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>
117 #include <linux/uaccess.h>
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/net_namespace.h>
123 #include <net/request_sock.h>
124 #include <net/sock.h>
125 #include <linux/net_tstamp.h>
126 #include <net/xfrm.h>
127 #include <linux/ipsec.h>
128 #include <net/cls_cgroup.h>
129 #include <net/netprio_cgroup.h>
130 #include <linux/sock_diag.h>
132 #include <linux/filter.h>
133 #include <net/sock_reuseport.h>
134 #include <net/bpf_sk_storage.h>
136 #include <trace/events/sock.h>
139 #include <net/busy_poll.h>
141 static DEFINE_MUTEX(proto_list_mutex);
142 static LIST_HEAD(proto_list);
144 static void sock_inuse_add(struct net *net, int val);
147 * sk_ns_capable - General socket capability test
148 * @sk: Socket to use a capability on or through
149 * @user_ns: The user namespace of the capability to use
150 * @cap: The capability to use
152 * Test to see if the opener of the socket had when the socket was
153 * created and the current process has the capability @cap in the user
154 * namespace @user_ns.
156 bool sk_ns_capable(const struct sock *sk,
157 struct user_namespace *user_ns, int cap)
159 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
160 ns_capable(user_ns, cap);
162 EXPORT_SYMBOL(sk_ns_capable);
165 * sk_capable - Socket global capability test
166 * @sk: Socket to use a capability on or through
167 * @cap: The global capability to use
169 * Test to see if the opener of the socket had when the socket was
170 * created and the current process has the capability @cap in all user
173 bool sk_capable(const struct sock *sk, int cap)
175 return sk_ns_capable(sk, &init_user_ns, cap);
177 EXPORT_SYMBOL(sk_capable);
180 * sk_net_capable - Network namespace socket capability test
181 * @sk: Socket to use a capability on or through
182 * @cap: The capability to use
184 * Test to see if the opener of the socket had when the socket was created
185 * and the current process has the capability @cap over the network namespace
186 * the socket is a member of.
188 bool sk_net_capable(const struct sock *sk, int cap)
190 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
192 EXPORT_SYMBOL(sk_net_capable);
195 * Each address family might have different locking rules, so we have
196 * one slock key per address family and separate keys for internal and
199 static struct lock_class_key af_family_keys[AF_MAX];
200 static struct lock_class_key af_family_kern_keys[AF_MAX];
201 static struct lock_class_key af_family_slock_keys[AF_MAX];
202 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
205 * Make lock validator output more readable. (we pre-construct these
206 * strings build-time, so that runtime initialization of socket
210 #define _sock_locks(x) \
211 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
212 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
213 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
214 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
215 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
216 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
217 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
218 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
219 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
220 x "27" , x "28" , x "AF_CAN" , \
221 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
222 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
223 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
224 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
225 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
228 static const char *const af_family_key_strings[AF_MAX+1] = {
229 _sock_locks("sk_lock-")
231 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
232 _sock_locks("slock-")
234 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
235 _sock_locks("clock-")
238 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
239 _sock_locks("k-sk_lock-")
241 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
242 _sock_locks("k-slock-")
244 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
245 _sock_locks("k-clock-")
247 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
248 _sock_locks("rlock-")
250 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
251 _sock_locks("wlock-")
253 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
254 _sock_locks("elock-")
258 * sk_callback_lock and sk queues locking rules are per-address-family,
259 * so split the lock classes by using a per-AF key:
261 static struct lock_class_key af_callback_keys[AF_MAX];
262 static struct lock_class_key af_rlock_keys[AF_MAX];
263 static struct lock_class_key af_wlock_keys[AF_MAX];
264 static struct lock_class_key af_elock_keys[AF_MAX];
265 static struct lock_class_key af_kern_callback_keys[AF_MAX];
267 /* Run time adjustable parameters. */
268 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
269 EXPORT_SYMBOL(sysctl_wmem_max);
270 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
271 EXPORT_SYMBOL(sysctl_rmem_max);
272 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
273 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
275 /* Maximal space eaten by iovec or ancillary data plus some space */
276 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
277 EXPORT_SYMBOL(sysctl_optmem_max);
279 int sysctl_tstamp_allow_data __read_mostly = 1;
281 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
282 EXPORT_SYMBOL_GPL(memalloc_socks_key);
285 * sk_set_memalloc - sets %SOCK_MEMALLOC
286 * @sk: socket to set it on
288 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
289 * It's the responsibility of the admin to adjust min_free_kbytes
290 * to meet the requirements
292 void sk_set_memalloc(struct sock *sk)
294 sock_set_flag(sk, SOCK_MEMALLOC);
295 sk->sk_allocation |= __GFP_MEMALLOC;
296 static_branch_inc(&memalloc_socks_key);
298 EXPORT_SYMBOL_GPL(sk_set_memalloc);
300 void sk_clear_memalloc(struct sock *sk)
302 sock_reset_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation &= ~__GFP_MEMALLOC;
304 static_branch_dec(&memalloc_socks_key);
307 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
308 * progress of swapping. SOCK_MEMALLOC may be cleared while
309 * it has rmem allocations due to the last swapfile being deactivated
310 * but there is a risk that the socket is unusable due to exceeding
311 * the rmem limits. Reclaim the reserves and obey rmem limits again.
315 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
317 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
320 unsigned int noreclaim_flag;
322 /* these should have been dropped before queueing */
323 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
325 noreclaim_flag = memalloc_noreclaim_save();
326 ret = sk->sk_backlog_rcv(sk, skb);
327 memalloc_noreclaim_restore(noreclaim_flag);
331 EXPORT_SYMBOL(__sk_backlog_rcv);
333 static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
335 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 size = sizeof(old_tv);
359 *(struct __kernel_sock_timeval *)optval = tv;
366 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval)
368 struct __kernel_sock_timeval tv;
370 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
371 struct old_timeval32 tv32;
373 if (optlen < sizeof(tv32))
376 if (copy_from_user(&tv32, optval, sizeof(tv32)))
378 tv.tv_sec = tv32.tv_sec;
379 tv.tv_usec = tv32.tv_usec;
380 } else if (old_timeval) {
381 struct __kernel_old_timeval old_tv;
383 if (optlen < sizeof(old_tv))
385 if (copy_from_user(&old_tv, optval, sizeof(old_tv)))
387 tv.tv_sec = old_tv.tv_sec;
388 tv.tv_usec = old_tv.tv_usec;
390 if (optlen < sizeof(tv))
392 if (copy_from_user(&tv, optval, sizeof(tv)))
395 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
399 static int warned __read_mostly;
402 if (warned < 10 && net_ratelimit()) {
404 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
405 __func__, current->comm, task_pid_nr(current));
409 *timeo_p = MAX_SCHEDULE_TIMEOUT;
410 if (tv.tv_sec == 0 && tv.tv_usec == 0)
412 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
413 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
417 static void sock_warn_obsolete_bsdism(const char *name)
420 static char warncomm[TASK_COMM_LEN];
421 if (strcmp(warncomm, current->comm) && warned < 5) {
422 strcpy(warncomm, current->comm);
423 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
429 static bool sock_needs_netstamp(const struct sock *sk)
431 switch (sk->sk_family) {
440 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
442 if (sk->sk_flags & flags) {
443 sk->sk_flags &= ~flags;
444 if (sock_needs_netstamp(sk) &&
445 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
446 net_disable_timestamp();
451 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
454 struct sk_buff_head *list = &sk->sk_receive_queue;
456 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
457 atomic_inc(&sk->sk_drops);
458 trace_sock_rcvqueue_full(sk, skb);
462 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
463 atomic_inc(&sk->sk_drops);
468 skb_set_owner_r(skb, sk);
470 /* we escape from rcu protected region, make sure we dont leak
475 spin_lock_irqsave(&list->lock, flags);
476 sock_skb_set_dropcount(sk, skb);
477 __skb_queue_tail(list, skb);
478 spin_unlock_irqrestore(&list->lock, flags);
480 if (!sock_flag(sk, SOCK_DEAD))
481 sk->sk_data_ready(sk);
484 EXPORT_SYMBOL(__sock_queue_rcv_skb);
486 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
490 err = sk_filter(sk, skb);
494 return __sock_queue_rcv_skb(sk, skb);
496 EXPORT_SYMBOL(sock_queue_rcv_skb);
498 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
499 const int nested, unsigned int trim_cap, bool refcounted)
501 int rc = NET_RX_SUCCESS;
503 if (sk_filter_trim_cap(sk, skb, trim_cap))
504 goto discard_and_relse;
508 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
509 atomic_inc(&sk->sk_drops);
510 goto discard_and_relse;
513 bh_lock_sock_nested(sk);
516 if (!sock_owned_by_user(sk)) {
518 * trylock + unlock semantics:
520 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
522 rc = sk_backlog_rcv(sk, skb);
524 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
525 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
527 atomic_inc(&sk->sk_drops);
528 goto discard_and_relse;
540 EXPORT_SYMBOL(__sk_receive_skb);
542 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
544 struct dst_entry *dst = __sk_dst_get(sk);
546 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
547 sk_tx_queue_clear(sk);
548 sk->sk_dst_pending_confirm = 0;
549 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
556 EXPORT_SYMBOL(__sk_dst_check);
558 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
560 struct dst_entry *dst = sk_dst_get(sk);
562 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
570 EXPORT_SYMBOL(sk_dst_check);
572 static int sock_setbindtodevice_locked(struct sock *sk, int ifindex)
574 int ret = -ENOPROTOOPT;
575 #ifdef CONFIG_NETDEVICES
576 struct net *net = sock_net(sk);
580 if (!ns_capable(net->user_ns, CAP_NET_RAW))
587 sk->sk_bound_dev_if = ifindex;
588 if (sk->sk_prot->rehash)
589 sk->sk_prot->rehash(sk);
600 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
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_user(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;
641 ret = sock_setbindtodevice_locked(sk, index);
650 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
651 int __user *optlen, int len)
653 int ret = -ENOPROTOOPT;
654 #ifdef CONFIG_NETDEVICES
655 struct net *net = sock_net(sk);
656 char devname[IFNAMSIZ];
658 if (sk->sk_bound_dev_if == 0) {
667 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
671 len = strlen(devname) + 1;
674 if (copy_to_user(optval, devname, len))
679 if (put_user(len, optlen))
690 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
693 sock_set_flag(sk, bit);
695 sock_reset_flag(sk, bit);
698 bool sk_mc_loop(struct sock *sk)
700 if (dev_recursion_level())
704 switch (sk->sk_family) {
706 return inet_sk(sk)->mc_loop;
707 #if IS_ENABLED(CONFIG_IPV6)
709 return inet6_sk(sk)->mc_loop;
715 EXPORT_SYMBOL(sk_mc_loop);
718 * This is meant for all protocols to use and covers goings on
719 * at the socket level. Everything here is generic.
722 int sock_setsockopt(struct socket *sock, int level, int optname,
723 char __user *optval, unsigned int optlen)
725 struct sock_txtime sk_txtime;
726 struct sock *sk = sock->sk;
733 * Options without arguments
736 if (optname == SO_BINDTODEVICE)
737 return sock_setbindtodevice(sk, optval, optlen);
739 if (optlen < sizeof(int))
742 if (get_user(val, (int __user *)optval))
745 valbool = val ? 1 : 0;
751 if (val && !capable(CAP_NET_ADMIN))
754 sock_valbool_flag(sk, SOCK_DBG, valbool);
757 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
760 sk->sk_reuseport = valbool;
769 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
773 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
776 /* Don't error on this BSD doesn't and if you think
777 * about it this is right. Otherwise apps have to
778 * play 'guess the biggest size' games. RCVBUF/SNDBUF
779 * are treated in BSD as hints
781 val = min_t(u32, val, sysctl_wmem_max);
783 /* Ensure val * 2 fits into an int, to prevent max_t()
784 * from treating it as a negative value.
786 val = min_t(int, val, INT_MAX / 2);
787 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
788 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
789 /* Wake up sending tasks if we upped the value. */
790 sk->sk_write_space(sk);
794 if (!capable(CAP_NET_ADMIN)) {
799 /* No negative values (to prevent underflow, as val will be
807 /* Don't error on this BSD doesn't and if you think
808 * about it this is right. Otherwise apps have to
809 * play 'guess the biggest size' games. RCVBUF/SNDBUF
810 * are treated in BSD as hints
812 val = min_t(u32, val, sysctl_rmem_max);
814 /* Ensure val * 2 fits into an int, to prevent max_t()
815 * from treating it as a negative value.
817 val = min_t(int, val, INT_MAX / 2);
818 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
820 * We double it on the way in to account for
821 * "struct sk_buff" etc. overhead. Applications
822 * assume that the SO_RCVBUF setting they make will
823 * allow that much actual data to be received on that
826 * Applications are unaware that "struct sk_buff" and
827 * other overheads allocate from the receive buffer
828 * during socket buffer allocation.
830 * And after considering the possible alternatives,
831 * returning the value we actually used in getsockopt
832 * is the most desirable behavior.
834 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
838 if (!capable(CAP_NET_ADMIN)) {
843 /* No negative values (to prevent underflow, as val will be
851 if (sk->sk_prot->keepalive)
852 sk->sk_prot->keepalive(sk, valbool);
853 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
857 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
861 sk->sk_no_check_tx = valbool;
865 if ((val >= 0 && val <= 6) ||
866 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
867 sk->sk_priority = val;
873 if (optlen < sizeof(ling)) {
874 ret = -EINVAL; /* 1003.1g */
877 if (copy_from_user(&ling, optval, sizeof(ling))) {
882 sock_reset_flag(sk, SOCK_LINGER);
884 #if (BITS_PER_LONG == 32)
885 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
886 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
889 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
890 sock_set_flag(sk, SOCK_LINGER);
895 sock_warn_obsolete_bsdism("setsockopt");
900 set_bit(SOCK_PASSCRED, &sock->flags);
902 clear_bit(SOCK_PASSCRED, &sock->flags);
905 case SO_TIMESTAMP_OLD:
906 case SO_TIMESTAMP_NEW:
907 case SO_TIMESTAMPNS_OLD:
908 case SO_TIMESTAMPNS_NEW:
910 if (optname == SO_TIMESTAMP_NEW || optname == SO_TIMESTAMPNS_NEW)
911 sock_set_flag(sk, SOCK_TSTAMP_NEW);
913 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
915 if (optname == SO_TIMESTAMP_OLD || optname == SO_TIMESTAMP_NEW)
916 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
918 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
919 sock_set_flag(sk, SOCK_RCVTSTAMP);
920 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
922 sock_reset_flag(sk, SOCK_RCVTSTAMP);
923 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
924 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
928 case SO_TIMESTAMPING_NEW:
929 sock_set_flag(sk, SOCK_TSTAMP_NEW);
931 case SO_TIMESTAMPING_OLD:
932 if (val & ~SOF_TIMESTAMPING_MASK) {
937 if (val & SOF_TIMESTAMPING_OPT_ID &&
938 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
939 if (sk->sk_protocol == IPPROTO_TCP &&
940 sk->sk_type == SOCK_STREAM) {
941 if ((1 << sk->sk_state) &
942 (TCPF_CLOSE | TCPF_LISTEN)) {
946 sk->sk_tskey = tcp_sk(sk)->snd_una;
952 if (val & SOF_TIMESTAMPING_OPT_STATS &&
953 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
958 sk->sk_tsflags = val;
959 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
960 sock_enable_timestamp(sk,
961 SOCK_TIMESTAMPING_RX_SOFTWARE);
963 if (optname == SO_TIMESTAMPING_NEW)
964 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
966 sock_disable_timestamp(sk,
967 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
974 if (sock->ops->set_rcvlowat)
975 ret = sock->ops->set_rcvlowat(sk, val);
977 sk->sk_rcvlowat = val ? : 1;
980 case SO_RCVTIMEO_OLD:
981 case SO_RCVTIMEO_NEW:
982 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD);
985 case SO_SNDTIMEO_OLD:
986 case SO_SNDTIMEO_NEW:
987 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD);
990 case SO_ATTACH_FILTER:
992 if (optlen == sizeof(struct sock_fprog)) {
993 struct sock_fprog fprog;
996 if (copy_from_user(&fprog, optval, sizeof(fprog)))
999 ret = sk_attach_filter(&fprog, sk);
1005 if (optlen == sizeof(u32)) {
1009 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1012 ret = sk_attach_bpf(ufd, sk);
1016 case SO_ATTACH_REUSEPORT_CBPF:
1018 if (optlen == sizeof(struct sock_fprog)) {
1019 struct sock_fprog fprog;
1022 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1025 ret = sk_reuseport_attach_filter(&fprog, sk);
1029 case SO_ATTACH_REUSEPORT_EBPF:
1031 if (optlen == sizeof(u32)) {
1035 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1038 ret = sk_reuseport_attach_bpf(ufd, sk);
1042 case SO_DETACH_FILTER:
1043 ret = sk_detach_filter(sk);
1046 case SO_LOCK_FILTER:
1047 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1050 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1055 set_bit(SOCK_PASSSEC, &sock->flags);
1057 clear_bit(SOCK_PASSSEC, &sock->flags);
1060 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1062 } else if (val != sk->sk_mark) {
1069 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1072 case SO_WIFI_STATUS:
1073 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1077 if (sock->ops->set_peek_off)
1078 ret = sock->ops->set_peek_off(sk, val);
1084 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1087 case SO_SELECT_ERR_QUEUE:
1088 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1091 #ifdef CONFIG_NET_RX_BUSY_POLL
1093 /* allow unprivileged users to decrease the value */
1094 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1100 sk->sk_ll_usec = val;
1105 case SO_MAX_PACING_RATE:
1107 unsigned long ulval = (val == ~0U) ? ~0UL : val;
1109 if (sizeof(ulval) != sizeof(val) &&
1110 optlen >= sizeof(ulval) &&
1111 get_user(ulval, (unsigned long __user *)optval)) {
1116 cmpxchg(&sk->sk_pacing_status,
1119 sk->sk_max_pacing_rate = ulval;
1120 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1123 case SO_INCOMING_CPU:
1124 sk->sk_incoming_cpu = val;
1129 dst_negative_advice(sk);
1133 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1134 if (!((sk->sk_type == SOCK_STREAM &&
1135 sk->sk_protocol == IPPROTO_TCP) ||
1136 (sk->sk_type == SOCK_DGRAM &&
1137 sk->sk_protocol == IPPROTO_UDP)))
1139 } else if (sk->sk_family != PF_RDS) {
1143 if (val < 0 || val > 1)
1146 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1151 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1153 } else if (optlen != sizeof(struct sock_txtime)) {
1155 } else if (copy_from_user(&sk_txtime, optval,
1156 sizeof(struct sock_txtime))) {
1158 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1161 sock_valbool_flag(sk, SOCK_TXTIME, true);
1162 sk->sk_clockid = sk_txtime.clockid;
1163 sk->sk_txtime_deadline_mode =
1164 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1165 sk->sk_txtime_report_errors =
1166 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1170 case SO_BINDTOIFINDEX:
1171 ret = sock_setbindtodevice_locked(sk, val);
1181 EXPORT_SYMBOL(sock_setsockopt);
1184 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1185 struct ucred *ucred)
1187 ucred->pid = pid_vnr(pid);
1188 ucred->uid = ucred->gid = -1;
1190 struct user_namespace *current_ns = current_user_ns();
1192 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1193 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1197 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1199 struct user_namespace *user_ns = current_user_ns();
1202 for (i = 0; i < src->ngroups; i++)
1203 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1209 int sock_getsockopt(struct socket *sock, int level, int optname,
1210 char __user *optval, int __user *optlen)
1212 struct sock *sk = sock->sk;
1217 unsigned long ulval;
1219 struct old_timeval32 tm32;
1220 struct __kernel_old_timeval tm;
1221 struct __kernel_sock_timeval stm;
1222 struct sock_txtime txtime;
1225 int lv = sizeof(int);
1228 if (get_user(len, optlen))
1233 memset(&v, 0, sizeof(v));
1237 v.val = sock_flag(sk, SOCK_DBG);
1241 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1245 v.val = sock_flag(sk, SOCK_BROADCAST);
1249 v.val = sk->sk_sndbuf;
1253 v.val = sk->sk_rcvbuf;
1257 v.val = sk->sk_reuse;
1261 v.val = sk->sk_reuseport;
1265 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1269 v.val = sk->sk_type;
1273 v.val = sk->sk_protocol;
1277 v.val = sk->sk_family;
1281 v.val = -sock_error(sk);
1283 v.val = xchg(&sk->sk_err_soft, 0);
1287 v.val = sock_flag(sk, SOCK_URGINLINE);
1291 v.val = sk->sk_no_check_tx;
1295 v.val = sk->sk_priority;
1299 lv = sizeof(v.ling);
1300 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1301 v.ling.l_linger = sk->sk_lingertime / HZ;
1305 sock_warn_obsolete_bsdism("getsockopt");
1308 case SO_TIMESTAMP_OLD:
1309 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1310 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1311 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1314 case SO_TIMESTAMPNS_OLD:
1315 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1318 case SO_TIMESTAMP_NEW:
1319 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1322 case SO_TIMESTAMPNS_NEW:
1323 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1326 case SO_TIMESTAMPING_OLD:
1327 v.val = sk->sk_tsflags;
1330 case SO_RCVTIMEO_OLD:
1331 case SO_RCVTIMEO_NEW:
1332 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1335 case SO_SNDTIMEO_OLD:
1336 case SO_SNDTIMEO_NEW:
1337 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1341 v.val = sk->sk_rcvlowat;
1349 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1354 struct ucred peercred;
1355 if (len > sizeof(peercred))
1356 len = sizeof(peercred);
1357 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1358 if (copy_to_user(optval, &peercred, len))
1367 if (!sk->sk_peer_cred)
1370 n = sk->sk_peer_cred->group_info->ngroups;
1371 if (len < n * sizeof(gid_t)) {
1372 len = n * sizeof(gid_t);
1373 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1375 len = n * sizeof(gid_t);
1377 ret = groups_to_user((gid_t __user *)optval,
1378 sk->sk_peer_cred->group_info);
1388 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1393 if (copy_to_user(optval, address, len))
1398 /* Dubious BSD thing... Probably nobody even uses it, but
1399 * the UNIX standard wants it for whatever reason... -DaveM
1402 v.val = sk->sk_state == TCP_LISTEN;
1406 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1410 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1413 v.val = sk->sk_mark;
1417 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1420 case SO_WIFI_STATUS:
1421 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1425 if (!sock->ops->set_peek_off)
1428 v.val = sk->sk_peek_off;
1431 v.val = sock_flag(sk, SOCK_NOFCS);
1434 case SO_BINDTODEVICE:
1435 return sock_getbindtodevice(sk, optval, optlen, len);
1438 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1444 case SO_LOCK_FILTER:
1445 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1448 case SO_BPF_EXTENSIONS:
1449 v.val = bpf_tell_extensions();
1452 case SO_SELECT_ERR_QUEUE:
1453 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1456 #ifdef CONFIG_NET_RX_BUSY_POLL
1458 v.val = sk->sk_ll_usec;
1462 case SO_MAX_PACING_RATE:
1463 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1464 lv = sizeof(v.ulval);
1465 v.ulval = sk->sk_max_pacing_rate;
1468 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1472 case SO_INCOMING_CPU:
1473 v.val = sk->sk_incoming_cpu;
1478 u32 meminfo[SK_MEMINFO_VARS];
1480 sk_get_meminfo(sk, meminfo);
1482 len = min_t(unsigned int, len, sizeof(meminfo));
1483 if (copy_to_user(optval, &meminfo, len))
1489 #ifdef CONFIG_NET_RX_BUSY_POLL
1490 case SO_INCOMING_NAPI_ID:
1491 v.val = READ_ONCE(sk->sk_napi_id);
1493 /* aggregate non-NAPI IDs down to 0 */
1494 if (v.val < MIN_NAPI_ID)
1504 v.val64 = sock_gen_cookie(sk);
1508 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1512 lv = sizeof(v.txtime);
1513 v.txtime.clockid = sk->sk_clockid;
1514 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1515 SOF_TXTIME_DEADLINE_MODE : 0;
1516 v.txtime.flags |= sk->sk_txtime_report_errors ?
1517 SOF_TXTIME_REPORT_ERRORS : 0;
1520 case SO_BINDTOIFINDEX:
1521 v.val = sk->sk_bound_dev_if;
1525 /* We implement the SO_SNDLOWAT etc to not be settable
1528 return -ENOPROTOOPT;
1533 if (copy_to_user(optval, &v, len))
1536 if (put_user(len, optlen))
1542 * Initialize an sk_lock.
1544 * (We also register the sk_lock with the lock validator.)
1546 static inline void sock_lock_init(struct sock *sk)
1548 if (sk->sk_kern_sock)
1549 sock_lock_init_class_and_name(
1551 af_family_kern_slock_key_strings[sk->sk_family],
1552 af_family_kern_slock_keys + sk->sk_family,
1553 af_family_kern_key_strings[sk->sk_family],
1554 af_family_kern_keys + sk->sk_family);
1556 sock_lock_init_class_and_name(
1558 af_family_slock_key_strings[sk->sk_family],
1559 af_family_slock_keys + sk->sk_family,
1560 af_family_key_strings[sk->sk_family],
1561 af_family_keys + sk->sk_family);
1565 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1566 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1567 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1569 static void sock_copy(struct sock *nsk, const struct sock *osk)
1571 #ifdef CONFIG_SECURITY_NETWORK
1572 void *sptr = nsk->sk_security;
1574 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1576 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1577 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1579 #ifdef CONFIG_SECURITY_NETWORK
1580 nsk->sk_security = sptr;
1581 security_sk_clone(osk, nsk);
1585 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1589 struct kmem_cache *slab;
1593 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1596 if (priority & __GFP_ZERO)
1597 sk_prot_clear_nulls(sk, prot->obj_size);
1599 sk = kmalloc(prot->obj_size, priority);
1602 if (security_sk_alloc(sk, family, priority))
1605 if (!try_module_get(prot->owner))
1607 sk_tx_queue_clear(sk);
1613 security_sk_free(sk);
1616 kmem_cache_free(slab, sk);
1622 static void sk_prot_free(struct proto *prot, struct sock *sk)
1624 struct kmem_cache *slab;
1625 struct module *owner;
1627 owner = prot->owner;
1630 cgroup_sk_free(&sk->sk_cgrp_data);
1631 mem_cgroup_sk_free(sk);
1632 security_sk_free(sk);
1634 kmem_cache_free(slab, sk);
1641 * sk_alloc - All socket objects are allocated here
1642 * @net: the applicable net namespace
1643 * @family: protocol family
1644 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1645 * @prot: struct proto associated with this new sock instance
1646 * @kern: is this to be a kernel socket?
1648 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1649 struct proto *prot, int kern)
1653 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1655 sk->sk_family = family;
1657 * See comment in struct sock definition to understand
1658 * why we need sk_prot_creator -acme
1660 sk->sk_prot = sk->sk_prot_creator = prot;
1661 sk->sk_kern_sock = kern;
1663 sk->sk_net_refcnt = kern ? 0 : 1;
1664 if (likely(sk->sk_net_refcnt)) {
1666 sock_inuse_add(net, 1);
1669 sock_net_set(sk, net);
1670 refcount_set(&sk->sk_wmem_alloc, 1);
1672 mem_cgroup_sk_alloc(sk);
1673 cgroup_sk_alloc(&sk->sk_cgrp_data);
1674 sock_update_classid(&sk->sk_cgrp_data);
1675 sock_update_netprioidx(&sk->sk_cgrp_data);
1680 EXPORT_SYMBOL(sk_alloc);
1682 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1683 * grace period. This is the case for UDP sockets and TCP listeners.
1685 static void __sk_destruct(struct rcu_head *head)
1687 struct sock *sk = container_of(head, struct sock, sk_rcu);
1688 struct sk_filter *filter;
1690 if (sk->sk_destruct)
1691 sk->sk_destruct(sk);
1693 filter = rcu_dereference_check(sk->sk_filter,
1694 refcount_read(&sk->sk_wmem_alloc) == 0);
1696 sk_filter_uncharge(sk, filter);
1697 RCU_INIT_POINTER(sk->sk_filter, NULL);
1699 if (rcu_access_pointer(sk->sk_reuseport_cb))
1700 reuseport_detach_sock(sk);
1702 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1704 #ifdef CONFIG_BPF_SYSCALL
1705 bpf_sk_storage_free(sk);
1708 if (atomic_read(&sk->sk_omem_alloc))
1709 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1710 __func__, atomic_read(&sk->sk_omem_alloc));
1712 if (sk->sk_frag.page) {
1713 put_page(sk->sk_frag.page);
1714 sk->sk_frag.page = NULL;
1717 if (sk->sk_peer_cred)
1718 put_cred(sk->sk_peer_cred);
1719 put_pid(sk->sk_peer_pid);
1720 if (likely(sk->sk_net_refcnt))
1721 put_net(sock_net(sk));
1722 sk_prot_free(sk->sk_prot_creator, sk);
1725 void sk_destruct(struct sock *sk)
1727 if (sock_flag(sk, SOCK_RCU_FREE))
1728 call_rcu(&sk->sk_rcu, __sk_destruct);
1730 __sk_destruct(&sk->sk_rcu);
1733 static void __sk_free(struct sock *sk)
1735 if (likely(sk->sk_net_refcnt))
1736 sock_inuse_add(sock_net(sk), -1);
1738 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1739 sock_diag_broadcast_destroy(sk);
1744 void sk_free(struct sock *sk)
1747 * We subtract one from sk_wmem_alloc and can know if
1748 * some packets are still in some tx queue.
1749 * If not null, sock_wfree() will call __sk_free(sk) later
1751 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1754 EXPORT_SYMBOL(sk_free);
1756 static void sk_init_common(struct sock *sk)
1758 skb_queue_head_init(&sk->sk_receive_queue);
1759 skb_queue_head_init(&sk->sk_write_queue);
1760 skb_queue_head_init(&sk->sk_error_queue);
1762 rwlock_init(&sk->sk_callback_lock);
1763 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1764 af_rlock_keys + sk->sk_family,
1765 af_family_rlock_key_strings[sk->sk_family]);
1766 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1767 af_wlock_keys + sk->sk_family,
1768 af_family_wlock_key_strings[sk->sk_family]);
1769 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1770 af_elock_keys + sk->sk_family,
1771 af_family_elock_key_strings[sk->sk_family]);
1772 lockdep_set_class_and_name(&sk->sk_callback_lock,
1773 af_callback_keys + sk->sk_family,
1774 af_family_clock_key_strings[sk->sk_family]);
1778 * sk_clone_lock - clone a socket, and lock its clone
1779 * @sk: the socket to clone
1780 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1782 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1784 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1787 bool is_charged = true;
1789 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1790 if (newsk != NULL) {
1791 struct sk_filter *filter;
1793 sock_copy(newsk, sk);
1795 newsk->sk_prot_creator = sk->sk_prot;
1798 if (likely(newsk->sk_net_refcnt))
1799 get_net(sock_net(newsk));
1800 sk_node_init(&newsk->sk_node);
1801 sock_lock_init(newsk);
1802 bh_lock_sock(newsk);
1803 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1804 newsk->sk_backlog.len = 0;
1806 atomic_set(&newsk->sk_rmem_alloc, 0);
1808 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1810 refcount_set(&newsk->sk_wmem_alloc, 1);
1811 atomic_set(&newsk->sk_omem_alloc, 0);
1812 sk_init_common(newsk);
1814 newsk->sk_dst_cache = NULL;
1815 newsk->sk_dst_pending_confirm = 0;
1816 newsk->sk_wmem_queued = 0;
1817 newsk->sk_forward_alloc = 0;
1818 atomic_set(&newsk->sk_drops, 0);
1819 newsk->sk_send_head = NULL;
1820 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1821 atomic_set(&newsk->sk_zckey, 0);
1823 sock_reset_flag(newsk, SOCK_DONE);
1824 mem_cgroup_sk_alloc(newsk);
1825 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1828 filter = rcu_dereference(sk->sk_filter);
1830 /* though it's an empty new sock, the charging may fail
1831 * if sysctl_optmem_max was changed between creation of
1832 * original socket and cloning
1834 is_charged = sk_filter_charge(newsk, filter);
1835 RCU_INIT_POINTER(newsk->sk_filter, filter);
1838 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1839 /* We need to make sure that we don't uncharge the new
1840 * socket if we couldn't charge it in the first place
1841 * as otherwise we uncharge the parent's filter.
1844 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1845 sk_free_unlock_clone(newsk);
1849 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1850 #ifdef CONFIG_BPF_SYSCALL
1851 RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL);
1855 newsk->sk_err_soft = 0;
1856 newsk->sk_priority = 0;
1857 newsk->sk_incoming_cpu = raw_smp_processor_id();
1858 if (likely(newsk->sk_net_refcnt))
1859 sock_inuse_add(sock_net(newsk), 1);
1862 * Before updating sk_refcnt, we must commit prior changes to memory
1863 * (Documentation/RCU/rculist_nulls.txt for details)
1866 refcount_set(&newsk->sk_refcnt, 2);
1869 * Increment the counter in the same struct proto as the master
1870 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1871 * is the same as sk->sk_prot->socks, as this field was copied
1874 * This _changes_ the previous behaviour, where
1875 * tcp_create_openreq_child always was incrementing the
1876 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1877 * to be taken into account in all callers. -acme
1879 sk_refcnt_debug_inc(newsk);
1880 sk_set_socket(newsk, NULL);
1881 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1883 if (newsk->sk_prot->sockets_allocated)
1884 sk_sockets_allocated_inc(newsk);
1886 if (sock_needs_netstamp(sk) &&
1887 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1888 net_enable_timestamp();
1893 EXPORT_SYMBOL_GPL(sk_clone_lock);
1895 void sk_free_unlock_clone(struct sock *sk)
1897 /* It is still raw copy of parent, so invalidate
1898 * destructor and make plain sk_free() */
1899 sk->sk_destruct = NULL;
1903 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1905 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1909 sk_dst_set(sk, dst);
1910 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1911 if (sk->sk_route_caps & NETIF_F_GSO)
1912 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1913 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1914 if (sk_can_gso(sk)) {
1915 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1916 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1918 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1919 sk->sk_gso_max_size = dst->dev->gso_max_size;
1920 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1923 sk->sk_gso_max_segs = max_segs;
1925 EXPORT_SYMBOL_GPL(sk_setup_caps);
1928 * Simple resource managers for sockets.
1933 * Write buffer destructor automatically called from kfree_skb.
1935 void sock_wfree(struct sk_buff *skb)
1937 struct sock *sk = skb->sk;
1938 unsigned int len = skb->truesize;
1940 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1942 * Keep a reference on sk_wmem_alloc, this will be released
1943 * after sk_write_space() call
1945 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1946 sk->sk_write_space(sk);
1950 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1951 * could not do because of in-flight packets
1953 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1956 EXPORT_SYMBOL(sock_wfree);
1958 /* This variant of sock_wfree() is used by TCP,
1959 * since it sets SOCK_USE_WRITE_QUEUE.
1961 void __sock_wfree(struct sk_buff *skb)
1963 struct sock *sk = skb->sk;
1965 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1969 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1974 if (unlikely(!sk_fullsock(sk))) {
1975 skb->destructor = sock_edemux;
1980 skb->destructor = sock_wfree;
1981 skb_set_hash_from_sk(skb, sk);
1983 * We used to take a refcount on sk, but following operation
1984 * is enough to guarantee sk_free() wont free this sock until
1985 * all in-flight packets are completed
1987 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1989 EXPORT_SYMBOL(skb_set_owner_w);
1991 /* This helper is used by netem, as it can hold packets in its
1992 * delay queue. We want to allow the owner socket to send more
1993 * packets, as if they were already TX completed by a typical driver.
1994 * But we also want to keep skb->sk set because some packet schedulers
1995 * rely on it (sch_fq for example).
1997 void skb_orphan_partial(struct sk_buff *skb)
1999 if (skb_is_tcp_pure_ack(skb))
2002 if (skb->destructor == sock_wfree
2004 || skb->destructor == tcp_wfree
2007 struct sock *sk = skb->sk;
2009 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2010 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2011 skb->destructor = sock_efree;
2017 EXPORT_SYMBOL(skb_orphan_partial);
2020 * Read buffer destructor automatically called from kfree_skb.
2022 void sock_rfree(struct sk_buff *skb)
2024 struct sock *sk = skb->sk;
2025 unsigned int len = skb->truesize;
2027 atomic_sub(len, &sk->sk_rmem_alloc);
2028 sk_mem_uncharge(sk, len);
2030 EXPORT_SYMBOL(sock_rfree);
2033 * Buffer destructor for skbs that are not used directly in read or write
2034 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2036 void sock_efree(struct sk_buff *skb)
2040 EXPORT_SYMBOL(sock_efree);
2042 kuid_t sock_i_uid(struct sock *sk)
2046 read_lock_bh(&sk->sk_callback_lock);
2047 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2048 read_unlock_bh(&sk->sk_callback_lock);
2051 EXPORT_SYMBOL(sock_i_uid);
2053 unsigned long sock_i_ino(struct sock *sk)
2057 read_lock_bh(&sk->sk_callback_lock);
2058 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2059 read_unlock_bh(&sk->sk_callback_lock);
2062 EXPORT_SYMBOL(sock_i_ino);
2065 * Allocate a skb from the socket's send buffer.
2067 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2070 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2071 struct sk_buff *skb = alloc_skb(size, priority);
2073 skb_set_owner_w(skb, sk);
2079 EXPORT_SYMBOL(sock_wmalloc);
2081 static void sock_ofree(struct sk_buff *skb)
2083 struct sock *sk = skb->sk;
2085 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2088 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2091 struct sk_buff *skb;
2093 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2094 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2098 skb = alloc_skb(size, priority);
2102 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2104 skb->destructor = sock_ofree;
2109 * Allocate a memory block from the socket's option memory buffer.
2111 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2113 if ((unsigned int)size <= sysctl_optmem_max &&
2114 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2116 /* First do the add, to avoid the race if kmalloc
2119 atomic_add(size, &sk->sk_omem_alloc);
2120 mem = kmalloc(size, priority);
2123 atomic_sub(size, &sk->sk_omem_alloc);
2127 EXPORT_SYMBOL(sock_kmalloc);
2129 /* Free an option memory block. Note, we actually want the inline
2130 * here as this allows gcc to detect the nullify and fold away the
2131 * condition entirely.
2133 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2136 if (WARN_ON_ONCE(!mem))
2142 atomic_sub(size, &sk->sk_omem_alloc);
2145 void sock_kfree_s(struct sock *sk, void *mem, int size)
2147 __sock_kfree_s(sk, mem, size, false);
2149 EXPORT_SYMBOL(sock_kfree_s);
2151 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2153 __sock_kfree_s(sk, mem, size, true);
2155 EXPORT_SYMBOL(sock_kzfree_s);
2157 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2158 I think, these locks should be removed for datagram sockets.
2160 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2164 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2168 if (signal_pending(current))
2170 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2171 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2172 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2174 if (sk->sk_shutdown & SEND_SHUTDOWN)
2178 timeo = schedule_timeout(timeo);
2180 finish_wait(sk_sleep(sk), &wait);
2186 * Generic send/receive buffer handlers
2189 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2190 unsigned long data_len, int noblock,
2191 int *errcode, int max_page_order)
2193 struct sk_buff *skb;
2197 timeo = sock_sndtimeo(sk, noblock);
2199 err = sock_error(sk);
2204 if (sk->sk_shutdown & SEND_SHUTDOWN)
2207 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2210 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2211 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2215 if (signal_pending(current))
2217 timeo = sock_wait_for_wmem(sk, timeo);
2219 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2220 errcode, sk->sk_allocation);
2222 skb_set_owner_w(skb, sk);
2226 err = sock_intr_errno(timeo);
2231 EXPORT_SYMBOL(sock_alloc_send_pskb);
2233 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2234 int noblock, int *errcode)
2236 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2238 EXPORT_SYMBOL(sock_alloc_send_skb);
2240 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2241 struct sockcm_cookie *sockc)
2245 switch (cmsg->cmsg_type) {
2247 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2249 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2251 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2253 case SO_TIMESTAMPING_OLD:
2254 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2257 tsflags = *(u32 *)CMSG_DATA(cmsg);
2258 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2261 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2262 sockc->tsflags |= tsflags;
2265 if (!sock_flag(sk, SOCK_TXTIME))
2267 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2269 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2271 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2273 case SCM_CREDENTIALS:
2280 EXPORT_SYMBOL(__sock_cmsg_send);
2282 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2283 struct sockcm_cookie *sockc)
2285 struct cmsghdr *cmsg;
2288 for_each_cmsghdr(cmsg, msg) {
2289 if (!CMSG_OK(msg, cmsg))
2291 if (cmsg->cmsg_level != SOL_SOCKET)
2293 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2299 EXPORT_SYMBOL(sock_cmsg_send);
2301 static void sk_enter_memory_pressure(struct sock *sk)
2303 if (!sk->sk_prot->enter_memory_pressure)
2306 sk->sk_prot->enter_memory_pressure(sk);
2309 static void sk_leave_memory_pressure(struct sock *sk)
2311 if (sk->sk_prot->leave_memory_pressure) {
2312 sk->sk_prot->leave_memory_pressure(sk);
2314 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2316 if (memory_pressure && *memory_pressure)
2317 *memory_pressure = 0;
2321 /* On 32bit arches, an skb frag is limited to 2^15 */
2322 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2323 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2326 * skb_page_frag_refill - check that a page_frag contains enough room
2327 * @sz: minimum size of the fragment we want to get
2328 * @pfrag: pointer to page_frag
2329 * @gfp: priority for memory allocation
2331 * Note: While this allocator tries to use high order pages, there is
2332 * no guarantee that allocations succeed. Therefore, @sz MUST be
2333 * less or equal than PAGE_SIZE.
2335 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2338 if (page_ref_count(pfrag->page) == 1) {
2342 if (pfrag->offset + sz <= pfrag->size)
2344 put_page(pfrag->page);
2348 if (SKB_FRAG_PAGE_ORDER &&
2349 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2350 /* Avoid direct reclaim but allow kswapd to wake */
2351 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2352 __GFP_COMP | __GFP_NOWARN |
2354 SKB_FRAG_PAGE_ORDER);
2355 if (likely(pfrag->page)) {
2356 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2360 pfrag->page = alloc_page(gfp);
2361 if (likely(pfrag->page)) {
2362 pfrag->size = PAGE_SIZE;
2367 EXPORT_SYMBOL(skb_page_frag_refill);
2369 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2371 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2374 sk_enter_memory_pressure(sk);
2375 sk_stream_moderate_sndbuf(sk);
2378 EXPORT_SYMBOL(sk_page_frag_refill);
2380 static void __lock_sock(struct sock *sk)
2381 __releases(&sk->sk_lock.slock)
2382 __acquires(&sk->sk_lock.slock)
2387 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2388 TASK_UNINTERRUPTIBLE);
2389 spin_unlock_bh(&sk->sk_lock.slock);
2391 spin_lock_bh(&sk->sk_lock.slock);
2392 if (!sock_owned_by_user(sk))
2395 finish_wait(&sk->sk_lock.wq, &wait);
2398 void __release_sock(struct sock *sk)
2399 __releases(&sk->sk_lock.slock)
2400 __acquires(&sk->sk_lock.slock)
2402 struct sk_buff *skb, *next;
2404 while ((skb = sk->sk_backlog.head) != NULL) {
2405 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2407 spin_unlock_bh(&sk->sk_lock.slock);
2412 WARN_ON_ONCE(skb_dst_is_noref(skb));
2413 skb_mark_not_on_list(skb);
2414 sk_backlog_rcv(sk, skb);
2419 } while (skb != NULL);
2421 spin_lock_bh(&sk->sk_lock.slock);
2425 * Doing the zeroing here guarantee we can not loop forever
2426 * while a wild producer attempts to flood us.
2428 sk->sk_backlog.len = 0;
2431 void __sk_flush_backlog(struct sock *sk)
2433 spin_lock_bh(&sk->sk_lock.slock);
2435 spin_unlock_bh(&sk->sk_lock.slock);
2439 * sk_wait_data - wait for data to arrive at sk_receive_queue
2440 * @sk: sock to wait on
2441 * @timeo: for how long
2442 * @skb: last skb seen on sk_receive_queue
2444 * Now socket state including sk->sk_err is changed only under lock,
2445 * hence we may omit checks after joining wait queue.
2446 * We check receive queue before schedule() only as optimization;
2447 * it is very likely that release_sock() added new data.
2449 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2451 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2454 add_wait_queue(sk_sleep(sk), &wait);
2455 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2456 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2457 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2458 remove_wait_queue(sk_sleep(sk), &wait);
2461 EXPORT_SYMBOL(sk_wait_data);
2464 * __sk_mem_raise_allocated - increase memory_allocated
2466 * @size: memory size to allocate
2467 * @amt: pages to allocate
2468 * @kind: allocation type
2470 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2472 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2474 struct proto *prot = sk->sk_prot;
2475 long allocated = sk_memory_allocated_add(sk, amt);
2476 bool charged = true;
2478 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2479 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2480 goto suppress_allocation;
2483 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2484 sk_leave_memory_pressure(sk);
2488 /* Under pressure. */
2489 if (allocated > sk_prot_mem_limits(sk, 1))
2490 sk_enter_memory_pressure(sk);
2492 /* Over hard limit. */
2493 if (allocated > sk_prot_mem_limits(sk, 2))
2494 goto suppress_allocation;
2496 /* guarantee minimum buffer size under pressure */
2497 if (kind == SK_MEM_RECV) {
2498 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2501 } else { /* SK_MEM_SEND */
2502 int wmem0 = sk_get_wmem0(sk, prot);
2504 if (sk->sk_type == SOCK_STREAM) {
2505 if (sk->sk_wmem_queued < wmem0)
2507 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2512 if (sk_has_memory_pressure(sk)) {
2515 if (!sk_under_memory_pressure(sk))
2517 alloc = sk_sockets_allocated_read_positive(sk);
2518 if (sk_prot_mem_limits(sk, 2) > alloc *
2519 sk_mem_pages(sk->sk_wmem_queued +
2520 atomic_read(&sk->sk_rmem_alloc) +
2521 sk->sk_forward_alloc))
2525 suppress_allocation:
2527 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2528 sk_stream_moderate_sndbuf(sk);
2530 /* Fail only if socket is _under_ its sndbuf.
2531 * In this case we cannot block, so that we have to fail.
2533 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2537 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2538 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2540 sk_memory_allocated_sub(sk, amt);
2542 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2543 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2547 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2550 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2552 * @size: memory size to allocate
2553 * @kind: allocation type
2555 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2556 * rmem allocation. This function assumes that protocols which have
2557 * memory_pressure use sk_wmem_queued as write buffer accounting.
2559 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2561 int ret, amt = sk_mem_pages(size);
2563 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2564 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2566 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2569 EXPORT_SYMBOL(__sk_mem_schedule);
2572 * __sk_mem_reduce_allocated - reclaim memory_allocated
2574 * @amount: number of quanta
2576 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2578 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2580 sk_memory_allocated_sub(sk, amount);
2582 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2583 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2585 if (sk_under_memory_pressure(sk) &&
2586 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2587 sk_leave_memory_pressure(sk);
2589 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2592 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2594 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2596 void __sk_mem_reclaim(struct sock *sk, int amount)
2598 amount >>= SK_MEM_QUANTUM_SHIFT;
2599 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2600 __sk_mem_reduce_allocated(sk, amount);
2602 EXPORT_SYMBOL(__sk_mem_reclaim);
2604 int sk_set_peek_off(struct sock *sk, int val)
2606 sk->sk_peek_off = val;
2609 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2612 * Set of default routines for initialising struct proto_ops when
2613 * the protocol does not support a particular function. In certain
2614 * cases where it makes no sense for a protocol to have a "do nothing"
2615 * function, some default processing is provided.
2618 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2622 EXPORT_SYMBOL(sock_no_bind);
2624 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2629 EXPORT_SYMBOL(sock_no_connect);
2631 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2635 EXPORT_SYMBOL(sock_no_socketpair);
2637 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2642 EXPORT_SYMBOL(sock_no_accept);
2644 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2649 EXPORT_SYMBOL(sock_no_getname);
2651 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2655 EXPORT_SYMBOL(sock_no_ioctl);
2657 int sock_no_listen(struct socket *sock, int backlog)
2661 EXPORT_SYMBOL(sock_no_listen);
2663 int sock_no_shutdown(struct socket *sock, int how)
2667 EXPORT_SYMBOL(sock_no_shutdown);
2669 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2670 char __user *optval, unsigned int optlen)
2674 EXPORT_SYMBOL(sock_no_setsockopt);
2676 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2677 char __user *optval, int __user *optlen)
2681 EXPORT_SYMBOL(sock_no_getsockopt);
2683 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2687 EXPORT_SYMBOL(sock_no_sendmsg);
2689 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2693 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2695 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2700 EXPORT_SYMBOL(sock_no_recvmsg);
2702 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2704 /* Mirror missing mmap method error code */
2707 EXPORT_SYMBOL(sock_no_mmap);
2709 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2712 struct msghdr msg = {.msg_flags = flags};
2714 char *kaddr = kmap(page);
2715 iov.iov_base = kaddr + offset;
2717 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2721 EXPORT_SYMBOL(sock_no_sendpage);
2723 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2724 int offset, size_t size, int flags)
2727 struct msghdr msg = {.msg_flags = flags};
2729 char *kaddr = kmap(page);
2731 iov.iov_base = kaddr + offset;
2733 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2737 EXPORT_SYMBOL(sock_no_sendpage_locked);
2740 * Default Socket Callbacks
2743 static void sock_def_wakeup(struct sock *sk)
2745 struct socket_wq *wq;
2748 wq = rcu_dereference(sk->sk_wq);
2749 if (skwq_has_sleeper(wq))
2750 wake_up_interruptible_all(&wq->wait);
2754 static void sock_def_error_report(struct sock *sk)
2756 struct socket_wq *wq;
2759 wq = rcu_dereference(sk->sk_wq);
2760 if (skwq_has_sleeper(wq))
2761 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2762 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2766 static void sock_def_readable(struct sock *sk)
2768 struct socket_wq *wq;
2771 wq = rcu_dereference(sk->sk_wq);
2772 if (skwq_has_sleeper(wq))
2773 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2774 EPOLLRDNORM | EPOLLRDBAND);
2775 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2779 static void sock_def_write_space(struct sock *sk)
2781 struct socket_wq *wq;
2785 /* Do not wake up a writer until he can make "significant"
2788 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2789 wq = rcu_dereference(sk->sk_wq);
2790 if (skwq_has_sleeper(wq))
2791 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2792 EPOLLWRNORM | EPOLLWRBAND);
2794 /* Should agree with poll, otherwise some programs break */
2795 if (sock_writeable(sk))
2796 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2802 static void sock_def_destruct(struct sock *sk)
2806 void sk_send_sigurg(struct sock *sk)
2808 if (sk->sk_socket && sk->sk_socket->file)
2809 if (send_sigurg(&sk->sk_socket->file->f_owner))
2810 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2812 EXPORT_SYMBOL(sk_send_sigurg);
2814 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2815 unsigned long expires)
2817 if (!mod_timer(timer, expires))
2820 EXPORT_SYMBOL(sk_reset_timer);
2822 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2824 if (del_timer(timer))
2827 EXPORT_SYMBOL(sk_stop_timer);
2829 void sock_init_data(struct socket *sock, struct sock *sk)
2832 sk->sk_send_head = NULL;
2834 timer_setup(&sk->sk_timer, NULL, 0);
2836 sk->sk_allocation = GFP_KERNEL;
2837 sk->sk_rcvbuf = sysctl_rmem_default;
2838 sk->sk_sndbuf = sysctl_wmem_default;
2839 sk->sk_state = TCP_CLOSE;
2840 sk_set_socket(sk, sock);
2842 sock_set_flag(sk, SOCK_ZAPPED);
2845 sk->sk_type = sock->type;
2846 RCU_INIT_POINTER(sk->sk_wq, sock->wq);
2848 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2850 RCU_INIT_POINTER(sk->sk_wq, NULL);
2851 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2854 rwlock_init(&sk->sk_callback_lock);
2855 if (sk->sk_kern_sock)
2856 lockdep_set_class_and_name(
2857 &sk->sk_callback_lock,
2858 af_kern_callback_keys + sk->sk_family,
2859 af_family_kern_clock_key_strings[sk->sk_family]);
2861 lockdep_set_class_and_name(
2862 &sk->sk_callback_lock,
2863 af_callback_keys + sk->sk_family,
2864 af_family_clock_key_strings[sk->sk_family]);
2866 sk->sk_state_change = sock_def_wakeup;
2867 sk->sk_data_ready = sock_def_readable;
2868 sk->sk_write_space = sock_def_write_space;
2869 sk->sk_error_report = sock_def_error_report;
2870 sk->sk_destruct = sock_def_destruct;
2872 sk->sk_frag.page = NULL;
2873 sk->sk_frag.offset = 0;
2874 sk->sk_peek_off = -1;
2876 sk->sk_peer_pid = NULL;
2877 sk->sk_peer_cred = NULL;
2878 sk->sk_write_pending = 0;
2879 sk->sk_rcvlowat = 1;
2880 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2881 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2883 sk->sk_stamp = SK_DEFAULT_STAMP;
2884 #if BITS_PER_LONG==32
2885 seqlock_init(&sk->sk_stamp_seq);
2887 atomic_set(&sk->sk_zckey, 0);
2889 #ifdef CONFIG_NET_RX_BUSY_POLL
2891 sk->sk_ll_usec = sysctl_net_busy_read;
2894 sk->sk_max_pacing_rate = ~0UL;
2895 sk->sk_pacing_rate = ~0UL;
2896 sk->sk_pacing_shift = 10;
2897 sk->sk_incoming_cpu = -1;
2899 sk_rx_queue_clear(sk);
2901 * Before updating sk_refcnt, we must commit prior changes to memory
2902 * (Documentation/RCU/rculist_nulls.txt for details)
2905 refcount_set(&sk->sk_refcnt, 1);
2906 atomic_set(&sk->sk_drops, 0);
2908 EXPORT_SYMBOL(sock_init_data);
2910 void lock_sock_nested(struct sock *sk, int subclass)
2913 spin_lock_bh(&sk->sk_lock.slock);
2914 if (sk->sk_lock.owned)
2916 sk->sk_lock.owned = 1;
2917 spin_unlock(&sk->sk_lock.slock);
2919 * The sk_lock has mutex_lock() semantics here:
2921 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2924 EXPORT_SYMBOL(lock_sock_nested);
2926 void release_sock(struct sock *sk)
2928 spin_lock_bh(&sk->sk_lock.slock);
2929 if (sk->sk_backlog.tail)
2932 /* Warning : release_cb() might need to release sk ownership,
2933 * ie call sock_release_ownership(sk) before us.
2935 if (sk->sk_prot->release_cb)
2936 sk->sk_prot->release_cb(sk);
2938 sock_release_ownership(sk);
2939 if (waitqueue_active(&sk->sk_lock.wq))
2940 wake_up(&sk->sk_lock.wq);
2941 spin_unlock_bh(&sk->sk_lock.slock);
2943 EXPORT_SYMBOL(release_sock);
2946 * lock_sock_fast - fast version of lock_sock
2949 * This version should be used for very small section, where process wont block
2950 * return false if fast path is taken:
2952 * sk_lock.slock locked, owned = 0, BH disabled
2954 * return true if slow path is taken:
2956 * sk_lock.slock unlocked, owned = 1, BH enabled
2958 bool lock_sock_fast(struct sock *sk)
2961 spin_lock_bh(&sk->sk_lock.slock);
2963 if (!sk->sk_lock.owned)
2965 * Note : We must disable BH
2970 sk->sk_lock.owned = 1;
2971 spin_unlock(&sk->sk_lock.slock);
2973 * The sk_lock has mutex_lock() semantics here:
2975 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2979 EXPORT_SYMBOL(lock_sock_fast);
2981 int sock_gettstamp(struct socket *sock, void __user *userstamp,
2982 bool timeval, bool time32)
2984 struct sock *sk = sock->sk;
2985 struct timespec64 ts;
2987 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2988 ts = ktime_to_timespec64(sock_read_timestamp(sk));
2989 if (ts.tv_sec == -1)
2991 if (ts.tv_sec == 0) {
2992 ktime_t kt = ktime_get_real();
2993 sock_write_timestamp(sk, kt);;
2994 ts = ktime_to_timespec64(kt);
3000 #ifdef CONFIG_COMPAT_32BIT_TIME
3002 return put_old_timespec32(&ts, userstamp);
3004 #ifdef CONFIG_SPARC64
3005 /* beware of padding in sparc64 timeval */
3006 if (timeval && !in_compat_syscall()) {
3007 struct __kernel_old_timeval __user tv = {
3008 .tv_sec = ts.tv_sec,
3009 .tv_usec = ts.tv_nsec,
3011 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3016 return put_timespec64(&ts, userstamp);
3018 EXPORT_SYMBOL(sock_gettstamp);
3020 void sock_enable_timestamp(struct sock *sk, int flag)
3022 if (!sock_flag(sk, flag)) {
3023 unsigned long previous_flags = sk->sk_flags;
3025 sock_set_flag(sk, flag);
3027 * we just set one of the two flags which require net
3028 * time stamping, but time stamping might have been on
3029 * already because of the other one
3031 if (sock_needs_netstamp(sk) &&
3032 !(previous_flags & SK_FLAGS_TIMESTAMP))
3033 net_enable_timestamp();
3037 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3038 int level, int type)
3040 struct sock_exterr_skb *serr;
3041 struct sk_buff *skb;
3045 skb = sock_dequeue_err_skb(sk);
3051 msg->msg_flags |= MSG_TRUNC;
3054 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3058 sock_recv_timestamp(msg, sk, skb);
3060 serr = SKB_EXT_ERR(skb);
3061 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3063 msg->msg_flags |= MSG_ERRQUEUE;
3071 EXPORT_SYMBOL(sock_recv_errqueue);
3074 * Get a socket option on an socket.
3076 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3077 * asynchronous errors should be reported by getsockopt. We assume
3078 * this means if you specify SO_ERROR (otherwise whats the point of it).
3080 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3081 char __user *optval, int __user *optlen)
3083 struct sock *sk = sock->sk;
3085 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3087 EXPORT_SYMBOL(sock_common_getsockopt);
3089 #ifdef CONFIG_COMPAT
3090 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3091 char __user *optval, int __user *optlen)
3093 struct sock *sk = sock->sk;
3095 if (sk->sk_prot->compat_getsockopt != NULL)
3096 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3098 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3100 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3103 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3106 struct sock *sk = sock->sk;
3110 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3111 flags & ~MSG_DONTWAIT, &addr_len);
3113 msg->msg_namelen = addr_len;
3116 EXPORT_SYMBOL(sock_common_recvmsg);
3119 * Set socket options on an inet socket.
3121 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3122 char __user *optval, unsigned int optlen)
3124 struct sock *sk = sock->sk;
3126 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3128 EXPORT_SYMBOL(sock_common_setsockopt);
3130 #ifdef CONFIG_COMPAT
3131 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3132 char __user *optval, unsigned int optlen)
3134 struct sock *sk = sock->sk;
3136 if (sk->sk_prot->compat_setsockopt != NULL)
3137 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3139 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3141 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3144 void sk_common_release(struct sock *sk)
3146 if (sk->sk_prot->destroy)
3147 sk->sk_prot->destroy(sk);
3150 * Observation: when sock_common_release is called, processes have
3151 * no access to socket. But net still has.
3152 * Step one, detach it from networking:
3154 * A. Remove from hash tables.
3157 sk->sk_prot->unhash(sk);
3160 * In this point socket cannot receive new packets, but it is possible
3161 * that some packets are in flight because some CPU runs receiver and
3162 * did hash table lookup before we unhashed socket. They will achieve
3163 * receive queue and will be purged by socket destructor.
3165 * Also we still have packets pending on receive queue and probably,
3166 * our own packets waiting in device queues. sock_destroy will drain
3167 * receive queue, but transmitted packets will delay socket destruction
3168 * until the last reference will be released.
3173 xfrm_sk_free_policy(sk);
3175 sk_refcnt_debug_release(sk);
3179 EXPORT_SYMBOL(sk_common_release);
3181 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3183 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3185 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3186 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3187 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3188 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3189 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3190 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3191 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3192 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3193 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3196 #ifdef CONFIG_PROC_FS
3197 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3199 int val[PROTO_INUSE_NR];
3202 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3204 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3206 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3208 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3210 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3212 int cpu, idx = prot->inuse_idx;
3215 for_each_possible_cpu(cpu)
3216 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3218 return res >= 0 ? res : 0;
3220 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3222 static void sock_inuse_add(struct net *net, int val)
3224 this_cpu_add(*net->core.sock_inuse, val);
3227 int sock_inuse_get(struct net *net)
3231 for_each_possible_cpu(cpu)
3232 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3237 EXPORT_SYMBOL_GPL(sock_inuse_get);
3239 static int __net_init sock_inuse_init_net(struct net *net)
3241 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3242 if (net->core.prot_inuse == NULL)
3245 net->core.sock_inuse = alloc_percpu(int);
3246 if (net->core.sock_inuse == NULL)
3252 free_percpu(net->core.prot_inuse);
3256 static void __net_exit sock_inuse_exit_net(struct net *net)
3258 free_percpu(net->core.prot_inuse);
3259 free_percpu(net->core.sock_inuse);
3262 static struct pernet_operations net_inuse_ops = {
3263 .init = sock_inuse_init_net,
3264 .exit = sock_inuse_exit_net,
3267 static __init int net_inuse_init(void)
3269 if (register_pernet_subsys(&net_inuse_ops))
3270 panic("Cannot initialize net inuse counters");
3275 core_initcall(net_inuse_init);
3277 static void assign_proto_idx(struct proto *prot)
3279 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3281 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3282 pr_err("PROTO_INUSE_NR exhausted\n");
3286 set_bit(prot->inuse_idx, proto_inuse_idx);
3289 static void release_proto_idx(struct proto *prot)
3291 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3292 clear_bit(prot->inuse_idx, proto_inuse_idx);
3295 static inline void assign_proto_idx(struct proto *prot)
3299 static inline void release_proto_idx(struct proto *prot)
3303 static void sock_inuse_add(struct net *net, int val)
3308 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3312 kfree(rsk_prot->slab_name);
3313 rsk_prot->slab_name = NULL;
3314 kmem_cache_destroy(rsk_prot->slab);
3315 rsk_prot->slab = NULL;
3318 static int req_prot_init(const struct proto *prot)
3320 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3325 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3327 if (!rsk_prot->slab_name)
3330 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3331 rsk_prot->obj_size, 0,
3332 SLAB_ACCOUNT | prot->slab_flags,
3335 if (!rsk_prot->slab) {
3336 pr_crit("%s: Can't create request sock SLAB cache!\n",
3343 int proto_register(struct proto *prot, int alloc_slab)
3346 prot->slab = kmem_cache_create_usercopy(prot->name,
3348 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3350 prot->useroffset, prot->usersize,
3353 if (prot->slab == NULL) {
3354 pr_crit("%s: Can't create sock SLAB cache!\n",
3359 if (req_prot_init(prot))
3360 goto out_free_request_sock_slab;
3362 if (prot->twsk_prot != NULL) {
3363 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3365 if (prot->twsk_prot->twsk_slab_name == NULL)
3366 goto out_free_request_sock_slab;
3368 prot->twsk_prot->twsk_slab =
3369 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3370 prot->twsk_prot->twsk_obj_size,
3375 if (prot->twsk_prot->twsk_slab == NULL)
3376 goto out_free_timewait_sock_slab_name;
3380 mutex_lock(&proto_list_mutex);
3381 list_add(&prot->node, &proto_list);
3382 assign_proto_idx(prot);
3383 mutex_unlock(&proto_list_mutex);
3386 out_free_timewait_sock_slab_name:
3387 kfree(prot->twsk_prot->twsk_slab_name);
3388 out_free_request_sock_slab:
3389 req_prot_cleanup(prot->rsk_prot);
3391 kmem_cache_destroy(prot->slab);
3396 EXPORT_SYMBOL(proto_register);
3398 void proto_unregister(struct proto *prot)
3400 mutex_lock(&proto_list_mutex);
3401 release_proto_idx(prot);
3402 list_del(&prot->node);
3403 mutex_unlock(&proto_list_mutex);
3405 kmem_cache_destroy(prot->slab);
3408 req_prot_cleanup(prot->rsk_prot);
3410 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3411 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3412 kfree(prot->twsk_prot->twsk_slab_name);
3413 prot->twsk_prot->twsk_slab = NULL;
3416 EXPORT_SYMBOL(proto_unregister);
3418 int sock_load_diag_module(int family, int protocol)
3421 if (!sock_is_registered(family))
3424 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3425 NETLINK_SOCK_DIAG, family);
3429 if (family == AF_INET &&
3430 protocol != IPPROTO_RAW &&
3431 !rcu_access_pointer(inet_protos[protocol]))
3435 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3436 NETLINK_SOCK_DIAG, family, protocol);
3438 EXPORT_SYMBOL(sock_load_diag_module);
3440 #ifdef CONFIG_PROC_FS
3441 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3442 __acquires(proto_list_mutex)
3444 mutex_lock(&proto_list_mutex);
3445 return seq_list_start_head(&proto_list, *pos);
3448 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3450 return seq_list_next(v, &proto_list, pos);
3453 static void proto_seq_stop(struct seq_file *seq, void *v)
3454 __releases(proto_list_mutex)
3456 mutex_unlock(&proto_list_mutex);
3459 static char proto_method_implemented(const void *method)
3461 return method == NULL ? 'n' : 'y';
3463 static long sock_prot_memory_allocated(struct proto *proto)
3465 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3468 static char *sock_prot_memory_pressure(struct proto *proto)
3470 return proto->memory_pressure != NULL ?
3471 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3474 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3477 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3478 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3481 sock_prot_inuse_get(seq_file_net(seq), proto),
3482 sock_prot_memory_allocated(proto),
3483 sock_prot_memory_pressure(proto),
3485 proto->slab == NULL ? "no" : "yes",
3486 module_name(proto->owner),
3487 proto_method_implemented(proto->close),
3488 proto_method_implemented(proto->connect),
3489 proto_method_implemented(proto->disconnect),
3490 proto_method_implemented(proto->accept),
3491 proto_method_implemented(proto->ioctl),
3492 proto_method_implemented(proto->init),
3493 proto_method_implemented(proto->destroy),
3494 proto_method_implemented(proto->shutdown),
3495 proto_method_implemented(proto->setsockopt),
3496 proto_method_implemented(proto->getsockopt),
3497 proto_method_implemented(proto->sendmsg),
3498 proto_method_implemented(proto->recvmsg),
3499 proto_method_implemented(proto->sendpage),
3500 proto_method_implemented(proto->bind),
3501 proto_method_implemented(proto->backlog_rcv),
3502 proto_method_implemented(proto->hash),
3503 proto_method_implemented(proto->unhash),
3504 proto_method_implemented(proto->get_port),
3505 proto_method_implemented(proto->enter_memory_pressure));
3508 static int proto_seq_show(struct seq_file *seq, void *v)
3510 if (v == &proto_list)
3511 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3520 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3522 proto_seq_printf(seq, list_entry(v, struct proto, node));
3526 static const struct seq_operations proto_seq_ops = {
3527 .start = proto_seq_start,
3528 .next = proto_seq_next,
3529 .stop = proto_seq_stop,
3530 .show = proto_seq_show,
3533 static __net_init int proto_init_net(struct net *net)
3535 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3536 sizeof(struct seq_net_private)))
3542 static __net_exit void proto_exit_net(struct net *net)
3544 remove_proc_entry("protocols", net->proc_net);
3548 static __net_initdata struct pernet_operations proto_net_ops = {
3549 .init = proto_init_net,
3550 .exit = proto_exit_net,
3553 static int __init proto_init(void)
3555 return register_pernet_subsys(&proto_net_ops);
3558 subsys_initcall(proto_init);
3560 #endif /* PROC_FS */
3562 #ifdef CONFIG_NET_RX_BUSY_POLL
3563 bool sk_busy_loop_end(void *p, unsigned long start_time)
3565 struct sock *sk = p;
3567 return !skb_queue_empty(&sk->sk_receive_queue) ||
3568 sk_busy_loop_timeout(sk, start_time);
3570 EXPORT_SYMBOL(sk_busy_loop_end);
3571 #endif /* CONFIG_NET_RX_BUSY_POLL */