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.
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116 #include <linux/compat.h>
118 #include <linux/uaccess.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
137 #include <trace/events/sock.h>
140 #include <net/busy_poll.h>
142 static DEFINE_MUTEX(proto_list_mutex);
143 static LIST_HEAD(proto_list);
145 static void sock_inuse_add(struct net *net, int val);
148 * sk_ns_capable - General socket capability test
149 * @sk: Socket to use a capability on or through
150 * @user_ns: The user namespace of the capability to use
151 * @cap: The capability to use
153 * Test to see if the opener of the socket had when the socket was
154 * created and the current process has the capability @cap in the user
155 * namespace @user_ns.
157 bool sk_ns_capable(const struct sock *sk,
158 struct user_namespace *user_ns, int cap)
160 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
161 ns_capable(user_ns, cap);
163 EXPORT_SYMBOL(sk_ns_capable);
166 * sk_capable - Socket global capability test
167 * @sk: Socket to use a capability on or through
168 * @cap: The global capability to use
170 * Test to see if the opener of the socket had when the socket was
171 * created and the current process has the capability @cap in all user
174 bool sk_capable(const struct sock *sk, int cap)
176 return sk_ns_capable(sk, &init_user_ns, cap);
178 EXPORT_SYMBOL(sk_capable);
181 * sk_net_capable - Network namespace socket capability test
182 * @sk: Socket to use a capability on or through
183 * @cap: The capability to use
185 * Test to see if the opener of the socket had when the socket was created
186 * and the current process has the capability @cap over the network namespace
187 * the socket is a member of.
189 bool sk_net_capable(const struct sock *sk, int cap)
191 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193 EXPORT_SYMBOL(sk_net_capable);
196 * Each address family might have different locking rules, so we have
197 * one slock key per address family and separate keys for internal and
200 static struct lock_class_key af_family_keys[AF_MAX];
201 static struct lock_class_key af_family_kern_keys[AF_MAX];
202 static struct lock_class_key af_family_slock_keys[AF_MAX];
203 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
211 #define _sock_locks(x) \
212 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
213 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
214 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
215 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
216 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
217 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
218 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
219 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
220 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
221 x "27" , x "28" , x "AF_CAN" , \
222 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
223 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
224 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
225 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
226 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
229 static const char *const af_family_key_strings[AF_MAX+1] = {
230 _sock_locks("sk_lock-")
232 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
233 _sock_locks("slock-")
235 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
236 _sock_locks("clock-")
239 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
240 _sock_locks("k-sk_lock-")
242 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("k-slock-")
245 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-clock-")
248 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
249 _sock_locks("rlock-")
251 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
252 _sock_locks("wlock-")
254 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
255 _sock_locks("elock-")
259 * sk_callback_lock and sk queues locking rules are per-address-family,
260 * so split the lock classes by using a per-AF key:
262 static struct lock_class_key af_callback_keys[AF_MAX];
263 static struct lock_class_key af_rlock_keys[AF_MAX];
264 static struct lock_class_key af_wlock_keys[AF_MAX];
265 static struct lock_class_key af_elock_keys[AF_MAX];
266 static struct lock_class_key af_kern_callback_keys[AF_MAX];
268 /* Run time adjustable parameters. */
269 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
270 EXPORT_SYMBOL(sysctl_wmem_max);
271 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
272 EXPORT_SYMBOL(sysctl_rmem_max);
273 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
274 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
276 /* Maximal space eaten by iovec or ancillary data plus some space */
277 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
278 EXPORT_SYMBOL(sysctl_optmem_max);
280 int sysctl_tstamp_allow_data __read_mostly = 1;
282 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
283 EXPORT_SYMBOL_GPL(memalloc_socks_key);
286 * sk_set_memalloc - sets %SOCK_MEMALLOC
287 * @sk: socket to set it on
289 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
290 * It's the responsibility of the admin to adjust min_free_kbytes
291 * to meet the requirements
293 void sk_set_memalloc(struct sock *sk)
295 sock_set_flag(sk, SOCK_MEMALLOC);
296 sk->sk_allocation |= __GFP_MEMALLOC;
297 static_branch_inc(&memalloc_socks_key);
299 EXPORT_SYMBOL_GPL(sk_set_memalloc);
301 void sk_clear_memalloc(struct sock *sk)
303 sock_reset_flag(sk, SOCK_MEMALLOC);
304 sk->sk_allocation &= ~__GFP_MEMALLOC;
305 static_branch_dec(&memalloc_socks_key);
308 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
309 * progress of swapping. SOCK_MEMALLOC may be cleared while
310 * it has rmem allocations due to the last swapfile being deactivated
311 * but there is a risk that the socket is unusable due to exceeding
312 * the rmem limits. Reclaim the reserves and obey rmem limits again.
316 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
318 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
321 unsigned int noreclaim_flag;
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
326 noreclaim_flag = memalloc_noreclaim_save();
327 ret = sk->sk_backlog_rcv(sk, skb);
328 memalloc_noreclaim_restore(noreclaim_flag);
332 EXPORT_SYMBOL(__sk_backlog_rcv);
334 static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
336 struct __kernel_sock_timeval tv;
338 if (timeo == MAX_SCHEDULE_TIMEOUT) {
342 tv.tv_sec = timeo / HZ;
343 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
346 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
347 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
348 *(struct old_timeval32 *)optval = tv32;
353 struct __kernel_old_timeval old_tv;
354 old_tv.tv_sec = tv.tv_sec;
355 old_tv.tv_usec = tv.tv_usec;
356 *(struct __kernel_old_timeval *)optval = old_tv;
357 return sizeof(old_tv);
360 *(struct __kernel_sock_timeval *)optval = tv;
364 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
367 struct __kernel_sock_timeval tv;
369 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
370 struct old_timeval32 tv32;
372 if (optlen < sizeof(tv32))
375 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
377 tv.tv_sec = tv32.tv_sec;
378 tv.tv_usec = tv32.tv_usec;
379 } else if (old_timeval) {
380 struct __kernel_old_timeval old_tv;
382 if (optlen < sizeof(old_tv))
384 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
386 tv.tv_sec = old_tv.tv_sec;
387 tv.tv_usec = old_tv.tv_usec;
389 if (optlen < sizeof(tv))
391 if (copy_from_sockptr(&tv, optval, sizeof(tv)))
394 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
398 static int warned __read_mostly;
401 if (warned < 10 && net_ratelimit()) {
403 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
404 __func__, current->comm, task_pid_nr(current));
408 *timeo_p = MAX_SCHEDULE_TIMEOUT;
409 if (tv.tv_sec == 0 && tv.tv_usec == 0)
411 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
412 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
416 static bool sock_needs_netstamp(const struct sock *sk)
418 switch (sk->sk_family) {
427 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
429 if (sk->sk_flags & flags) {
430 sk->sk_flags &= ~flags;
431 if (sock_needs_netstamp(sk) &&
432 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
433 net_disable_timestamp();
438 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
441 struct sk_buff_head *list = &sk->sk_receive_queue;
443 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
444 atomic_inc(&sk->sk_drops);
445 trace_sock_rcvqueue_full(sk, skb);
449 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
450 atomic_inc(&sk->sk_drops);
455 skb_set_owner_r(skb, sk);
457 /* we escape from rcu protected region, make sure we dont leak
462 spin_lock_irqsave(&list->lock, flags);
463 sock_skb_set_dropcount(sk, skb);
464 __skb_queue_tail(list, skb);
465 spin_unlock_irqrestore(&list->lock, flags);
467 if (!sock_flag(sk, SOCK_DEAD))
468 sk->sk_data_ready(sk);
471 EXPORT_SYMBOL(__sock_queue_rcv_skb);
473 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
477 err = sk_filter(sk, skb);
481 return __sock_queue_rcv_skb(sk, skb);
483 EXPORT_SYMBOL(sock_queue_rcv_skb);
485 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
486 const int nested, unsigned int trim_cap, bool refcounted)
488 int rc = NET_RX_SUCCESS;
490 if (sk_filter_trim_cap(sk, skb, trim_cap))
491 goto discard_and_relse;
495 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
496 atomic_inc(&sk->sk_drops);
497 goto discard_and_relse;
500 bh_lock_sock_nested(sk);
503 if (!sock_owned_by_user(sk)) {
505 * trylock + unlock semantics:
507 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
509 rc = sk_backlog_rcv(sk, skb);
511 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
512 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
514 atomic_inc(&sk->sk_drops);
515 goto discard_and_relse;
527 EXPORT_SYMBOL(__sk_receive_skb);
529 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
531 struct dst_entry *dst = __sk_dst_get(sk);
533 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
534 sk_tx_queue_clear(sk);
535 sk->sk_dst_pending_confirm = 0;
536 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
543 EXPORT_SYMBOL(__sk_dst_check);
545 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
547 struct dst_entry *dst = sk_dst_get(sk);
549 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
557 EXPORT_SYMBOL(sk_dst_check);
559 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
561 int ret = -ENOPROTOOPT;
562 #ifdef CONFIG_NETDEVICES
563 struct net *net = sock_net(sk);
567 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
574 sk->sk_bound_dev_if = ifindex;
575 if (sk->sk_prot->rehash)
576 sk->sk_prot->rehash(sk);
587 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
593 ret = sock_bindtoindex_locked(sk, ifindex);
599 EXPORT_SYMBOL(sock_bindtoindex);
601 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
603 int ret = -ENOPROTOOPT;
604 #ifdef CONFIG_NETDEVICES
605 struct net *net = sock_net(sk);
606 char devname[IFNAMSIZ];
613 /* Bind this socket to a particular device like "eth0",
614 * as specified in the passed interface name. If the
615 * name is "" or the option length is zero the socket
618 if (optlen > IFNAMSIZ - 1)
619 optlen = IFNAMSIZ - 1;
620 memset(devname, 0, sizeof(devname));
623 if (copy_from_sockptr(devname, optval, optlen))
627 if (devname[0] != '\0') {
628 struct net_device *dev;
631 dev = dev_get_by_name_rcu(net, devname);
633 index = dev->ifindex;
640 return sock_bindtoindex(sk, index, true);
647 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
648 int __user *optlen, int len)
650 int ret = -ENOPROTOOPT;
651 #ifdef CONFIG_NETDEVICES
652 struct net *net = sock_net(sk);
653 char devname[IFNAMSIZ];
655 if (sk->sk_bound_dev_if == 0) {
664 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
668 len = strlen(devname) + 1;
671 if (copy_to_user(optval, devname, len))
676 if (put_user(len, optlen))
687 bool sk_mc_loop(struct sock *sk)
689 if (dev_recursion_level())
693 switch (sk->sk_family) {
695 return inet_sk(sk)->mc_loop;
696 #if IS_ENABLED(CONFIG_IPV6)
698 return inet6_sk(sk)->mc_loop;
704 EXPORT_SYMBOL(sk_mc_loop);
706 void sock_set_reuseaddr(struct sock *sk)
709 sk->sk_reuse = SK_CAN_REUSE;
712 EXPORT_SYMBOL(sock_set_reuseaddr);
714 void sock_set_reuseport(struct sock *sk)
717 sk->sk_reuseport = true;
720 EXPORT_SYMBOL(sock_set_reuseport);
722 void sock_no_linger(struct sock *sk)
725 sk->sk_lingertime = 0;
726 sock_set_flag(sk, SOCK_LINGER);
729 EXPORT_SYMBOL(sock_no_linger);
731 void sock_set_priority(struct sock *sk, u32 priority)
734 sk->sk_priority = priority;
737 EXPORT_SYMBOL(sock_set_priority);
739 void sock_set_sndtimeo(struct sock *sk, s64 secs)
742 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
743 sk->sk_sndtimeo = secs * HZ;
745 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
748 EXPORT_SYMBOL(sock_set_sndtimeo);
750 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
753 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
754 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
755 sock_set_flag(sk, SOCK_RCVTSTAMP);
756 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
758 sock_reset_flag(sk, SOCK_RCVTSTAMP);
759 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
760 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
764 void sock_enable_timestamps(struct sock *sk)
767 __sock_set_timestamps(sk, true, false, true);
770 EXPORT_SYMBOL(sock_enable_timestamps);
772 void sock_set_keepalive(struct sock *sk)
775 if (sk->sk_prot->keepalive)
776 sk->sk_prot->keepalive(sk, true);
777 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
780 EXPORT_SYMBOL(sock_set_keepalive);
782 static void __sock_set_rcvbuf(struct sock *sk, int val)
784 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
785 * as a negative value.
787 val = min_t(int, val, INT_MAX / 2);
788 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
790 /* We double it on the way in to account for "struct sk_buff" etc.
791 * overhead. Applications assume that the SO_RCVBUF setting they make
792 * will allow that much actual data to be received on that socket.
794 * Applications are unaware that "struct sk_buff" and other overheads
795 * allocate from the receive buffer during socket buffer allocation.
797 * And after considering the possible alternatives, returning the value
798 * we actually used in getsockopt is the most desirable behavior.
800 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
803 void sock_set_rcvbuf(struct sock *sk, int val)
806 __sock_set_rcvbuf(sk, val);
809 EXPORT_SYMBOL(sock_set_rcvbuf);
811 void sock_set_mark(struct sock *sk, u32 val)
817 EXPORT_SYMBOL(sock_set_mark);
820 * This is meant for all protocols to use and covers goings on
821 * at the socket level. Everything here is generic.
824 int sock_setsockopt(struct socket *sock, int level, int optname,
825 sockptr_t optval, unsigned int optlen)
827 struct sock_txtime sk_txtime;
828 struct sock *sk = sock->sk;
835 * Options without arguments
838 if (optname == SO_BINDTODEVICE)
839 return sock_setbindtodevice(sk, optval, optlen);
841 if (optlen < sizeof(int))
844 if (copy_from_sockptr(&val, optval, sizeof(val)))
847 valbool = val ? 1 : 0;
853 if (val && !capable(CAP_NET_ADMIN))
856 sock_valbool_flag(sk, SOCK_DBG, valbool);
859 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
862 sk->sk_reuseport = valbool;
871 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
875 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
878 /* Don't error on this BSD doesn't and if you think
879 * about it this is right. Otherwise apps have to
880 * play 'guess the biggest size' games. RCVBUF/SNDBUF
881 * are treated in BSD as hints
883 val = min_t(u32, val, sysctl_wmem_max);
885 /* Ensure val * 2 fits into an int, to prevent max_t()
886 * from treating it as a negative value.
888 val = min_t(int, val, INT_MAX / 2);
889 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
890 WRITE_ONCE(sk->sk_sndbuf,
891 max_t(int, val * 2, SOCK_MIN_SNDBUF));
892 /* Wake up sending tasks if we upped the value. */
893 sk->sk_write_space(sk);
897 if (!capable(CAP_NET_ADMIN)) {
902 /* No negative values (to prevent underflow, as val will be
910 /* Don't error on this BSD doesn't and if you think
911 * about it this is right. Otherwise apps have to
912 * play 'guess the biggest size' games. RCVBUF/SNDBUF
913 * are treated in BSD as hints
915 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
919 if (!capable(CAP_NET_ADMIN)) {
924 /* No negative values (to prevent underflow, as val will be
927 __sock_set_rcvbuf(sk, max(val, 0));
931 if (sk->sk_prot->keepalive)
932 sk->sk_prot->keepalive(sk, valbool);
933 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
937 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
941 sk->sk_no_check_tx = valbool;
945 if ((val >= 0 && val <= 6) ||
946 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
947 sk->sk_priority = val;
953 if (optlen < sizeof(ling)) {
954 ret = -EINVAL; /* 1003.1g */
957 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
962 sock_reset_flag(sk, SOCK_LINGER);
964 #if (BITS_PER_LONG == 32)
965 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
966 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
969 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
970 sock_set_flag(sk, SOCK_LINGER);
979 set_bit(SOCK_PASSCRED, &sock->flags);
981 clear_bit(SOCK_PASSCRED, &sock->flags);
984 case SO_TIMESTAMP_OLD:
985 __sock_set_timestamps(sk, valbool, false, false);
987 case SO_TIMESTAMP_NEW:
988 __sock_set_timestamps(sk, valbool, true, false);
990 case SO_TIMESTAMPNS_OLD:
991 __sock_set_timestamps(sk, valbool, false, true);
993 case SO_TIMESTAMPNS_NEW:
994 __sock_set_timestamps(sk, valbool, true, true);
996 case SO_TIMESTAMPING_NEW:
997 sock_set_flag(sk, SOCK_TSTAMP_NEW);
999 case SO_TIMESTAMPING_OLD:
1000 if (val & ~SOF_TIMESTAMPING_MASK) {
1005 if (val & SOF_TIMESTAMPING_OPT_ID &&
1006 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
1007 if (sk->sk_protocol == IPPROTO_TCP &&
1008 sk->sk_type == SOCK_STREAM) {
1009 if ((1 << sk->sk_state) &
1010 (TCPF_CLOSE | TCPF_LISTEN)) {
1014 sk->sk_tskey = tcp_sk(sk)->snd_una;
1020 if (val & SOF_TIMESTAMPING_OPT_STATS &&
1021 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
1026 sk->sk_tsflags = val;
1027 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
1028 sock_enable_timestamp(sk,
1029 SOCK_TIMESTAMPING_RX_SOFTWARE);
1031 if (optname == SO_TIMESTAMPING_NEW)
1032 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
1034 sock_disable_timestamp(sk,
1035 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
1042 if (sock->ops->set_rcvlowat)
1043 ret = sock->ops->set_rcvlowat(sk, val);
1045 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1048 case SO_RCVTIMEO_OLD:
1049 case SO_RCVTIMEO_NEW:
1050 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1051 optlen, optname == SO_RCVTIMEO_OLD);
1054 case SO_SNDTIMEO_OLD:
1055 case SO_SNDTIMEO_NEW:
1056 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1057 optlen, optname == SO_SNDTIMEO_OLD);
1060 case SO_ATTACH_FILTER: {
1061 struct sock_fprog fprog;
1063 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1065 ret = sk_attach_filter(&fprog, sk);
1070 if (optlen == sizeof(u32)) {
1074 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1077 ret = sk_attach_bpf(ufd, sk);
1081 case SO_ATTACH_REUSEPORT_CBPF: {
1082 struct sock_fprog fprog;
1084 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1086 ret = sk_reuseport_attach_filter(&fprog, sk);
1089 case SO_ATTACH_REUSEPORT_EBPF:
1091 if (optlen == sizeof(u32)) {
1095 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1098 ret = sk_reuseport_attach_bpf(ufd, sk);
1102 case SO_DETACH_REUSEPORT_BPF:
1103 ret = reuseport_detach_prog(sk);
1106 case SO_DETACH_FILTER:
1107 ret = sk_detach_filter(sk);
1110 case SO_LOCK_FILTER:
1111 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1114 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1119 set_bit(SOCK_PASSSEC, &sock->flags);
1121 clear_bit(SOCK_PASSSEC, &sock->flags);
1124 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1126 } else if (val != sk->sk_mark) {
1133 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1136 case SO_WIFI_STATUS:
1137 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1141 if (sock->ops->set_peek_off)
1142 ret = sock->ops->set_peek_off(sk, val);
1148 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1151 case SO_SELECT_ERR_QUEUE:
1152 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1155 #ifdef CONFIG_NET_RX_BUSY_POLL
1157 /* allow unprivileged users to decrease the value */
1158 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1164 sk->sk_ll_usec = val;
1169 case SO_MAX_PACING_RATE:
1171 unsigned long ulval = (val == ~0U) ? ~0UL : val;
1173 if (sizeof(ulval) != sizeof(val) &&
1174 optlen >= sizeof(ulval) &&
1175 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1180 cmpxchg(&sk->sk_pacing_status,
1183 sk->sk_max_pacing_rate = ulval;
1184 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1187 case SO_INCOMING_CPU:
1188 WRITE_ONCE(sk->sk_incoming_cpu, val);
1193 dst_negative_advice(sk);
1197 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1198 if (!((sk->sk_type == SOCK_STREAM &&
1199 sk->sk_protocol == IPPROTO_TCP) ||
1200 (sk->sk_type == SOCK_DGRAM &&
1201 sk->sk_protocol == IPPROTO_UDP)))
1203 } else if (sk->sk_family != PF_RDS) {
1207 if (val < 0 || val > 1)
1210 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1215 if (optlen != sizeof(struct sock_txtime)) {
1218 } else if (copy_from_sockptr(&sk_txtime, optval,
1219 sizeof(struct sock_txtime))) {
1222 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1226 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1227 * scheduler has enough safe guards.
1229 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1230 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1234 sock_valbool_flag(sk, SOCK_TXTIME, true);
1235 sk->sk_clockid = sk_txtime.clockid;
1236 sk->sk_txtime_deadline_mode =
1237 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1238 sk->sk_txtime_report_errors =
1239 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1242 case SO_BINDTOIFINDEX:
1243 ret = sock_bindtoindex_locked(sk, val);
1253 EXPORT_SYMBOL(sock_setsockopt);
1256 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1257 struct ucred *ucred)
1259 ucred->pid = pid_vnr(pid);
1260 ucred->uid = ucred->gid = -1;
1262 struct user_namespace *current_ns = current_user_ns();
1264 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1265 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1269 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1271 struct user_namespace *user_ns = current_user_ns();
1274 for (i = 0; i < src->ngroups; i++)
1275 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1281 int sock_getsockopt(struct socket *sock, int level, int optname,
1282 char __user *optval, int __user *optlen)
1284 struct sock *sk = sock->sk;
1289 unsigned long ulval;
1291 struct old_timeval32 tm32;
1292 struct __kernel_old_timeval tm;
1293 struct __kernel_sock_timeval stm;
1294 struct sock_txtime txtime;
1297 int lv = sizeof(int);
1300 if (get_user(len, optlen))
1305 memset(&v, 0, sizeof(v));
1309 v.val = sock_flag(sk, SOCK_DBG);
1313 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1317 v.val = sock_flag(sk, SOCK_BROADCAST);
1321 v.val = sk->sk_sndbuf;
1325 v.val = sk->sk_rcvbuf;
1329 v.val = sk->sk_reuse;
1333 v.val = sk->sk_reuseport;
1337 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1341 v.val = sk->sk_type;
1345 v.val = sk->sk_protocol;
1349 v.val = sk->sk_family;
1353 v.val = -sock_error(sk);
1355 v.val = xchg(&sk->sk_err_soft, 0);
1359 v.val = sock_flag(sk, SOCK_URGINLINE);
1363 v.val = sk->sk_no_check_tx;
1367 v.val = sk->sk_priority;
1371 lv = sizeof(v.ling);
1372 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1373 v.ling.l_linger = sk->sk_lingertime / HZ;
1379 case SO_TIMESTAMP_OLD:
1380 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1381 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1382 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1385 case SO_TIMESTAMPNS_OLD:
1386 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1389 case SO_TIMESTAMP_NEW:
1390 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1393 case SO_TIMESTAMPNS_NEW:
1394 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1397 case SO_TIMESTAMPING_OLD:
1398 v.val = sk->sk_tsflags;
1401 case SO_RCVTIMEO_OLD:
1402 case SO_RCVTIMEO_NEW:
1403 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1406 case SO_SNDTIMEO_OLD:
1407 case SO_SNDTIMEO_NEW:
1408 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1412 v.val = sk->sk_rcvlowat;
1420 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1425 struct ucred peercred;
1426 if (len > sizeof(peercred))
1427 len = sizeof(peercred);
1428 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1429 if (copy_to_user(optval, &peercred, len))
1438 if (!sk->sk_peer_cred)
1441 n = sk->sk_peer_cred->group_info->ngroups;
1442 if (len < n * sizeof(gid_t)) {
1443 len = n * sizeof(gid_t);
1444 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1446 len = n * sizeof(gid_t);
1448 ret = groups_to_user((gid_t __user *)optval,
1449 sk->sk_peer_cred->group_info);
1459 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1464 if (copy_to_user(optval, address, len))
1469 /* Dubious BSD thing... Probably nobody even uses it, but
1470 * the UNIX standard wants it for whatever reason... -DaveM
1473 v.val = sk->sk_state == TCP_LISTEN;
1477 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1481 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1484 v.val = sk->sk_mark;
1488 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1491 case SO_WIFI_STATUS:
1492 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1496 if (!sock->ops->set_peek_off)
1499 v.val = sk->sk_peek_off;
1502 v.val = sock_flag(sk, SOCK_NOFCS);
1505 case SO_BINDTODEVICE:
1506 return sock_getbindtodevice(sk, optval, optlen, len);
1509 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1515 case SO_LOCK_FILTER:
1516 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1519 case SO_BPF_EXTENSIONS:
1520 v.val = bpf_tell_extensions();
1523 case SO_SELECT_ERR_QUEUE:
1524 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1527 #ifdef CONFIG_NET_RX_BUSY_POLL
1529 v.val = sk->sk_ll_usec;
1533 case SO_MAX_PACING_RATE:
1534 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1535 lv = sizeof(v.ulval);
1536 v.ulval = sk->sk_max_pacing_rate;
1539 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1543 case SO_INCOMING_CPU:
1544 v.val = READ_ONCE(sk->sk_incoming_cpu);
1549 u32 meminfo[SK_MEMINFO_VARS];
1551 sk_get_meminfo(sk, meminfo);
1553 len = min_t(unsigned int, len, sizeof(meminfo));
1554 if (copy_to_user(optval, &meminfo, len))
1560 #ifdef CONFIG_NET_RX_BUSY_POLL
1561 case SO_INCOMING_NAPI_ID:
1562 v.val = READ_ONCE(sk->sk_napi_id);
1564 /* aggregate non-NAPI IDs down to 0 */
1565 if (v.val < MIN_NAPI_ID)
1575 v.val64 = sock_gen_cookie(sk);
1579 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1583 lv = sizeof(v.txtime);
1584 v.txtime.clockid = sk->sk_clockid;
1585 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1586 SOF_TXTIME_DEADLINE_MODE : 0;
1587 v.txtime.flags |= sk->sk_txtime_report_errors ?
1588 SOF_TXTIME_REPORT_ERRORS : 0;
1591 case SO_BINDTOIFINDEX:
1592 v.val = sk->sk_bound_dev_if;
1596 /* We implement the SO_SNDLOWAT etc to not be settable
1599 return -ENOPROTOOPT;
1604 if (copy_to_user(optval, &v, len))
1607 if (put_user(len, optlen))
1613 * Initialize an sk_lock.
1615 * (We also register the sk_lock with the lock validator.)
1617 static inline void sock_lock_init(struct sock *sk)
1619 if (sk->sk_kern_sock)
1620 sock_lock_init_class_and_name(
1622 af_family_kern_slock_key_strings[sk->sk_family],
1623 af_family_kern_slock_keys + sk->sk_family,
1624 af_family_kern_key_strings[sk->sk_family],
1625 af_family_kern_keys + sk->sk_family);
1627 sock_lock_init_class_and_name(
1629 af_family_slock_key_strings[sk->sk_family],
1630 af_family_slock_keys + sk->sk_family,
1631 af_family_key_strings[sk->sk_family],
1632 af_family_keys + sk->sk_family);
1636 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1637 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1638 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1640 static void sock_copy(struct sock *nsk, const struct sock *osk)
1642 const struct proto *prot = READ_ONCE(osk->sk_prot);
1643 #ifdef CONFIG_SECURITY_NETWORK
1644 void *sptr = nsk->sk_security;
1646 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1648 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1649 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1651 #ifdef CONFIG_SECURITY_NETWORK
1652 nsk->sk_security = sptr;
1653 security_sk_clone(osk, nsk);
1657 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1661 struct kmem_cache *slab;
1665 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1668 if (want_init_on_alloc(priority))
1669 sk_prot_clear_nulls(sk, prot->obj_size);
1671 sk = kmalloc(prot->obj_size, priority);
1674 if (security_sk_alloc(sk, family, priority))
1677 if (!try_module_get(prot->owner))
1679 sk_tx_queue_clear(sk);
1685 security_sk_free(sk);
1688 kmem_cache_free(slab, sk);
1694 static void sk_prot_free(struct proto *prot, struct sock *sk)
1696 struct kmem_cache *slab;
1697 struct module *owner;
1699 owner = prot->owner;
1702 cgroup_sk_free(&sk->sk_cgrp_data);
1703 mem_cgroup_sk_free(sk);
1704 security_sk_free(sk);
1706 kmem_cache_free(slab, sk);
1713 * sk_alloc - All socket objects are allocated here
1714 * @net: the applicable net namespace
1715 * @family: protocol family
1716 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1717 * @prot: struct proto associated with this new sock instance
1718 * @kern: is this to be a kernel socket?
1720 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1721 struct proto *prot, int kern)
1725 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1727 sk->sk_family = family;
1729 * See comment in struct sock definition to understand
1730 * why we need sk_prot_creator -acme
1732 sk->sk_prot = sk->sk_prot_creator = prot;
1733 sk->sk_kern_sock = kern;
1735 sk->sk_net_refcnt = kern ? 0 : 1;
1736 if (likely(sk->sk_net_refcnt)) {
1738 sock_inuse_add(net, 1);
1741 sock_net_set(sk, net);
1742 refcount_set(&sk->sk_wmem_alloc, 1);
1744 mem_cgroup_sk_alloc(sk);
1745 cgroup_sk_alloc(&sk->sk_cgrp_data);
1746 sock_update_classid(&sk->sk_cgrp_data);
1747 sock_update_netprioidx(&sk->sk_cgrp_data);
1748 sk_tx_queue_clear(sk);
1753 EXPORT_SYMBOL(sk_alloc);
1755 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1756 * grace period. This is the case for UDP sockets and TCP listeners.
1758 static void __sk_destruct(struct rcu_head *head)
1760 struct sock *sk = container_of(head, struct sock, sk_rcu);
1761 struct sk_filter *filter;
1763 if (sk->sk_destruct)
1764 sk->sk_destruct(sk);
1766 filter = rcu_dereference_check(sk->sk_filter,
1767 refcount_read(&sk->sk_wmem_alloc) == 0);
1769 sk_filter_uncharge(sk, filter);
1770 RCU_INIT_POINTER(sk->sk_filter, NULL);
1773 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1775 #ifdef CONFIG_BPF_SYSCALL
1776 bpf_sk_storage_free(sk);
1779 if (atomic_read(&sk->sk_omem_alloc))
1780 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1781 __func__, atomic_read(&sk->sk_omem_alloc));
1783 if (sk->sk_frag.page) {
1784 put_page(sk->sk_frag.page);
1785 sk->sk_frag.page = NULL;
1788 if (sk->sk_peer_cred)
1789 put_cred(sk->sk_peer_cred);
1790 put_pid(sk->sk_peer_pid);
1791 if (likely(sk->sk_net_refcnt))
1792 put_net(sock_net(sk));
1793 sk_prot_free(sk->sk_prot_creator, sk);
1796 void sk_destruct(struct sock *sk)
1798 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1800 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1801 reuseport_detach_sock(sk);
1802 use_call_rcu = true;
1806 call_rcu(&sk->sk_rcu, __sk_destruct);
1808 __sk_destruct(&sk->sk_rcu);
1811 static void __sk_free(struct sock *sk)
1813 if (likely(sk->sk_net_refcnt))
1814 sock_inuse_add(sock_net(sk), -1);
1816 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1817 sock_diag_broadcast_destroy(sk);
1822 void sk_free(struct sock *sk)
1825 * We subtract one from sk_wmem_alloc and can know if
1826 * some packets are still in some tx queue.
1827 * If not null, sock_wfree() will call __sk_free(sk) later
1829 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1832 EXPORT_SYMBOL(sk_free);
1834 static void sk_init_common(struct sock *sk)
1836 skb_queue_head_init(&sk->sk_receive_queue);
1837 skb_queue_head_init(&sk->sk_write_queue);
1838 skb_queue_head_init(&sk->sk_error_queue);
1840 rwlock_init(&sk->sk_callback_lock);
1841 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1842 af_rlock_keys + sk->sk_family,
1843 af_family_rlock_key_strings[sk->sk_family]);
1844 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1845 af_wlock_keys + sk->sk_family,
1846 af_family_wlock_key_strings[sk->sk_family]);
1847 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1848 af_elock_keys + sk->sk_family,
1849 af_family_elock_key_strings[sk->sk_family]);
1850 lockdep_set_class_and_name(&sk->sk_callback_lock,
1851 af_callback_keys + sk->sk_family,
1852 af_family_clock_key_strings[sk->sk_family]);
1856 * sk_clone_lock - clone a socket, and lock its clone
1857 * @sk: the socket to clone
1858 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1860 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1862 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1864 struct proto *prot = READ_ONCE(sk->sk_prot);
1866 bool is_charged = true;
1868 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
1869 if (newsk != NULL) {
1870 struct sk_filter *filter;
1872 sock_copy(newsk, sk);
1874 newsk->sk_prot_creator = prot;
1877 if (likely(newsk->sk_net_refcnt))
1878 get_net(sock_net(newsk));
1879 sk_node_init(&newsk->sk_node);
1880 sock_lock_init(newsk);
1881 bh_lock_sock(newsk);
1882 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1883 newsk->sk_backlog.len = 0;
1885 atomic_set(&newsk->sk_rmem_alloc, 0);
1887 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1889 refcount_set(&newsk->sk_wmem_alloc, 1);
1890 atomic_set(&newsk->sk_omem_alloc, 0);
1891 sk_init_common(newsk);
1893 newsk->sk_dst_cache = NULL;
1894 newsk->sk_dst_pending_confirm = 0;
1895 newsk->sk_wmem_queued = 0;
1896 newsk->sk_forward_alloc = 0;
1897 atomic_set(&newsk->sk_drops, 0);
1898 newsk->sk_send_head = NULL;
1899 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1900 atomic_set(&newsk->sk_zckey, 0);
1902 sock_reset_flag(newsk, SOCK_DONE);
1904 /* sk->sk_memcg will be populated at accept() time */
1905 newsk->sk_memcg = NULL;
1907 cgroup_sk_clone(&newsk->sk_cgrp_data);
1910 filter = rcu_dereference(sk->sk_filter);
1912 /* though it's an empty new sock, the charging may fail
1913 * if sysctl_optmem_max was changed between creation of
1914 * original socket and cloning
1916 is_charged = sk_filter_charge(newsk, filter);
1917 RCU_INIT_POINTER(newsk->sk_filter, filter);
1920 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1921 /* We need to make sure that we don't uncharge the new
1922 * socket if we couldn't charge it in the first place
1923 * as otherwise we uncharge the parent's filter.
1926 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1927 sk_free_unlock_clone(newsk);
1931 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1933 if (bpf_sk_storage_clone(sk, newsk)) {
1934 sk_free_unlock_clone(newsk);
1939 /* Clear sk_user_data if parent had the pointer tagged
1940 * as not suitable for copying when cloning.
1942 if (sk_user_data_is_nocopy(newsk))
1943 newsk->sk_user_data = NULL;
1946 newsk->sk_err_soft = 0;
1947 newsk->sk_priority = 0;
1948 newsk->sk_incoming_cpu = raw_smp_processor_id();
1949 if (likely(newsk->sk_net_refcnt))
1950 sock_inuse_add(sock_net(newsk), 1);
1953 * Before updating sk_refcnt, we must commit prior changes to memory
1954 * (Documentation/RCU/rculist_nulls.rst for details)
1957 refcount_set(&newsk->sk_refcnt, 2);
1960 * Increment the counter in the same struct proto as the master
1961 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1962 * is the same as sk->sk_prot->socks, as this field was copied
1965 * This _changes_ the previous behaviour, where
1966 * tcp_create_openreq_child always was incrementing the
1967 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1968 * to be taken into account in all callers. -acme
1970 sk_refcnt_debug_inc(newsk);
1971 sk_set_socket(newsk, NULL);
1972 sk_tx_queue_clear(newsk);
1973 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1975 if (newsk->sk_prot->sockets_allocated)
1976 sk_sockets_allocated_inc(newsk);
1978 if (sock_needs_netstamp(sk) &&
1979 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1980 net_enable_timestamp();
1985 EXPORT_SYMBOL_GPL(sk_clone_lock);
1987 void sk_free_unlock_clone(struct sock *sk)
1989 /* It is still raw copy of parent, so invalidate
1990 * destructor and make plain sk_free() */
1991 sk->sk_destruct = NULL;
1995 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1997 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2001 sk_dst_set(sk, dst);
2002 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2003 if (sk->sk_route_caps & NETIF_F_GSO)
2004 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2005 sk->sk_route_caps &= ~sk->sk_route_nocaps;
2006 if (sk_can_gso(sk)) {
2007 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2008 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2010 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2011 sk->sk_gso_max_size = dst->dev->gso_max_size;
2012 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2015 sk->sk_gso_max_segs = max_segs;
2017 EXPORT_SYMBOL_GPL(sk_setup_caps);
2020 * Simple resource managers for sockets.
2025 * Write buffer destructor automatically called from kfree_skb.
2027 void sock_wfree(struct sk_buff *skb)
2029 struct sock *sk = skb->sk;
2030 unsigned int len = skb->truesize;
2032 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2034 * Keep a reference on sk_wmem_alloc, this will be released
2035 * after sk_write_space() call
2037 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2038 sk->sk_write_space(sk);
2042 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2043 * could not do because of in-flight packets
2045 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2048 EXPORT_SYMBOL(sock_wfree);
2050 /* This variant of sock_wfree() is used by TCP,
2051 * since it sets SOCK_USE_WRITE_QUEUE.
2053 void __sock_wfree(struct sk_buff *skb)
2055 struct sock *sk = skb->sk;
2057 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2061 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2066 if (unlikely(!sk_fullsock(sk))) {
2067 skb->destructor = sock_edemux;
2072 skb->destructor = sock_wfree;
2073 skb_set_hash_from_sk(skb, sk);
2075 * We used to take a refcount on sk, but following operation
2076 * is enough to guarantee sk_free() wont free this sock until
2077 * all in-flight packets are completed
2079 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2081 EXPORT_SYMBOL(skb_set_owner_w);
2083 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2085 #ifdef CONFIG_TLS_DEVICE
2086 /* Drivers depend on in-order delivery for crypto offload,
2087 * partial orphan breaks out-of-order-OK logic.
2092 return (skb->destructor == sock_wfree ||
2093 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2096 /* This helper is used by netem, as it can hold packets in its
2097 * delay queue. We want to allow the owner socket to send more
2098 * packets, as if they were already TX completed by a typical driver.
2099 * But we also want to keep skb->sk set because some packet schedulers
2100 * rely on it (sch_fq for example).
2102 void skb_orphan_partial(struct sk_buff *skb)
2104 if (skb_is_tcp_pure_ack(skb))
2107 if (can_skb_orphan_partial(skb)) {
2108 struct sock *sk = skb->sk;
2110 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2111 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2112 skb->destructor = sock_efree;
2118 EXPORT_SYMBOL(skb_orphan_partial);
2121 * Read buffer destructor automatically called from kfree_skb.
2123 void sock_rfree(struct sk_buff *skb)
2125 struct sock *sk = skb->sk;
2126 unsigned int len = skb->truesize;
2128 atomic_sub(len, &sk->sk_rmem_alloc);
2129 sk_mem_uncharge(sk, len);
2131 EXPORT_SYMBOL(sock_rfree);
2134 * Buffer destructor for skbs that are not used directly in read or write
2135 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2137 void sock_efree(struct sk_buff *skb)
2141 EXPORT_SYMBOL(sock_efree);
2143 /* Buffer destructor for prefetch/receive path where reference count may
2144 * not be held, e.g. for listen sockets.
2147 void sock_pfree(struct sk_buff *skb)
2149 if (sk_is_refcounted(skb->sk))
2150 sock_gen_put(skb->sk);
2152 EXPORT_SYMBOL(sock_pfree);
2153 #endif /* CONFIG_INET */
2155 kuid_t sock_i_uid(struct sock *sk)
2159 read_lock_bh(&sk->sk_callback_lock);
2160 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2161 read_unlock_bh(&sk->sk_callback_lock);
2164 EXPORT_SYMBOL(sock_i_uid);
2166 unsigned long sock_i_ino(struct sock *sk)
2170 read_lock_bh(&sk->sk_callback_lock);
2171 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2172 read_unlock_bh(&sk->sk_callback_lock);
2175 EXPORT_SYMBOL(sock_i_ino);
2178 * Allocate a skb from the socket's send buffer.
2180 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2184 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2185 struct sk_buff *skb = alloc_skb(size, priority);
2188 skb_set_owner_w(skb, sk);
2194 EXPORT_SYMBOL(sock_wmalloc);
2196 static void sock_ofree(struct sk_buff *skb)
2198 struct sock *sk = skb->sk;
2200 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2203 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2206 struct sk_buff *skb;
2208 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2209 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2213 skb = alloc_skb(size, priority);
2217 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2219 skb->destructor = sock_ofree;
2224 * Allocate a memory block from the socket's option memory buffer.
2226 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2228 if ((unsigned int)size <= sysctl_optmem_max &&
2229 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2231 /* First do the add, to avoid the race if kmalloc
2234 atomic_add(size, &sk->sk_omem_alloc);
2235 mem = kmalloc(size, priority);
2238 atomic_sub(size, &sk->sk_omem_alloc);
2242 EXPORT_SYMBOL(sock_kmalloc);
2244 /* Free an option memory block. Note, we actually want the inline
2245 * here as this allows gcc to detect the nullify and fold away the
2246 * condition entirely.
2248 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2251 if (WARN_ON_ONCE(!mem))
2254 kfree_sensitive(mem);
2257 atomic_sub(size, &sk->sk_omem_alloc);
2260 void sock_kfree_s(struct sock *sk, void *mem, int size)
2262 __sock_kfree_s(sk, mem, size, false);
2264 EXPORT_SYMBOL(sock_kfree_s);
2266 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2268 __sock_kfree_s(sk, mem, size, true);
2270 EXPORT_SYMBOL(sock_kzfree_s);
2272 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2273 I think, these locks should be removed for datagram sockets.
2275 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2279 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2283 if (signal_pending(current))
2285 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2286 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2287 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2289 if (sk->sk_shutdown & SEND_SHUTDOWN)
2293 timeo = schedule_timeout(timeo);
2295 finish_wait(sk_sleep(sk), &wait);
2301 * Generic send/receive buffer handlers
2304 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2305 unsigned long data_len, int noblock,
2306 int *errcode, int max_page_order)
2308 struct sk_buff *skb;
2312 timeo = sock_sndtimeo(sk, noblock);
2314 err = sock_error(sk);
2319 if (sk->sk_shutdown & SEND_SHUTDOWN)
2322 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2325 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2326 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2330 if (signal_pending(current))
2332 timeo = sock_wait_for_wmem(sk, timeo);
2334 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2335 errcode, sk->sk_allocation);
2337 skb_set_owner_w(skb, sk);
2341 err = sock_intr_errno(timeo);
2346 EXPORT_SYMBOL(sock_alloc_send_pskb);
2348 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2349 int noblock, int *errcode)
2351 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2353 EXPORT_SYMBOL(sock_alloc_send_skb);
2355 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2356 struct sockcm_cookie *sockc)
2360 switch (cmsg->cmsg_type) {
2362 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2364 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2366 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2368 case SO_TIMESTAMPING_OLD:
2369 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2372 tsflags = *(u32 *)CMSG_DATA(cmsg);
2373 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2376 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2377 sockc->tsflags |= tsflags;
2380 if (!sock_flag(sk, SOCK_TXTIME))
2382 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2384 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2386 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2388 case SCM_CREDENTIALS:
2395 EXPORT_SYMBOL(__sock_cmsg_send);
2397 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2398 struct sockcm_cookie *sockc)
2400 struct cmsghdr *cmsg;
2403 for_each_cmsghdr(cmsg, msg) {
2404 if (!CMSG_OK(msg, cmsg))
2406 if (cmsg->cmsg_level != SOL_SOCKET)
2408 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2414 EXPORT_SYMBOL(sock_cmsg_send);
2416 static void sk_enter_memory_pressure(struct sock *sk)
2418 if (!sk->sk_prot->enter_memory_pressure)
2421 sk->sk_prot->enter_memory_pressure(sk);
2424 static void sk_leave_memory_pressure(struct sock *sk)
2426 if (sk->sk_prot->leave_memory_pressure) {
2427 sk->sk_prot->leave_memory_pressure(sk);
2429 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2431 if (memory_pressure && READ_ONCE(*memory_pressure))
2432 WRITE_ONCE(*memory_pressure, 0);
2436 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2437 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2440 * skb_page_frag_refill - check that a page_frag contains enough room
2441 * @sz: minimum size of the fragment we want to get
2442 * @pfrag: pointer to page_frag
2443 * @gfp: priority for memory allocation
2445 * Note: While this allocator tries to use high order pages, there is
2446 * no guarantee that allocations succeed. Therefore, @sz MUST be
2447 * less or equal than PAGE_SIZE.
2449 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2452 if (page_ref_count(pfrag->page) == 1) {
2456 if (pfrag->offset + sz <= pfrag->size)
2458 put_page(pfrag->page);
2462 if (SKB_FRAG_PAGE_ORDER &&
2463 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2464 /* Avoid direct reclaim but allow kswapd to wake */
2465 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2466 __GFP_COMP | __GFP_NOWARN |
2468 SKB_FRAG_PAGE_ORDER);
2469 if (likely(pfrag->page)) {
2470 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2474 pfrag->page = alloc_page(gfp);
2475 if (likely(pfrag->page)) {
2476 pfrag->size = PAGE_SIZE;
2481 EXPORT_SYMBOL(skb_page_frag_refill);
2483 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2485 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2488 sk_enter_memory_pressure(sk);
2489 sk_stream_moderate_sndbuf(sk);
2492 EXPORT_SYMBOL(sk_page_frag_refill);
2494 static void __lock_sock(struct sock *sk)
2495 __releases(&sk->sk_lock.slock)
2496 __acquires(&sk->sk_lock.slock)
2501 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2502 TASK_UNINTERRUPTIBLE);
2503 spin_unlock_bh(&sk->sk_lock.slock);
2505 spin_lock_bh(&sk->sk_lock.slock);
2506 if (!sock_owned_by_user(sk))
2509 finish_wait(&sk->sk_lock.wq, &wait);
2512 void __release_sock(struct sock *sk)
2513 __releases(&sk->sk_lock.slock)
2514 __acquires(&sk->sk_lock.slock)
2516 struct sk_buff *skb, *next;
2518 while ((skb = sk->sk_backlog.head) != NULL) {
2519 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2521 spin_unlock_bh(&sk->sk_lock.slock);
2526 WARN_ON_ONCE(skb_dst_is_noref(skb));
2527 skb_mark_not_on_list(skb);
2528 sk_backlog_rcv(sk, skb);
2533 } while (skb != NULL);
2535 spin_lock_bh(&sk->sk_lock.slock);
2539 * Doing the zeroing here guarantee we can not loop forever
2540 * while a wild producer attempts to flood us.
2542 sk->sk_backlog.len = 0;
2545 void __sk_flush_backlog(struct sock *sk)
2547 spin_lock_bh(&sk->sk_lock.slock);
2549 spin_unlock_bh(&sk->sk_lock.slock);
2553 * sk_wait_data - wait for data to arrive at sk_receive_queue
2554 * @sk: sock to wait on
2555 * @timeo: for how long
2556 * @skb: last skb seen on sk_receive_queue
2558 * Now socket state including sk->sk_err is changed only under lock,
2559 * hence we may omit checks after joining wait queue.
2560 * We check receive queue before schedule() only as optimization;
2561 * it is very likely that release_sock() added new data.
2563 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2565 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2568 add_wait_queue(sk_sleep(sk), &wait);
2569 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2570 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2571 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2572 remove_wait_queue(sk_sleep(sk), &wait);
2575 EXPORT_SYMBOL(sk_wait_data);
2578 * __sk_mem_raise_allocated - increase memory_allocated
2580 * @size: memory size to allocate
2581 * @amt: pages to allocate
2582 * @kind: allocation type
2584 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2586 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2588 struct proto *prot = sk->sk_prot;
2589 long allocated = sk_memory_allocated_add(sk, amt);
2590 bool charged = true;
2592 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2593 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2594 goto suppress_allocation;
2597 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2598 sk_leave_memory_pressure(sk);
2602 /* Under pressure. */
2603 if (allocated > sk_prot_mem_limits(sk, 1))
2604 sk_enter_memory_pressure(sk);
2606 /* Over hard limit. */
2607 if (allocated > sk_prot_mem_limits(sk, 2))
2608 goto suppress_allocation;
2610 /* guarantee minimum buffer size under pressure */
2611 if (kind == SK_MEM_RECV) {
2612 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2615 } else { /* SK_MEM_SEND */
2616 int wmem0 = sk_get_wmem0(sk, prot);
2618 if (sk->sk_type == SOCK_STREAM) {
2619 if (sk->sk_wmem_queued < wmem0)
2621 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2626 if (sk_has_memory_pressure(sk)) {
2629 if (!sk_under_memory_pressure(sk))
2631 alloc = sk_sockets_allocated_read_positive(sk);
2632 if (sk_prot_mem_limits(sk, 2) > alloc *
2633 sk_mem_pages(sk->sk_wmem_queued +
2634 atomic_read(&sk->sk_rmem_alloc) +
2635 sk->sk_forward_alloc))
2639 suppress_allocation:
2641 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2642 sk_stream_moderate_sndbuf(sk);
2644 /* Fail only if socket is _under_ its sndbuf.
2645 * In this case we cannot block, so that we have to fail.
2647 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2651 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2652 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2654 sk_memory_allocated_sub(sk, amt);
2656 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2657 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2661 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2664 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2666 * @size: memory size to allocate
2667 * @kind: allocation type
2669 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2670 * rmem allocation. This function assumes that protocols which have
2671 * memory_pressure use sk_wmem_queued as write buffer accounting.
2673 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2675 int ret, amt = sk_mem_pages(size);
2677 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2678 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2680 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2683 EXPORT_SYMBOL(__sk_mem_schedule);
2686 * __sk_mem_reduce_allocated - reclaim memory_allocated
2688 * @amount: number of quanta
2690 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2692 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2694 sk_memory_allocated_sub(sk, amount);
2696 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2697 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2699 if (sk_under_memory_pressure(sk) &&
2700 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2701 sk_leave_memory_pressure(sk);
2703 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2706 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2708 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2710 void __sk_mem_reclaim(struct sock *sk, int amount)
2712 amount >>= SK_MEM_QUANTUM_SHIFT;
2713 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2714 __sk_mem_reduce_allocated(sk, amount);
2716 EXPORT_SYMBOL(__sk_mem_reclaim);
2718 int sk_set_peek_off(struct sock *sk, int val)
2720 sk->sk_peek_off = val;
2723 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2726 * Set of default routines for initialising struct proto_ops when
2727 * the protocol does not support a particular function. In certain
2728 * cases where it makes no sense for a protocol to have a "do nothing"
2729 * function, some default processing is provided.
2732 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2736 EXPORT_SYMBOL(sock_no_bind);
2738 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2743 EXPORT_SYMBOL(sock_no_connect);
2745 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2749 EXPORT_SYMBOL(sock_no_socketpair);
2751 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2756 EXPORT_SYMBOL(sock_no_accept);
2758 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2763 EXPORT_SYMBOL(sock_no_getname);
2765 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2769 EXPORT_SYMBOL(sock_no_ioctl);
2771 int sock_no_listen(struct socket *sock, int backlog)
2775 EXPORT_SYMBOL(sock_no_listen);
2777 int sock_no_shutdown(struct socket *sock, int how)
2781 EXPORT_SYMBOL(sock_no_shutdown);
2783 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2787 EXPORT_SYMBOL(sock_no_sendmsg);
2789 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2793 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2795 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2800 EXPORT_SYMBOL(sock_no_recvmsg);
2802 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2804 /* Mirror missing mmap method error code */
2807 EXPORT_SYMBOL(sock_no_mmap);
2810 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2811 * various sock-based usage counts.
2813 void __receive_sock(struct file *file)
2815 struct socket *sock;
2819 * The resulting value of "error" is ignored here since we only
2820 * need to take action when the file is a socket and testing
2821 * "sock" for NULL is sufficient.
2823 sock = sock_from_file(file, &error);
2825 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2826 sock_update_classid(&sock->sk->sk_cgrp_data);
2830 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2833 struct msghdr msg = {.msg_flags = flags};
2835 char *kaddr = kmap(page);
2836 iov.iov_base = kaddr + offset;
2838 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2842 EXPORT_SYMBOL(sock_no_sendpage);
2844 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2845 int offset, size_t size, int flags)
2848 struct msghdr msg = {.msg_flags = flags};
2850 char *kaddr = kmap(page);
2852 iov.iov_base = kaddr + offset;
2854 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2858 EXPORT_SYMBOL(sock_no_sendpage_locked);
2861 * Default Socket Callbacks
2864 static void sock_def_wakeup(struct sock *sk)
2866 struct socket_wq *wq;
2869 wq = rcu_dereference(sk->sk_wq);
2870 if (skwq_has_sleeper(wq))
2871 wake_up_interruptible_all(&wq->wait);
2875 static void sock_def_error_report(struct sock *sk)
2877 struct socket_wq *wq;
2880 wq = rcu_dereference(sk->sk_wq);
2881 if (skwq_has_sleeper(wq))
2882 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2883 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2887 void sock_def_readable(struct sock *sk)
2889 struct socket_wq *wq;
2892 wq = rcu_dereference(sk->sk_wq);
2893 if (skwq_has_sleeper(wq))
2894 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2895 EPOLLRDNORM | EPOLLRDBAND);
2896 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2900 static void sock_def_write_space(struct sock *sk)
2902 struct socket_wq *wq;
2906 /* Do not wake up a writer until he can make "significant"
2909 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2910 wq = rcu_dereference(sk->sk_wq);
2911 if (skwq_has_sleeper(wq))
2912 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2913 EPOLLWRNORM | EPOLLWRBAND);
2915 /* Should agree with poll, otherwise some programs break */
2916 if (sock_writeable(sk))
2917 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2923 static void sock_def_destruct(struct sock *sk)
2927 void sk_send_sigurg(struct sock *sk)
2929 if (sk->sk_socket && sk->sk_socket->file)
2930 if (send_sigurg(&sk->sk_socket->file->f_owner))
2931 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2933 EXPORT_SYMBOL(sk_send_sigurg);
2935 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2936 unsigned long expires)
2938 if (!mod_timer(timer, expires))
2941 EXPORT_SYMBOL(sk_reset_timer);
2943 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2945 if (del_timer(timer))
2948 EXPORT_SYMBOL(sk_stop_timer);
2950 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
2952 if (del_timer_sync(timer))
2955 EXPORT_SYMBOL(sk_stop_timer_sync);
2957 void sock_init_data(struct socket *sock, struct sock *sk)
2960 sk->sk_send_head = NULL;
2962 timer_setup(&sk->sk_timer, NULL, 0);
2964 sk->sk_allocation = GFP_KERNEL;
2965 sk->sk_rcvbuf = sysctl_rmem_default;
2966 sk->sk_sndbuf = sysctl_wmem_default;
2967 sk->sk_state = TCP_CLOSE;
2968 sk_set_socket(sk, sock);
2970 sock_set_flag(sk, SOCK_ZAPPED);
2973 sk->sk_type = sock->type;
2974 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2976 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2978 RCU_INIT_POINTER(sk->sk_wq, NULL);
2979 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2982 rwlock_init(&sk->sk_callback_lock);
2983 if (sk->sk_kern_sock)
2984 lockdep_set_class_and_name(
2985 &sk->sk_callback_lock,
2986 af_kern_callback_keys + sk->sk_family,
2987 af_family_kern_clock_key_strings[sk->sk_family]);
2989 lockdep_set_class_and_name(
2990 &sk->sk_callback_lock,
2991 af_callback_keys + sk->sk_family,
2992 af_family_clock_key_strings[sk->sk_family]);
2994 sk->sk_state_change = sock_def_wakeup;
2995 sk->sk_data_ready = sock_def_readable;
2996 sk->sk_write_space = sock_def_write_space;
2997 sk->sk_error_report = sock_def_error_report;
2998 sk->sk_destruct = sock_def_destruct;
3000 sk->sk_frag.page = NULL;
3001 sk->sk_frag.offset = 0;
3002 sk->sk_peek_off = -1;
3004 sk->sk_peer_pid = NULL;
3005 sk->sk_peer_cred = NULL;
3006 sk->sk_write_pending = 0;
3007 sk->sk_rcvlowat = 1;
3008 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3009 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3011 sk->sk_stamp = SK_DEFAULT_STAMP;
3012 #if BITS_PER_LONG==32
3013 seqlock_init(&sk->sk_stamp_seq);
3015 atomic_set(&sk->sk_zckey, 0);
3017 #ifdef CONFIG_NET_RX_BUSY_POLL
3019 sk->sk_ll_usec = sysctl_net_busy_read;
3022 sk->sk_max_pacing_rate = ~0UL;
3023 sk->sk_pacing_rate = ~0UL;
3024 WRITE_ONCE(sk->sk_pacing_shift, 10);
3025 sk->sk_incoming_cpu = -1;
3027 sk_rx_queue_clear(sk);
3029 * Before updating sk_refcnt, we must commit prior changes to memory
3030 * (Documentation/RCU/rculist_nulls.rst for details)
3033 refcount_set(&sk->sk_refcnt, 1);
3034 atomic_set(&sk->sk_drops, 0);
3036 EXPORT_SYMBOL(sock_init_data);
3038 void lock_sock_nested(struct sock *sk, int subclass)
3041 spin_lock_bh(&sk->sk_lock.slock);
3042 if (sk->sk_lock.owned)
3044 sk->sk_lock.owned = 1;
3045 spin_unlock(&sk->sk_lock.slock);
3047 * The sk_lock has mutex_lock() semantics here:
3049 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3052 EXPORT_SYMBOL(lock_sock_nested);
3054 void release_sock(struct sock *sk)
3056 spin_lock_bh(&sk->sk_lock.slock);
3057 if (sk->sk_backlog.tail)
3060 /* Warning : release_cb() might need to release sk ownership,
3061 * ie call sock_release_ownership(sk) before us.
3063 if (sk->sk_prot->release_cb)
3064 sk->sk_prot->release_cb(sk);
3066 sock_release_ownership(sk);
3067 if (waitqueue_active(&sk->sk_lock.wq))
3068 wake_up(&sk->sk_lock.wq);
3069 spin_unlock_bh(&sk->sk_lock.slock);
3071 EXPORT_SYMBOL(release_sock);
3074 * lock_sock_fast - fast version of lock_sock
3077 * This version should be used for very small section, where process wont block
3078 * return false if fast path is taken:
3080 * sk_lock.slock locked, owned = 0, BH disabled
3082 * return true if slow path is taken:
3084 * sk_lock.slock unlocked, owned = 1, BH enabled
3086 bool lock_sock_fast(struct sock *sk)
3089 spin_lock_bh(&sk->sk_lock.slock);
3091 if (!sk->sk_lock.owned)
3093 * Note : We must disable BH
3098 sk->sk_lock.owned = 1;
3099 spin_unlock(&sk->sk_lock.slock);
3101 * The sk_lock has mutex_lock() semantics here:
3103 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3107 EXPORT_SYMBOL(lock_sock_fast);
3109 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3110 bool timeval, bool time32)
3112 struct sock *sk = sock->sk;
3113 struct timespec64 ts;
3115 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3116 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3117 if (ts.tv_sec == -1)
3119 if (ts.tv_sec == 0) {
3120 ktime_t kt = ktime_get_real();
3121 sock_write_timestamp(sk, kt);
3122 ts = ktime_to_timespec64(kt);
3128 #ifdef CONFIG_COMPAT_32BIT_TIME
3130 return put_old_timespec32(&ts, userstamp);
3132 #ifdef CONFIG_SPARC64
3133 /* beware of padding in sparc64 timeval */
3134 if (timeval && !in_compat_syscall()) {
3135 struct __kernel_old_timeval __user tv = {
3136 .tv_sec = ts.tv_sec,
3137 .tv_usec = ts.tv_nsec,
3139 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3144 return put_timespec64(&ts, userstamp);
3146 EXPORT_SYMBOL(sock_gettstamp);
3148 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3150 if (!sock_flag(sk, flag)) {
3151 unsigned long previous_flags = sk->sk_flags;
3153 sock_set_flag(sk, flag);
3155 * we just set one of the two flags which require net
3156 * time stamping, but time stamping might have been on
3157 * already because of the other one
3159 if (sock_needs_netstamp(sk) &&
3160 !(previous_flags & SK_FLAGS_TIMESTAMP))
3161 net_enable_timestamp();
3165 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3166 int level, int type)
3168 struct sock_exterr_skb *serr;
3169 struct sk_buff *skb;
3173 skb = sock_dequeue_err_skb(sk);
3179 msg->msg_flags |= MSG_TRUNC;
3182 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3186 sock_recv_timestamp(msg, sk, skb);
3188 serr = SKB_EXT_ERR(skb);
3189 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3191 msg->msg_flags |= MSG_ERRQUEUE;
3199 EXPORT_SYMBOL(sock_recv_errqueue);
3202 * Get a socket option on an socket.
3204 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3205 * asynchronous errors should be reported by getsockopt. We assume
3206 * this means if you specify SO_ERROR (otherwise whats the point of it).
3208 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3209 char __user *optval, int __user *optlen)
3211 struct sock *sk = sock->sk;
3213 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3215 EXPORT_SYMBOL(sock_common_getsockopt);
3217 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3220 struct sock *sk = sock->sk;
3224 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3225 flags & ~MSG_DONTWAIT, &addr_len);
3227 msg->msg_namelen = addr_len;
3230 EXPORT_SYMBOL(sock_common_recvmsg);
3233 * Set socket options on an inet socket.
3235 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3236 sockptr_t optval, unsigned int optlen)
3238 struct sock *sk = sock->sk;
3240 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3242 EXPORT_SYMBOL(sock_common_setsockopt);
3244 void sk_common_release(struct sock *sk)
3246 if (sk->sk_prot->destroy)
3247 sk->sk_prot->destroy(sk);
3250 * Observation: when sk_common_release is called, processes have
3251 * no access to socket. But net still has.
3252 * Step one, detach it from networking:
3254 * A. Remove from hash tables.
3257 sk->sk_prot->unhash(sk);
3260 * In this point socket cannot receive new packets, but it is possible
3261 * that some packets are in flight because some CPU runs receiver and
3262 * did hash table lookup before we unhashed socket. They will achieve
3263 * receive queue and will be purged by socket destructor.
3265 * Also we still have packets pending on receive queue and probably,
3266 * our own packets waiting in device queues. sock_destroy will drain
3267 * receive queue, but transmitted packets will delay socket destruction
3268 * until the last reference will be released.
3273 xfrm_sk_free_policy(sk);
3275 sk_refcnt_debug_release(sk);
3279 EXPORT_SYMBOL(sk_common_release);
3281 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3283 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3285 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3286 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3287 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3288 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3289 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3290 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3291 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3292 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3293 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3296 #ifdef CONFIG_PROC_FS
3297 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3299 int val[PROTO_INUSE_NR];
3302 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3304 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3306 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3308 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3310 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3312 int cpu, idx = prot->inuse_idx;
3315 for_each_possible_cpu(cpu)
3316 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3318 return res >= 0 ? res : 0;
3320 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3322 static void sock_inuse_add(struct net *net, int val)
3324 this_cpu_add(*net->core.sock_inuse, val);
3327 int sock_inuse_get(struct net *net)
3331 for_each_possible_cpu(cpu)
3332 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3337 EXPORT_SYMBOL_GPL(sock_inuse_get);
3339 static int __net_init sock_inuse_init_net(struct net *net)
3341 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3342 if (net->core.prot_inuse == NULL)
3345 net->core.sock_inuse = alloc_percpu(int);
3346 if (net->core.sock_inuse == NULL)
3352 free_percpu(net->core.prot_inuse);
3356 static void __net_exit sock_inuse_exit_net(struct net *net)
3358 free_percpu(net->core.prot_inuse);
3359 free_percpu(net->core.sock_inuse);
3362 static struct pernet_operations net_inuse_ops = {
3363 .init = sock_inuse_init_net,
3364 .exit = sock_inuse_exit_net,
3367 static __init int net_inuse_init(void)
3369 if (register_pernet_subsys(&net_inuse_ops))
3370 panic("Cannot initialize net inuse counters");
3375 core_initcall(net_inuse_init);
3377 static int assign_proto_idx(struct proto *prot)
3379 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3381 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3382 pr_err("PROTO_INUSE_NR exhausted\n");
3386 set_bit(prot->inuse_idx, proto_inuse_idx);
3390 static void release_proto_idx(struct proto *prot)
3392 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3393 clear_bit(prot->inuse_idx, proto_inuse_idx);
3396 static inline int assign_proto_idx(struct proto *prot)
3401 static inline void release_proto_idx(struct proto *prot)
3405 static void sock_inuse_add(struct net *net, int val)
3410 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3414 kfree(twsk_prot->twsk_slab_name);
3415 twsk_prot->twsk_slab_name = NULL;
3416 kmem_cache_destroy(twsk_prot->twsk_slab);
3417 twsk_prot->twsk_slab = NULL;
3420 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3424 kfree(rsk_prot->slab_name);
3425 rsk_prot->slab_name = NULL;
3426 kmem_cache_destroy(rsk_prot->slab);
3427 rsk_prot->slab = NULL;
3430 static int req_prot_init(const struct proto *prot)
3432 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3437 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3439 if (!rsk_prot->slab_name)
3442 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3443 rsk_prot->obj_size, 0,
3444 SLAB_ACCOUNT | prot->slab_flags,
3447 if (!rsk_prot->slab) {
3448 pr_crit("%s: Can't create request sock SLAB cache!\n",
3455 int proto_register(struct proto *prot, int alloc_slab)
3460 prot->slab = kmem_cache_create_usercopy(prot->name,
3462 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3464 prot->useroffset, prot->usersize,
3467 if (prot->slab == NULL) {
3468 pr_crit("%s: Can't create sock SLAB cache!\n",
3473 if (req_prot_init(prot))
3474 goto out_free_request_sock_slab;
3476 if (prot->twsk_prot != NULL) {
3477 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3479 if (prot->twsk_prot->twsk_slab_name == NULL)
3480 goto out_free_request_sock_slab;
3482 prot->twsk_prot->twsk_slab =
3483 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3484 prot->twsk_prot->twsk_obj_size,
3489 if (prot->twsk_prot->twsk_slab == NULL)
3490 goto out_free_timewait_sock_slab;
3494 mutex_lock(&proto_list_mutex);
3495 ret = assign_proto_idx(prot);
3497 mutex_unlock(&proto_list_mutex);
3498 goto out_free_timewait_sock_slab;
3500 list_add(&prot->node, &proto_list);
3501 mutex_unlock(&proto_list_mutex);
3504 out_free_timewait_sock_slab:
3505 if (alloc_slab && prot->twsk_prot)
3506 tw_prot_cleanup(prot->twsk_prot);
3507 out_free_request_sock_slab:
3509 req_prot_cleanup(prot->rsk_prot);
3511 kmem_cache_destroy(prot->slab);
3517 EXPORT_SYMBOL(proto_register);
3519 void proto_unregister(struct proto *prot)
3521 mutex_lock(&proto_list_mutex);
3522 release_proto_idx(prot);
3523 list_del(&prot->node);
3524 mutex_unlock(&proto_list_mutex);
3526 kmem_cache_destroy(prot->slab);
3529 req_prot_cleanup(prot->rsk_prot);
3530 tw_prot_cleanup(prot->twsk_prot);
3532 EXPORT_SYMBOL(proto_unregister);
3534 int sock_load_diag_module(int family, int protocol)
3537 if (!sock_is_registered(family))
3540 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3541 NETLINK_SOCK_DIAG, family);
3545 if (family == AF_INET &&
3546 protocol != IPPROTO_RAW &&
3547 protocol < MAX_INET_PROTOS &&
3548 !rcu_access_pointer(inet_protos[protocol]))
3552 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3553 NETLINK_SOCK_DIAG, family, protocol);
3555 EXPORT_SYMBOL(sock_load_diag_module);
3557 #ifdef CONFIG_PROC_FS
3558 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3559 __acquires(proto_list_mutex)
3561 mutex_lock(&proto_list_mutex);
3562 return seq_list_start_head(&proto_list, *pos);
3565 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3567 return seq_list_next(v, &proto_list, pos);
3570 static void proto_seq_stop(struct seq_file *seq, void *v)
3571 __releases(proto_list_mutex)
3573 mutex_unlock(&proto_list_mutex);
3576 static char proto_method_implemented(const void *method)
3578 return method == NULL ? 'n' : 'y';
3580 static long sock_prot_memory_allocated(struct proto *proto)
3582 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3585 static const char *sock_prot_memory_pressure(struct proto *proto)
3587 return proto->memory_pressure != NULL ?
3588 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3591 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3594 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3595 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3598 sock_prot_inuse_get(seq_file_net(seq), proto),
3599 sock_prot_memory_allocated(proto),
3600 sock_prot_memory_pressure(proto),
3602 proto->slab == NULL ? "no" : "yes",
3603 module_name(proto->owner),
3604 proto_method_implemented(proto->close),
3605 proto_method_implemented(proto->connect),
3606 proto_method_implemented(proto->disconnect),
3607 proto_method_implemented(proto->accept),
3608 proto_method_implemented(proto->ioctl),
3609 proto_method_implemented(proto->init),
3610 proto_method_implemented(proto->destroy),
3611 proto_method_implemented(proto->shutdown),
3612 proto_method_implemented(proto->setsockopt),
3613 proto_method_implemented(proto->getsockopt),
3614 proto_method_implemented(proto->sendmsg),
3615 proto_method_implemented(proto->recvmsg),
3616 proto_method_implemented(proto->sendpage),
3617 proto_method_implemented(proto->bind),
3618 proto_method_implemented(proto->backlog_rcv),
3619 proto_method_implemented(proto->hash),
3620 proto_method_implemented(proto->unhash),
3621 proto_method_implemented(proto->get_port),
3622 proto_method_implemented(proto->enter_memory_pressure));
3625 static int proto_seq_show(struct seq_file *seq, void *v)
3627 if (v == &proto_list)
3628 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3637 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3639 proto_seq_printf(seq, list_entry(v, struct proto, node));
3643 static const struct seq_operations proto_seq_ops = {
3644 .start = proto_seq_start,
3645 .next = proto_seq_next,
3646 .stop = proto_seq_stop,
3647 .show = proto_seq_show,
3650 static __net_init int proto_init_net(struct net *net)
3652 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3653 sizeof(struct seq_net_private)))
3659 static __net_exit void proto_exit_net(struct net *net)
3661 remove_proc_entry("protocols", net->proc_net);
3665 static __net_initdata struct pernet_operations proto_net_ops = {
3666 .init = proto_init_net,
3667 .exit = proto_exit_net,
3670 static int __init proto_init(void)
3672 return register_pernet_subsys(&proto_net_ops);
3675 subsys_initcall(proto_init);
3677 #endif /* PROC_FS */
3679 #ifdef CONFIG_NET_RX_BUSY_POLL
3680 bool sk_busy_loop_end(void *p, unsigned long start_time)
3682 struct sock *sk = p;
3684 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3685 sk_busy_loop_timeout(sk, start_time);
3687 EXPORT_SYMBOL(sk_busy_loop_end);
3688 #endif /* CONFIG_NET_RX_BUSY_POLL */
3690 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3692 if (!sk->sk_prot->bind_add)
3694 return sk->sk_prot->bind_add(sk, addr, addr_len);
3696 EXPORT_SYMBOL(sock_bind_add);