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 #include <linux/ethtool.h>
144 static DEFINE_MUTEX(proto_list_mutex);
145 static LIST_HEAD(proto_list);
147 static void sock_inuse_add(struct net *net, int val);
150 * sk_ns_capable - General socket capability test
151 * @sk: Socket to use a capability on or through
152 * @user_ns: The user namespace of the capability to use
153 * @cap: The capability to use
155 * Test to see if the opener of the socket had when the socket was
156 * created and the current process has the capability @cap in the user
157 * namespace @user_ns.
159 bool sk_ns_capable(const struct sock *sk,
160 struct user_namespace *user_ns, int cap)
162 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
163 ns_capable(user_ns, cap);
165 EXPORT_SYMBOL(sk_ns_capable);
168 * sk_capable - Socket global capability test
169 * @sk: Socket to use a capability on or through
170 * @cap: The global capability to use
172 * Test to see if the opener of the socket had when the socket was
173 * created and the current process has the capability @cap in all user
176 bool sk_capable(const struct sock *sk, int cap)
178 return sk_ns_capable(sk, &init_user_ns, cap);
180 EXPORT_SYMBOL(sk_capable);
183 * sk_net_capable - Network namespace socket capability test
184 * @sk: Socket to use a capability on or through
185 * @cap: The capability to use
187 * Test to see if the opener of the socket had when the socket was created
188 * and the current process has the capability @cap over the network namespace
189 * the socket is a member of.
191 bool sk_net_capable(const struct sock *sk, int cap)
193 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195 EXPORT_SYMBOL(sk_net_capable);
198 * Each address family might have different locking rules, so we have
199 * one slock key per address family and separate keys for internal and
202 static struct lock_class_key af_family_keys[AF_MAX];
203 static struct lock_class_key af_family_kern_keys[AF_MAX];
204 static struct lock_class_key af_family_slock_keys[AF_MAX];
205 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
208 * Make lock validator output more readable. (we pre-construct these
209 * strings build-time, so that runtime initialization of socket
213 #define _sock_locks(x) \
214 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
215 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
216 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
217 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
218 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
219 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
220 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
221 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
222 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
223 x "27" , x "28" , x "AF_CAN" , \
224 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
225 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
226 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
227 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
228 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
232 static const char *const af_family_key_strings[AF_MAX+1] = {
233 _sock_locks("sk_lock-")
235 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
236 _sock_locks("slock-")
238 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
239 _sock_locks("clock-")
242 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
243 _sock_locks("k-sk_lock-")
245 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-slock-")
248 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
249 _sock_locks("k-clock-")
251 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
252 _sock_locks("rlock-")
254 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
255 _sock_locks("wlock-")
257 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
258 _sock_locks("elock-")
262 * sk_callback_lock and sk queues locking rules are per-address-family,
263 * so split the lock classes by using a per-AF key:
265 static struct lock_class_key af_callback_keys[AF_MAX];
266 static struct lock_class_key af_rlock_keys[AF_MAX];
267 static struct lock_class_key af_wlock_keys[AF_MAX];
268 static struct lock_class_key af_elock_keys[AF_MAX];
269 static struct lock_class_key af_kern_callback_keys[AF_MAX];
271 /* Run time adjustable parameters. */
272 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
273 EXPORT_SYMBOL(sysctl_wmem_max);
274 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
275 EXPORT_SYMBOL(sysctl_rmem_max);
276 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
277 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
279 /* Maximal space eaten by iovec or ancillary data plus some space */
280 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
281 EXPORT_SYMBOL(sysctl_optmem_max);
283 int sysctl_tstamp_allow_data __read_mostly = 1;
285 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
286 EXPORT_SYMBOL_GPL(memalloc_socks_key);
289 * sk_set_memalloc - sets %SOCK_MEMALLOC
290 * @sk: socket to set it on
292 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
293 * It's the responsibility of the admin to adjust min_free_kbytes
294 * to meet the requirements
296 void sk_set_memalloc(struct sock *sk)
298 sock_set_flag(sk, SOCK_MEMALLOC);
299 sk->sk_allocation |= __GFP_MEMALLOC;
300 static_branch_inc(&memalloc_socks_key);
302 EXPORT_SYMBOL_GPL(sk_set_memalloc);
304 void sk_clear_memalloc(struct sock *sk)
306 sock_reset_flag(sk, SOCK_MEMALLOC);
307 sk->sk_allocation &= ~__GFP_MEMALLOC;
308 static_branch_dec(&memalloc_socks_key);
311 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
312 * progress of swapping. SOCK_MEMALLOC may be cleared while
313 * it has rmem allocations due to the last swapfile being deactivated
314 * but there is a risk that the socket is unusable due to exceeding
315 * the rmem limits. Reclaim the reserves and obey rmem limits again.
319 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
321 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
324 unsigned int noreclaim_flag;
326 /* these should have been dropped before queueing */
327 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
329 noreclaim_flag = memalloc_noreclaim_save();
330 ret = sk->sk_backlog_rcv(sk, skb);
331 memalloc_noreclaim_restore(noreclaim_flag);
335 EXPORT_SYMBOL(__sk_backlog_rcv);
337 void sk_error_report(struct sock *sk)
339 sk->sk_error_report(sk);
341 switch (sk->sk_family) {
345 trace_inet_sk_error_report(sk);
351 EXPORT_SYMBOL(sk_error_report);
353 static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
355 struct __kernel_sock_timeval tv;
357 if (timeo == MAX_SCHEDULE_TIMEOUT) {
361 tv.tv_sec = timeo / HZ;
362 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
365 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
366 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
367 *(struct old_timeval32 *)optval = tv32;
372 struct __kernel_old_timeval old_tv;
373 old_tv.tv_sec = tv.tv_sec;
374 old_tv.tv_usec = tv.tv_usec;
375 *(struct __kernel_old_timeval *)optval = old_tv;
376 return sizeof(old_tv);
379 *(struct __kernel_sock_timeval *)optval = tv;
383 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
386 struct __kernel_sock_timeval tv;
388 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
389 struct old_timeval32 tv32;
391 if (optlen < sizeof(tv32))
394 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
396 tv.tv_sec = tv32.tv_sec;
397 tv.tv_usec = tv32.tv_usec;
398 } else if (old_timeval) {
399 struct __kernel_old_timeval old_tv;
401 if (optlen < sizeof(old_tv))
403 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
405 tv.tv_sec = old_tv.tv_sec;
406 tv.tv_usec = old_tv.tv_usec;
408 if (optlen < sizeof(tv))
410 if (copy_from_sockptr(&tv, optval, sizeof(tv)))
413 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
417 static int warned __read_mostly;
420 if (warned < 10 && net_ratelimit()) {
422 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
423 __func__, current->comm, task_pid_nr(current));
427 *timeo_p = MAX_SCHEDULE_TIMEOUT;
428 if (tv.tv_sec == 0 && tv.tv_usec == 0)
430 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
431 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
435 static bool sock_needs_netstamp(const struct sock *sk)
437 switch (sk->sk_family) {
446 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
448 if (sk->sk_flags & flags) {
449 sk->sk_flags &= ~flags;
450 if (sock_needs_netstamp(sk) &&
451 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
452 net_disable_timestamp();
457 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
460 struct sk_buff_head *list = &sk->sk_receive_queue;
462 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
463 atomic_inc(&sk->sk_drops);
464 trace_sock_rcvqueue_full(sk, skb);
468 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
469 atomic_inc(&sk->sk_drops);
474 skb_set_owner_r(skb, sk);
476 /* we escape from rcu protected region, make sure we dont leak
481 spin_lock_irqsave(&list->lock, flags);
482 sock_skb_set_dropcount(sk, skb);
483 __skb_queue_tail(list, skb);
484 spin_unlock_irqrestore(&list->lock, flags);
486 if (!sock_flag(sk, SOCK_DEAD))
487 sk->sk_data_ready(sk);
490 EXPORT_SYMBOL(__sock_queue_rcv_skb);
492 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
496 err = sk_filter(sk, skb);
500 return __sock_queue_rcv_skb(sk, skb);
502 EXPORT_SYMBOL(sock_queue_rcv_skb);
504 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
505 const int nested, unsigned int trim_cap, bool refcounted)
507 int rc = NET_RX_SUCCESS;
509 if (sk_filter_trim_cap(sk, skb, trim_cap))
510 goto discard_and_relse;
514 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
515 atomic_inc(&sk->sk_drops);
516 goto discard_and_relse;
519 bh_lock_sock_nested(sk);
522 if (!sock_owned_by_user(sk)) {
524 * trylock + unlock semantics:
526 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
528 rc = sk_backlog_rcv(sk, skb);
530 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
531 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
533 atomic_inc(&sk->sk_drops);
534 goto discard_and_relse;
546 EXPORT_SYMBOL(__sk_receive_skb);
548 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
550 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
552 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
554 struct dst_entry *dst = __sk_dst_get(sk);
556 if (dst && dst->obsolete &&
557 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
558 dst, cookie) == NULL) {
559 sk_tx_queue_clear(sk);
560 sk->sk_dst_pending_confirm = 0;
561 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
568 EXPORT_SYMBOL(__sk_dst_check);
570 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
572 struct dst_entry *dst = sk_dst_get(sk);
574 if (dst && dst->obsolete &&
575 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
576 dst, cookie) == NULL) {
584 EXPORT_SYMBOL(sk_dst_check);
586 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
588 int ret = -ENOPROTOOPT;
589 #ifdef CONFIG_NETDEVICES
590 struct net *net = sock_net(sk);
594 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
601 sk->sk_bound_dev_if = ifindex;
602 if (sk->sk_prot->rehash)
603 sk->sk_prot->rehash(sk);
614 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
620 ret = sock_bindtoindex_locked(sk, ifindex);
626 EXPORT_SYMBOL(sock_bindtoindex);
628 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
630 int ret = -ENOPROTOOPT;
631 #ifdef CONFIG_NETDEVICES
632 struct net *net = sock_net(sk);
633 char devname[IFNAMSIZ];
640 /* Bind this socket to a particular device like "eth0",
641 * as specified in the passed interface name. If the
642 * name is "" or the option length is zero the socket
645 if (optlen > IFNAMSIZ - 1)
646 optlen = IFNAMSIZ - 1;
647 memset(devname, 0, sizeof(devname));
650 if (copy_from_sockptr(devname, optval, optlen))
654 if (devname[0] != '\0') {
655 struct net_device *dev;
658 dev = dev_get_by_name_rcu(net, devname);
660 index = dev->ifindex;
667 return sock_bindtoindex(sk, index, true);
674 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
675 int __user *optlen, int len)
677 int ret = -ENOPROTOOPT;
678 #ifdef CONFIG_NETDEVICES
679 struct net *net = sock_net(sk);
680 char devname[IFNAMSIZ];
682 if (sk->sk_bound_dev_if == 0) {
691 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
695 len = strlen(devname) + 1;
698 if (copy_to_user(optval, devname, len))
703 if (put_user(len, optlen))
714 bool sk_mc_loop(struct sock *sk)
716 if (dev_recursion_level())
720 /* IPV6_ADDRFORM can change sk->sk_family under us. */
721 switch (READ_ONCE(sk->sk_family)) {
723 return inet_sk(sk)->mc_loop;
724 #if IS_ENABLED(CONFIG_IPV6)
726 return inet6_sk(sk)->mc_loop;
732 EXPORT_SYMBOL(sk_mc_loop);
734 void sock_set_reuseaddr(struct sock *sk)
737 sk->sk_reuse = SK_CAN_REUSE;
740 EXPORT_SYMBOL(sock_set_reuseaddr);
742 void sock_set_reuseport(struct sock *sk)
745 sk->sk_reuseport = true;
748 EXPORT_SYMBOL(sock_set_reuseport);
750 void sock_no_linger(struct sock *sk)
753 sk->sk_lingertime = 0;
754 sock_set_flag(sk, SOCK_LINGER);
757 EXPORT_SYMBOL(sock_no_linger);
759 void sock_set_priority(struct sock *sk, u32 priority)
762 sk->sk_priority = priority;
765 EXPORT_SYMBOL(sock_set_priority);
767 void sock_set_sndtimeo(struct sock *sk, s64 secs)
770 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
771 sk->sk_sndtimeo = secs * HZ;
773 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
776 EXPORT_SYMBOL(sock_set_sndtimeo);
778 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
781 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
782 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
783 sock_set_flag(sk, SOCK_RCVTSTAMP);
784 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
786 sock_reset_flag(sk, SOCK_RCVTSTAMP);
787 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
791 void sock_enable_timestamps(struct sock *sk)
794 __sock_set_timestamps(sk, true, false, true);
797 EXPORT_SYMBOL(sock_enable_timestamps);
799 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
802 case SO_TIMESTAMP_OLD:
803 __sock_set_timestamps(sk, valbool, false, false);
805 case SO_TIMESTAMP_NEW:
806 __sock_set_timestamps(sk, valbool, true, false);
808 case SO_TIMESTAMPNS_OLD:
809 __sock_set_timestamps(sk, valbool, false, true);
811 case SO_TIMESTAMPNS_NEW:
812 __sock_set_timestamps(sk, valbool, true, true);
817 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
819 struct net *net = sock_net(sk);
820 struct net_device *dev = NULL;
825 if (sk->sk_bound_dev_if)
826 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
829 pr_err("%s: sock not bind to device\n", __func__);
833 num = ethtool_get_phc_vclocks(dev, &vclock_index);
836 for (i = 0; i < num; i++) {
837 if (*(vclock_index + i) == phc_index) {
849 sk->sk_bind_phc = phc_index;
854 int sock_set_timestamping(struct sock *sk, int optname,
855 struct so_timestamping timestamping)
857 int val = timestamping.flags;
860 if (val & ~SOF_TIMESTAMPING_MASK)
863 if (val & SOF_TIMESTAMPING_OPT_ID &&
864 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
865 if (sk->sk_protocol == IPPROTO_TCP &&
866 sk->sk_type == SOCK_STREAM) {
867 if ((1 << sk->sk_state) &
868 (TCPF_CLOSE | TCPF_LISTEN))
870 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
872 atomic_set(&sk->sk_tskey, 0);
876 if (val & SOF_TIMESTAMPING_OPT_STATS &&
877 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
880 if (val & SOF_TIMESTAMPING_BIND_PHC) {
881 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
886 sk->sk_tsflags = val;
887 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
889 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
890 sock_enable_timestamp(sk,
891 SOCK_TIMESTAMPING_RX_SOFTWARE);
893 sock_disable_timestamp(sk,
894 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
898 void sock_set_keepalive(struct sock *sk)
901 if (sk->sk_prot->keepalive)
902 sk->sk_prot->keepalive(sk, true);
903 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
906 EXPORT_SYMBOL(sock_set_keepalive);
908 static void __sock_set_rcvbuf(struct sock *sk, int val)
910 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
911 * as a negative value.
913 val = min_t(int, val, INT_MAX / 2);
914 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
916 /* We double it on the way in to account for "struct sk_buff" etc.
917 * overhead. Applications assume that the SO_RCVBUF setting they make
918 * will allow that much actual data to be received on that socket.
920 * Applications are unaware that "struct sk_buff" and other overheads
921 * allocate from the receive buffer during socket buffer allocation.
923 * And after considering the possible alternatives, returning the value
924 * we actually used in getsockopt is the most desirable behavior.
926 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
929 void sock_set_rcvbuf(struct sock *sk, int val)
932 __sock_set_rcvbuf(sk, val);
935 EXPORT_SYMBOL(sock_set_rcvbuf);
937 static void __sock_set_mark(struct sock *sk, u32 val)
939 if (val != sk->sk_mark) {
945 void sock_set_mark(struct sock *sk, u32 val)
948 __sock_set_mark(sk, val);
951 EXPORT_SYMBOL(sock_set_mark);
954 * This is meant for all protocols to use and covers goings on
955 * at the socket level. Everything here is generic.
958 int sock_setsockopt(struct socket *sock, int level, int optname,
959 sockptr_t optval, unsigned int optlen)
961 struct so_timestamping timestamping;
962 struct sock_txtime sk_txtime;
963 struct sock *sk = sock->sk;
970 * Options without arguments
973 if (optname == SO_BINDTODEVICE)
974 return sock_setbindtodevice(sk, optval, optlen);
976 if (optlen < sizeof(int))
979 if (copy_from_sockptr(&val, optval, sizeof(val)))
982 valbool = val ? 1 : 0;
988 if (val && !capable(CAP_NET_ADMIN))
991 sock_valbool_flag(sk, SOCK_DBG, valbool);
994 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
997 sk->sk_reuseport = valbool;
1006 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1010 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1013 /* Don't error on this BSD doesn't and if you think
1014 * about it this is right. Otherwise apps have to
1015 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1016 * are treated in BSD as hints
1018 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1020 /* Ensure val * 2 fits into an int, to prevent max_t()
1021 * from treating it as a negative value.
1023 val = min_t(int, val, INT_MAX / 2);
1024 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1025 WRITE_ONCE(sk->sk_sndbuf,
1026 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1027 /* Wake up sending tasks if we upped the value. */
1028 sk->sk_write_space(sk);
1031 case SO_SNDBUFFORCE:
1032 if (!capable(CAP_NET_ADMIN)) {
1037 /* No negative values (to prevent underflow, as val will be
1045 /* Don't error on this BSD doesn't and if you think
1046 * about it this is right. Otherwise apps have to
1047 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1048 * are treated in BSD as hints
1050 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1053 case SO_RCVBUFFORCE:
1054 if (!capable(CAP_NET_ADMIN)) {
1059 /* No negative values (to prevent underflow, as val will be
1062 __sock_set_rcvbuf(sk, max(val, 0));
1066 if (sk->sk_prot->keepalive)
1067 sk->sk_prot->keepalive(sk, valbool);
1068 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1072 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1076 sk->sk_no_check_tx = valbool;
1080 if ((val >= 0 && val <= 6) ||
1081 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1082 sk->sk_priority = val;
1088 if (optlen < sizeof(ling)) {
1089 ret = -EINVAL; /* 1003.1g */
1092 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1097 sock_reset_flag(sk, SOCK_LINGER);
1099 #if (BITS_PER_LONG == 32)
1100 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1101 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1104 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1105 sock_set_flag(sk, SOCK_LINGER);
1114 set_bit(SOCK_PASSCRED, &sock->flags);
1116 clear_bit(SOCK_PASSCRED, &sock->flags);
1119 case SO_TIMESTAMP_OLD:
1120 case SO_TIMESTAMP_NEW:
1121 case SO_TIMESTAMPNS_OLD:
1122 case SO_TIMESTAMPNS_NEW:
1123 sock_set_timestamp(sk, optname, valbool);
1126 case SO_TIMESTAMPING_NEW:
1127 case SO_TIMESTAMPING_OLD:
1128 if (optlen == sizeof(timestamping)) {
1129 if (copy_from_sockptr(×tamping, optval,
1130 sizeof(timestamping))) {
1135 memset(×tamping, 0, sizeof(timestamping));
1136 timestamping.flags = val;
1138 ret = sock_set_timestamping(sk, optname, timestamping);
1144 if (sock->ops->set_rcvlowat)
1145 ret = sock->ops->set_rcvlowat(sk, val);
1147 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1150 case SO_RCVTIMEO_OLD:
1151 case SO_RCVTIMEO_NEW:
1152 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1153 optlen, optname == SO_RCVTIMEO_OLD);
1156 case SO_SNDTIMEO_OLD:
1157 case SO_SNDTIMEO_NEW:
1158 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1159 optlen, optname == SO_SNDTIMEO_OLD);
1162 case SO_ATTACH_FILTER: {
1163 struct sock_fprog fprog;
1165 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1167 ret = sk_attach_filter(&fprog, sk);
1172 if (optlen == sizeof(u32)) {
1176 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1179 ret = sk_attach_bpf(ufd, sk);
1183 case SO_ATTACH_REUSEPORT_CBPF: {
1184 struct sock_fprog fprog;
1186 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1188 ret = sk_reuseport_attach_filter(&fprog, sk);
1191 case SO_ATTACH_REUSEPORT_EBPF:
1193 if (optlen == sizeof(u32)) {
1197 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1200 ret = sk_reuseport_attach_bpf(ufd, sk);
1204 case SO_DETACH_REUSEPORT_BPF:
1205 ret = reuseport_detach_prog(sk);
1208 case SO_DETACH_FILTER:
1209 ret = sk_detach_filter(sk);
1212 case SO_LOCK_FILTER:
1213 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1216 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1221 set_bit(SOCK_PASSSEC, &sock->flags);
1223 clear_bit(SOCK_PASSSEC, &sock->flags);
1226 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1231 __sock_set_mark(sk, val);
1235 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1238 case SO_WIFI_STATUS:
1239 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1243 if (sock->ops->set_peek_off)
1244 ret = sock->ops->set_peek_off(sk, val);
1250 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1253 case SO_SELECT_ERR_QUEUE:
1254 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1257 #ifdef CONFIG_NET_RX_BUSY_POLL
1259 /* allow unprivileged users to decrease the value */
1260 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1266 WRITE_ONCE(sk->sk_ll_usec, val);
1269 case SO_PREFER_BUSY_POLL:
1270 if (valbool && !capable(CAP_NET_ADMIN))
1273 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1275 case SO_BUSY_POLL_BUDGET:
1276 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1279 if (val < 0 || val > U16_MAX)
1282 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1287 case SO_MAX_PACING_RATE:
1289 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1291 if (sizeof(ulval) != sizeof(val) &&
1292 optlen >= sizeof(ulval) &&
1293 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1298 cmpxchg(&sk->sk_pacing_status,
1301 /* Pairs with READ_ONCE() from sk_getsockopt() */
1302 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1303 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1306 case SO_INCOMING_CPU:
1307 reuseport_update_incoming_cpu(sk, val);
1312 dst_negative_advice(sk);
1316 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1317 if (!((sk->sk_type == SOCK_STREAM &&
1318 sk->sk_protocol == IPPROTO_TCP) ||
1319 (sk->sk_type == SOCK_DGRAM &&
1320 sk->sk_protocol == IPPROTO_UDP)))
1322 } else if (sk->sk_family != PF_RDS) {
1326 if (val < 0 || val > 1)
1329 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1334 if (optlen != sizeof(struct sock_txtime)) {
1337 } else if (copy_from_sockptr(&sk_txtime, optval,
1338 sizeof(struct sock_txtime))) {
1341 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1345 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1346 * scheduler has enough safe guards.
1348 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1349 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1353 sock_valbool_flag(sk, SOCK_TXTIME, true);
1354 sk->sk_clockid = sk_txtime.clockid;
1355 sk->sk_txtime_deadline_mode =
1356 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1357 sk->sk_txtime_report_errors =
1358 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1361 case SO_BINDTOIFINDEX:
1362 ret = sock_bindtoindex_locked(sk, val);
1366 if (val & ~SOCK_BUF_LOCK_MASK) {
1370 sk->sk_userlocks = val | (sk->sk_userlocks &
1371 ~SOCK_BUF_LOCK_MASK);
1381 EXPORT_SYMBOL(sock_setsockopt);
1383 static const struct cred *sk_get_peer_cred(struct sock *sk)
1385 const struct cred *cred;
1387 spin_lock(&sk->sk_peer_lock);
1388 cred = get_cred(sk->sk_peer_cred);
1389 spin_unlock(&sk->sk_peer_lock);
1394 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1395 struct ucred *ucred)
1397 ucred->pid = pid_vnr(pid);
1398 ucred->uid = ucred->gid = -1;
1400 struct user_namespace *current_ns = current_user_ns();
1402 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1403 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1407 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1409 struct user_namespace *user_ns = current_user_ns();
1412 for (i = 0; i < src->ngroups; i++)
1413 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1419 int sock_getsockopt(struct socket *sock, int level, int optname,
1420 char __user *optval, int __user *optlen)
1422 struct sock *sk = sock->sk;
1427 unsigned long ulval;
1429 struct old_timeval32 tm32;
1430 struct __kernel_old_timeval tm;
1431 struct __kernel_sock_timeval stm;
1432 struct sock_txtime txtime;
1433 struct so_timestamping timestamping;
1436 int lv = sizeof(int);
1439 if (get_user(len, optlen))
1444 memset(&v, 0, sizeof(v));
1448 v.val = sock_flag(sk, SOCK_DBG);
1452 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1456 v.val = sock_flag(sk, SOCK_BROADCAST);
1460 v.val = READ_ONCE(sk->sk_sndbuf);
1464 v.val = READ_ONCE(sk->sk_rcvbuf);
1468 v.val = sk->sk_reuse;
1472 v.val = sk->sk_reuseport;
1476 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1480 v.val = sk->sk_type;
1484 v.val = sk->sk_protocol;
1488 v.val = sk->sk_family;
1492 v.val = -sock_error(sk);
1494 v.val = xchg(&sk->sk_err_soft, 0);
1498 v.val = sock_flag(sk, SOCK_URGINLINE);
1502 v.val = sk->sk_no_check_tx;
1506 v.val = sk->sk_priority;
1510 lv = sizeof(v.ling);
1511 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1512 v.ling.l_linger = sk->sk_lingertime / HZ;
1518 case SO_TIMESTAMP_OLD:
1519 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1520 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1521 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1524 case SO_TIMESTAMPNS_OLD:
1525 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1528 case SO_TIMESTAMP_NEW:
1529 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1532 case SO_TIMESTAMPNS_NEW:
1533 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1536 case SO_TIMESTAMPING_OLD:
1537 lv = sizeof(v.timestamping);
1538 v.timestamping.flags = sk->sk_tsflags;
1539 v.timestamping.bind_phc = sk->sk_bind_phc;
1542 case SO_RCVTIMEO_OLD:
1543 case SO_RCVTIMEO_NEW:
1544 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1547 case SO_SNDTIMEO_OLD:
1548 case SO_SNDTIMEO_NEW:
1549 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1553 v.val = READ_ONCE(sk->sk_rcvlowat);
1561 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1566 struct ucred peercred;
1567 if (len > sizeof(peercred))
1568 len = sizeof(peercred);
1570 spin_lock(&sk->sk_peer_lock);
1571 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1572 spin_unlock(&sk->sk_peer_lock);
1574 if (copy_to_user(optval, &peercred, len))
1581 const struct cred *cred;
1584 cred = sk_get_peer_cred(sk);
1588 n = cred->group_info->ngroups;
1589 if (len < n * sizeof(gid_t)) {
1590 len = n * sizeof(gid_t);
1592 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1594 len = n * sizeof(gid_t);
1596 ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1607 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1612 if (copy_to_user(optval, address, len))
1617 /* Dubious BSD thing... Probably nobody even uses it, but
1618 * the UNIX standard wants it for whatever reason... -DaveM
1621 v.val = sk->sk_state == TCP_LISTEN;
1625 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1629 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1632 v.val = sk->sk_mark;
1636 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1639 case SO_WIFI_STATUS:
1640 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1644 if (!sock->ops->set_peek_off)
1647 v.val = READ_ONCE(sk->sk_peek_off);
1650 v.val = sock_flag(sk, SOCK_NOFCS);
1653 case SO_BINDTODEVICE:
1654 return sock_getbindtodevice(sk, optval, optlen, len);
1657 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1663 case SO_LOCK_FILTER:
1664 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1667 case SO_BPF_EXTENSIONS:
1668 v.val = bpf_tell_extensions();
1671 case SO_SELECT_ERR_QUEUE:
1672 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1675 #ifdef CONFIG_NET_RX_BUSY_POLL
1677 v.val = READ_ONCE(sk->sk_ll_usec);
1679 case SO_PREFER_BUSY_POLL:
1680 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1684 case SO_MAX_PACING_RATE:
1685 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1686 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1687 lv = sizeof(v.ulval);
1688 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1691 v.val = min_t(unsigned long, ~0U,
1692 READ_ONCE(sk->sk_max_pacing_rate));
1696 case SO_INCOMING_CPU:
1697 v.val = READ_ONCE(sk->sk_incoming_cpu);
1702 u32 meminfo[SK_MEMINFO_VARS];
1704 sk_get_meminfo(sk, meminfo);
1706 len = min_t(unsigned int, len, sizeof(meminfo));
1707 if (copy_to_user(optval, &meminfo, len))
1713 #ifdef CONFIG_NET_RX_BUSY_POLL
1714 case SO_INCOMING_NAPI_ID:
1715 v.val = READ_ONCE(sk->sk_napi_id);
1717 /* aggregate non-NAPI IDs down to 0 */
1718 if (v.val < MIN_NAPI_ID)
1728 v.val64 = sock_gen_cookie(sk);
1732 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1736 lv = sizeof(v.txtime);
1737 v.txtime.clockid = sk->sk_clockid;
1738 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1739 SOF_TXTIME_DEADLINE_MODE : 0;
1740 v.txtime.flags |= sk->sk_txtime_report_errors ?
1741 SOF_TXTIME_REPORT_ERRORS : 0;
1744 case SO_BINDTOIFINDEX:
1745 v.val = sk->sk_bound_dev_if;
1748 case SO_NETNS_COOKIE:
1752 v.val64 = sock_net(sk)->net_cookie;
1756 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1760 /* We implement the SO_SNDLOWAT etc to not be settable
1763 return -ENOPROTOOPT;
1768 if (copy_to_user(optval, &v, len))
1771 if (put_user(len, optlen))
1777 * Initialize an sk_lock.
1779 * (We also register the sk_lock with the lock validator.)
1781 static inline void sock_lock_init(struct sock *sk)
1783 if (sk->sk_kern_sock)
1784 sock_lock_init_class_and_name(
1786 af_family_kern_slock_key_strings[sk->sk_family],
1787 af_family_kern_slock_keys + sk->sk_family,
1788 af_family_kern_key_strings[sk->sk_family],
1789 af_family_kern_keys + sk->sk_family);
1791 sock_lock_init_class_and_name(
1793 af_family_slock_key_strings[sk->sk_family],
1794 af_family_slock_keys + sk->sk_family,
1795 af_family_key_strings[sk->sk_family],
1796 af_family_keys + sk->sk_family);
1800 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1801 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1802 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1804 static void sock_copy(struct sock *nsk, const struct sock *osk)
1806 const struct proto *prot = READ_ONCE(osk->sk_prot);
1807 #ifdef CONFIG_SECURITY_NETWORK
1808 void *sptr = nsk->sk_security;
1811 /* If we move sk_tx_queue_mapping out of the private section,
1812 * we must check if sk_tx_queue_clear() is called after
1813 * sock_copy() in sk_clone_lock().
1815 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1816 offsetof(struct sock, sk_dontcopy_begin) ||
1817 offsetof(struct sock, sk_tx_queue_mapping) >=
1818 offsetof(struct sock, sk_dontcopy_end));
1820 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1822 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1823 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1825 #ifdef CONFIG_SECURITY_NETWORK
1826 nsk->sk_security = sptr;
1827 security_sk_clone(osk, nsk);
1831 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1835 struct kmem_cache *slab;
1839 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1842 if (want_init_on_alloc(priority))
1843 sk_prot_clear_nulls(sk, prot->obj_size);
1845 sk = kmalloc(prot->obj_size, priority);
1848 if (security_sk_alloc(sk, family, priority))
1851 if (!try_module_get(prot->owner))
1858 security_sk_free(sk);
1861 kmem_cache_free(slab, sk);
1867 static void sk_prot_free(struct proto *prot, struct sock *sk)
1869 struct kmem_cache *slab;
1870 struct module *owner;
1872 owner = prot->owner;
1875 cgroup_sk_free(&sk->sk_cgrp_data);
1876 mem_cgroup_sk_free(sk);
1877 security_sk_free(sk);
1879 kmem_cache_free(slab, sk);
1886 * sk_alloc - All socket objects are allocated here
1887 * @net: the applicable net namespace
1888 * @family: protocol family
1889 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1890 * @prot: struct proto associated with this new sock instance
1891 * @kern: is this to be a kernel socket?
1893 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1894 struct proto *prot, int kern)
1898 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1900 sk->sk_family = family;
1902 * See comment in struct sock definition to understand
1903 * why we need sk_prot_creator -acme
1905 sk->sk_prot = sk->sk_prot_creator = prot;
1906 sk->sk_kern_sock = kern;
1908 sk->sk_net_refcnt = kern ? 0 : 1;
1909 if (likely(sk->sk_net_refcnt)) {
1911 sock_inuse_add(net, 1);
1914 sock_net_set(sk, net);
1915 refcount_set(&sk->sk_wmem_alloc, 1);
1917 mem_cgroup_sk_alloc(sk);
1918 cgroup_sk_alloc(&sk->sk_cgrp_data);
1919 sock_update_classid(&sk->sk_cgrp_data);
1920 sock_update_netprioidx(&sk->sk_cgrp_data);
1921 sk_tx_queue_clear(sk);
1926 EXPORT_SYMBOL(sk_alloc);
1928 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1929 * grace period. This is the case for UDP sockets and TCP listeners.
1931 static void __sk_destruct(struct rcu_head *head)
1933 struct sock *sk = container_of(head, struct sock, sk_rcu);
1934 struct sk_filter *filter;
1936 if (sk->sk_destruct)
1937 sk->sk_destruct(sk);
1939 filter = rcu_dereference_check(sk->sk_filter,
1940 refcount_read(&sk->sk_wmem_alloc) == 0);
1942 sk_filter_uncharge(sk, filter);
1943 RCU_INIT_POINTER(sk->sk_filter, NULL);
1946 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1948 #ifdef CONFIG_BPF_SYSCALL
1949 bpf_sk_storage_free(sk);
1952 if (atomic_read(&sk->sk_omem_alloc))
1953 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1954 __func__, atomic_read(&sk->sk_omem_alloc));
1956 if (sk->sk_frag.page) {
1957 put_page(sk->sk_frag.page);
1958 sk->sk_frag.page = NULL;
1961 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
1962 put_cred(sk->sk_peer_cred);
1963 put_pid(sk->sk_peer_pid);
1965 if (likely(sk->sk_net_refcnt))
1966 put_net(sock_net(sk));
1967 sk_prot_free(sk->sk_prot_creator, sk);
1970 void sk_destruct(struct sock *sk)
1972 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1974 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1975 reuseport_detach_sock(sk);
1976 use_call_rcu = true;
1980 call_rcu(&sk->sk_rcu, __sk_destruct);
1982 __sk_destruct(&sk->sk_rcu);
1985 static void __sk_free(struct sock *sk)
1987 if (likely(sk->sk_net_refcnt))
1988 sock_inuse_add(sock_net(sk), -1);
1990 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1991 sock_diag_broadcast_destroy(sk);
1996 void sk_free(struct sock *sk)
1999 * We subtract one from sk_wmem_alloc and can know if
2000 * some packets are still in some tx queue.
2001 * If not null, sock_wfree() will call __sk_free(sk) later
2003 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2006 EXPORT_SYMBOL(sk_free);
2008 static void sk_init_common(struct sock *sk)
2010 skb_queue_head_init(&sk->sk_receive_queue);
2011 skb_queue_head_init(&sk->sk_write_queue);
2012 skb_queue_head_init(&sk->sk_error_queue);
2014 rwlock_init(&sk->sk_callback_lock);
2015 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2016 af_rlock_keys + sk->sk_family,
2017 af_family_rlock_key_strings[sk->sk_family]);
2018 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2019 af_wlock_keys + sk->sk_family,
2020 af_family_wlock_key_strings[sk->sk_family]);
2021 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2022 af_elock_keys + sk->sk_family,
2023 af_family_elock_key_strings[sk->sk_family]);
2024 lockdep_set_class_and_name(&sk->sk_callback_lock,
2025 af_callback_keys + sk->sk_family,
2026 af_family_clock_key_strings[sk->sk_family]);
2030 * sk_clone_lock - clone a socket, and lock its clone
2031 * @sk: the socket to clone
2032 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2034 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2036 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2038 struct proto *prot = READ_ONCE(sk->sk_prot);
2039 struct sk_filter *filter;
2040 bool is_charged = true;
2043 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2047 sock_copy(newsk, sk);
2049 newsk->sk_prot_creator = prot;
2052 if (likely(newsk->sk_net_refcnt)) {
2053 get_net(sock_net(newsk));
2054 sock_inuse_add(sock_net(newsk), 1);
2056 sk_node_init(&newsk->sk_node);
2057 sock_lock_init(newsk);
2058 bh_lock_sock(newsk);
2059 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2060 newsk->sk_backlog.len = 0;
2062 atomic_set(&newsk->sk_rmem_alloc, 0);
2064 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2065 refcount_set(&newsk->sk_wmem_alloc, 1);
2067 atomic_set(&newsk->sk_omem_alloc, 0);
2068 sk_init_common(newsk);
2070 newsk->sk_dst_cache = NULL;
2071 newsk->sk_dst_pending_confirm = 0;
2072 newsk->sk_wmem_queued = 0;
2073 newsk->sk_forward_alloc = 0;
2074 atomic_set(&newsk->sk_drops, 0);
2075 newsk->sk_send_head = NULL;
2076 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2077 atomic_set(&newsk->sk_zckey, 0);
2079 sock_reset_flag(newsk, SOCK_DONE);
2081 /* sk->sk_memcg will be populated at accept() time */
2082 newsk->sk_memcg = NULL;
2084 cgroup_sk_clone(&newsk->sk_cgrp_data);
2087 filter = rcu_dereference(sk->sk_filter);
2089 /* though it's an empty new sock, the charging may fail
2090 * if sysctl_optmem_max was changed between creation of
2091 * original socket and cloning
2093 is_charged = sk_filter_charge(newsk, filter);
2094 RCU_INIT_POINTER(newsk->sk_filter, filter);
2097 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2098 /* We need to make sure that we don't uncharge the new
2099 * socket if we couldn't charge it in the first place
2100 * as otherwise we uncharge the parent's filter.
2103 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2104 sk_free_unlock_clone(newsk);
2108 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2110 if (bpf_sk_storage_clone(sk, newsk)) {
2111 sk_free_unlock_clone(newsk);
2116 /* Clear sk_user_data if parent had the pointer tagged
2117 * as not suitable for copying when cloning.
2119 if (sk_user_data_is_nocopy(newsk))
2120 newsk->sk_user_data = NULL;
2123 newsk->sk_err_soft = 0;
2124 newsk->sk_priority = 0;
2125 newsk->sk_incoming_cpu = raw_smp_processor_id();
2127 /* Before updating sk_refcnt, we must commit prior changes to memory
2128 * (Documentation/RCU/rculist_nulls.rst for details)
2131 refcount_set(&newsk->sk_refcnt, 2);
2133 /* Increment the counter in the same struct proto as the master
2134 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2135 * is the same as sk->sk_prot->socks, as this field was copied
2138 * This _changes_ the previous behaviour, where
2139 * tcp_create_openreq_child always was incrementing the
2140 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2141 * to be taken into account in all callers. -acme
2143 sk_refcnt_debug_inc(newsk);
2144 sk_set_socket(newsk, NULL);
2145 sk_tx_queue_clear(newsk);
2146 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2148 if (newsk->sk_prot->sockets_allocated)
2149 sk_sockets_allocated_inc(newsk);
2151 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2152 net_enable_timestamp();
2156 EXPORT_SYMBOL_GPL(sk_clone_lock);
2158 void sk_free_unlock_clone(struct sock *sk)
2160 /* It is still raw copy of parent, so invalidate
2161 * destructor and make plain sk_free() */
2162 sk->sk_destruct = NULL;
2166 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2168 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2172 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2173 if (sk->sk_route_caps & NETIF_F_GSO)
2174 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2175 sk->sk_route_caps &= ~sk->sk_route_nocaps;
2176 if (sk_can_gso(sk)) {
2177 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2178 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2180 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2181 sk->sk_gso_max_size = dst->dev->gso_max_size;
2182 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2185 sk->sk_gso_max_segs = max_segs;
2186 sk_dst_set(sk, dst);
2188 EXPORT_SYMBOL_GPL(sk_setup_caps);
2191 * Simple resource managers for sockets.
2196 * Write buffer destructor automatically called from kfree_skb.
2198 void sock_wfree(struct sk_buff *skb)
2200 struct sock *sk = skb->sk;
2201 unsigned int len = skb->truesize;
2203 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2205 * Keep a reference on sk_wmem_alloc, this will be released
2206 * after sk_write_space() call
2208 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2209 sk->sk_write_space(sk);
2213 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2214 * could not do because of in-flight packets
2216 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2219 EXPORT_SYMBOL(sock_wfree);
2221 /* This variant of sock_wfree() is used by TCP,
2222 * since it sets SOCK_USE_WRITE_QUEUE.
2224 void __sock_wfree(struct sk_buff *skb)
2226 struct sock *sk = skb->sk;
2228 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2232 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2237 if (unlikely(!sk_fullsock(sk))) {
2238 skb->destructor = sock_edemux;
2243 skb->destructor = sock_wfree;
2244 skb_set_hash_from_sk(skb, sk);
2246 * We used to take a refcount on sk, but following operation
2247 * is enough to guarantee sk_free() wont free this sock until
2248 * all in-flight packets are completed
2250 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2252 EXPORT_SYMBOL(skb_set_owner_w);
2254 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2256 #ifdef CONFIG_TLS_DEVICE
2257 /* Drivers depend on in-order delivery for crypto offload,
2258 * partial orphan breaks out-of-order-OK logic.
2263 return (skb->destructor == sock_wfree ||
2264 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2267 /* This helper is used by netem, as it can hold packets in its
2268 * delay queue. We want to allow the owner socket to send more
2269 * packets, as if they were already TX completed by a typical driver.
2270 * But we also want to keep skb->sk set because some packet schedulers
2271 * rely on it (sch_fq for example).
2273 void skb_orphan_partial(struct sk_buff *skb)
2275 if (skb_is_tcp_pure_ack(skb))
2278 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2283 EXPORT_SYMBOL(skb_orphan_partial);
2286 * Read buffer destructor automatically called from kfree_skb.
2288 void sock_rfree(struct sk_buff *skb)
2290 struct sock *sk = skb->sk;
2291 unsigned int len = skb->truesize;
2293 atomic_sub(len, &sk->sk_rmem_alloc);
2294 sk_mem_uncharge(sk, len);
2296 EXPORT_SYMBOL(sock_rfree);
2299 * Buffer destructor for skbs that are not used directly in read or write
2300 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2302 void sock_efree(struct sk_buff *skb)
2306 EXPORT_SYMBOL(sock_efree);
2308 /* Buffer destructor for prefetch/receive path where reference count may
2309 * not be held, e.g. for listen sockets.
2312 void sock_pfree(struct sk_buff *skb)
2314 if (sk_is_refcounted(skb->sk))
2315 sock_gen_put(skb->sk);
2317 EXPORT_SYMBOL(sock_pfree);
2318 #endif /* CONFIG_INET */
2320 kuid_t sock_i_uid(struct sock *sk)
2324 read_lock_bh(&sk->sk_callback_lock);
2325 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2326 read_unlock_bh(&sk->sk_callback_lock);
2329 EXPORT_SYMBOL(sock_i_uid);
2331 unsigned long __sock_i_ino(struct sock *sk)
2335 read_lock(&sk->sk_callback_lock);
2336 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2337 read_unlock(&sk->sk_callback_lock);
2340 EXPORT_SYMBOL(__sock_i_ino);
2342 unsigned long sock_i_ino(struct sock *sk)
2347 ino = __sock_i_ino(sk);
2351 EXPORT_SYMBOL(sock_i_ino);
2354 * Allocate a skb from the socket's send buffer.
2356 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2360 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2361 struct sk_buff *skb = alloc_skb(size, priority);
2364 skb_set_owner_w(skb, sk);
2370 EXPORT_SYMBOL(sock_wmalloc);
2372 static void sock_ofree(struct sk_buff *skb)
2374 struct sock *sk = skb->sk;
2376 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2379 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2382 struct sk_buff *skb;
2384 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2385 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2386 READ_ONCE(sysctl_optmem_max))
2389 skb = alloc_skb(size, priority);
2393 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2395 skb->destructor = sock_ofree;
2400 * Allocate a memory block from the socket's option memory buffer.
2402 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2404 int optmem_max = READ_ONCE(sysctl_optmem_max);
2406 if ((unsigned int)size <= optmem_max &&
2407 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2409 /* First do the add, to avoid the race if kmalloc
2412 atomic_add(size, &sk->sk_omem_alloc);
2413 mem = kmalloc(size, priority);
2416 atomic_sub(size, &sk->sk_omem_alloc);
2420 EXPORT_SYMBOL(sock_kmalloc);
2422 /* Free an option memory block. Note, we actually want the inline
2423 * here as this allows gcc to detect the nullify and fold away the
2424 * condition entirely.
2426 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2429 if (WARN_ON_ONCE(!mem))
2432 kfree_sensitive(mem);
2435 atomic_sub(size, &sk->sk_omem_alloc);
2438 void sock_kfree_s(struct sock *sk, void *mem, int size)
2440 __sock_kfree_s(sk, mem, size, false);
2442 EXPORT_SYMBOL(sock_kfree_s);
2444 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2446 __sock_kfree_s(sk, mem, size, true);
2448 EXPORT_SYMBOL(sock_kzfree_s);
2450 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2451 I think, these locks should be removed for datagram sockets.
2453 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2457 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2461 if (signal_pending(current))
2463 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2464 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2465 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2467 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2469 if (READ_ONCE(sk->sk_err))
2471 timeo = schedule_timeout(timeo);
2473 finish_wait(sk_sleep(sk), &wait);
2479 * Generic send/receive buffer handlers
2482 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2483 unsigned long data_len, int noblock,
2484 int *errcode, int max_page_order)
2486 struct sk_buff *skb;
2490 timeo = sock_sndtimeo(sk, noblock);
2492 err = sock_error(sk);
2497 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2500 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2503 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2504 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2508 if (signal_pending(current))
2510 timeo = sock_wait_for_wmem(sk, timeo);
2512 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2513 errcode, sk->sk_allocation);
2515 skb_set_owner_w(skb, sk);
2519 err = sock_intr_errno(timeo);
2524 EXPORT_SYMBOL(sock_alloc_send_pskb);
2526 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2527 int noblock, int *errcode)
2529 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2531 EXPORT_SYMBOL(sock_alloc_send_skb);
2533 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2534 struct sockcm_cookie *sockc)
2538 switch (cmsg->cmsg_type) {
2540 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2542 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2544 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2546 case SO_TIMESTAMPING_OLD:
2547 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2550 tsflags = *(u32 *)CMSG_DATA(cmsg);
2551 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2554 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2555 sockc->tsflags |= tsflags;
2558 if (!sock_flag(sk, SOCK_TXTIME))
2560 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2562 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2564 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2566 case SCM_CREDENTIALS:
2573 EXPORT_SYMBOL(__sock_cmsg_send);
2575 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2576 struct sockcm_cookie *sockc)
2578 struct cmsghdr *cmsg;
2581 for_each_cmsghdr(cmsg, msg) {
2582 if (!CMSG_OK(msg, cmsg))
2584 if (cmsg->cmsg_level != SOL_SOCKET)
2586 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2592 EXPORT_SYMBOL(sock_cmsg_send);
2594 static void sk_enter_memory_pressure(struct sock *sk)
2596 if (!sk->sk_prot->enter_memory_pressure)
2599 sk->sk_prot->enter_memory_pressure(sk);
2602 static void sk_leave_memory_pressure(struct sock *sk)
2604 if (sk->sk_prot->leave_memory_pressure) {
2605 sk->sk_prot->leave_memory_pressure(sk);
2607 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2609 if (memory_pressure && READ_ONCE(*memory_pressure))
2610 WRITE_ONCE(*memory_pressure, 0);
2614 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2617 * skb_page_frag_refill - check that a page_frag contains enough room
2618 * @sz: minimum size of the fragment we want to get
2619 * @pfrag: pointer to page_frag
2620 * @gfp: priority for memory allocation
2622 * Note: While this allocator tries to use high order pages, there is
2623 * no guarantee that allocations succeed. Therefore, @sz MUST be
2624 * less or equal than PAGE_SIZE.
2626 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2629 if (page_ref_count(pfrag->page) == 1) {
2633 if (pfrag->offset + sz <= pfrag->size)
2635 put_page(pfrag->page);
2639 if (SKB_FRAG_PAGE_ORDER &&
2640 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2641 /* Avoid direct reclaim but allow kswapd to wake */
2642 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2643 __GFP_COMP | __GFP_NOWARN |
2645 SKB_FRAG_PAGE_ORDER);
2646 if (likely(pfrag->page)) {
2647 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2651 pfrag->page = alloc_page(gfp);
2652 if (likely(pfrag->page)) {
2653 pfrag->size = PAGE_SIZE;
2658 EXPORT_SYMBOL(skb_page_frag_refill);
2660 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2662 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2665 sk_enter_memory_pressure(sk);
2666 sk_stream_moderate_sndbuf(sk);
2669 EXPORT_SYMBOL(sk_page_frag_refill);
2671 void __lock_sock(struct sock *sk)
2672 __releases(&sk->sk_lock.slock)
2673 __acquires(&sk->sk_lock.slock)
2678 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2679 TASK_UNINTERRUPTIBLE);
2680 spin_unlock_bh(&sk->sk_lock.slock);
2682 spin_lock_bh(&sk->sk_lock.slock);
2683 if (!sock_owned_by_user(sk))
2686 finish_wait(&sk->sk_lock.wq, &wait);
2689 void __release_sock(struct sock *sk)
2690 __releases(&sk->sk_lock.slock)
2691 __acquires(&sk->sk_lock.slock)
2693 struct sk_buff *skb, *next;
2695 while ((skb = sk->sk_backlog.head) != NULL) {
2696 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2698 spin_unlock_bh(&sk->sk_lock.slock);
2703 WARN_ON_ONCE(skb_dst_is_noref(skb));
2704 skb_mark_not_on_list(skb);
2705 sk_backlog_rcv(sk, skb);
2710 } while (skb != NULL);
2712 spin_lock_bh(&sk->sk_lock.slock);
2716 * Doing the zeroing here guarantee we can not loop forever
2717 * while a wild producer attempts to flood us.
2719 sk->sk_backlog.len = 0;
2722 void __sk_flush_backlog(struct sock *sk)
2724 spin_lock_bh(&sk->sk_lock.slock);
2726 spin_unlock_bh(&sk->sk_lock.slock);
2730 * sk_wait_data - wait for data to arrive at sk_receive_queue
2731 * @sk: sock to wait on
2732 * @timeo: for how long
2733 * @skb: last skb seen on sk_receive_queue
2735 * Now socket state including sk->sk_err is changed only under lock,
2736 * hence we may omit checks after joining wait queue.
2737 * We check receive queue before schedule() only as optimization;
2738 * it is very likely that release_sock() added new data.
2740 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2742 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2745 add_wait_queue(sk_sleep(sk), &wait);
2746 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2747 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2748 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2749 remove_wait_queue(sk_sleep(sk), &wait);
2752 EXPORT_SYMBOL(sk_wait_data);
2755 * __sk_mem_raise_allocated - increase memory_allocated
2757 * @size: memory size to allocate
2758 * @amt: pages to allocate
2759 * @kind: allocation type
2761 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2763 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2765 struct proto *prot = sk->sk_prot;
2766 long allocated = sk_memory_allocated_add(sk, amt);
2767 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2768 bool charged = true;
2771 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2772 gfp_memcg_charge())))
2773 goto suppress_allocation;
2776 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2777 sk_leave_memory_pressure(sk);
2781 /* Under pressure. */
2782 if (allocated > sk_prot_mem_limits(sk, 1))
2783 sk_enter_memory_pressure(sk);
2785 /* Over hard limit. */
2786 if (allocated > sk_prot_mem_limits(sk, 2))
2787 goto suppress_allocation;
2789 /* guarantee minimum buffer size under pressure */
2790 if (kind == SK_MEM_RECV) {
2791 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2794 } else { /* SK_MEM_SEND */
2795 int wmem0 = sk_get_wmem0(sk, prot);
2797 if (sk->sk_type == SOCK_STREAM) {
2798 if (sk->sk_wmem_queued < wmem0)
2800 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2805 if (sk_has_memory_pressure(sk)) {
2808 if (!sk_under_memory_pressure(sk))
2810 alloc = sk_sockets_allocated_read_positive(sk);
2811 if (sk_prot_mem_limits(sk, 2) > alloc *
2812 sk_mem_pages(sk->sk_wmem_queued +
2813 atomic_read(&sk->sk_rmem_alloc) +
2814 sk->sk_forward_alloc))
2818 suppress_allocation:
2820 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2821 sk_stream_moderate_sndbuf(sk);
2823 /* Fail only if socket is _under_ its sndbuf.
2824 * In this case we cannot block, so that we have to fail.
2826 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
2827 /* Force charge with __GFP_NOFAIL */
2828 if (memcg_charge && !charged) {
2829 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2830 gfp_memcg_charge() | __GFP_NOFAIL);
2836 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2837 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2839 sk_memory_allocated_sub(sk, amt);
2841 if (memcg_charge && charged)
2842 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2846 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2849 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2851 * @size: memory size to allocate
2852 * @kind: allocation type
2854 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2855 * rmem allocation. This function assumes that protocols which have
2856 * memory_pressure use sk_wmem_queued as write buffer accounting.
2858 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2860 int ret, amt = sk_mem_pages(size);
2862 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2863 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2865 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2868 EXPORT_SYMBOL(__sk_mem_schedule);
2871 * __sk_mem_reduce_allocated - reclaim memory_allocated
2873 * @amount: number of quanta
2875 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2877 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2879 sk_memory_allocated_sub(sk, amount);
2881 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2882 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2884 if (sk_under_global_memory_pressure(sk) &&
2885 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2886 sk_leave_memory_pressure(sk);
2888 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2891 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2893 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2895 void __sk_mem_reclaim(struct sock *sk, int amount)
2897 amount >>= SK_MEM_QUANTUM_SHIFT;
2898 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2899 __sk_mem_reduce_allocated(sk, amount);
2901 EXPORT_SYMBOL(__sk_mem_reclaim);
2903 int sk_set_peek_off(struct sock *sk, int val)
2905 WRITE_ONCE(sk->sk_peek_off, val);
2908 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2911 * Set of default routines for initialising struct proto_ops when
2912 * the protocol does not support a particular function. In certain
2913 * cases where it makes no sense for a protocol to have a "do nothing"
2914 * function, some default processing is provided.
2917 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2921 EXPORT_SYMBOL(sock_no_bind);
2923 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2928 EXPORT_SYMBOL(sock_no_connect);
2930 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2934 EXPORT_SYMBOL(sock_no_socketpair);
2936 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2941 EXPORT_SYMBOL(sock_no_accept);
2943 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2948 EXPORT_SYMBOL(sock_no_getname);
2950 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2954 EXPORT_SYMBOL(sock_no_ioctl);
2956 int sock_no_listen(struct socket *sock, int backlog)
2960 EXPORT_SYMBOL(sock_no_listen);
2962 int sock_no_shutdown(struct socket *sock, int how)
2966 EXPORT_SYMBOL(sock_no_shutdown);
2968 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2972 EXPORT_SYMBOL(sock_no_sendmsg);
2974 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2978 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2980 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2985 EXPORT_SYMBOL(sock_no_recvmsg);
2987 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2989 /* Mirror missing mmap method error code */
2992 EXPORT_SYMBOL(sock_no_mmap);
2995 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2996 * various sock-based usage counts.
2998 void __receive_sock(struct file *file)
3000 struct socket *sock;
3002 sock = sock_from_file(file);
3004 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3005 sock_update_classid(&sock->sk->sk_cgrp_data);
3009 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3012 struct msghdr msg = {.msg_flags = flags};
3014 char *kaddr = kmap(page);
3015 iov.iov_base = kaddr + offset;
3017 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3021 EXPORT_SYMBOL(sock_no_sendpage);
3023 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3024 int offset, size_t size, int flags)
3027 struct msghdr msg = {.msg_flags = flags};
3029 char *kaddr = kmap(page);
3031 iov.iov_base = kaddr + offset;
3033 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3037 EXPORT_SYMBOL(sock_no_sendpage_locked);
3040 * Default Socket Callbacks
3043 static void sock_def_wakeup(struct sock *sk)
3045 struct socket_wq *wq;
3048 wq = rcu_dereference(sk->sk_wq);
3049 if (skwq_has_sleeper(wq))
3050 wake_up_interruptible_all(&wq->wait);
3054 static void sock_def_error_report(struct sock *sk)
3056 struct socket_wq *wq;
3059 wq = rcu_dereference(sk->sk_wq);
3060 if (skwq_has_sleeper(wq))
3061 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3062 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3066 void sock_def_readable(struct sock *sk)
3068 struct socket_wq *wq;
3071 wq = rcu_dereference(sk->sk_wq);
3072 if (skwq_has_sleeper(wq))
3073 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3074 EPOLLRDNORM | EPOLLRDBAND);
3075 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3079 static void sock_def_write_space(struct sock *sk)
3081 struct socket_wq *wq;
3085 /* Do not wake up a writer until he can make "significant"
3088 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
3089 wq = rcu_dereference(sk->sk_wq);
3090 if (skwq_has_sleeper(wq))
3091 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3092 EPOLLWRNORM | EPOLLWRBAND);
3094 /* Should agree with poll, otherwise some programs break */
3095 if (sock_writeable(sk))
3096 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3102 static void sock_def_destruct(struct sock *sk)
3106 void sk_send_sigurg(struct sock *sk)
3108 if (sk->sk_socket && sk->sk_socket->file)
3109 if (send_sigurg(&sk->sk_socket->file->f_owner))
3110 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3112 EXPORT_SYMBOL(sk_send_sigurg);
3114 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3115 unsigned long expires)
3117 if (!mod_timer(timer, expires))
3120 EXPORT_SYMBOL(sk_reset_timer);
3122 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3124 if (del_timer(timer))
3127 EXPORT_SYMBOL(sk_stop_timer);
3129 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3131 if (del_timer_sync(timer))
3134 EXPORT_SYMBOL(sk_stop_timer_sync);
3136 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3139 sk->sk_send_head = NULL;
3141 timer_setup(&sk->sk_timer, NULL, 0);
3143 sk->sk_allocation = GFP_KERNEL;
3144 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3145 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3146 sk->sk_state = TCP_CLOSE;
3147 sk_set_socket(sk, sock);
3149 sock_set_flag(sk, SOCK_ZAPPED);
3152 sk->sk_type = sock->type;
3153 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3156 RCU_INIT_POINTER(sk->sk_wq, NULL);
3160 rwlock_init(&sk->sk_callback_lock);
3161 if (sk->sk_kern_sock)
3162 lockdep_set_class_and_name(
3163 &sk->sk_callback_lock,
3164 af_kern_callback_keys + sk->sk_family,
3165 af_family_kern_clock_key_strings[sk->sk_family]);
3167 lockdep_set_class_and_name(
3168 &sk->sk_callback_lock,
3169 af_callback_keys + sk->sk_family,
3170 af_family_clock_key_strings[sk->sk_family]);
3172 sk->sk_state_change = sock_def_wakeup;
3173 sk->sk_data_ready = sock_def_readable;
3174 sk->sk_write_space = sock_def_write_space;
3175 sk->sk_error_report = sock_def_error_report;
3176 sk->sk_destruct = sock_def_destruct;
3178 sk->sk_frag.page = NULL;
3179 sk->sk_frag.offset = 0;
3180 sk->sk_peek_off = -1;
3182 sk->sk_peer_pid = NULL;
3183 sk->sk_peer_cred = NULL;
3184 spin_lock_init(&sk->sk_peer_lock);
3186 sk->sk_write_pending = 0;
3187 sk->sk_rcvlowat = 1;
3188 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3189 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3191 sk->sk_stamp = SK_DEFAULT_STAMP;
3192 #if BITS_PER_LONG==32
3193 seqlock_init(&sk->sk_stamp_seq);
3195 atomic_set(&sk->sk_zckey, 0);
3197 #ifdef CONFIG_NET_RX_BUSY_POLL
3199 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3202 sk->sk_max_pacing_rate = ~0UL;
3203 sk->sk_pacing_rate = ~0UL;
3204 WRITE_ONCE(sk->sk_pacing_shift, 10);
3205 sk->sk_incoming_cpu = -1;
3207 sk_rx_queue_clear(sk);
3209 * Before updating sk_refcnt, we must commit prior changes to memory
3210 * (Documentation/RCU/rculist_nulls.rst for details)
3213 refcount_set(&sk->sk_refcnt, 1);
3214 atomic_set(&sk->sk_drops, 0);
3216 EXPORT_SYMBOL(sock_init_data_uid);
3218 void sock_init_data(struct socket *sock, struct sock *sk)
3221 SOCK_INODE(sock)->i_uid :
3222 make_kuid(sock_net(sk)->user_ns, 0);
3224 sock_init_data_uid(sock, sk, uid);
3226 EXPORT_SYMBOL(sock_init_data);
3228 void lock_sock_nested(struct sock *sk, int subclass)
3230 /* The sk_lock has mutex_lock() semantics here. */
3231 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3234 spin_lock_bh(&sk->sk_lock.slock);
3235 if (sk->sk_lock.owned)
3237 sk->sk_lock.owned = 1;
3238 spin_unlock_bh(&sk->sk_lock.slock);
3240 EXPORT_SYMBOL(lock_sock_nested);
3242 void release_sock(struct sock *sk)
3244 spin_lock_bh(&sk->sk_lock.slock);
3245 if (sk->sk_backlog.tail)
3248 /* Warning : release_cb() might need to release sk ownership,
3249 * ie call sock_release_ownership(sk) before us.
3251 if (sk->sk_prot->release_cb)
3252 sk->sk_prot->release_cb(sk);
3254 sock_release_ownership(sk);
3255 if (waitqueue_active(&sk->sk_lock.wq))
3256 wake_up(&sk->sk_lock.wq);
3257 spin_unlock_bh(&sk->sk_lock.slock);
3259 EXPORT_SYMBOL(release_sock);
3261 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3264 spin_lock_bh(&sk->sk_lock.slock);
3266 if (!sk->sk_lock.owned) {
3268 * Fast path return with bottom halves disabled and
3269 * sock::sk_lock.slock held.
3271 * The 'mutex' is not contended and holding
3272 * sock::sk_lock.slock prevents all other lockers to
3273 * proceed so the corresponding unlock_sock_fast() can
3274 * avoid the slow path of release_sock() completely and
3275 * just release slock.
3277 * From a semantical POV this is equivalent to 'acquiring'
3278 * the 'mutex', hence the corresponding lockdep
3279 * mutex_release() has to happen in the fast path of
3280 * unlock_sock_fast().
3286 sk->sk_lock.owned = 1;
3287 __acquire(&sk->sk_lock.slock);
3288 spin_unlock_bh(&sk->sk_lock.slock);
3291 EXPORT_SYMBOL(__lock_sock_fast);
3293 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3294 bool timeval, bool time32)
3296 struct sock *sk = sock->sk;
3297 struct timespec64 ts;
3299 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3300 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3301 if (ts.tv_sec == -1)
3303 if (ts.tv_sec == 0) {
3304 ktime_t kt = ktime_get_real();
3305 sock_write_timestamp(sk, kt);
3306 ts = ktime_to_timespec64(kt);
3312 #ifdef CONFIG_COMPAT_32BIT_TIME
3314 return put_old_timespec32(&ts, userstamp);
3316 #ifdef CONFIG_SPARC64
3317 /* beware of padding in sparc64 timeval */
3318 if (timeval && !in_compat_syscall()) {
3319 struct __kernel_old_timeval __user tv = {
3320 .tv_sec = ts.tv_sec,
3321 .tv_usec = ts.tv_nsec,
3323 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3328 return put_timespec64(&ts, userstamp);
3330 EXPORT_SYMBOL(sock_gettstamp);
3332 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3334 if (!sock_flag(sk, flag)) {
3335 unsigned long previous_flags = sk->sk_flags;
3337 sock_set_flag(sk, flag);
3339 * we just set one of the two flags which require net
3340 * time stamping, but time stamping might have been on
3341 * already because of the other one
3343 if (sock_needs_netstamp(sk) &&
3344 !(previous_flags & SK_FLAGS_TIMESTAMP))
3345 net_enable_timestamp();
3349 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3350 int level, int type)
3352 struct sock_exterr_skb *serr;
3353 struct sk_buff *skb;
3357 skb = sock_dequeue_err_skb(sk);
3363 msg->msg_flags |= MSG_TRUNC;
3366 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3370 sock_recv_timestamp(msg, sk, skb);
3372 serr = SKB_EXT_ERR(skb);
3373 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3375 msg->msg_flags |= MSG_ERRQUEUE;
3383 EXPORT_SYMBOL(sock_recv_errqueue);
3386 * Get a socket option on an socket.
3388 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3389 * asynchronous errors should be reported by getsockopt. We assume
3390 * this means if you specify SO_ERROR (otherwise whats the point of it).
3392 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3393 char __user *optval, int __user *optlen)
3395 struct sock *sk = sock->sk;
3397 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3399 EXPORT_SYMBOL(sock_common_getsockopt);
3401 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3404 struct sock *sk = sock->sk;
3408 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3409 flags & ~MSG_DONTWAIT, &addr_len);
3411 msg->msg_namelen = addr_len;
3414 EXPORT_SYMBOL(sock_common_recvmsg);
3417 * Set socket options on an inet socket.
3419 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3420 sockptr_t optval, unsigned int optlen)
3422 struct sock *sk = sock->sk;
3424 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3426 EXPORT_SYMBOL(sock_common_setsockopt);
3428 void sk_common_release(struct sock *sk)
3430 if (sk->sk_prot->destroy)
3431 sk->sk_prot->destroy(sk);
3434 * Observation: when sk_common_release is called, processes have
3435 * no access to socket. But net still has.
3436 * Step one, detach it from networking:
3438 * A. Remove from hash tables.
3441 sk->sk_prot->unhash(sk);
3444 * In this point socket cannot receive new packets, but it is possible
3445 * that some packets are in flight because some CPU runs receiver and
3446 * did hash table lookup before we unhashed socket. They will achieve
3447 * receive queue and will be purged by socket destructor.
3449 * Also we still have packets pending on receive queue and probably,
3450 * our own packets waiting in device queues. sock_destroy will drain
3451 * receive queue, but transmitted packets will delay socket destruction
3452 * until the last reference will be released.
3457 xfrm_sk_free_policy(sk);
3459 sk_refcnt_debug_release(sk);
3463 EXPORT_SYMBOL(sk_common_release);
3465 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3467 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3469 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3470 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3471 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3472 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3473 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3474 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3475 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3476 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3477 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3480 #ifdef CONFIG_PROC_FS
3481 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3483 int val[PROTO_INUSE_NR];
3486 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3488 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3490 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3492 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3494 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3496 int cpu, idx = prot->inuse_idx;
3499 for_each_possible_cpu(cpu)
3500 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3502 return res >= 0 ? res : 0;
3504 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3506 static void sock_inuse_add(struct net *net, int val)
3508 this_cpu_add(*net->core.sock_inuse, val);
3511 int sock_inuse_get(struct net *net)
3515 for_each_possible_cpu(cpu)
3516 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3521 EXPORT_SYMBOL_GPL(sock_inuse_get);
3523 static int __net_init sock_inuse_init_net(struct net *net)
3525 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3526 if (net->core.prot_inuse == NULL)
3529 net->core.sock_inuse = alloc_percpu(int);
3530 if (net->core.sock_inuse == NULL)
3536 free_percpu(net->core.prot_inuse);
3540 static void __net_exit sock_inuse_exit_net(struct net *net)
3542 free_percpu(net->core.prot_inuse);
3543 free_percpu(net->core.sock_inuse);
3546 static struct pernet_operations net_inuse_ops = {
3547 .init = sock_inuse_init_net,
3548 .exit = sock_inuse_exit_net,
3551 static __init int net_inuse_init(void)
3553 if (register_pernet_subsys(&net_inuse_ops))
3554 panic("Cannot initialize net inuse counters");
3559 core_initcall(net_inuse_init);
3561 static int assign_proto_idx(struct proto *prot)
3563 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3565 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3566 pr_err("PROTO_INUSE_NR exhausted\n");
3570 set_bit(prot->inuse_idx, proto_inuse_idx);
3574 static void release_proto_idx(struct proto *prot)
3576 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3577 clear_bit(prot->inuse_idx, proto_inuse_idx);
3580 static inline int assign_proto_idx(struct proto *prot)
3585 static inline void release_proto_idx(struct proto *prot)
3589 static void sock_inuse_add(struct net *net, int val)
3594 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3598 kfree(twsk_prot->twsk_slab_name);
3599 twsk_prot->twsk_slab_name = NULL;
3600 kmem_cache_destroy(twsk_prot->twsk_slab);
3601 twsk_prot->twsk_slab = NULL;
3604 static int tw_prot_init(const struct proto *prot)
3606 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3611 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3613 if (!twsk_prot->twsk_slab_name)
3616 twsk_prot->twsk_slab =
3617 kmem_cache_create(twsk_prot->twsk_slab_name,
3618 twsk_prot->twsk_obj_size, 0,
3619 SLAB_ACCOUNT | prot->slab_flags,
3621 if (!twsk_prot->twsk_slab) {
3622 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3630 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3634 kfree(rsk_prot->slab_name);
3635 rsk_prot->slab_name = NULL;
3636 kmem_cache_destroy(rsk_prot->slab);
3637 rsk_prot->slab = NULL;
3640 static int req_prot_init(const struct proto *prot)
3642 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3647 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3649 if (!rsk_prot->slab_name)
3652 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3653 rsk_prot->obj_size, 0,
3654 SLAB_ACCOUNT | prot->slab_flags,
3657 if (!rsk_prot->slab) {
3658 pr_crit("%s: Can't create request sock SLAB cache!\n",
3665 int proto_register(struct proto *prot, int alloc_slab)
3670 prot->slab = kmem_cache_create_usercopy(prot->name,
3672 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3674 prot->useroffset, prot->usersize,
3677 if (prot->slab == NULL) {
3678 pr_crit("%s: Can't create sock SLAB cache!\n",
3683 if (req_prot_init(prot))
3684 goto out_free_request_sock_slab;
3686 if (tw_prot_init(prot))
3687 goto out_free_timewait_sock_slab;
3690 mutex_lock(&proto_list_mutex);
3691 ret = assign_proto_idx(prot);
3693 mutex_unlock(&proto_list_mutex);
3694 goto out_free_timewait_sock_slab;
3696 list_add(&prot->node, &proto_list);
3697 mutex_unlock(&proto_list_mutex);
3700 out_free_timewait_sock_slab:
3702 tw_prot_cleanup(prot->twsk_prot);
3703 out_free_request_sock_slab:
3705 req_prot_cleanup(prot->rsk_prot);
3707 kmem_cache_destroy(prot->slab);
3713 EXPORT_SYMBOL(proto_register);
3715 void proto_unregister(struct proto *prot)
3717 mutex_lock(&proto_list_mutex);
3718 release_proto_idx(prot);
3719 list_del(&prot->node);
3720 mutex_unlock(&proto_list_mutex);
3722 kmem_cache_destroy(prot->slab);
3725 req_prot_cleanup(prot->rsk_prot);
3726 tw_prot_cleanup(prot->twsk_prot);
3728 EXPORT_SYMBOL(proto_unregister);
3730 int sock_load_diag_module(int family, int protocol)
3733 if (!sock_is_registered(family))
3736 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3737 NETLINK_SOCK_DIAG, family);
3741 if (family == AF_INET &&
3742 protocol != IPPROTO_RAW &&
3743 protocol < MAX_INET_PROTOS &&
3744 !rcu_access_pointer(inet_protos[protocol]))
3748 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3749 NETLINK_SOCK_DIAG, family, protocol);
3751 EXPORT_SYMBOL(sock_load_diag_module);
3753 #ifdef CONFIG_PROC_FS
3754 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3755 __acquires(proto_list_mutex)
3757 mutex_lock(&proto_list_mutex);
3758 return seq_list_start_head(&proto_list, *pos);
3761 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3763 return seq_list_next(v, &proto_list, pos);
3766 static void proto_seq_stop(struct seq_file *seq, void *v)
3767 __releases(proto_list_mutex)
3769 mutex_unlock(&proto_list_mutex);
3772 static char proto_method_implemented(const void *method)
3774 return method == NULL ? 'n' : 'y';
3776 static long sock_prot_memory_allocated(struct proto *proto)
3778 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3781 static const char *sock_prot_memory_pressure(struct proto *proto)
3783 return proto->memory_pressure != NULL ?
3784 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3787 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3790 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3791 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3794 sock_prot_inuse_get(seq_file_net(seq), proto),
3795 sock_prot_memory_allocated(proto),
3796 sock_prot_memory_pressure(proto),
3798 proto->slab == NULL ? "no" : "yes",
3799 module_name(proto->owner),
3800 proto_method_implemented(proto->close),
3801 proto_method_implemented(proto->connect),
3802 proto_method_implemented(proto->disconnect),
3803 proto_method_implemented(proto->accept),
3804 proto_method_implemented(proto->ioctl),
3805 proto_method_implemented(proto->init),
3806 proto_method_implemented(proto->destroy),
3807 proto_method_implemented(proto->shutdown),
3808 proto_method_implemented(proto->setsockopt),
3809 proto_method_implemented(proto->getsockopt),
3810 proto_method_implemented(proto->sendmsg),
3811 proto_method_implemented(proto->recvmsg),
3812 proto_method_implemented(proto->sendpage),
3813 proto_method_implemented(proto->bind),
3814 proto_method_implemented(proto->backlog_rcv),
3815 proto_method_implemented(proto->hash),
3816 proto_method_implemented(proto->unhash),
3817 proto_method_implemented(proto->get_port),
3818 proto_method_implemented(proto->enter_memory_pressure));
3821 static int proto_seq_show(struct seq_file *seq, void *v)
3823 if (v == &proto_list)
3824 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3833 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3835 proto_seq_printf(seq, list_entry(v, struct proto, node));
3839 static const struct seq_operations proto_seq_ops = {
3840 .start = proto_seq_start,
3841 .next = proto_seq_next,
3842 .stop = proto_seq_stop,
3843 .show = proto_seq_show,
3846 static __net_init int proto_init_net(struct net *net)
3848 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3849 sizeof(struct seq_net_private)))
3855 static __net_exit void proto_exit_net(struct net *net)
3857 remove_proc_entry("protocols", net->proc_net);
3861 static __net_initdata struct pernet_operations proto_net_ops = {
3862 .init = proto_init_net,
3863 .exit = proto_exit_net,
3866 static int __init proto_init(void)
3868 return register_pernet_subsys(&proto_net_ops);
3871 subsys_initcall(proto_init);
3873 #endif /* PROC_FS */
3875 #ifdef CONFIG_NET_RX_BUSY_POLL
3876 bool sk_busy_loop_end(void *p, unsigned long start_time)
3878 struct sock *sk = p;
3880 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3881 sk_busy_loop_timeout(sk, start_time);
3883 EXPORT_SYMBOL(sk_busy_loop_end);
3884 #endif /* CONFIG_NET_RX_BUSY_POLL */
3886 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3888 if (!sk->sk_prot->bind_add)
3890 return sk->sk_prot->bind_add(sk, addr, addr_len);
3892 EXPORT_SYMBOL(sock_bind_add);