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 switch (sk->sk_family) {
722 return inet_sk(sk)->mc_loop;
723 #if IS_ENABLED(CONFIG_IPV6)
725 return inet6_sk(sk)->mc_loop;
731 EXPORT_SYMBOL(sk_mc_loop);
733 void sock_set_reuseaddr(struct sock *sk)
736 sk->sk_reuse = SK_CAN_REUSE;
739 EXPORT_SYMBOL(sock_set_reuseaddr);
741 void sock_set_reuseport(struct sock *sk)
744 sk->sk_reuseport = true;
747 EXPORT_SYMBOL(sock_set_reuseport);
749 void sock_no_linger(struct sock *sk)
752 sk->sk_lingertime = 0;
753 sock_set_flag(sk, SOCK_LINGER);
756 EXPORT_SYMBOL(sock_no_linger);
758 void sock_set_priority(struct sock *sk, u32 priority)
761 sk->sk_priority = priority;
764 EXPORT_SYMBOL(sock_set_priority);
766 void sock_set_sndtimeo(struct sock *sk, s64 secs)
769 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
770 sk->sk_sndtimeo = secs * HZ;
772 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
775 EXPORT_SYMBOL(sock_set_sndtimeo);
777 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
780 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
781 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
782 sock_set_flag(sk, SOCK_RCVTSTAMP);
783 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
785 sock_reset_flag(sk, SOCK_RCVTSTAMP);
786 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
790 void sock_enable_timestamps(struct sock *sk)
793 __sock_set_timestamps(sk, true, false, true);
796 EXPORT_SYMBOL(sock_enable_timestamps);
798 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
801 case SO_TIMESTAMP_OLD:
802 __sock_set_timestamps(sk, valbool, false, false);
804 case SO_TIMESTAMP_NEW:
805 __sock_set_timestamps(sk, valbool, true, false);
807 case SO_TIMESTAMPNS_OLD:
808 __sock_set_timestamps(sk, valbool, false, true);
810 case SO_TIMESTAMPNS_NEW:
811 __sock_set_timestamps(sk, valbool, true, true);
816 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
818 struct net *net = sock_net(sk);
819 struct net_device *dev = NULL;
824 if (sk->sk_bound_dev_if)
825 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
828 pr_err("%s: sock not bind to device\n", __func__);
832 num = ethtool_get_phc_vclocks(dev, &vclock_index);
835 for (i = 0; i < num; i++) {
836 if (*(vclock_index + i) == phc_index) {
848 sk->sk_bind_phc = phc_index;
853 int sock_set_timestamping(struct sock *sk, int optname,
854 struct so_timestamping timestamping)
856 int val = timestamping.flags;
859 if (val & ~SOF_TIMESTAMPING_MASK)
862 if (val & SOF_TIMESTAMPING_OPT_ID &&
863 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
864 if (sk->sk_protocol == IPPROTO_TCP &&
865 sk->sk_type == SOCK_STREAM) {
866 if ((1 << sk->sk_state) &
867 (TCPF_CLOSE | TCPF_LISTEN))
869 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
871 atomic_set(&sk->sk_tskey, 0);
875 if (val & SOF_TIMESTAMPING_OPT_STATS &&
876 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
879 if (val & SOF_TIMESTAMPING_BIND_PHC) {
880 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
885 sk->sk_tsflags = val;
886 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
888 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
889 sock_enable_timestamp(sk,
890 SOCK_TIMESTAMPING_RX_SOFTWARE);
892 sock_disable_timestamp(sk,
893 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
897 void sock_set_keepalive(struct sock *sk)
900 if (sk->sk_prot->keepalive)
901 sk->sk_prot->keepalive(sk, true);
902 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
905 EXPORT_SYMBOL(sock_set_keepalive);
907 static void __sock_set_rcvbuf(struct sock *sk, int val)
909 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
910 * as a negative value.
912 val = min_t(int, val, INT_MAX / 2);
913 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
915 /* We double it on the way in to account for "struct sk_buff" etc.
916 * overhead. Applications assume that the SO_RCVBUF setting they make
917 * will allow that much actual data to be received on that socket.
919 * Applications are unaware that "struct sk_buff" and other overheads
920 * allocate from the receive buffer during socket buffer allocation.
922 * And after considering the possible alternatives, returning the value
923 * we actually used in getsockopt is the most desirable behavior.
925 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
928 void sock_set_rcvbuf(struct sock *sk, int val)
931 __sock_set_rcvbuf(sk, val);
934 EXPORT_SYMBOL(sock_set_rcvbuf);
936 static void __sock_set_mark(struct sock *sk, u32 val)
938 if (val != sk->sk_mark) {
944 void sock_set_mark(struct sock *sk, u32 val)
947 __sock_set_mark(sk, val);
950 EXPORT_SYMBOL(sock_set_mark);
953 * This is meant for all protocols to use and covers goings on
954 * at the socket level. Everything here is generic.
957 int sock_setsockopt(struct socket *sock, int level, int optname,
958 sockptr_t optval, unsigned int optlen)
960 struct so_timestamping timestamping;
961 struct sock_txtime sk_txtime;
962 struct sock *sk = sock->sk;
969 * Options without arguments
972 if (optname == SO_BINDTODEVICE)
973 return sock_setbindtodevice(sk, optval, optlen);
975 if (optlen < sizeof(int))
978 if (copy_from_sockptr(&val, optval, sizeof(val)))
981 valbool = val ? 1 : 0;
987 if (val && !capable(CAP_NET_ADMIN))
990 sock_valbool_flag(sk, SOCK_DBG, valbool);
993 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
996 sk->sk_reuseport = valbool;
1005 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1009 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1012 /* Don't error on this BSD doesn't and if you think
1013 * about it this is right. Otherwise apps have to
1014 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1015 * are treated in BSD as hints
1017 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1019 /* Ensure val * 2 fits into an int, to prevent max_t()
1020 * from treating it as a negative value.
1022 val = min_t(int, val, INT_MAX / 2);
1023 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1024 WRITE_ONCE(sk->sk_sndbuf,
1025 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1026 /* Wake up sending tasks if we upped the value. */
1027 sk->sk_write_space(sk);
1030 case SO_SNDBUFFORCE:
1031 if (!capable(CAP_NET_ADMIN)) {
1036 /* No negative values (to prevent underflow, as val will be
1044 /* Don't error on this BSD doesn't and if you think
1045 * about it this is right. Otherwise apps have to
1046 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1047 * are treated in BSD as hints
1049 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1052 case SO_RCVBUFFORCE:
1053 if (!capable(CAP_NET_ADMIN)) {
1058 /* No negative values (to prevent underflow, as val will be
1061 __sock_set_rcvbuf(sk, max(val, 0));
1065 if (sk->sk_prot->keepalive)
1066 sk->sk_prot->keepalive(sk, valbool);
1067 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1071 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1075 sk->sk_no_check_tx = valbool;
1079 if ((val >= 0 && val <= 6) ||
1080 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1081 sk->sk_priority = val;
1087 if (optlen < sizeof(ling)) {
1088 ret = -EINVAL; /* 1003.1g */
1091 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1096 sock_reset_flag(sk, SOCK_LINGER);
1098 #if (BITS_PER_LONG == 32)
1099 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1100 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1103 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1104 sock_set_flag(sk, SOCK_LINGER);
1113 set_bit(SOCK_PASSCRED, &sock->flags);
1115 clear_bit(SOCK_PASSCRED, &sock->flags);
1118 case SO_TIMESTAMP_OLD:
1119 case SO_TIMESTAMP_NEW:
1120 case SO_TIMESTAMPNS_OLD:
1121 case SO_TIMESTAMPNS_NEW:
1122 sock_set_timestamp(sk, optname, valbool);
1125 case SO_TIMESTAMPING_NEW:
1126 case SO_TIMESTAMPING_OLD:
1127 if (optlen == sizeof(timestamping)) {
1128 if (copy_from_sockptr(×tamping, optval,
1129 sizeof(timestamping))) {
1134 memset(×tamping, 0, sizeof(timestamping));
1135 timestamping.flags = val;
1137 ret = sock_set_timestamping(sk, optname, timestamping);
1143 if (sock->ops->set_rcvlowat)
1144 ret = sock->ops->set_rcvlowat(sk, val);
1146 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1149 case SO_RCVTIMEO_OLD:
1150 case SO_RCVTIMEO_NEW:
1151 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1152 optlen, optname == SO_RCVTIMEO_OLD);
1155 case SO_SNDTIMEO_OLD:
1156 case SO_SNDTIMEO_NEW:
1157 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1158 optlen, optname == SO_SNDTIMEO_OLD);
1161 case SO_ATTACH_FILTER: {
1162 struct sock_fprog fprog;
1164 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1166 ret = sk_attach_filter(&fprog, sk);
1171 if (optlen == sizeof(u32)) {
1175 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1178 ret = sk_attach_bpf(ufd, sk);
1182 case SO_ATTACH_REUSEPORT_CBPF: {
1183 struct sock_fprog fprog;
1185 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1187 ret = sk_reuseport_attach_filter(&fprog, sk);
1190 case SO_ATTACH_REUSEPORT_EBPF:
1192 if (optlen == sizeof(u32)) {
1196 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1199 ret = sk_reuseport_attach_bpf(ufd, sk);
1203 case SO_DETACH_REUSEPORT_BPF:
1204 ret = reuseport_detach_prog(sk);
1207 case SO_DETACH_FILTER:
1208 ret = sk_detach_filter(sk);
1211 case SO_LOCK_FILTER:
1212 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1215 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1220 set_bit(SOCK_PASSSEC, &sock->flags);
1222 clear_bit(SOCK_PASSSEC, &sock->flags);
1225 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1230 __sock_set_mark(sk, val);
1234 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1237 case SO_WIFI_STATUS:
1238 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1242 if (sock->ops->set_peek_off)
1243 ret = sock->ops->set_peek_off(sk, val);
1249 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1252 case SO_SELECT_ERR_QUEUE:
1253 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1256 #ifdef CONFIG_NET_RX_BUSY_POLL
1258 /* allow unprivileged users to decrease the value */
1259 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1265 WRITE_ONCE(sk->sk_ll_usec, val);
1268 case SO_PREFER_BUSY_POLL:
1269 if (valbool && !capable(CAP_NET_ADMIN))
1272 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1274 case SO_BUSY_POLL_BUDGET:
1275 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1278 if (val < 0 || val > U16_MAX)
1281 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1286 case SO_MAX_PACING_RATE:
1288 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1290 if (sizeof(ulval) != sizeof(val) &&
1291 optlen >= sizeof(ulval) &&
1292 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1297 cmpxchg(&sk->sk_pacing_status,
1300 sk->sk_max_pacing_rate = ulval;
1301 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1304 case SO_INCOMING_CPU:
1305 reuseport_update_incoming_cpu(sk, val);
1310 dst_negative_advice(sk);
1314 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1315 if (!((sk->sk_type == SOCK_STREAM &&
1316 sk->sk_protocol == IPPROTO_TCP) ||
1317 (sk->sk_type == SOCK_DGRAM &&
1318 sk->sk_protocol == IPPROTO_UDP)))
1320 } else if (sk->sk_family != PF_RDS) {
1324 if (val < 0 || val > 1)
1327 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1332 if (optlen != sizeof(struct sock_txtime)) {
1335 } else if (copy_from_sockptr(&sk_txtime, optval,
1336 sizeof(struct sock_txtime))) {
1339 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1343 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1344 * scheduler has enough safe guards.
1346 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1347 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1351 sock_valbool_flag(sk, SOCK_TXTIME, true);
1352 sk->sk_clockid = sk_txtime.clockid;
1353 sk->sk_txtime_deadline_mode =
1354 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1355 sk->sk_txtime_report_errors =
1356 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1359 case SO_BINDTOIFINDEX:
1360 ret = sock_bindtoindex_locked(sk, val);
1364 if (val & ~SOCK_BUF_LOCK_MASK) {
1368 sk->sk_userlocks = val | (sk->sk_userlocks &
1369 ~SOCK_BUF_LOCK_MASK);
1379 EXPORT_SYMBOL(sock_setsockopt);
1381 static const struct cred *sk_get_peer_cred(struct sock *sk)
1383 const struct cred *cred;
1385 spin_lock(&sk->sk_peer_lock);
1386 cred = get_cred(sk->sk_peer_cred);
1387 spin_unlock(&sk->sk_peer_lock);
1392 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1393 struct ucred *ucred)
1395 ucred->pid = pid_vnr(pid);
1396 ucred->uid = ucred->gid = -1;
1398 struct user_namespace *current_ns = current_user_ns();
1400 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1401 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1405 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1407 struct user_namespace *user_ns = current_user_ns();
1410 for (i = 0; i < src->ngroups; i++)
1411 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1417 int sock_getsockopt(struct socket *sock, int level, int optname,
1418 char __user *optval, int __user *optlen)
1420 struct sock *sk = sock->sk;
1425 unsigned long ulval;
1427 struct old_timeval32 tm32;
1428 struct __kernel_old_timeval tm;
1429 struct __kernel_sock_timeval stm;
1430 struct sock_txtime txtime;
1431 struct so_timestamping timestamping;
1434 int lv = sizeof(int);
1437 if (get_user(len, optlen))
1442 memset(&v, 0, sizeof(v));
1446 v.val = sock_flag(sk, SOCK_DBG);
1450 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1454 v.val = sock_flag(sk, SOCK_BROADCAST);
1458 v.val = sk->sk_sndbuf;
1462 v.val = sk->sk_rcvbuf;
1466 v.val = sk->sk_reuse;
1470 v.val = sk->sk_reuseport;
1474 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1478 v.val = sk->sk_type;
1482 v.val = sk->sk_protocol;
1486 v.val = sk->sk_family;
1490 v.val = -sock_error(sk);
1492 v.val = xchg(&sk->sk_err_soft, 0);
1496 v.val = sock_flag(sk, SOCK_URGINLINE);
1500 v.val = sk->sk_no_check_tx;
1504 v.val = sk->sk_priority;
1508 lv = sizeof(v.ling);
1509 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1510 v.ling.l_linger = sk->sk_lingertime / HZ;
1516 case SO_TIMESTAMP_OLD:
1517 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1518 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1519 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1522 case SO_TIMESTAMPNS_OLD:
1523 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1526 case SO_TIMESTAMP_NEW:
1527 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1530 case SO_TIMESTAMPNS_NEW:
1531 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1534 case SO_TIMESTAMPING_OLD:
1535 lv = sizeof(v.timestamping);
1536 v.timestamping.flags = sk->sk_tsflags;
1537 v.timestamping.bind_phc = sk->sk_bind_phc;
1540 case SO_RCVTIMEO_OLD:
1541 case SO_RCVTIMEO_NEW:
1542 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1545 case SO_SNDTIMEO_OLD:
1546 case SO_SNDTIMEO_NEW:
1547 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1551 v.val = sk->sk_rcvlowat;
1559 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1564 struct ucred peercred;
1565 if (len > sizeof(peercred))
1566 len = sizeof(peercred);
1568 spin_lock(&sk->sk_peer_lock);
1569 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1570 spin_unlock(&sk->sk_peer_lock);
1572 if (copy_to_user(optval, &peercred, len))
1579 const struct cred *cred;
1582 cred = sk_get_peer_cred(sk);
1586 n = cred->group_info->ngroups;
1587 if (len < n * sizeof(gid_t)) {
1588 len = n * sizeof(gid_t);
1590 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1592 len = n * sizeof(gid_t);
1594 ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1605 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1610 if (copy_to_user(optval, address, len))
1615 /* Dubious BSD thing... Probably nobody even uses it, but
1616 * the UNIX standard wants it for whatever reason... -DaveM
1619 v.val = sk->sk_state == TCP_LISTEN;
1623 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1627 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1630 v.val = sk->sk_mark;
1634 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1637 case SO_WIFI_STATUS:
1638 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1642 if (!sock->ops->set_peek_off)
1645 v.val = sk->sk_peek_off;
1648 v.val = sock_flag(sk, SOCK_NOFCS);
1651 case SO_BINDTODEVICE:
1652 return sock_getbindtodevice(sk, optval, optlen, len);
1655 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1661 case SO_LOCK_FILTER:
1662 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1665 case SO_BPF_EXTENSIONS:
1666 v.val = bpf_tell_extensions();
1669 case SO_SELECT_ERR_QUEUE:
1670 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1673 #ifdef CONFIG_NET_RX_BUSY_POLL
1675 v.val = sk->sk_ll_usec;
1677 case SO_PREFER_BUSY_POLL:
1678 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1682 case SO_MAX_PACING_RATE:
1683 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1684 lv = sizeof(v.ulval);
1685 v.ulval = sk->sk_max_pacing_rate;
1688 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1692 case SO_INCOMING_CPU:
1693 v.val = READ_ONCE(sk->sk_incoming_cpu);
1698 u32 meminfo[SK_MEMINFO_VARS];
1700 sk_get_meminfo(sk, meminfo);
1702 len = min_t(unsigned int, len, sizeof(meminfo));
1703 if (copy_to_user(optval, &meminfo, len))
1709 #ifdef CONFIG_NET_RX_BUSY_POLL
1710 case SO_INCOMING_NAPI_ID:
1711 v.val = READ_ONCE(sk->sk_napi_id);
1713 /* aggregate non-NAPI IDs down to 0 */
1714 if (v.val < MIN_NAPI_ID)
1724 v.val64 = sock_gen_cookie(sk);
1728 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1732 lv = sizeof(v.txtime);
1733 v.txtime.clockid = sk->sk_clockid;
1734 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1735 SOF_TXTIME_DEADLINE_MODE : 0;
1736 v.txtime.flags |= sk->sk_txtime_report_errors ?
1737 SOF_TXTIME_REPORT_ERRORS : 0;
1740 case SO_BINDTOIFINDEX:
1741 v.val = sk->sk_bound_dev_if;
1744 case SO_NETNS_COOKIE:
1748 v.val64 = sock_net(sk)->net_cookie;
1752 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1756 /* We implement the SO_SNDLOWAT etc to not be settable
1759 return -ENOPROTOOPT;
1764 if (copy_to_user(optval, &v, len))
1767 if (put_user(len, optlen))
1773 * Initialize an sk_lock.
1775 * (We also register the sk_lock with the lock validator.)
1777 static inline void sock_lock_init(struct sock *sk)
1779 if (sk->sk_kern_sock)
1780 sock_lock_init_class_and_name(
1782 af_family_kern_slock_key_strings[sk->sk_family],
1783 af_family_kern_slock_keys + sk->sk_family,
1784 af_family_kern_key_strings[sk->sk_family],
1785 af_family_kern_keys + sk->sk_family);
1787 sock_lock_init_class_and_name(
1789 af_family_slock_key_strings[sk->sk_family],
1790 af_family_slock_keys + sk->sk_family,
1791 af_family_key_strings[sk->sk_family],
1792 af_family_keys + sk->sk_family);
1796 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1797 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1798 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1800 static void sock_copy(struct sock *nsk, const struct sock *osk)
1802 const struct proto *prot = READ_ONCE(osk->sk_prot);
1803 #ifdef CONFIG_SECURITY_NETWORK
1804 void *sptr = nsk->sk_security;
1807 /* If we move sk_tx_queue_mapping out of the private section,
1808 * we must check if sk_tx_queue_clear() is called after
1809 * sock_copy() in sk_clone_lock().
1811 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1812 offsetof(struct sock, sk_dontcopy_begin) ||
1813 offsetof(struct sock, sk_tx_queue_mapping) >=
1814 offsetof(struct sock, sk_dontcopy_end));
1816 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1818 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1819 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1821 #ifdef CONFIG_SECURITY_NETWORK
1822 nsk->sk_security = sptr;
1823 security_sk_clone(osk, nsk);
1827 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1831 struct kmem_cache *slab;
1835 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1838 if (want_init_on_alloc(priority))
1839 sk_prot_clear_nulls(sk, prot->obj_size);
1841 sk = kmalloc(prot->obj_size, priority);
1844 if (security_sk_alloc(sk, family, priority))
1847 if (!try_module_get(prot->owner))
1854 security_sk_free(sk);
1857 kmem_cache_free(slab, sk);
1863 static void sk_prot_free(struct proto *prot, struct sock *sk)
1865 struct kmem_cache *slab;
1866 struct module *owner;
1868 owner = prot->owner;
1871 cgroup_sk_free(&sk->sk_cgrp_data);
1872 mem_cgroup_sk_free(sk);
1873 security_sk_free(sk);
1875 kmem_cache_free(slab, sk);
1882 * sk_alloc - All socket objects are allocated here
1883 * @net: the applicable net namespace
1884 * @family: protocol family
1885 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1886 * @prot: struct proto associated with this new sock instance
1887 * @kern: is this to be a kernel socket?
1889 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1890 struct proto *prot, int kern)
1894 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1896 sk->sk_family = family;
1898 * See comment in struct sock definition to understand
1899 * why we need sk_prot_creator -acme
1901 sk->sk_prot = sk->sk_prot_creator = prot;
1902 sk->sk_kern_sock = kern;
1904 sk->sk_net_refcnt = kern ? 0 : 1;
1905 if (likely(sk->sk_net_refcnt)) {
1907 sock_inuse_add(net, 1);
1910 sock_net_set(sk, net);
1911 refcount_set(&sk->sk_wmem_alloc, 1);
1913 mem_cgroup_sk_alloc(sk);
1914 cgroup_sk_alloc(&sk->sk_cgrp_data);
1915 sock_update_classid(&sk->sk_cgrp_data);
1916 sock_update_netprioidx(&sk->sk_cgrp_data);
1917 sk_tx_queue_clear(sk);
1922 EXPORT_SYMBOL(sk_alloc);
1924 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1925 * grace period. This is the case for UDP sockets and TCP listeners.
1927 static void __sk_destruct(struct rcu_head *head)
1929 struct sock *sk = container_of(head, struct sock, sk_rcu);
1930 struct sk_filter *filter;
1932 if (sk->sk_destruct)
1933 sk->sk_destruct(sk);
1935 filter = rcu_dereference_check(sk->sk_filter,
1936 refcount_read(&sk->sk_wmem_alloc) == 0);
1938 sk_filter_uncharge(sk, filter);
1939 RCU_INIT_POINTER(sk->sk_filter, NULL);
1942 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1944 #ifdef CONFIG_BPF_SYSCALL
1945 bpf_sk_storage_free(sk);
1948 if (atomic_read(&sk->sk_omem_alloc))
1949 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1950 __func__, atomic_read(&sk->sk_omem_alloc));
1952 if (sk->sk_frag.page) {
1953 put_page(sk->sk_frag.page);
1954 sk->sk_frag.page = NULL;
1957 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
1958 put_cred(sk->sk_peer_cred);
1959 put_pid(sk->sk_peer_pid);
1961 if (likely(sk->sk_net_refcnt))
1962 put_net(sock_net(sk));
1963 sk_prot_free(sk->sk_prot_creator, sk);
1966 void sk_destruct(struct sock *sk)
1968 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1970 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1971 reuseport_detach_sock(sk);
1972 use_call_rcu = true;
1976 call_rcu(&sk->sk_rcu, __sk_destruct);
1978 __sk_destruct(&sk->sk_rcu);
1981 static void __sk_free(struct sock *sk)
1983 if (likely(sk->sk_net_refcnt))
1984 sock_inuse_add(sock_net(sk), -1);
1986 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1987 sock_diag_broadcast_destroy(sk);
1992 void sk_free(struct sock *sk)
1995 * We subtract one from sk_wmem_alloc and can know if
1996 * some packets are still in some tx queue.
1997 * If not null, sock_wfree() will call __sk_free(sk) later
1999 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2002 EXPORT_SYMBOL(sk_free);
2004 static void sk_init_common(struct sock *sk)
2006 skb_queue_head_init(&sk->sk_receive_queue);
2007 skb_queue_head_init(&sk->sk_write_queue);
2008 skb_queue_head_init(&sk->sk_error_queue);
2010 rwlock_init(&sk->sk_callback_lock);
2011 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2012 af_rlock_keys + sk->sk_family,
2013 af_family_rlock_key_strings[sk->sk_family]);
2014 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2015 af_wlock_keys + sk->sk_family,
2016 af_family_wlock_key_strings[sk->sk_family]);
2017 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2018 af_elock_keys + sk->sk_family,
2019 af_family_elock_key_strings[sk->sk_family]);
2020 lockdep_set_class_and_name(&sk->sk_callback_lock,
2021 af_callback_keys + sk->sk_family,
2022 af_family_clock_key_strings[sk->sk_family]);
2026 * sk_clone_lock - clone a socket, and lock its clone
2027 * @sk: the socket to clone
2028 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2030 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2032 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2034 struct proto *prot = READ_ONCE(sk->sk_prot);
2035 struct sk_filter *filter;
2036 bool is_charged = true;
2039 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2043 sock_copy(newsk, sk);
2045 newsk->sk_prot_creator = prot;
2048 if (likely(newsk->sk_net_refcnt)) {
2049 get_net(sock_net(newsk));
2050 sock_inuse_add(sock_net(newsk), 1);
2052 sk_node_init(&newsk->sk_node);
2053 sock_lock_init(newsk);
2054 bh_lock_sock(newsk);
2055 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2056 newsk->sk_backlog.len = 0;
2058 atomic_set(&newsk->sk_rmem_alloc, 0);
2060 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2061 refcount_set(&newsk->sk_wmem_alloc, 1);
2063 atomic_set(&newsk->sk_omem_alloc, 0);
2064 sk_init_common(newsk);
2066 newsk->sk_dst_cache = NULL;
2067 newsk->sk_dst_pending_confirm = 0;
2068 newsk->sk_wmem_queued = 0;
2069 newsk->sk_forward_alloc = 0;
2070 atomic_set(&newsk->sk_drops, 0);
2071 newsk->sk_send_head = NULL;
2072 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2073 atomic_set(&newsk->sk_zckey, 0);
2075 sock_reset_flag(newsk, SOCK_DONE);
2077 /* sk->sk_memcg will be populated at accept() time */
2078 newsk->sk_memcg = NULL;
2080 cgroup_sk_clone(&newsk->sk_cgrp_data);
2083 filter = rcu_dereference(sk->sk_filter);
2085 /* though it's an empty new sock, the charging may fail
2086 * if sysctl_optmem_max was changed between creation of
2087 * original socket and cloning
2089 is_charged = sk_filter_charge(newsk, filter);
2090 RCU_INIT_POINTER(newsk->sk_filter, filter);
2093 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2094 /* We need to make sure that we don't uncharge the new
2095 * socket if we couldn't charge it in the first place
2096 * as otherwise we uncharge the parent's filter.
2099 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2100 sk_free_unlock_clone(newsk);
2104 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2106 if (bpf_sk_storage_clone(sk, newsk)) {
2107 sk_free_unlock_clone(newsk);
2112 /* Clear sk_user_data if parent had the pointer tagged
2113 * as not suitable for copying when cloning.
2115 if (sk_user_data_is_nocopy(newsk))
2116 newsk->sk_user_data = NULL;
2119 newsk->sk_err_soft = 0;
2120 newsk->sk_priority = 0;
2121 newsk->sk_incoming_cpu = raw_smp_processor_id();
2123 /* Before updating sk_refcnt, we must commit prior changes to memory
2124 * (Documentation/RCU/rculist_nulls.rst for details)
2127 refcount_set(&newsk->sk_refcnt, 2);
2129 /* Increment the counter in the same struct proto as the master
2130 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2131 * is the same as sk->sk_prot->socks, as this field was copied
2134 * This _changes_ the previous behaviour, where
2135 * tcp_create_openreq_child always was incrementing the
2136 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2137 * to be taken into account in all callers. -acme
2139 sk_refcnt_debug_inc(newsk);
2140 sk_set_socket(newsk, NULL);
2141 sk_tx_queue_clear(newsk);
2142 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2144 if (newsk->sk_prot->sockets_allocated)
2145 sk_sockets_allocated_inc(newsk);
2147 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2148 net_enable_timestamp();
2152 EXPORT_SYMBOL_GPL(sk_clone_lock);
2154 void sk_free_unlock_clone(struct sock *sk)
2156 /* It is still raw copy of parent, so invalidate
2157 * destructor and make plain sk_free() */
2158 sk->sk_destruct = NULL;
2162 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2164 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2168 sk_dst_set(sk, dst);
2169 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2170 if (sk->sk_route_caps & NETIF_F_GSO)
2171 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2172 sk->sk_route_caps &= ~sk->sk_route_nocaps;
2173 if (sk_can_gso(sk)) {
2174 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2175 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2177 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2178 sk->sk_gso_max_size = dst->dev->gso_max_size;
2179 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2182 sk->sk_gso_max_segs = max_segs;
2184 EXPORT_SYMBOL_GPL(sk_setup_caps);
2187 * Simple resource managers for sockets.
2192 * Write buffer destructor automatically called from kfree_skb.
2194 void sock_wfree(struct sk_buff *skb)
2196 struct sock *sk = skb->sk;
2197 unsigned int len = skb->truesize;
2199 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2201 * Keep a reference on sk_wmem_alloc, this will be released
2202 * after sk_write_space() call
2204 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2205 sk->sk_write_space(sk);
2209 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2210 * could not do because of in-flight packets
2212 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2215 EXPORT_SYMBOL(sock_wfree);
2217 /* This variant of sock_wfree() is used by TCP,
2218 * since it sets SOCK_USE_WRITE_QUEUE.
2220 void __sock_wfree(struct sk_buff *skb)
2222 struct sock *sk = skb->sk;
2224 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2228 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2233 if (unlikely(!sk_fullsock(sk))) {
2234 skb->destructor = sock_edemux;
2239 skb->destructor = sock_wfree;
2240 skb_set_hash_from_sk(skb, sk);
2242 * We used to take a refcount on sk, but following operation
2243 * is enough to guarantee sk_free() wont free this sock until
2244 * all in-flight packets are completed
2246 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2248 EXPORT_SYMBOL(skb_set_owner_w);
2250 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2252 #ifdef CONFIG_TLS_DEVICE
2253 /* Drivers depend on in-order delivery for crypto offload,
2254 * partial orphan breaks out-of-order-OK logic.
2259 return (skb->destructor == sock_wfree ||
2260 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2263 /* This helper is used by netem, as it can hold packets in its
2264 * delay queue. We want to allow the owner socket to send more
2265 * packets, as if they were already TX completed by a typical driver.
2266 * But we also want to keep skb->sk set because some packet schedulers
2267 * rely on it (sch_fq for example).
2269 void skb_orphan_partial(struct sk_buff *skb)
2271 if (skb_is_tcp_pure_ack(skb))
2274 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2279 EXPORT_SYMBOL(skb_orphan_partial);
2282 * Read buffer destructor automatically called from kfree_skb.
2284 void sock_rfree(struct sk_buff *skb)
2286 struct sock *sk = skb->sk;
2287 unsigned int len = skb->truesize;
2289 atomic_sub(len, &sk->sk_rmem_alloc);
2290 sk_mem_uncharge(sk, len);
2292 EXPORT_SYMBOL(sock_rfree);
2295 * Buffer destructor for skbs that are not used directly in read or write
2296 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2298 void sock_efree(struct sk_buff *skb)
2302 EXPORT_SYMBOL(sock_efree);
2304 /* Buffer destructor for prefetch/receive path where reference count may
2305 * not be held, e.g. for listen sockets.
2308 void sock_pfree(struct sk_buff *skb)
2310 if (sk_is_refcounted(skb->sk))
2311 sock_gen_put(skb->sk);
2313 EXPORT_SYMBOL(sock_pfree);
2314 #endif /* CONFIG_INET */
2316 kuid_t sock_i_uid(struct sock *sk)
2320 read_lock_bh(&sk->sk_callback_lock);
2321 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2322 read_unlock_bh(&sk->sk_callback_lock);
2325 EXPORT_SYMBOL(sock_i_uid);
2327 unsigned long sock_i_ino(struct sock *sk)
2331 read_lock_bh(&sk->sk_callback_lock);
2332 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2333 read_unlock_bh(&sk->sk_callback_lock);
2336 EXPORT_SYMBOL(sock_i_ino);
2339 * Allocate a skb from the socket's send buffer.
2341 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2345 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2346 struct sk_buff *skb = alloc_skb(size, priority);
2349 skb_set_owner_w(skb, sk);
2355 EXPORT_SYMBOL(sock_wmalloc);
2357 static void sock_ofree(struct sk_buff *skb)
2359 struct sock *sk = skb->sk;
2361 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2364 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2367 struct sk_buff *skb;
2369 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2370 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2371 READ_ONCE(sysctl_optmem_max))
2374 skb = alloc_skb(size, priority);
2378 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2380 skb->destructor = sock_ofree;
2385 * Allocate a memory block from the socket's option memory buffer.
2387 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2389 int optmem_max = READ_ONCE(sysctl_optmem_max);
2391 if ((unsigned int)size <= optmem_max &&
2392 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2394 /* First do the add, to avoid the race if kmalloc
2397 atomic_add(size, &sk->sk_omem_alloc);
2398 mem = kmalloc(size, priority);
2401 atomic_sub(size, &sk->sk_omem_alloc);
2405 EXPORT_SYMBOL(sock_kmalloc);
2407 /* Free an option memory block. Note, we actually want the inline
2408 * here as this allows gcc to detect the nullify and fold away the
2409 * condition entirely.
2411 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2414 if (WARN_ON_ONCE(!mem))
2417 kfree_sensitive(mem);
2420 atomic_sub(size, &sk->sk_omem_alloc);
2423 void sock_kfree_s(struct sock *sk, void *mem, int size)
2425 __sock_kfree_s(sk, mem, size, false);
2427 EXPORT_SYMBOL(sock_kfree_s);
2429 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2431 __sock_kfree_s(sk, mem, size, true);
2433 EXPORT_SYMBOL(sock_kzfree_s);
2435 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2436 I think, these locks should be removed for datagram sockets.
2438 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2442 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2446 if (signal_pending(current))
2448 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2449 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2450 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2452 if (sk->sk_shutdown & SEND_SHUTDOWN)
2456 timeo = schedule_timeout(timeo);
2458 finish_wait(sk_sleep(sk), &wait);
2464 * Generic send/receive buffer handlers
2467 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2468 unsigned long data_len, int noblock,
2469 int *errcode, int max_page_order)
2471 struct sk_buff *skb;
2475 timeo = sock_sndtimeo(sk, noblock);
2477 err = sock_error(sk);
2482 if (sk->sk_shutdown & SEND_SHUTDOWN)
2485 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2488 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2489 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2493 if (signal_pending(current))
2495 timeo = sock_wait_for_wmem(sk, timeo);
2497 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2498 errcode, sk->sk_allocation);
2500 skb_set_owner_w(skb, sk);
2504 err = sock_intr_errno(timeo);
2509 EXPORT_SYMBOL(sock_alloc_send_pskb);
2511 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2512 int noblock, int *errcode)
2514 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2516 EXPORT_SYMBOL(sock_alloc_send_skb);
2518 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2519 struct sockcm_cookie *sockc)
2523 switch (cmsg->cmsg_type) {
2525 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2527 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2529 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2531 case SO_TIMESTAMPING_OLD:
2532 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2535 tsflags = *(u32 *)CMSG_DATA(cmsg);
2536 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2539 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2540 sockc->tsflags |= tsflags;
2543 if (!sock_flag(sk, SOCK_TXTIME))
2545 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2547 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2549 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2551 case SCM_CREDENTIALS:
2558 EXPORT_SYMBOL(__sock_cmsg_send);
2560 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2561 struct sockcm_cookie *sockc)
2563 struct cmsghdr *cmsg;
2566 for_each_cmsghdr(cmsg, msg) {
2567 if (!CMSG_OK(msg, cmsg))
2569 if (cmsg->cmsg_level != SOL_SOCKET)
2571 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2577 EXPORT_SYMBOL(sock_cmsg_send);
2579 static void sk_enter_memory_pressure(struct sock *sk)
2581 if (!sk->sk_prot->enter_memory_pressure)
2584 sk->sk_prot->enter_memory_pressure(sk);
2587 static void sk_leave_memory_pressure(struct sock *sk)
2589 if (sk->sk_prot->leave_memory_pressure) {
2590 sk->sk_prot->leave_memory_pressure(sk);
2592 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2594 if (memory_pressure && READ_ONCE(*memory_pressure))
2595 WRITE_ONCE(*memory_pressure, 0);
2599 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2602 * skb_page_frag_refill - check that a page_frag contains enough room
2603 * @sz: minimum size of the fragment we want to get
2604 * @pfrag: pointer to page_frag
2605 * @gfp: priority for memory allocation
2607 * Note: While this allocator tries to use high order pages, there is
2608 * no guarantee that allocations succeed. Therefore, @sz MUST be
2609 * less or equal than PAGE_SIZE.
2611 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2614 if (page_ref_count(pfrag->page) == 1) {
2618 if (pfrag->offset + sz <= pfrag->size)
2620 put_page(pfrag->page);
2624 if (SKB_FRAG_PAGE_ORDER &&
2625 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2626 /* Avoid direct reclaim but allow kswapd to wake */
2627 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2628 __GFP_COMP | __GFP_NOWARN |
2630 SKB_FRAG_PAGE_ORDER);
2631 if (likely(pfrag->page)) {
2632 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2636 pfrag->page = alloc_page(gfp);
2637 if (likely(pfrag->page)) {
2638 pfrag->size = PAGE_SIZE;
2643 EXPORT_SYMBOL(skb_page_frag_refill);
2645 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2647 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2650 sk_enter_memory_pressure(sk);
2651 sk_stream_moderate_sndbuf(sk);
2654 EXPORT_SYMBOL(sk_page_frag_refill);
2656 void __lock_sock(struct sock *sk)
2657 __releases(&sk->sk_lock.slock)
2658 __acquires(&sk->sk_lock.slock)
2663 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2664 TASK_UNINTERRUPTIBLE);
2665 spin_unlock_bh(&sk->sk_lock.slock);
2667 spin_lock_bh(&sk->sk_lock.slock);
2668 if (!sock_owned_by_user(sk))
2671 finish_wait(&sk->sk_lock.wq, &wait);
2674 void __release_sock(struct sock *sk)
2675 __releases(&sk->sk_lock.slock)
2676 __acquires(&sk->sk_lock.slock)
2678 struct sk_buff *skb, *next;
2680 while ((skb = sk->sk_backlog.head) != NULL) {
2681 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2683 spin_unlock_bh(&sk->sk_lock.slock);
2688 WARN_ON_ONCE(skb_dst_is_noref(skb));
2689 skb_mark_not_on_list(skb);
2690 sk_backlog_rcv(sk, skb);
2695 } while (skb != NULL);
2697 spin_lock_bh(&sk->sk_lock.slock);
2701 * Doing the zeroing here guarantee we can not loop forever
2702 * while a wild producer attempts to flood us.
2704 sk->sk_backlog.len = 0;
2707 void __sk_flush_backlog(struct sock *sk)
2709 spin_lock_bh(&sk->sk_lock.slock);
2711 spin_unlock_bh(&sk->sk_lock.slock);
2715 * sk_wait_data - wait for data to arrive at sk_receive_queue
2716 * @sk: sock to wait on
2717 * @timeo: for how long
2718 * @skb: last skb seen on sk_receive_queue
2720 * Now socket state including sk->sk_err is changed only under lock,
2721 * hence we may omit checks after joining wait queue.
2722 * We check receive queue before schedule() only as optimization;
2723 * it is very likely that release_sock() added new data.
2725 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2727 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2730 add_wait_queue(sk_sleep(sk), &wait);
2731 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2732 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2733 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2734 remove_wait_queue(sk_sleep(sk), &wait);
2737 EXPORT_SYMBOL(sk_wait_data);
2740 * __sk_mem_raise_allocated - increase memory_allocated
2742 * @size: memory size to allocate
2743 * @amt: pages to allocate
2744 * @kind: allocation type
2746 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2748 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2750 struct proto *prot = sk->sk_prot;
2751 long allocated = sk_memory_allocated_add(sk, amt);
2752 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2753 bool charged = true;
2756 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2757 gfp_memcg_charge())))
2758 goto suppress_allocation;
2761 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2762 sk_leave_memory_pressure(sk);
2766 /* Under pressure. */
2767 if (allocated > sk_prot_mem_limits(sk, 1))
2768 sk_enter_memory_pressure(sk);
2770 /* Over hard limit. */
2771 if (allocated > sk_prot_mem_limits(sk, 2))
2772 goto suppress_allocation;
2774 /* guarantee minimum buffer size under pressure */
2775 if (kind == SK_MEM_RECV) {
2776 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2779 } else { /* SK_MEM_SEND */
2780 int wmem0 = sk_get_wmem0(sk, prot);
2782 if (sk->sk_type == SOCK_STREAM) {
2783 if (sk->sk_wmem_queued < wmem0)
2785 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2790 if (sk_has_memory_pressure(sk)) {
2793 if (!sk_under_memory_pressure(sk))
2795 alloc = sk_sockets_allocated_read_positive(sk);
2796 if (sk_prot_mem_limits(sk, 2) > alloc *
2797 sk_mem_pages(sk->sk_wmem_queued +
2798 atomic_read(&sk->sk_rmem_alloc) +
2799 sk->sk_forward_alloc))
2803 suppress_allocation:
2805 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2806 sk_stream_moderate_sndbuf(sk);
2808 /* Fail only if socket is _under_ its sndbuf.
2809 * In this case we cannot block, so that we have to fail.
2811 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
2812 /* Force charge with __GFP_NOFAIL */
2813 if (memcg_charge && !charged) {
2814 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
2815 gfp_memcg_charge() | __GFP_NOFAIL);
2821 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2822 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2824 sk_memory_allocated_sub(sk, amt);
2826 if (memcg_charge && charged)
2827 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2831 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2834 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2836 * @size: memory size to allocate
2837 * @kind: allocation type
2839 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2840 * rmem allocation. This function assumes that protocols which have
2841 * memory_pressure use sk_wmem_queued as write buffer accounting.
2843 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2845 int ret, amt = sk_mem_pages(size);
2847 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2848 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2850 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2853 EXPORT_SYMBOL(__sk_mem_schedule);
2856 * __sk_mem_reduce_allocated - reclaim memory_allocated
2858 * @amount: number of quanta
2860 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2862 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2864 sk_memory_allocated_sub(sk, amount);
2866 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2867 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2869 if (sk_under_memory_pressure(sk) &&
2870 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2871 sk_leave_memory_pressure(sk);
2873 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2876 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2878 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2880 void __sk_mem_reclaim(struct sock *sk, int amount)
2882 amount >>= SK_MEM_QUANTUM_SHIFT;
2883 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2884 __sk_mem_reduce_allocated(sk, amount);
2886 EXPORT_SYMBOL(__sk_mem_reclaim);
2888 int sk_set_peek_off(struct sock *sk, int val)
2890 sk->sk_peek_off = val;
2893 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2896 * Set of default routines for initialising struct proto_ops when
2897 * the protocol does not support a particular function. In certain
2898 * cases where it makes no sense for a protocol to have a "do nothing"
2899 * function, some default processing is provided.
2902 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2906 EXPORT_SYMBOL(sock_no_bind);
2908 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2913 EXPORT_SYMBOL(sock_no_connect);
2915 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2919 EXPORT_SYMBOL(sock_no_socketpair);
2921 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2926 EXPORT_SYMBOL(sock_no_accept);
2928 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2933 EXPORT_SYMBOL(sock_no_getname);
2935 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2939 EXPORT_SYMBOL(sock_no_ioctl);
2941 int sock_no_listen(struct socket *sock, int backlog)
2945 EXPORT_SYMBOL(sock_no_listen);
2947 int sock_no_shutdown(struct socket *sock, int how)
2951 EXPORT_SYMBOL(sock_no_shutdown);
2953 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2957 EXPORT_SYMBOL(sock_no_sendmsg);
2959 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2963 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2965 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2970 EXPORT_SYMBOL(sock_no_recvmsg);
2972 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2974 /* Mirror missing mmap method error code */
2977 EXPORT_SYMBOL(sock_no_mmap);
2980 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2981 * various sock-based usage counts.
2983 void __receive_sock(struct file *file)
2985 struct socket *sock;
2987 sock = sock_from_file(file);
2989 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2990 sock_update_classid(&sock->sk->sk_cgrp_data);
2994 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2997 struct msghdr msg = {.msg_flags = flags};
2999 char *kaddr = kmap(page);
3000 iov.iov_base = kaddr + offset;
3002 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3006 EXPORT_SYMBOL(sock_no_sendpage);
3008 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3009 int offset, size_t size, int flags)
3012 struct msghdr msg = {.msg_flags = flags};
3014 char *kaddr = kmap(page);
3016 iov.iov_base = kaddr + offset;
3018 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3022 EXPORT_SYMBOL(sock_no_sendpage_locked);
3025 * Default Socket Callbacks
3028 static void sock_def_wakeup(struct sock *sk)
3030 struct socket_wq *wq;
3033 wq = rcu_dereference(sk->sk_wq);
3034 if (skwq_has_sleeper(wq))
3035 wake_up_interruptible_all(&wq->wait);
3039 static void sock_def_error_report(struct sock *sk)
3041 struct socket_wq *wq;
3044 wq = rcu_dereference(sk->sk_wq);
3045 if (skwq_has_sleeper(wq))
3046 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3047 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3051 void sock_def_readable(struct sock *sk)
3053 struct socket_wq *wq;
3056 wq = rcu_dereference(sk->sk_wq);
3057 if (skwq_has_sleeper(wq))
3058 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3059 EPOLLRDNORM | EPOLLRDBAND);
3060 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3064 static void sock_def_write_space(struct sock *sk)
3066 struct socket_wq *wq;
3070 /* Do not wake up a writer until he can make "significant"
3073 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
3074 wq = rcu_dereference(sk->sk_wq);
3075 if (skwq_has_sleeper(wq))
3076 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3077 EPOLLWRNORM | EPOLLWRBAND);
3079 /* Should agree with poll, otherwise some programs break */
3080 if (sock_writeable(sk))
3081 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3087 static void sock_def_destruct(struct sock *sk)
3091 void sk_send_sigurg(struct sock *sk)
3093 if (sk->sk_socket && sk->sk_socket->file)
3094 if (send_sigurg(&sk->sk_socket->file->f_owner))
3095 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3097 EXPORT_SYMBOL(sk_send_sigurg);
3099 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3100 unsigned long expires)
3102 if (!mod_timer(timer, expires))
3105 EXPORT_SYMBOL(sk_reset_timer);
3107 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3109 if (del_timer(timer))
3112 EXPORT_SYMBOL(sk_stop_timer);
3114 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3116 if (del_timer_sync(timer))
3119 EXPORT_SYMBOL(sk_stop_timer_sync);
3121 void sock_init_data(struct socket *sock, struct sock *sk)
3124 sk->sk_send_head = NULL;
3126 timer_setup(&sk->sk_timer, NULL, 0);
3128 sk->sk_allocation = GFP_KERNEL;
3129 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3130 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3131 sk->sk_state = TCP_CLOSE;
3132 sk_set_socket(sk, sock);
3134 sock_set_flag(sk, SOCK_ZAPPED);
3137 sk->sk_type = sock->type;
3138 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3140 sk->sk_uid = SOCK_INODE(sock)->i_uid;
3142 RCU_INIT_POINTER(sk->sk_wq, NULL);
3143 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3146 rwlock_init(&sk->sk_callback_lock);
3147 if (sk->sk_kern_sock)
3148 lockdep_set_class_and_name(
3149 &sk->sk_callback_lock,
3150 af_kern_callback_keys + sk->sk_family,
3151 af_family_kern_clock_key_strings[sk->sk_family]);
3153 lockdep_set_class_and_name(
3154 &sk->sk_callback_lock,
3155 af_callback_keys + sk->sk_family,
3156 af_family_clock_key_strings[sk->sk_family]);
3158 sk->sk_state_change = sock_def_wakeup;
3159 sk->sk_data_ready = sock_def_readable;
3160 sk->sk_write_space = sock_def_write_space;
3161 sk->sk_error_report = sock_def_error_report;
3162 sk->sk_destruct = sock_def_destruct;
3164 sk->sk_frag.page = NULL;
3165 sk->sk_frag.offset = 0;
3166 sk->sk_peek_off = -1;
3168 sk->sk_peer_pid = NULL;
3169 sk->sk_peer_cred = NULL;
3170 spin_lock_init(&sk->sk_peer_lock);
3172 sk->sk_write_pending = 0;
3173 sk->sk_rcvlowat = 1;
3174 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3175 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3177 sk->sk_stamp = SK_DEFAULT_STAMP;
3178 #if BITS_PER_LONG==32
3179 seqlock_init(&sk->sk_stamp_seq);
3181 atomic_set(&sk->sk_zckey, 0);
3183 #ifdef CONFIG_NET_RX_BUSY_POLL
3185 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3188 sk->sk_max_pacing_rate = ~0UL;
3189 sk->sk_pacing_rate = ~0UL;
3190 WRITE_ONCE(sk->sk_pacing_shift, 10);
3191 sk->sk_incoming_cpu = -1;
3193 sk_rx_queue_clear(sk);
3195 * Before updating sk_refcnt, we must commit prior changes to memory
3196 * (Documentation/RCU/rculist_nulls.rst for details)
3199 refcount_set(&sk->sk_refcnt, 1);
3200 atomic_set(&sk->sk_drops, 0);
3202 EXPORT_SYMBOL(sock_init_data);
3204 void lock_sock_nested(struct sock *sk, int subclass)
3206 /* The sk_lock has mutex_lock() semantics here. */
3207 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3210 spin_lock_bh(&sk->sk_lock.slock);
3211 if (sk->sk_lock.owned)
3213 sk->sk_lock.owned = 1;
3214 spin_unlock_bh(&sk->sk_lock.slock);
3216 EXPORT_SYMBOL(lock_sock_nested);
3218 void release_sock(struct sock *sk)
3220 spin_lock_bh(&sk->sk_lock.slock);
3221 if (sk->sk_backlog.tail)
3224 /* Warning : release_cb() might need to release sk ownership,
3225 * ie call sock_release_ownership(sk) before us.
3227 if (sk->sk_prot->release_cb)
3228 sk->sk_prot->release_cb(sk);
3230 sock_release_ownership(sk);
3231 if (waitqueue_active(&sk->sk_lock.wq))
3232 wake_up(&sk->sk_lock.wq);
3233 spin_unlock_bh(&sk->sk_lock.slock);
3235 EXPORT_SYMBOL(release_sock);
3237 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3240 spin_lock_bh(&sk->sk_lock.slock);
3242 if (!sk->sk_lock.owned) {
3244 * Fast path return with bottom halves disabled and
3245 * sock::sk_lock.slock held.
3247 * The 'mutex' is not contended and holding
3248 * sock::sk_lock.slock prevents all other lockers to
3249 * proceed so the corresponding unlock_sock_fast() can
3250 * avoid the slow path of release_sock() completely and
3251 * just release slock.
3253 * From a semantical POV this is equivalent to 'acquiring'
3254 * the 'mutex', hence the corresponding lockdep
3255 * mutex_release() has to happen in the fast path of
3256 * unlock_sock_fast().
3262 sk->sk_lock.owned = 1;
3263 __acquire(&sk->sk_lock.slock);
3264 spin_unlock_bh(&sk->sk_lock.slock);
3267 EXPORT_SYMBOL(__lock_sock_fast);
3269 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3270 bool timeval, bool time32)
3272 struct sock *sk = sock->sk;
3273 struct timespec64 ts;
3275 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3276 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3277 if (ts.tv_sec == -1)
3279 if (ts.tv_sec == 0) {
3280 ktime_t kt = ktime_get_real();
3281 sock_write_timestamp(sk, kt);
3282 ts = ktime_to_timespec64(kt);
3288 #ifdef CONFIG_COMPAT_32BIT_TIME
3290 return put_old_timespec32(&ts, userstamp);
3292 #ifdef CONFIG_SPARC64
3293 /* beware of padding in sparc64 timeval */
3294 if (timeval && !in_compat_syscall()) {
3295 struct __kernel_old_timeval __user tv = {
3296 .tv_sec = ts.tv_sec,
3297 .tv_usec = ts.tv_nsec,
3299 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3304 return put_timespec64(&ts, userstamp);
3306 EXPORT_SYMBOL(sock_gettstamp);
3308 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3310 if (!sock_flag(sk, flag)) {
3311 unsigned long previous_flags = sk->sk_flags;
3313 sock_set_flag(sk, flag);
3315 * we just set one of the two flags which require net
3316 * time stamping, but time stamping might have been on
3317 * already because of the other one
3319 if (sock_needs_netstamp(sk) &&
3320 !(previous_flags & SK_FLAGS_TIMESTAMP))
3321 net_enable_timestamp();
3325 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3326 int level, int type)
3328 struct sock_exterr_skb *serr;
3329 struct sk_buff *skb;
3333 skb = sock_dequeue_err_skb(sk);
3339 msg->msg_flags |= MSG_TRUNC;
3342 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3346 sock_recv_timestamp(msg, sk, skb);
3348 serr = SKB_EXT_ERR(skb);
3349 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3351 msg->msg_flags |= MSG_ERRQUEUE;
3359 EXPORT_SYMBOL(sock_recv_errqueue);
3362 * Get a socket option on an socket.
3364 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3365 * asynchronous errors should be reported by getsockopt. We assume
3366 * this means if you specify SO_ERROR (otherwise whats the point of it).
3368 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3369 char __user *optval, int __user *optlen)
3371 struct sock *sk = sock->sk;
3373 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3375 EXPORT_SYMBOL(sock_common_getsockopt);
3377 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3380 struct sock *sk = sock->sk;
3384 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3385 flags & ~MSG_DONTWAIT, &addr_len);
3387 msg->msg_namelen = addr_len;
3390 EXPORT_SYMBOL(sock_common_recvmsg);
3393 * Set socket options on an inet socket.
3395 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3396 sockptr_t optval, unsigned int optlen)
3398 struct sock *sk = sock->sk;
3400 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3402 EXPORT_SYMBOL(sock_common_setsockopt);
3404 void sk_common_release(struct sock *sk)
3406 if (sk->sk_prot->destroy)
3407 sk->sk_prot->destroy(sk);
3410 * Observation: when sk_common_release is called, processes have
3411 * no access to socket. But net still has.
3412 * Step one, detach it from networking:
3414 * A. Remove from hash tables.
3417 sk->sk_prot->unhash(sk);
3420 * In this point socket cannot receive new packets, but it is possible
3421 * that some packets are in flight because some CPU runs receiver and
3422 * did hash table lookup before we unhashed socket. They will achieve
3423 * receive queue and will be purged by socket destructor.
3425 * Also we still have packets pending on receive queue and probably,
3426 * our own packets waiting in device queues. sock_destroy will drain
3427 * receive queue, but transmitted packets will delay socket destruction
3428 * until the last reference will be released.
3433 xfrm_sk_free_policy(sk);
3435 sk_refcnt_debug_release(sk);
3439 EXPORT_SYMBOL(sk_common_release);
3441 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3443 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3445 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3446 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3447 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3448 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3449 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3450 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3451 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3452 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3453 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3456 #ifdef CONFIG_PROC_FS
3457 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3459 int val[PROTO_INUSE_NR];
3462 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3464 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3466 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3468 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3470 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3472 int cpu, idx = prot->inuse_idx;
3475 for_each_possible_cpu(cpu)
3476 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3478 return res >= 0 ? res : 0;
3480 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3482 static void sock_inuse_add(struct net *net, int val)
3484 this_cpu_add(*net->core.sock_inuse, val);
3487 int sock_inuse_get(struct net *net)
3491 for_each_possible_cpu(cpu)
3492 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3497 EXPORT_SYMBOL_GPL(sock_inuse_get);
3499 static int __net_init sock_inuse_init_net(struct net *net)
3501 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3502 if (net->core.prot_inuse == NULL)
3505 net->core.sock_inuse = alloc_percpu(int);
3506 if (net->core.sock_inuse == NULL)
3512 free_percpu(net->core.prot_inuse);
3516 static void __net_exit sock_inuse_exit_net(struct net *net)
3518 free_percpu(net->core.prot_inuse);
3519 free_percpu(net->core.sock_inuse);
3522 static struct pernet_operations net_inuse_ops = {
3523 .init = sock_inuse_init_net,
3524 .exit = sock_inuse_exit_net,
3527 static __init int net_inuse_init(void)
3529 if (register_pernet_subsys(&net_inuse_ops))
3530 panic("Cannot initialize net inuse counters");
3535 core_initcall(net_inuse_init);
3537 static int assign_proto_idx(struct proto *prot)
3539 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3541 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3542 pr_err("PROTO_INUSE_NR exhausted\n");
3546 set_bit(prot->inuse_idx, proto_inuse_idx);
3550 static void release_proto_idx(struct proto *prot)
3552 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3553 clear_bit(prot->inuse_idx, proto_inuse_idx);
3556 static inline int assign_proto_idx(struct proto *prot)
3561 static inline void release_proto_idx(struct proto *prot)
3565 static void sock_inuse_add(struct net *net, int val)
3570 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3574 kfree(twsk_prot->twsk_slab_name);
3575 twsk_prot->twsk_slab_name = NULL;
3576 kmem_cache_destroy(twsk_prot->twsk_slab);
3577 twsk_prot->twsk_slab = NULL;
3580 static int tw_prot_init(const struct proto *prot)
3582 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3587 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3589 if (!twsk_prot->twsk_slab_name)
3592 twsk_prot->twsk_slab =
3593 kmem_cache_create(twsk_prot->twsk_slab_name,
3594 twsk_prot->twsk_obj_size, 0,
3595 SLAB_ACCOUNT | prot->slab_flags,
3597 if (!twsk_prot->twsk_slab) {
3598 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3606 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3610 kfree(rsk_prot->slab_name);
3611 rsk_prot->slab_name = NULL;
3612 kmem_cache_destroy(rsk_prot->slab);
3613 rsk_prot->slab = NULL;
3616 static int req_prot_init(const struct proto *prot)
3618 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3623 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3625 if (!rsk_prot->slab_name)
3628 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3629 rsk_prot->obj_size, 0,
3630 SLAB_ACCOUNT | prot->slab_flags,
3633 if (!rsk_prot->slab) {
3634 pr_crit("%s: Can't create request sock SLAB cache!\n",
3641 int proto_register(struct proto *prot, int alloc_slab)
3646 prot->slab = kmem_cache_create_usercopy(prot->name,
3648 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3650 prot->useroffset, prot->usersize,
3653 if (prot->slab == NULL) {
3654 pr_crit("%s: Can't create sock SLAB cache!\n",
3659 if (req_prot_init(prot))
3660 goto out_free_request_sock_slab;
3662 if (tw_prot_init(prot))
3663 goto out_free_timewait_sock_slab;
3666 mutex_lock(&proto_list_mutex);
3667 ret = assign_proto_idx(prot);
3669 mutex_unlock(&proto_list_mutex);
3670 goto out_free_timewait_sock_slab;
3672 list_add(&prot->node, &proto_list);
3673 mutex_unlock(&proto_list_mutex);
3676 out_free_timewait_sock_slab:
3678 tw_prot_cleanup(prot->twsk_prot);
3679 out_free_request_sock_slab:
3681 req_prot_cleanup(prot->rsk_prot);
3683 kmem_cache_destroy(prot->slab);
3689 EXPORT_SYMBOL(proto_register);
3691 void proto_unregister(struct proto *prot)
3693 mutex_lock(&proto_list_mutex);
3694 release_proto_idx(prot);
3695 list_del(&prot->node);
3696 mutex_unlock(&proto_list_mutex);
3698 kmem_cache_destroy(prot->slab);
3701 req_prot_cleanup(prot->rsk_prot);
3702 tw_prot_cleanup(prot->twsk_prot);
3704 EXPORT_SYMBOL(proto_unregister);
3706 int sock_load_diag_module(int family, int protocol)
3709 if (!sock_is_registered(family))
3712 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3713 NETLINK_SOCK_DIAG, family);
3717 if (family == AF_INET &&
3718 protocol != IPPROTO_RAW &&
3719 protocol < MAX_INET_PROTOS &&
3720 !rcu_access_pointer(inet_protos[protocol]))
3724 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3725 NETLINK_SOCK_DIAG, family, protocol);
3727 EXPORT_SYMBOL(sock_load_diag_module);
3729 #ifdef CONFIG_PROC_FS
3730 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3731 __acquires(proto_list_mutex)
3733 mutex_lock(&proto_list_mutex);
3734 return seq_list_start_head(&proto_list, *pos);
3737 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3739 return seq_list_next(v, &proto_list, pos);
3742 static void proto_seq_stop(struct seq_file *seq, void *v)
3743 __releases(proto_list_mutex)
3745 mutex_unlock(&proto_list_mutex);
3748 static char proto_method_implemented(const void *method)
3750 return method == NULL ? 'n' : 'y';
3752 static long sock_prot_memory_allocated(struct proto *proto)
3754 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3757 static const char *sock_prot_memory_pressure(struct proto *proto)
3759 return proto->memory_pressure != NULL ?
3760 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3763 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3766 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3767 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3770 sock_prot_inuse_get(seq_file_net(seq), proto),
3771 sock_prot_memory_allocated(proto),
3772 sock_prot_memory_pressure(proto),
3774 proto->slab == NULL ? "no" : "yes",
3775 module_name(proto->owner),
3776 proto_method_implemented(proto->close),
3777 proto_method_implemented(proto->connect),
3778 proto_method_implemented(proto->disconnect),
3779 proto_method_implemented(proto->accept),
3780 proto_method_implemented(proto->ioctl),
3781 proto_method_implemented(proto->init),
3782 proto_method_implemented(proto->destroy),
3783 proto_method_implemented(proto->shutdown),
3784 proto_method_implemented(proto->setsockopt),
3785 proto_method_implemented(proto->getsockopt),
3786 proto_method_implemented(proto->sendmsg),
3787 proto_method_implemented(proto->recvmsg),
3788 proto_method_implemented(proto->sendpage),
3789 proto_method_implemented(proto->bind),
3790 proto_method_implemented(proto->backlog_rcv),
3791 proto_method_implemented(proto->hash),
3792 proto_method_implemented(proto->unhash),
3793 proto_method_implemented(proto->get_port),
3794 proto_method_implemented(proto->enter_memory_pressure));
3797 static int proto_seq_show(struct seq_file *seq, void *v)
3799 if (v == &proto_list)
3800 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3809 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3811 proto_seq_printf(seq, list_entry(v, struct proto, node));
3815 static const struct seq_operations proto_seq_ops = {
3816 .start = proto_seq_start,
3817 .next = proto_seq_next,
3818 .stop = proto_seq_stop,
3819 .show = proto_seq_show,
3822 static __net_init int proto_init_net(struct net *net)
3824 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3825 sizeof(struct seq_net_private)))
3831 static __net_exit void proto_exit_net(struct net *net)
3833 remove_proc_entry("protocols", net->proc_net);
3837 static __net_initdata struct pernet_operations proto_net_ops = {
3838 .init = proto_init_net,
3839 .exit = proto_exit_net,
3842 static int __init proto_init(void)
3844 return register_pernet_subsys(&proto_net_ops);
3847 subsys_initcall(proto_init);
3849 #endif /* PROC_FS */
3851 #ifdef CONFIG_NET_RX_BUSY_POLL
3852 bool sk_busy_loop_end(void *p, unsigned long start_time)
3854 struct sock *sk = p;
3856 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3857 sk_busy_loop_timeout(sk, start_time);
3859 EXPORT_SYMBOL(sk_busy_loop_end);
3860 #endif /* CONFIG_NET_RX_BUSY_POLL */
3862 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3864 if (!sk->sk_prot->bind_add)
3866 return sk->sk_prot->bind_add(sk, addr, addr_len);
3868 EXPORT_SYMBOL(sock_bind_add);