);
3008 sk->sk_state_change = sock_def_wakeup;
3009 sk->sk_data_ready = sock_def_readable;
3010 sk->sk_write_space = sock_def_write_space;
3011 sk->sk_error_report = sock_def_error_report;
3012 sk->sk_destruct = sock_def_destruct;
3014 sk->sk_frag.page = NULL;
3015 sk->sk_frag.offset = 0;
3016 sk->sk_peek_off = -1;
3018 sk->sk_peer_pid = NULL;
3019 sk->sk_peer_cred = NULL;
3020 sk->sk_write_pending = 0;
3021 sk->sk_rcvlowat = 1;
3022 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3023 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3025 sk->sk_stamp = SK_DEFAULT_STAMP;
3026 #if BITS_PER_LONG==32
3027 seqlock_init(&sk->sk_stamp_seq);
3029 atomic_set(&sk->sk_zckey, 0);
3031 #ifdef CONFIG_NET_RX_BUSY_POLL
3033 sk->sk_ll_usec = sysctl_net_busy_read;
3036 sk->sk_max_pacing_rate = ~0UL;
3037 sk->sk_pacing_rate = ~0UL;
3038 WRITE_ONCE(sk->sk_pacing_shift, 10);
3039 sk->sk_incoming_cpu = -1;
3041 sk_rx_queue_clear(sk);
3043 * Before updating sk_refcnt, we must commit prior changes to memory
3044 * (Documentation/RCU/rculist_nulls.rst for details)
3047 refcount_set(&sk->sk_refcnt, 1);
3048 atomic_set(&sk->sk_drops, 0);
3050 EXPORT_SYMBOL(sock_init_data);
3052 void lock_sock_nested(struct sock *sk, int subclass)
3055 spin_lock_bh(&sk->sk_lock.slock);
3056 if (sk->sk_lock.owned)
3058 sk->sk_lock.owned = 1;
3059 spin_unlock(&sk->sk_lock.slock);
3061 * The sk_lock has mutex_lock() semantics here:
3063 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3066 EXPORT_SYMBOL(lock_sock_nested);
3068 void release_sock(struct sock *sk)
3070 spin_lock_bh(&sk->sk_lock.slock);
3071 if (sk->sk_backlog.tail)
3074 /* Warning : release_cb() might need to release sk ownership,
3075 * ie call sock_release_ownership(sk) before us.
3077 if (sk->sk_prot->release_cb)
3078 sk->sk_prot->release_cb(sk);
3080 sock_release_ownership(sk);
3081 if (waitqueue_active(&sk->sk_lock.wq))
3082 wake_up(&sk->sk_lock.wq);
3083 spin_unlock_bh(&sk->sk_lock.slock);
3085 EXPORT_SYMBOL(release_sock);
3088 * lock_sock_fast - fast version of lock_sock
3091 * This version should be used for very small section, where process wont block
3092 * return false if fast path is taken:
3094 * sk_lock.slock locked, owned = 0, BH disabled
3096 * return true if slow path is taken:
3098 * sk_lock.slock unlocked, owned = 1, BH enabled
3100 bool lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3103 spin_lock_bh(&sk->sk_lock.slock);
3105 if (!sk->sk_lock.owned)
3107 * Note : We must disable BH
3112 sk->sk_lock.owned = 1;
3113 spin_unlock(&sk->sk_lock.slock);
3115 * The sk_lock has mutex_lock() semantics here:
3117 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3118 __acquire(&sk->sk_lock.slock);
3122 EXPORT_SYMBOL(lock_sock_fast);
3124 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3125 bool timeval, bool time32)
3127 struct sock *sk = sock->sk;
3128 struct timespec64 ts;
3130 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3131 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3132 if (ts.tv_sec == -1)
3134 if (ts.tv_sec == 0) {
3135 ktime_t kt = ktime_get_real();
3136 sock_write_timestamp(sk, kt);
3137 ts = ktime_to_timespec64(kt);
3143 #ifdef CONFIG_COMPAT_32BIT_TIME
3145 return put_old_timespec32(&ts, userstamp);
3147 #ifdef CONFIG_SPARC64
3148 /* beware of padding in sparc64 timeval */
3149 if (timeval && !in_compat_syscall()) {
3150 struct __kernel_old_timeval __user tv = {
3151 .tv_sec = ts.tv_sec,
3152 .tv_usec = ts.tv_nsec,
3154 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3159 return put_timespec64(&ts, userstamp);
3161 EXPORT_SYMBOL(sock_gettstamp);
3163 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3165 if (!sock_flag(sk, flag)) {
3166 unsigned long previous_flags = sk->sk_flags;
3168 sock_set_flag(sk, flag);
3170 * we just set one of the two flags which require net
3171 * time stamping, but time stamping might have been on
3172 * already because of the other one
3174 if (sock_needs_netstamp(sk) &&
3175 !(previous_flags & SK_FLAGS_TIMESTAMP))
3176 net_enable_timestamp();
3180 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3181 int level, int type)
3183 struct sock_exterr_skb *serr;
3184 struct sk_buff *skb;
3188 skb = sock_dequeue_err_skb(sk);
3194 msg->msg_flags |= MSG_TRUNC;
3197 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3201 sock_recv_timestamp(msg, sk, skb);
3203 serr = SKB_EXT_ERR(skb);
3204 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3206 msg->msg_flags |= MSG_ERRQUEUE;
3214 EXPORT_SYMBOL(sock_recv_errqueue);
3217 * Get a socket option on an socket.
3219 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3220 * asynchronous errors should be reported by getsockopt. We assume
3221 * this means if you specify SO_ERROR (otherwise whats the point of it).
3223 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3224 char __user *optval, int __user *optlen)
3226 struct sock *sk = sock->sk;
3228 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3230 EXPORT_SYMBOL(sock_common_getsockopt);
3232 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3235 struct sock *sk = sock->sk;
3239 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3240 flags & ~MSG_DONTWAIT, &addr_len);
3242 msg->msg_namelen = addr_len;
3245 EXPORT_SYMBOL(sock_common_recvmsg);
3248 * Set socket options on an inet socket.
3250 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3251 sockptr_t optval, unsigned int optlen)
3253 struct sock *sk = sock->sk;
3255 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3257 EXPORT_SYMBOL(sock_common_setsockopt);
3259 void sk_common_release(struct sock *sk)
3261 if (sk->sk_prot->destroy)
3262 sk->sk_prot->destroy(sk);
3265 * Observation: when sk_common_release is called, processes have
3266 * no access to socket. But net still has.
3267 * Step one, detach it from networking:
3269 * A. Remove from hash tables.
3272 sk->sk_prot->unhash(sk);
3275 * In this point socket cannot receive new packets, but it is possible
3276 * that some packets are in flight because some CPU runs receiver and
3277 * did hash table lookup before we unhashed socket. They will achieve
3278 * receive queue and will be purged by socket destructor.
3280 * Also we still have packets pending on receive queue and probably,
3281 * our own packets waiting in device queues. sock_destroy will drain
3282 * receive queue, but transmitted packets will delay socket destruction
3283 * until the last reference will be released.
3288 xfrm_sk_free_policy(sk);
3290 sk_refcnt_debug_release(sk);
3294 EXPORT_SYMBOL(sk_common_release);
3296 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3298 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3300 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3301 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3302 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3303 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3304 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3305 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3306 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3307 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3308 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3311 #ifdef CONFIG_PROC_FS
3312 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3314 int val[PROTO_INUSE_NR];
3317 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3319 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3321 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3323 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3325 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3327 int cpu, idx = prot->inuse_idx;
3330 for_each_possible_cpu(cpu)
3331 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3333 return res >= 0 ? res : 0;
3335 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3337 static void sock_inuse_add(struct net *net, int val)
3339 this_cpu_add(*net->core.sock_inuse, val);
3342 int sock_inuse_get(struct net *net)
3346 for_each_possible_cpu(cpu)
3347 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3352 EXPORT_SYMBOL_GPL(sock_inuse_get);
3354 static int __net_init sock_inuse_init_net(struct net *net)
3356 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3357 if (net->core.prot_inuse == NULL)
3360 net->core.sock_inuse = alloc_percpu(int);
3361 if (net->core.sock_inuse == NULL)
3367 free_percpu(net->core.prot_inuse);
3371 static void __net_exit sock_inuse_exit_net(struct net *net)
3373 free_percpu(net->core.prot_inuse);
3374 free_percpu(net->core.sock_inuse);
3377 static struct pernet_operations net_inuse_ops = {
3378 .init = sock_inuse_init_net,
3379 .exit = sock_inuse_exit_net,
3382 static __init int net_inuse_init(void)
3384 if (register_pernet_subsys(&net_inuse_ops))
3385 panic("Cannot initialize net inuse counters");
3390 core_initcall(net_inuse_init);
3392 static int assign_proto_idx(struct proto *prot)
3394 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3396 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3397 pr_err("PROTO_INUSE_NR exhausted\n");
3401 set_bit(prot->inuse_idx, proto_inuse_idx);
3405 static void release_proto_idx(struct proto *prot)
3407 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3408 clear_bit(prot->inuse_idx, proto_inuse_idx);
3411 static inline int assign_proto_idx(struct proto *prot)
3416 static inline void release_proto_idx(struct proto *prot)
3420 static void sock_inuse_add(struct net *net, int val)
3425 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3429 kfree(twsk_prot->twsk_slab_name);
3430 twsk_prot->twsk_slab_name = NULL;
3431 kmem_cache_destroy(twsk_prot->twsk_slab);
3432 twsk_prot->twsk_slab = NULL;
3435 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3439 kfree(rsk_prot->slab_name);
3440 rsk_prot->slab_name = NULL;
3441 kmem_cache_destroy(rsk_prot->slab);
3442 rsk_prot->slab = NULL;
3445 static int req_prot_init(const struct proto *prot)
3447 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3452 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3454 if (!rsk_prot->slab_name)
3457 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3458 rsk_prot->obj_size, 0,
3459 SLAB_ACCOUNT | prot->slab_flags,
3462 if (!rsk_prot->slab) {
3463 pr_crit("%s: Can't create request sock SLAB cache!\n",
3470 int proto_register(struct proto *prot, int alloc_slab)
3475 prot->slab = kmem_cache_create_usercopy(prot->name,
3477 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3479 prot->useroffset, prot->usersize,
3482 if (prot->slab == NULL) {
3483 pr_crit("%s: Can't create sock SLAB cache!\n",
3488 if (req_prot_init(prot))
3489 goto out_free_request_sock_slab;
3491 if (prot->twsk_prot != NULL) {
3492 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3494 if (prot->twsk_prot->twsk_slab_name == NULL)
3495 goto out_free_request_sock_slab;
3497 prot->twsk_prot->twsk_slab =
3498 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3499 prot->twsk_prot->twsk_obj_size,
3504 if (prot->twsk_prot->twsk_slab == NULL)
3505 goto out_free_timewait_sock_slab;
3509 mutex_lock(&proto_list_mutex);
3510 ret = assign_proto_idx(prot);
3512 mutex_unlock(&proto_list_mutex);
3513 goto out_free_timewait_sock_slab;
3515 list_add(&prot->node, &proto_list);
3516 mutex_unlock(&proto_list_mutex);
3519 out_free_timewait_sock_slab:
3520 if (alloc_slab && prot->twsk_prot)
3521 tw_prot_cleanup(prot->twsk_prot);
3522 out_free_request_sock_slab:
3524 req_prot_cleanup(prot->rsk_prot);
3526 kmem_cache_destroy(prot->slab);
3532 EXPORT_SYMBOL(proto_register);
3534 void proto_unregister(struct proto *prot)
3536 mutex_lock(&proto_list_mutex);
3537 release_proto_idx(prot);
3538 list_del(&prot->node);
3539 mutex_unlock(&proto_list_mutex);
3541 kmem_cache_destroy(prot->slab);
3544 req_prot_cleanup(prot->rsk_prot);
3545 tw_prot_cleanup(prot->twsk_prot);
3547 EXPORT_SYMBOL(proto_unregister);
3549 int sock_load_diag_module(int family, int protocol)
3552 if (!sock_is_registered(family))
3555 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3556 NETLINK_SOCK_DIAG, family);
3560 if (family == AF_INET &&
3561 protocol != IPPROTO_RAW &&
3562 protocol < MAX_INET_PROTOS &&
3563 !rcu_access_pointer(inet_protos[protocol]))
3567 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3568 NETLINK_SOCK_DIAG, family, protocol);
3570 EXPORT_SYMBOL(sock_load_diag_module);
3572 #ifdef CONFIG_PROC_FS
3573 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3574 __acquires(proto_list_mutex)
3576 mutex_lock(&proto_list_mutex);
3577 return seq_list_start_head(&proto_list, *pos);
3580 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3582 return seq_list_next(v, &proto_list, pos);
3585 static void proto_seq_stop(struct seq_file *seq, void *v)
3586 __releases(proto_list_mutex)
3588 mutex_unlock(&proto_list_mutex);
3591 static char proto_method_implemented(const void *method)
3593 return method == NULL ? 'n' : 'y';
3595 static long sock_prot_memory_allocated(struct proto *proto)
3597 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3600 static const char *sock_prot_memory_pressure(struct proto *proto)
3602 return proto->memory_pressure != NULL ?
3603 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3606 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3609 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3610 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3613 sock_prot_inuse_get(seq_file_net(seq), proto),
3614 sock_prot_memory_allocated(proto),
3615 sock_prot_memory_pressure(proto),
3617 proto->slab == NULL ? "no" : "yes",
3618 module_name(proto->owner),
3619 proto_method_implemented(proto->close),
3620 proto_method_implemented(proto->connect),
3621 proto_method_implemented(proto->disconnect),
3622 proto_method_implemented(proto->accept),
3623 proto_method_implemented(proto->ioctl),
3624 proto_method_implemented(proto->init),
3625 proto_method_implemented(proto->destroy),
3626 proto_method_implemented(proto->shutdown),
3627 proto_method_implemented(proto->setsockopt),
3628 proto_method_implemented(proto->getsockopt),
3629 proto_method_implemented(proto->sendmsg),
3630 proto_method_implemented(proto->recvmsg),
3631 proto_method_implemented(proto->sendpage),
3632 proto_method_implemented(proto->bind),
3633 proto_method_implemented(proto->backlog_rcv),
3634 proto_method_implemented(proto->hash),
3635 proto_method_implemented(proto->unhash),
3636 proto_method_implemented(proto->get_port),
3637 proto_method_implemented(proto->enter_memory_pressure));
3640 static int proto_seq_show(struct seq_file *seq, void *v)
3642 if (v == &proto_list)
3643 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3652 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3654 proto_seq_printf(seq, list_entry(v, struct proto, node));
3658 static const struct seq_operations proto_seq_ops = {
3659 .start = proto_seq_start,
3660 .next = proto_seq_next,
3661 .stop = proto_seq_stop,
3662 .show = proto_seq_show,
3665 static __net_init int proto_init_net(struct net *net)
3667 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3668 sizeof(struct seq_net_private)))
3674 static __net_exit void proto_exit_net(struct net *net)
3676 remove_proc_entry("protocols", net->proc_net);
3680 static __net_initdata struct pernet_operations proto_net_ops = {
3681 .init = proto_init_net,
3682 .exit = proto_exit_net,
3685 static int __init proto_init(void)
3687 return register_pernet_subsys(&proto_net_ops);
3690 subsys_initcall(proto_init);
3692 #endif /* PROC_FS */
3694 #ifdef CONFIG_NET_RX_BUSY_POLL
3695 bool sk_busy_loop_end(void *p, unsigned long start_time)
3697 struct sock *sk = p;
3699 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3700 sk_busy_loop_timeout(sk, start_time);
3702 EXPORT_SYMBOL(sk_busy_loop_end);
3703 #endif /* CONFIG_NET_RX_BUSY_POLL */
3705 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3707 if (!sk->sk_prot->bind_add)
3709 return sk->sk_prot->bind_add(sk, addr, addr_len);
3711 EXPORT_SYMBOL(sock_bind_add);